2017-11-18 17:09:20 +01:00
|
|
|
/* SPDX-License-Identifier: LGPL-2.1+ */
|
2010-02-03 13:03:47 +01:00
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
#include <errno.h>
|
2010-01-27 00:15:56 +01:00
|
|
|
#include <stdlib.h>
|
2017-12-22 13:08:14 +01:00
|
|
|
#include <sys/prctl.h>
|
2015-10-23 18:52:53 +02:00
|
|
|
#include <unistd.h>
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2013-11-19 21:12:59 +01:00
|
|
|
#include "sd-id128.h"
|
|
|
|
#include "sd-messages.h"
|
2015-10-23 18:52:53 +02:00
|
|
|
|
2018-05-15 20:17:34 +02:00
|
|
|
#include "all-units.h"
|
2018-10-17 20:40:09 +02:00
|
|
|
#include "alloc-util.h"
|
2019-04-23 12:14:20 +02:00
|
|
|
#include "bpf-firewall.h"
|
2015-10-23 18:52:53 +02:00
|
|
|
#include "bus-common-errors.h"
|
|
|
|
#include "bus-util.h"
|
2019-08-01 13:14:45 +02:00
|
|
|
#include "cgroup-setup.h"
|
2010-07-11 00:50:49 +02:00
|
|
|
#include "cgroup-util.h"
|
2015-10-23 18:52:53 +02:00
|
|
|
#include "dbus-unit.h"
|
|
|
|
#include "dbus.h"
|
|
|
|
#include "dropin.h"
|
|
|
|
#include "escape.h"
|
|
|
|
#include "execute.h"
|
2016-11-11 19:59:19 +01:00
|
|
|
#include "fd-util.h"
|
2013-02-14 12:26:13 +01:00
|
|
|
#include "fileio-label.h"
|
2018-11-30 21:07:21 +01:00
|
|
|
#include "fileio.h"
|
2016-11-07 16:14:59 +01:00
|
|
|
#include "format-util.h"
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
#include "fs-util.h"
|
2016-08-30 23:18:46 +02:00
|
|
|
#include "id128-util.h"
|
2017-09-21 14:05:35 +02:00
|
|
|
#include "io-util.h"
|
2019-07-04 15:46:16 +02:00
|
|
|
#include "install.h"
|
2015-10-23 18:52:53 +02:00
|
|
|
#include "load-dropin.h"
|
|
|
|
#include "load-fragment.h"
|
|
|
|
#include "log.h"
|
|
|
|
#include "macro.h"
|
2019-10-31 03:07:23 +01:00
|
|
|
#include "missing_audit.h"
|
2015-10-23 18:52:53 +02:00
|
|
|
#include "mkdir.h"
|
2015-10-26 16:18:16 +01:00
|
|
|
#include "parse-util.h"
|
2015-10-23 18:52:53 +02:00
|
|
|
#include "path-util.h"
|
2015-04-10 19:10:00 +02:00
|
|
|
#include "process-util.h"
|
2019-08-25 10:57:08 +02:00
|
|
|
#include "rm-rf.h"
|
2018-10-17 20:40:09 +02:00
|
|
|
#include "serialize.h"
|
2015-10-23 18:52:53 +02:00
|
|
|
#include "set.h"
|
2016-04-08 11:27:28 +02:00
|
|
|
#include "signal-util.h"
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
#include "sparse-endian.h"
|
2015-09-01 17:25:59 +02:00
|
|
|
#include "special.h"
|
2017-11-22 15:03:51 +01:00
|
|
|
#include "specifier.h"
|
2015-10-26 22:01:44 +01:00
|
|
|
#include "stat-util.h"
|
2016-01-12 15:34:20 +01:00
|
|
|
#include "stdio-util.h"
|
2017-11-13 17:14:07 +01:00
|
|
|
#include "string-table.h"
|
2015-10-24 22:58:24 +02:00
|
|
|
#include "string-util.h"
|
2015-10-23 18:52:53 +02:00
|
|
|
#include "strv.h"
|
2018-11-23 17:46:25 +01:00
|
|
|
#include "terminal-util.h"
|
2018-11-30 21:05:27 +01:00
|
|
|
#include "tmpfile-util.h"
|
2016-04-07 15:43:59 +02:00
|
|
|
#include "umask-util.h"
|
2015-10-23 18:52:53 +02:00
|
|
|
#include "unit-name.h"
|
2015-09-01 17:25:59 +02:00
|
|
|
#include "unit.h"
|
2015-10-25 22:32:30 +01:00
|
|
|
#include "user-util.h"
|
|
|
|
#include "virt.h"
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2019-07-27 23:40:24 +02:00
|
|
|
/* Thresholds for logging at INFO level about resource consumption */
|
|
|
|
#define MENTIONWORTHY_CPU_NSEC (1 * NSEC_PER_SEC)
|
|
|
|
#define MENTIONWORTHY_IO_BYTES (1024 * 1024ULL)
|
|
|
|
#define MENTIONWORTHY_IP_BYTES (0ULL)
|
|
|
|
|
|
|
|
/* Thresholds for logging at INFO level about resource consumption */
|
|
|
|
#define NOTICEWORTHY_CPU_NSEC (10*60 * NSEC_PER_SEC) /* 10 minutes */
|
|
|
|
#define NOTICEWORTHY_IO_BYTES (10 * 1024 * 1024ULL) /* 10 MB */
|
|
|
|
#define NOTICEWORTHY_IP_BYTES (128 * 1024 * 1024ULL) /* 128 MB */
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
const UnitVTable * const unit_vtable[_UNIT_TYPE_MAX] = {
|
|
|
|
[UNIT_SERVICE] = &service_vtable,
|
|
|
|
[UNIT_SOCKET] = &socket_vtable,
|
|
|
|
[UNIT_TARGET] = &target_vtable,
|
|
|
|
[UNIT_DEVICE] = &device_vtable,
|
|
|
|
[UNIT_MOUNT] = &mount_vtable,
|
|
|
|
[UNIT_AUTOMOUNT] = &automount_vtable,
|
2010-05-24 05:25:33 +02:00
|
|
|
[UNIT_SWAP] = &swap_vtable,
|
2013-12-02 23:30:19 +01:00
|
|
|
[UNIT_TIMER] = &timer_vtable,
|
2013-06-17 21:33:26 +02:00
|
|
|
[UNIT_PATH] = &path_vtable,
|
2013-07-01 00:03:57 +02:00
|
|
|
[UNIT_SLICE] = &slice_vtable,
|
2017-11-13 17:14:07 +01:00
|
|
|
[UNIT_SCOPE] = &scope_vtable,
|
2010-01-26 21:39:06 +01:00
|
|
|
};
|
|
|
|
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
static void maybe_warn_about_dependency(Unit *u, const char *other, UnitDependency dependency);
|
2014-08-08 02:46:49 +02:00
|
|
|
|
2012-01-15 10:53:49 +01:00
|
|
|
Unit *unit_new(Manager *m, size_t size) {
|
2010-01-26 21:39:06 +01:00
|
|
|
Unit *u;
|
|
|
|
|
|
|
|
assert(m);
|
2012-01-15 12:04:08 +01:00
|
|
|
assert(size >= sizeof(Unit));
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2012-01-15 10:53:49 +01:00
|
|
|
u = malloc0(size);
|
|
|
|
if (!u)
|
2010-01-26 21:39:06 +01:00
|
|
|
return NULL;
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
u->manager = m;
|
|
|
|
u->type = _UNIT_TYPE_INVALID;
|
|
|
|
u->default_dependencies = true;
|
|
|
|
u->unit_file_state = _UNIT_FILE_STATE_INVALID;
|
2014-12-02 02:38:18 +01:00
|
|
|
u->unit_file_preset = -1;
|
2013-11-26 01:39:53 +01:00
|
|
|
u->on_failure_job_mode = JOB_REPLACE;
|
2019-03-19 17:17:31 +01:00
|
|
|
u->cgroup_control_inotify_wd = -1;
|
2019-03-19 19:05:19 +01:00
|
|
|
u->cgroup_memory_inotify_wd = -1;
|
core: rework unit timeout handling, and add new setting RuntimeMaxSec=
This clean-ups timeout handling in PID 1. Specifically, instead of storing 0 in internal timeout variables as
indication for a disabled timeout, use USEC_INFINITY which is in-line with how we do this in the rest of our code
(following the logic that 0 means "no", and USEC_INFINITY means "never").
This also replace all usec_t additions with invocations to usec_add(), so that USEC_INFINITY is properly propagated,
and sd-event considers it has indication for turning off the event source.
This also alters the deserialization of the units to restart timeouts from the time they were originally started from.
Before this patch timeouts would be restarted beginning with the time of the deserialization, which could lead to
artificially prolonged timeouts if a daemon reload took place.
Finally, a new RuntimeMaxSec= setting is introduced for service units, that specifies a maximum runtime after which a
specific service is forcibly terminated. This is useful to put time limits on time-intensive processing jobs.
This also simplifies the various xyz_spawn() calls of the various types in that explicit distruction of the timers is
removed, as that is done anyway by the state change handlers, and a state change is always done when the xyz_spawn()
calls fail.
Fixes: #2249
2016-02-01 21:48:10 +01:00
|
|
|
u->job_timeout = USEC_INFINITY;
|
2017-02-17 17:47:20 +01:00
|
|
|
u->job_running_timeout = USEC_INFINITY;
|
2016-08-01 19:24:40 +02:00
|
|
|
u->ref_uid = UID_INVALID;
|
|
|
|
u->ref_gid = GID_INVALID;
|
2016-08-18 20:58:10 +02:00
|
|
|
u->cpu_usage_last = NSEC_INFINITY;
|
2018-09-30 12:33:16 +02:00
|
|
|
u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL;
|
2018-11-16 11:41:18 +01:00
|
|
|
u->failure_action_exit_status = u->success_action_exit_status = -1;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2016-11-11 19:59:19 +01:00
|
|
|
u->ip_accounting_ingress_map_fd = -1;
|
|
|
|
u->ip_accounting_egress_map_fd = -1;
|
|
|
|
u->ipv4_allow_map_fd = -1;
|
|
|
|
u->ipv6_allow_map_fd = -1;
|
|
|
|
u->ipv4_deny_map_fd = -1;
|
|
|
|
u->ipv6_deny_map_fd = -1;
|
|
|
|
|
2017-11-22 15:03:51 +01:00
|
|
|
u->last_section_private = -1;
|
|
|
|
|
2019-09-19 17:45:41 +02:00
|
|
|
u->start_ratelimit = (RateLimit) { m->default_start_limit_interval, m->default_start_limit_burst };
|
2019-09-19 17:41:20 +02:00
|
|
|
u->auto_stop_ratelimit = (RateLimit) { 10 * USEC_PER_SEC, 16 };
|
2015-05-19 16:00:24 +02:00
|
|
|
|
2019-03-22 12:16:03 +01:00
|
|
|
for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++)
|
|
|
|
u->io_accounting_last[i] = UINT64_MAX;
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
return u;
|
|
|
|
}
|
|
|
|
|
2016-10-25 00:29:05 +02:00
|
|
|
int unit_new_for_name(Manager *m, size_t size, const char *name, Unit **ret) {
|
2018-03-09 21:34:28 +01:00
|
|
|
_cleanup_(unit_freep) Unit *u = NULL;
|
2016-10-25 00:29:05 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
u = unit_new(m, size);
|
|
|
|
if (!u)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
r = unit_add_name(u, name);
|
2018-03-09 21:34:28 +01:00
|
|
|
if (r < 0)
|
2016-10-25 00:29:05 +02:00
|
|
|
return r;
|
|
|
|
|
2018-04-05 07:26:26 +02:00
|
|
|
*ret = TAKE_PTR(u);
|
|
|
|
|
2016-10-25 00:29:05 +02:00
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2018-12-12 16:45:33 +01:00
|
|
|
bool unit_has_name(const Unit *u, const char *name) {
|
2010-04-06 16:32:07 +02:00
|
|
|
assert(u);
|
|
|
|
assert(name);
|
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
return streq_ptr(name, u->id) ||
|
|
|
|
set_contains(u->aliases, name);
|
2010-04-06 16:32:07 +02:00
|
|
|
}
|
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
static void unit_init(Unit *u) {
|
|
|
|
CGroupContext *cc;
|
|
|
|
ExecContext *ec;
|
|
|
|
KillContext *kc;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(u->manager);
|
|
|
|
assert(u->type >= 0);
|
|
|
|
|
|
|
|
cc = unit_get_cgroup_context(u);
|
|
|
|
if (cc) {
|
|
|
|
cgroup_context_init(cc);
|
|
|
|
|
|
|
|
/* Copy in the manager defaults into the cgroup
|
|
|
|
* context, _before_ the rest of the settings have
|
|
|
|
* been initialized */
|
|
|
|
|
|
|
|
cc->cpu_accounting = u->manager->default_cpu_accounting;
|
2016-05-05 22:42:55 +02:00
|
|
|
cc->io_accounting = u->manager->default_io_accounting;
|
2014-03-19 20:40:05 +01:00
|
|
|
cc->blockio_accounting = u->manager->default_blockio_accounting;
|
|
|
|
cc->memory_accounting = u->manager->default_memory_accounting;
|
2015-09-10 12:32:16 +02:00
|
|
|
cc->tasks_accounting = u->manager->default_tasks_accounting;
|
2016-11-11 19:59:19 +01:00
|
|
|
cc->ip_accounting = u->manager->default_ip_accounting;
|
2015-11-13 17:13:55 +01:00
|
|
|
|
|
|
|
if (u->type != UNIT_SLICE)
|
|
|
|
cc->tasks_max = u->manager->default_tasks_max;
|
2014-03-19 20:40:05 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
ec = unit_get_exec_context(u);
|
2017-09-14 21:19:05 +02:00
|
|
|
if (ec) {
|
2014-03-19 20:40:05 +01:00
|
|
|
exec_context_init(ec);
|
|
|
|
|
2020-04-03 10:00:25 +02:00
|
|
|
if (MANAGER_IS_SYSTEM(u->manager))
|
|
|
|
ec->keyring_mode = EXEC_KEYRING_SHARED;
|
|
|
|
else {
|
|
|
|
ec->keyring_mode = EXEC_KEYRING_INHERIT;
|
|
|
|
|
|
|
|
/* User manager might have its umask redefined by PAM or UMask=. In this
|
|
|
|
* case let the units it manages inherit this value by default. They can
|
|
|
|
* still tune this value through their own unit file */
|
|
|
|
(void) get_process_umask(getpid_cached(), &ec->umask);
|
|
|
|
}
|
2017-09-14 21:19:05 +02:00
|
|
|
}
|
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
kc = unit_get_kill_context(u);
|
|
|
|
if (kc)
|
|
|
|
kill_context_init(kc);
|
|
|
|
|
|
|
|
if (UNIT_VTABLE(u)->init)
|
|
|
|
UNIT_VTABLE(u)->init(u);
|
|
|
|
}
|
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
static int unit_add_alias(Unit *u, char *donated_name) {
|
|
|
|
int r;
|
|
|
|
|
|
|
|
/* Make sure that u->names is allocated. We may leave u->names
|
|
|
|
* empty if we fail later, but this is not a problem. */
|
|
|
|
r = set_ensure_allocated(&u->aliases, &string_hash_ops);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
r = set_put(u->aliases, donated_name);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
assert(r > 0);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
int unit_add_name(Unit *u, const char *text) {
|
2020-05-27 15:49:17 +02:00
|
|
|
_cleanup_free_ char *name = NULL, *instance = NULL;
|
2010-01-26 21:39:06 +01:00
|
|
|
UnitType t;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(text);
|
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
if (unit_name_is_valid(text, UNIT_NAME_TEMPLATE)) {
|
2012-01-15 12:04:08 +01:00
|
|
|
if (!u->instance)
|
2020-03-25 10:08:50 +01:00
|
|
|
return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
|
|
|
|
"instance is not set when adding name '%s': %m", text);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
r = unit_name_replace_instance(text, u->instance, &name);
|
2015-04-30 20:21:00 +02:00
|
|
|
if (r < 0)
|
2020-03-25 10:08:50 +01:00
|
|
|
return log_unit_debug_errno(u, r,
|
|
|
|
"failed to build instance name from '%s': %m", text);
|
2015-04-30 20:21:00 +02:00
|
|
|
} else {
|
2020-05-27 15:49:17 +02:00
|
|
|
name = strdup(text);
|
|
|
|
if (!name)
|
2015-04-30 20:21:00 +02:00
|
|
|
return -ENOMEM;
|
|
|
|
}
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
if (unit_has_name(u, name))
|
2015-04-30 20:21:00 +02:00
|
|
|
return 0;
|
2020-05-27 15:49:17 +02:00
|
|
|
|
|
|
|
if (hashmap_contains(u->manager->units, name))
|
2020-03-25 10:08:50 +01:00
|
|
|
return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EEXIST),
|
2020-05-27 15:49:17 +02:00
|
|
|
"unit already exist when adding name '%s': %m", name);
|
2015-04-30 20:21:00 +02:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
if (!unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
|
2020-03-25 10:08:50 +01:00
|
|
|
return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
|
2020-05-27 15:49:17 +02:00
|
|
|
"name '%s' is invalid: %m", name);
|
2010-04-10 17:53:17 +02:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
t = unit_name_to_type(name);
|
2015-04-30 20:21:00 +02:00
|
|
|
if (t < 0)
|
2020-03-25 10:08:50 +01:00
|
|
|
return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
|
2020-07-07 12:08:22 +02:00
|
|
|
"failed to derive unit type from name '%s': %m", name);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
if (u->type != _UNIT_TYPE_INVALID && t != u->type)
|
2020-03-25 10:08:50 +01:00
|
|
|
return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
|
|
|
|
"unit type is illegal: u->type(%d) and t(%d) for name '%s': %m",
|
2020-05-27 15:49:17 +02:00
|
|
|
u->type, t, name);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
r = unit_name_to_instance(name, &instance);
|
2013-10-22 01:54:10 +02:00
|
|
|
if (r < 0)
|
2020-05-27 15:49:17 +02:00
|
|
|
return log_unit_debug_errno(u, r, "failed to extract instance from name '%s': %m", name);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
if (instance && !unit_type_may_template(t))
|
|
|
|
return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL), "templates are not allowed for name '%s': %m", name);
|
2010-04-15 03:11:11 +02:00
|
|
|
|
2020-05-27 16:39:35 +02:00
|
|
|
/* Ensure that this unit either has no instance, or that the instance matches. */
|
|
|
|
if (u->type != _UNIT_TYPE_INVALID && !streq_ptr(u->instance, instance))
|
2020-03-25 10:08:50 +01:00
|
|
|
return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EINVAL),
|
2020-05-29 11:22:14 +02:00
|
|
|
"cannot add name %s, the instances don't match (\"%s\" != \"%s\").",
|
|
|
|
name, instance, u->instance);
|
2010-04-15 03:11:11 +02:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
if (u->id && !unit_type_may_alias(t))
|
2020-05-29 11:22:14 +02:00
|
|
|
return log_unit_debug_errno(u, SYNTHETIC_ERRNO(EEXIST),
|
|
|
|
"cannot add name %s, aliases are not allowed for %s units.",
|
|
|
|
name, unit_type_to_string(t));
|
2010-04-15 03:11:11 +02:00
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
if (hashmap_size(u->manager->units) >= MANAGER_MAX_NAMES)
|
2020-05-29 11:22:14 +02:00
|
|
|
return log_unit_warning_errno(u, SYNTHETIC_ERRNO(E2BIG), "cannot add name, manager has too many units: %m");
|
2010-04-22 02:56:42 +02:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
/* Add name to the global hashmap first, because that's easier to undo */
|
|
|
|
r = hashmap_put(u->manager->units, name, u);
|
2015-04-30 20:21:00 +02:00
|
|
|
if (r < 0)
|
2020-03-25 10:08:50 +01:00
|
|
|
return log_unit_debug_errno(u, r, "add unit to hashmap failed for name '%s': %m", text);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
if (u->id) {
|
|
|
|
r = unit_add_alias(u, name); /* unit_add_alias() takes ownership of the name on success */
|
|
|
|
if (r < 0) {
|
|
|
|
hashmap_remove(u->manager->units, name);
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
TAKE_PTR(name);
|
|
|
|
|
|
|
|
} else {
|
|
|
|
/* A new name, we don't need the set yet. */
|
|
|
|
assert(u->type == _UNIT_TYPE_INVALID);
|
|
|
|
assert(!u->instance);
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
u->type = t;
|
2020-05-27 15:49:17 +02:00
|
|
|
u->id = TAKE_PTR(name);
|
|
|
|
u->instance = TAKE_PTR(instance);
|
2010-04-15 03:11:11 +02:00
|
|
|
|
2013-10-14 06:10:14 +02:00
|
|
|
LIST_PREPEND(units_by_type, u->manager->units_by_type[t], u);
|
2014-03-19 20:40:05 +01:00
|
|
|
unit_init(u);
|
|
|
|
}
|
2010-04-15 03:11:11 +02:00
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
unit_add_to_dbus_queue(u);
|
|
|
|
return 0;
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
2010-01-29 01:49:34 +01:00
|
|
|
int unit_choose_id(Unit *u, const char *name) {
|
2013-07-25 17:36:01 +02:00
|
|
|
_cleanup_free_ char *t = NULL;
|
2020-05-27 15:49:17 +02:00
|
|
|
char *s;
|
2010-07-16 19:40:24 +02:00
|
|
|
int r;
|
2010-01-29 01:49:34 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(name);
|
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
|
2012-01-15 12:04:08 +01:00
|
|
|
if (!u->instance)
|
2010-04-15 03:11:11 +02:00
|
|
|
return -EINVAL;
|
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
r = unit_name_replace_instance(name, u->instance, &t);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2010-04-15 03:11:11 +02:00
|
|
|
|
|
|
|
name = t;
|
|
|
|
}
|
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
if (streq_ptr(u->id, name))
|
|
|
|
return 0; /* Nothing to do. */
|
|
|
|
|
|
|
|
/* Selects one of the aliases of this unit as the id */
|
|
|
|
s = set_get(u->aliases, (char*) name);
|
2010-04-15 03:11:11 +02:00
|
|
|
if (!s)
|
2010-01-29 01:49:34 +01:00
|
|
|
return -ENOENT;
|
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
if (u->id) {
|
|
|
|
r = set_remove_and_put(u->aliases, name, u->id);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
} else
|
|
|
|
assert_se(set_remove(u->aliases, name)); /* see set_get() above… */
|
2010-07-16 19:40:24 +02:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
u->id = s; /* Old u->id is now stored in the set, and s is not stored anywhere */
|
2010-02-05 00:38:41 +01:00
|
|
|
unit_add_to_dbus_queue(u);
|
2010-04-15 03:11:11 +02:00
|
|
|
|
2010-01-29 01:49:34 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2010-01-29 03:18:09 +01:00
|
|
|
int unit_set_description(Unit *u, const char *description) {
|
2017-09-26 22:49:23 +02:00
|
|
|
int r;
|
2010-01-29 03:18:09 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
2017-09-26 22:49:23 +02:00
|
|
|
r = free_and_strdup(&u->description, empty_to_null(description));
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
if (r > 0)
|
|
|
|
unit_add_to_dbus_queue(u);
|
2010-02-05 00:38:41 +01:00
|
|
|
|
2010-01-29 03:18:09 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2018-02-13 10:50:13 +01:00
|
|
|
bool unit_may_gc(Unit *u) {
|
2015-01-06 00:26:25 +01:00
|
|
|
UnitActiveState state;
|
2017-11-24 20:27:01 +01:00
|
|
|
int r;
|
2017-11-13 17:14:07 +01:00
|
|
|
|
2010-04-21 06:01:13 +02:00
|
|
|
assert(u);
|
|
|
|
|
2018-02-13 10:50:13 +01:00
|
|
|
/* Checks whether the unit is ready to be unloaded for garbage collection.
|
|
|
|
* Returns true when the unit may be collected, and false if there's some
|
2018-02-13 14:37:11 +01:00
|
|
|
* reason to keep it loaded.
|
|
|
|
*
|
|
|
|
* References from other units are *not* checked here. Instead, this is done
|
|
|
|
* in unit_gc_sweep(), but using markers to properly collect dependency loops.
|
|
|
|
*/
|
2017-11-13 17:14:07 +01:00
|
|
|
|
2015-01-06 00:26:25 +01:00
|
|
|
if (u->job)
|
2018-02-13 10:50:13 +01:00
|
|
|
return false;
|
2010-04-21 06:01:13 +02:00
|
|
|
|
2015-01-06 00:26:25 +01:00
|
|
|
if (u->nop_job)
|
2018-02-13 10:50:13 +01:00
|
|
|
return false;
|
2010-08-11 02:07:59 +02:00
|
|
|
|
2015-01-06 00:26:25 +01:00
|
|
|
state = unit_active_state(u);
|
|
|
|
|
2017-11-13 15:08:49 +01:00
|
|
|
/* If the unit is inactive and failed and no job is queued for it, then release its runtime resources */
|
2015-01-06 00:26:25 +01:00
|
|
|
if (UNIT_IS_INACTIVE_OR_FAILED(state) &&
|
|
|
|
UNIT_VTABLE(u)->release_resources)
|
2017-11-13 15:08:49 +01:00
|
|
|
UNIT_VTABLE(u)->release_resources(u);
|
2015-01-06 00:26:25 +01:00
|
|
|
|
2016-10-24 21:41:54 +02:00
|
|
|
if (u->perpetual)
|
2018-02-13 10:50:13 +01:00
|
|
|
return false;
|
2012-09-10 10:12:10 +02:00
|
|
|
|
2016-08-15 18:12:01 +02:00
|
|
|
if (sd_bus_track_count(u->bus_track) > 0)
|
2018-02-13 10:50:13 +01:00
|
|
|
return false;
|
2016-08-15 18:12:01 +02:00
|
|
|
|
2017-11-13 17:14:07 +01:00
|
|
|
/* But we keep the unit object around for longer when it is referenced or configured to not be gc'ed */
|
|
|
|
switch (u->collect_mode) {
|
|
|
|
|
|
|
|
case COLLECT_INACTIVE:
|
|
|
|
if (state != UNIT_INACTIVE)
|
2018-02-13 10:50:13 +01:00
|
|
|
return false;
|
2017-11-13 17:14:07 +01:00
|
|
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
case COLLECT_INACTIVE_OR_FAILED:
|
|
|
|
if (!IN_SET(state, UNIT_INACTIVE, UNIT_FAILED))
|
2018-02-13 10:50:13 +01:00
|
|
|
return false;
|
2017-11-13 17:14:07 +01:00
|
|
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
default:
|
|
|
|
assert_not_reached("Unknown garbage collection mode");
|
|
|
|
}
|
|
|
|
|
2017-11-24 20:27:01 +01:00
|
|
|
if (u->cgroup_path) {
|
|
|
|
/* If the unit has a cgroup, then check whether there's anything in it. If so, we should stay
|
|
|
|
* around. Units with active processes should never be collected. */
|
|
|
|
|
|
|
|
r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path);
|
|
|
|
if (r <= 0)
|
2018-02-13 10:50:13 +01:00
|
|
|
return false;
|
2017-11-24 20:27:01 +01:00
|
|
|
}
|
|
|
|
|
2018-02-13 10:50:13 +01:00
|
|
|
if (UNIT_VTABLE(u)->may_gc && !UNIT_VTABLE(u)->may_gc(u))
|
|
|
|
return false;
|
2010-04-21 06:01:13 +02:00
|
|
|
|
2018-02-13 10:50:13 +01:00
|
|
|
return true;
|
2010-04-21 06:01:13 +02:00
|
|
|
}
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
void unit_add_to_load_queue(Unit *u) {
|
|
|
|
assert(u);
|
2012-01-15 12:04:08 +01:00
|
|
|
assert(u->type != _UNIT_TYPE_INVALID);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->load_state != UNIT_STUB || u->in_load_queue)
|
2010-01-26 21:39:06 +01:00
|
|
|
return;
|
|
|
|
|
2013-10-14 06:10:14 +02:00
|
|
|
LIST_PREPEND(load_queue, u->manager->load_queue, u);
|
2012-01-15 12:04:08 +01:00
|
|
|
u->in_load_queue = true;
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
2010-04-06 02:43:58 +02:00
|
|
|
void unit_add_to_cleanup_queue(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->in_cleanup_queue)
|
2010-04-06 02:43:58 +02:00
|
|
|
return;
|
|
|
|
|
2013-10-14 06:10:14 +02:00
|
|
|
LIST_PREPEND(cleanup_queue, u->manager->cleanup_queue, u);
|
2012-01-15 12:04:08 +01:00
|
|
|
u->in_cleanup_queue = true;
|
2010-04-06 02:43:58 +02:00
|
|
|
}
|
|
|
|
|
2010-04-21 06:01:13 +02:00
|
|
|
void unit_add_to_gc_queue(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->in_gc_queue || u->in_cleanup_queue)
|
2010-04-21 06:01:13 +02:00
|
|
|
return;
|
|
|
|
|
2018-02-13 10:50:13 +01:00
|
|
|
if (!unit_may_gc(u))
|
2010-04-21 06:01:13 +02:00
|
|
|
return;
|
|
|
|
|
2016-11-15 19:32:50 +01:00
|
|
|
LIST_PREPEND(gc_queue, u->manager->gc_unit_queue, u);
|
2012-01-15 12:04:08 +01:00
|
|
|
u->in_gc_queue = true;
|
2010-04-21 06:01:13 +02:00
|
|
|
}
|
|
|
|
|
2010-02-05 00:38:41 +01:00
|
|
|
void unit_add_to_dbus_queue(Unit *u) {
|
|
|
|
assert(u);
|
2012-01-15 12:04:08 +01:00
|
|
|
assert(u->type != _UNIT_TYPE_INVALID);
|
2010-02-05 00:38:41 +01:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->load_state == UNIT_STUB || u->in_dbus_queue)
|
2010-02-05 00:38:41 +01:00
|
|
|
return;
|
|
|
|
|
2010-07-05 00:58:07 +02:00
|
|
|
/* Shortcut things if nobody cares */
|
2014-03-03 01:33:45 +01:00
|
|
|
if (sd_bus_track_count(u->manager->subscribed) <= 0 &&
|
2017-02-28 17:55:57 +01:00
|
|
|
sd_bus_track_count(u->bus_track) <= 0 &&
|
2014-03-03 01:33:45 +01:00
|
|
|
set_isempty(u->manager->private_buses)) {
|
2012-01-15 12:04:08 +01:00
|
|
|
u->sent_dbus_new_signal = true;
|
2010-05-16 03:57:07 +02:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2013-10-14 06:10:14 +02:00
|
|
|
LIST_PREPEND(dbus_queue, u->manager->dbus_unit_queue, u);
|
2012-01-15 12:04:08 +01:00
|
|
|
u->in_dbus_queue = true;
|
2010-02-05 00:38:41 +01:00
|
|
|
}
|
|
|
|
|
2018-08-20 10:43:31 +02:00
|
|
|
void unit_submit_to_stop_when_unneeded_queue(Unit *u) {
|
2018-08-09 16:26:27 +02:00
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (u->in_stop_when_unneeded_queue)
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (!u->stop_when_unneeded)
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
|
|
|
|
return;
|
|
|
|
|
|
|
|
LIST_PREPEND(stop_when_unneeded_queue, u->manager->stop_when_unneeded_queue, u);
|
|
|
|
u->in_stop_when_unneeded_queue = true;
|
|
|
|
}
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
static void bidi_set_free(Unit *u, Hashmap *h) {
|
2010-01-26 21:39:06 +01:00
|
|
|
Unit *other;
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
Iterator i;
|
|
|
|
void *v;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
/* Frees the hashmap and makes sure we are dropped from the inverse pointers */
|
2010-01-26 21:39:06 +01:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, h, i) {
|
2020-05-28 15:25:22 +02:00
|
|
|
for (UnitDependency d = 0; d < _UNIT_DEPENDENCY_MAX; d++)
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
hashmap_remove(other->dependencies[d], u);
|
2010-04-21 06:01:13 +02:00
|
|
|
|
|
|
|
unit_add_to_gc_queue(other);
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
hashmap_free(h);
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
2013-06-28 04:12:58 +02:00
|
|
|
static void unit_remove_transient(Unit *u) {
|
|
|
|
char **i;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (!u->transient)
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (u->fragment_path)
|
2015-08-28 16:05:32 +02:00
|
|
|
(void) unlink(u->fragment_path);
|
2013-06-28 04:12:58 +02:00
|
|
|
|
|
|
|
STRV_FOREACH(i, u->dropin_paths) {
|
2016-02-25 01:13:57 +01:00
|
|
|
_cleanup_free_ char *p = NULL, *pp = NULL;
|
2013-06-28 04:12:58 +02:00
|
|
|
|
2016-02-25 01:13:57 +01:00
|
|
|
p = dirname_malloc(*i); /* Get the drop-in directory from the drop-in file */
|
|
|
|
if (!p)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
pp = dirname_malloc(p); /* Get the config directory from the drop-in directory */
|
|
|
|
if (!pp)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/* Only drop transient drop-ins */
|
|
|
|
if (!path_equal(u->manager->lookup_paths.transient, pp))
|
|
|
|
continue;
|
2013-06-28 04:12:58 +02:00
|
|
|
|
2016-02-25 01:13:57 +01:00
|
|
|
(void) unlink(*i);
|
|
|
|
(void) rmdir(p);
|
2013-06-28 04:12:58 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-09-26 20:14:24 +02:00
|
|
|
static void unit_free_requires_mounts_for(Unit *u) {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
assert(u);
|
2013-09-26 20:14:24 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
for (;;) {
|
|
|
|
_cleanup_free_ char *path;
|
2013-09-26 20:14:24 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
path = hashmap_steal_first_key(u->requires_mounts_for);
|
|
|
|
if (!path)
|
|
|
|
break;
|
|
|
|
else {
|
|
|
|
char s[strlen(path) + 1];
|
2013-09-26 20:14:24 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
PATH_FOREACH_PREFIX_MORE(s, path) {
|
|
|
|
char *y;
|
|
|
|
Set *x;
|
|
|
|
|
|
|
|
x = hashmap_get2(u->manager->units_requiring_mounts_for, s, (void**) &y);
|
|
|
|
if (!x)
|
|
|
|
continue;
|
2013-09-26 20:14:24 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
(void) set_remove(x, u);
|
2013-09-26 20:14:24 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
if (set_isempty(x)) {
|
|
|
|
(void) hashmap_remove(u->manager->units_requiring_mounts_for, y);
|
|
|
|
free(y);
|
|
|
|
set_free(x);
|
|
|
|
}
|
2013-09-26 20:14:24 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
u->requires_mounts_for = hashmap_free(u->requires_mounts_for);
|
2013-09-26 20:14:24 +02:00
|
|
|
}
|
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
static void unit_done(Unit *u) {
|
|
|
|
ExecContext *ec;
|
|
|
|
CGroupContext *cc;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (u->type < 0)
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (UNIT_VTABLE(u)->done)
|
|
|
|
UNIT_VTABLE(u)->done(u);
|
|
|
|
|
|
|
|
ec = unit_get_exec_context(u);
|
|
|
|
if (ec)
|
|
|
|
exec_context_done(ec);
|
|
|
|
|
|
|
|
cc = unit_get_cgroup_context(u);
|
|
|
|
if (cc)
|
|
|
|
cgroup_context_done(cc);
|
|
|
|
}
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
void unit_free(Unit *u) {
|
|
|
|
Iterator i;
|
|
|
|
char *t;
|
|
|
|
|
2016-11-28 19:41:20 +01:00
|
|
|
if (!u)
|
|
|
|
return;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2017-11-28 21:24:20 +01:00
|
|
|
u->transient_file = safe_fclose(u->transient_file);
|
2016-04-07 15:43:59 +02:00
|
|
|
|
2016-02-24 21:36:09 +01:00
|
|
|
if (!MANAGER_IS_RELOADING(u->manager))
|
2013-06-28 04:12:58 +02:00
|
|
|
unit_remove_transient(u);
|
|
|
|
|
2010-02-05 00:38:41 +01:00
|
|
|
bus_unit_send_removed_signal(u);
|
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
unit_done(u);
|
2010-06-04 22:31:33 +02:00
|
|
|
|
core: rework how we track service and scope PIDs
This reworks how systemd tracks processes on cgroupv1 systems where
cgroup notification is not reliable. Previously, whenever we had reason
to believe that new processes showed up or got removed we'd scan the
cgroup of the scope or service unit for new processes, and would tidy up
the list of PIDs previously watched. This scanning is relatively slow,
and does not scale well. With this change behaviour is changed: instead
of scanning for new/removed processes right away we do this work in a
per-unit deferred event loop job. This event source is scheduled at a
very low priority, so that it is executed when we have time but does not
starve other event sources. This has two benefits: this expensive work is
coalesced, if events happen in quick succession, and we won't delay
SIGCHLD handling for too long.
This patch basically replaces all direct invocation of
unit_watch_all_pids() in scope.c and service.c with invocations of the
new unit_enqueue_rewatch_pids() call which just enqueues a request of
watching/tidying up the PID sets (with one exception: in
scope_enter_signal() and service_enter_signal() we'll still do
unit_watch_all_pids() synchronously first, since we really want to know
all processes we are about to kill so that we can track them properly.
Moreover, all direct invocations of unit_tidy_watch_pids() and
unit_synthesize_cgroup_empty_event() are removed too, when the
unit_enqueue_rewatch_pids() call is invoked, as the queued job will run
those operations too.
All of this is done on cgroupsv1 systems only, and is disabled on
cgroupsv2 systems as cgroup-empty notifications are reliable there, and
we do not need SIGCHLD events to track processes there.
Fixes: #9138
2018-05-31 15:41:59 +02:00
|
|
|
unit_dequeue_rewatch_pids(u);
|
2015-08-06 12:53:06 +02:00
|
|
|
|
core: rework how we track service and scope PIDs
This reworks how systemd tracks processes on cgroupv1 systems where
cgroup notification is not reliable. Previously, whenever we had reason
to believe that new processes showed up or got removed we'd scan the
cgroup of the scope or service unit for new processes, and would tidy up
the list of PIDs previously watched. This scanning is relatively slow,
and does not scale well. With this change behaviour is changed: instead
of scanning for new/removed processes right away we do this work in a
per-unit deferred event loop job. This event source is scheduled at a
very low priority, so that it is executed when we have time but does not
starve other event sources. This has two benefits: this expensive work is
coalesced, if events happen in quick succession, and we won't delay
SIGCHLD handling for too long.
This patch basically replaces all direct invocation of
unit_watch_all_pids() in scope.c and service.c with invocations of the
new unit_enqueue_rewatch_pids() call which just enqueues a request of
watching/tidying up the PID sets (with one exception: in
scope_enter_signal() and service_enter_signal() we'll still do
unit_watch_all_pids() synchronously first, since we really want to know
all processes we are about to kill so that we can track them properly.
Moreover, all direct invocations of unit_tidy_watch_pids() and
unit_synthesize_cgroup_empty_event() are removed too, when the
unit_enqueue_rewatch_pids() call is invoked, as the queued job will run
those operations too.
All of this is done on cgroupsv1 systems only, and is disabled on
cgroupsv2 systems as cgroup-empty notifications are reliable there, and
we do not need SIGCHLD events to track processes there.
Fixes: #9138
2018-05-31 15:41:59 +02:00
|
|
|
sd_bus_slot_unref(u->match_bus_slot);
|
2016-08-15 18:12:01 +02:00
|
|
|
sd_bus_track_unref(u->bus_track);
|
|
|
|
u->deserialized_refs = strv_free(u->deserialized_refs);
|
2020-04-29 17:53:43 +02:00
|
|
|
u->pending_freezer_message = sd_bus_message_unref(u->pending_freezer_message);
|
2016-08-15 18:12:01 +02:00
|
|
|
|
2013-09-26 20:14:24 +02:00
|
|
|
unit_free_requires_mounts_for(u);
|
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
SET_FOREACH(t, u->aliases, i)
|
2012-01-15 12:04:08 +01:00
|
|
|
hashmap_remove_value(u->manager->units, t, u);
|
2020-05-27 15:49:17 +02:00
|
|
|
if (u->id)
|
|
|
|
hashmap_remove_value(u->manager->units, u->id, u);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2016-08-30 23:18:46 +02:00
|
|
|
if (!sd_id128_is_null(u->invocation_id))
|
|
|
|
hashmap_remove_value(u->manager->units_by_invocation_id, &u->invocation_id, u);
|
|
|
|
|
2012-04-20 10:21:37 +02:00
|
|
|
if (u->job) {
|
|
|
|
Job *j = u->job;
|
|
|
|
job_uninstall(j);
|
|
|
|
job_free(j);
|
|
|
|
}
|
2010-06-05 02:16:20 +02:00
|
|
|
|
core: add NOP jobs, job type collapsing
Two of our current job types are special:
JOB_TRY_RESTART, JOB_RELOAD_OR_START.
They differ from other job types by being sensitive to the unit active state.
They perform some action when the unit is active and some other action
otherwise. This raises a question: when exactly should the unit state be
checked to make the decision?
Currently the unit state is checked when the job becomes runnable. It's more
sensible to check the state immediately when the job is added by the user.
When the user types "systemctl try-restart foo.service", he really intends
to restart the service if it's running right now. If it isn't running right
now, the restart is pointless.
Consider the example (from Bugzilla[1]):
sleep.service takes some time to start.
hello.service has After=sleep.service.
Both services get started. Two jobs will appear:
hello.service/start waiting
sleep.service/start running
Then someone runs "systemctl try-restart hello.service".
Currently the try-restart operation will block and wait for
sleep.service/start to complete.
The correct result is to complete the try-restart operation immediately
with success, because hello.service is not running. The two original
jobs must not be disturbed by this.
To fix this we introduce two new concepts:
- a new job type: JOB_NOP
A JOB_NOP job does not do anything to the unit. It does not pull in any
dependencies. It is always immediately runnable. When installed to a unit,
it sits in a special slot (u->nop_job) where it never conflicts with
the installed job (u->job) of a different type. It never merges with jobs
of other types, but it can merge into an already installed JOB_NOP job.
- "collapsing" of job types
When a job of one of the two special types is added, the state of the unit
is checked immediately and the job type changes:
JOB_TRY_RESTART -> JOB_RESTART or JOB_NOP
JOB_RELOAD_OR_START -> JOB_RELOAD or JOB_START
Should a job type JOB_RELOAD_OR_START appear later during job merging, it
collapses immediately afterwards.
Collapsing actually makes some things simpler, because there are now fewer
job types that are allowed in the transaction.
[1] Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=753586
2012-04-25 11:58:27 +02:00
|
|
|
if (u->nop_job) {
|
|
|
|
Job *j = u->nop_job;
|
|
|
|
job_uninstall(j);
|
|
|
|
job_free(j);
|
|
|
|
}
|
|
|
|
|
2020-05-28 15:25:22 +02:00
|
|
|
for (UnitDependency d = 0; d < _UNIT_DEPENDENCY_MAX; d++)
|
2012-01-15 12:04:08 +01:00
|
|
|
bidi_set_free(u, u->dependencies[d]);
|
2010-06-05 02:16:20 +02:00
|
|
|
|
2020-05-21 13:24:43 +02:00
|
|
|
/* A unit is being dropped from the tree, make sure our family is realized properly. Do this after we
|
|
|
|
* detach the unit from slice tree in order to eliminate its effect on controller masks. */
|
2020-06-01 17:33:51 +02:00
|
|
|
if (UNIT_ISSET(u->slice))
|
2020-05-21 13:24:43 +02:00
|
|
|
unit_add_family_to_cgroup_realize_queue(UNIT_DEREF(u->slice));
|
2020-06-01 17:30:35 +02:00
|
|
|
|
2018-01-24 19:59:55 +01:00
|
|
|
if (u->on_console)
|
|
|
|
manager_unref_console(u->manager);
|
|
|
|
|
core: unified cgroup hierarchy support
This patch set adds full support the new unified cgroup hierarchy logic
of modern kernels.
A new kernel command line option "systemd.unified_cgroup_hierarchy=1" is
added. If specified the unified hierarchy is mounted to /sys/fs/cgroup
instead of a tmpfs. No further hierarchies are mounted. The kernel
command line option defaults to off. We can turn it on by default as
soon as the kernel's APIs regarding this are stabilized (but even then
downstream distros might want to turn this off, as this will break any
tools that access cgroupfs directly).
It is possibly to choose for each boot individually whether the unified
or the legacy hierarchy is used. nspawn will by default provide the
legacy hierarchy to containers if the host is using it, and the unified
otherwise. However it is possible to run containers with the unified
hierarchy on a legacy host and vice versa, by setting the
$UNIFIED_CGROUP_HIERARCHY environment variable for nspawn to 1 or 0,
respectively.
The unified hierarchy provides reliable cgroup empty notifications for
the first time, via inotify. To make use of this we maintain one
manager-wide inotify fd, and each cgroup to it.
This patch also removes cg_delete() which is unused now.
On kernel 4.2 only the "memory" controller is compatible with the
unified hierarchy, hence that's the only controller systemd exposes when
booted in unified heirarchy mode.
This introduces a new enum for enumerating supported controllers, plus a
related enum for the mask bits mapping to it. The core is changed to
make use of this everywhere.
This moves PID 1 into a new "init.scope" implicit scope unit in the root
slice. This is necessary since on the unified hierarchy cgroups may
either contain subgroups or processes but not both. PID 1 hence has to
move out of the root cgroup (strictly speaking the root cgroup is the
only one where processes and subgroups are still allowed, but in order
to support containers nicey, we move PID 1 into the new scope in all
cases.) This new unit is also used on legacy hierarchy setups. It's
actually pretty useful on all systems, as it can then be used to filter
journal messages coming from PID 1, and so on.
The root slice ("-.slice") is now implicitly created and started (and
does not require a unit file on disk anymore), since
that's where "init.scope" is located and the slice needs to be started
before the scope can.
To check whether we are in unified or legacy hierarchy mode we use
statfs() on /sys/fs/cgroup. If the .f_type field reports tmpfs we are in
legacy mode, if it reports cgroupfs we are in unified mode.
This patch set carefuly makes sure that cgls and cgtop continue to work
as desired.
When invoking nspawn as a service it will implicitly create two
subcgroups in the cgroup it is using, one to move the nspawn process
into, the other to move the actual container processes into. This is
done because of the requirement that cgroups may either contain
processes or other subgroups.
2015-09-01 19:22:36 +02:00
|
|
|
unit_release_cgroup(u);
|
2013-07-10 21:17:37 +02:00
|
|
|
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
if (!MANAGER_IS_RELOADING(u->manager))
|
|
|
|
unit_unlink_state_files(u);
|
|
|
|
|
2016-08-01 19:24:40 +02:00
|
|
|
unit_unref_uid_gid(u, false);
|
|
|
|
|
2015-09-11 17:25:35 +02:00
|
|
|
(void) manager_update_failed_units(u->manager, u, false);
|
2014-05-22 00:06:16 +02:00
|
|
|
set_remove(u->manager->startup_units, u);
|
2014-03-12 20:55:13 +01:00
|
|
|
|
2014-02-06 17:17:51 +01:00
|
|
|
unit_unwatch_all_pids(u);
|
|
|
|
|
2013-06-18 02:07:35 +02:00
|
|
|
unit_ref_unset(&u->slice);
|
2018-02-13 13:12:43 +01:00
|
|
|
while (u->refs_by_target)
|
|
|
|
unit_ref_unset(u->refs_by_target);
|
2012-01-06 23:08:54 +01:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->type != _UNIT_TYPE_INVALID)
|
2013-10-14 06:10:14 +02:00
|
|
|
LIST_REMOVE(units_by_type, u->manager->units_by_type[u->type], u);
|
2010-01-29 06:04:08 +01:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->in_load_queue)
|
2013-10-14 06:10:14 +02:00
|
|
|
LIST_REMOVE(load_queue, u->manager->load_queue, u);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->in_dbus_queue)
|
2013-10-14 06:10:14 +02:00
|
|
|
LIST_REMOVE(dbus_queue, u->manager->dbus_unit_queue, u);
|
2010-02-05 00:38:41 +01:00
|
|
|
|
2016-11-15 19:23:29 +01:00
|
|
|
if (u->in_gc_queue)
|
2016-11-15 19:32:50 +01:00
|
|
|
LIST_REMOVE(gc_queue, u->manager->gc_unit_queue, u);
|
2010-04-21 06:01:13 +02:00
|
|
|
|
2017-09-26 22:15:02 +02:00
|
|
|
if (u->in_cgroup_realize_queue)
|
|
|
|
LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2017-09-26 22:43:08 +02:00
|
|
|
if (u->in_cgroup_empty_queue)
|
|
|
|
LIST_REMOVE(cgroup_empty_queue, u->manager->cgroup_empty_queue, u);
|
|
|
|
|
pid1: properly remove references to the unit from gc queue during final cleanup
When various references to the unit were dropped during cleanup in unit_free(),
add_to_gc_queue() could be called on this unit. If the unit was previously in
the gc queue (at the time when unit_free() was called on it), this wouldn't
matter, because it'd have in_gc_queue still set even though it was already
removed from the queue. But if it wasn't set, then the unit could be added to
the queue. Then after unit_free() would deallocate the unit, we would be left
with a dangling pointer in gc_queue.
A unit could be added to the gc queue in two places called from unit_free():
in the job_install calls, and in unit_ref_unset(). The first was OK, because
it was above the LIST_REMOVE(gc_queue,...) call, but the second was not, because
it was after that. Move the all LIST_REMOVE() calls down.
2018-02-13 23:57:43 +01:00
|
|
|
if (u->in_cleanup_queue)
|
|
|
|
LIST_REMOVE(cleanup_queue, u->manager->cleanup_queue, u);
|
2018-01-24 19:59:55 +01:00
|
|
|
|
2018-03-23 15:28:06 +01:00
|
|
|
if (u->in_target_deps_queue)
|
|
|
|
LIST_REMOVE(target_deps_queue, u->manager->target_deps_queue, u);
|
|
|
|
|
2018-08-09 16:26:27 +02:00
|
|
|
if (u->in_stop_when_unneeded_queue)
|
|
|
|
LIST_REMOVE(stop_when_unneeded_queue, u->manager->stop_when_unneeded_queue, u);
|
|
|
|
|
2016-11-11 19:59:19 +01:00
|
|
|
safe_close(u->ip_accounting_ingress_map_fd);
|
|
|
|
safe_close(u->ip_accounting_egress_map_fd);
|
2013-07-10 21:17:37 +02:00
|
|
|
|
2016-11-11 19:59:19 +01:00
|
|
|
safe_close(u->ipv4_allow_map_fd);
|
|
|
|
safe_close(u->ipv6_allow_map_fd);
|
|
|
|
safe_close(u->ipv4_deny_map_fd);
|
|
|
|
safe_close(u->ipv6_deny_map_fd);
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
|
2016-11-11 19:59:19 +01:00
|
|
|
bpf_program_unref(u->ip_bpf_ingress);
|
2018-02-20 19:28:24 +01:00
|
|
|
bpf_program_unref(u->ip_bpf_ingress_installed);
|
2016-11-11 19:59:19 +01:00
|
|
|
bpf_program_unref(u->ip_bpf_egress);
|
2018-02-20 19:28:24 +01:00
|
|
|
bpf_program_unref(u->ip_bpf_egress_installed);
|
2016-08-01 19:24:40 +02:00
|
|
|
|
2019-04-23 12:14:20 +02:00
|
|
|
set_free(u->ip_bpf_custom_ingress);
|
|
|
|
set_free(u->ip_bpf_custom_egress);
|
|
|
|
set_free(u->ip_bpf_custom_ingress_installed);
|
|
|
|
set_free(u->ip_bpf_custom_egress_installed);
|
|
|
|
|
2018-10-08 23:33:05 +02:00
|
|
|
bpf_program_unref(u->bpf_device_control_installed);
|
|
|
|
|
2018-02-14 00:01:05 +01:00
|
|
|
condition_free_list(u->conditions);
|
|
|
|
condition_free_list(u->asserts);
|
2014-03-12 20:55:13 +01:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
free(u->description);
|
2012-05-21 15:12:18 +02:00
|
|
|
strv_free(u->documentation);
|
2012-01-15 12:04:08 +01:00
|
|
|
free(u->fragment_path);
|
2012-05-22 23:08:24 +02:00
|
|
|
free(u->source_path);
|
2013-04-01 12:32:35 +02:00
|
|
|
strv_free(u->dropin_paths);
|
2012-01-15 12:04:08 +01:00
|
|
|
free(u->instance);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2014-10-28 01:49:07 +01:00
|
|
|
free(u->job_timeout_reboot_arg);
|
core: make the StartLimitXYZ= settings generic and apply to any kind of unit, not just services
This moves the StartLimitBurst=, StartLimitInterval=, StartLimitAction=, RebootArgument= from the [Service] section
into the [Unit] section of unit files, and thus support it in all unit types, not just in services.
This way we can enforce the start limit much earlier, in particular before testing the unit conditions, so that
repeated start-up failure due to failed conditions is also considered for the start limit logic.
For compatibility the four options may also be configured in the [Service] section still, but we only document them in
their new section [Unit].
This also renamed the socket unit failure code "service-failed-permanent" into "service-start-limit-hit" to express
more clearly what it is about, after all it's only triggered through the start limit being hit.
Finally, the code in busname_trigger_notify() and socket_trigger_notify() is altered to become more alike.
Fixes: #2467
2016-02-09 18:38:03 +01:00
|
|
|
free(u->reboot_arg);
|
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
set_free_free(u->aliases);
|
|
|
|
free(u->id);
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
free(u);
|
|
|
|
}
|
|
|
|
|
2020-04-29 17:53:43 +02:00
|
|
|
FreezerState unit_freezer_state(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
return u->freezer_state;
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_freezer_state_kernel(Unit *u, FreezerState *ret) {
|
|
|
|
char *values[1] = {};
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
r = cg_get_keyed_attribute(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events",
|
|
|
|
STRV_MAKE("frozen"), values);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
r = _FREEZER_STATE_INVALID;
|
|
|
|
|
|
|
|
if (values[0]) {
|
|
|
|
if (streq(values[0], "0"))
|
|
|
|
r = FREEZER_RUNNING;
|
|
|
|
else if (streq(values[0], "1"))
|
|
|
|
r = FREEZER_FROZEN;
|
|
|
|
}
|
|
|
|
|
|
|
|
free(values[0]);
|
|
|
|
*ret = r;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
UnitActiveState unit_active_state(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->load_state == UNIT_MERGED)
|
2010-07-01 00:31:53 +02:00
|
|
|
return unit_active_state(unit_follow_merge(u));
|
|
|
|
|
|
|
|
/* After a reload it might happen that a unit is not correctly
|
|
|
|
* loaded but still has a process around. That's why we won't
|
2010-08-31 00:23:34 +02:00
|
|
|
* shortcut failed loading to UNIT_INACTIVE_FAILED. */
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
return UNIT_VTABLE(u)->active_state(u);
|
|
|
|
}
|
|
|
|
|
2010-04-13 20:59:01 +02:00
|
|
|
const char* unit_sub_state_to_string(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
return UNIT_VTABLE(u)->sub_state_to_string(u);
|
|
|
|
}
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
static int hashmap_complete_move(Hashmap **s, Hashmap **other) {
|
2010-04-06 02:43:58 +02:00
|
|
|
assert(s);
|
|
|
|
assert(other);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2010-04-06 02:43:58 +02:00
|
|
|
if (!*other)
|
2014-10-15 00:23:21 +02:00
|
|
|
return 0;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
if (*s)
|
|
|
|
return hashmap_move(*s, *other);
|
2018-03-22 16:53:26 +01:00
|
|
|
else
|
|
|
|
*s = TAKE_PTR(*other);
|
2014-10-15 00:23:21 +02:00
|
|
|
|
|
|
|
return 0;
|
2010-04-06 02:43:58 +02:00
|
|
|
}
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2014-10-15 00:23:21 +02:00
|
|
|
static int merge_names(Unit *u, Unit *other) {
|
2020-05-27 15:49:17 +02:00
|
|
|
char *name;
|
2010-04-06 02:43:58 +02:00
|
|
|
Iterator i;
|
2014-10-15 00:23:21 +02:00
|
|
|
int r;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2010-04-06 02:43:58 +02:00
|
|
|
assert(u);
|
|
|
|
assert(other);
|
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
r = unit_add_alias(u, other->id);
|
2014-10-15 00:23:21 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
r = set_move(u->aliases, other->aliases);
|
|
|
|
if (r < 0) {
|
|
|
|
set_remove(u->aliases, other->id);
|
|
|
|
return r;
|
|
|
|
}
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
TAKE_PTR(other->id);
|
|
|
|
other->aliases = set_free_free(other->aliases);
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
SET_FOREACH(name, u->aliases, i)
|
|
|
|
assert_se(hashmap_replace(u->manager->units, name, u) == 0);
|
2014-10-15 00:23:21 +02:00
|
|
|
|
|
|
|
return 0;
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
2014-10-15 00:00:30 +02:00
|
|
|
static int reserve_dependencies(Unit *u, Unit *other, UnitDependency d) {
|
|
|
|
unsigned n_reserve;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(other);
|
|
|
|
assert(d < _UNIT_DEPENDENCY_MAX);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If u does not have this dependency set allocated, there is no need
|
2014-12-29 10:45:58 +01:00
|
|
|
* to reserve anything. In that case other's set will be transferred
|
2014-10-15 00:00:30 +02:00
|
|
|
* as a whole to u by complete_move().
|
|
|
|
*/
|
|
|
|
if (!u->dependencies[d])
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* merge_dependencies() will skip a u-on-u dependency */
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
n_reserve = hashmap_size(other->dependencies[d]) - !!hashmap_get(other->dependencies[d], u);
|
2014-10-15 00:00:30 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
return hashmap_reserve(u->dependencies[d], n_reserve);
|
2014-10-15 00:00:30 +02:00
|
|
|
}
|
|
|
|
|
2014-08-08 02:46:49 +02:00
|
|
|
static void merge_dependencies(Unit *u, Unit *other, const char *other_id, UnitDependency d) {
|
2010-04-06 02:43:58 +02:00
|
|
|
Iterator i;
|
|
|
|
Unit *back;
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
void *v;
|
2010-01-26 21:39:06 +01:00
|
|
|
int r;
|
2010-04-06 02:43:58 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
/* Merges all dependencies of type 'd' of the unit 'other' into the deps of the unit 'u' */
|
|
|
|
|
2010-04-06 02:43:58 +02:00
|
|
|
assert(u);
|
|
|
|
assert(other);
|
|
|
|
assert(d < _UNIT_DEPENDENCY_MAX);
|
|
|
|
|
2019-04-27 02:22:40 +02:00
|
|
|
/* Fix backwards pointers. Let's iterate through all dependent units of the other unit. */
|
2020-05-28 15:25:22 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, back, other->dependencies[d], i)
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
|
2020-05-28 15:25:22 +02:00
|
|
|
/* Let's now iterate through the dependencies of that dependencies of the other units,
|
|
|
|
* looking for pointers back, and let's fix them up, to instead point to 'u'. */
|
|
|
|
for (UnitDependency k = 0; k < _UNIT_DEPENDENCY_MAX; k++)
|
2014-08-08 02:42:58 +02:00
|
|
|
if (back == u) {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
/* Do not add dependencies between u and itself. */
|
|
|
|
if (hashmap_remove(back->dependencies[k], other))
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
maybe_warn_about_dependency(u, other_id, k);
|
2014-08-08 02:42:58 +02:00
|
|
|
} else {
|
2020-05-28 15:25:22 +02:00
|
|
|
UnitDependencyInfo di_u, di_other;
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
|
|
|
|
/* Let's drop this dependency between "back" and "other", and let's create it between
|
|
|
|
* "back" and "u" instead. Let's merge the bit masks of the dependency we are moving,
|
|
|
|
* and any such dependency which might already exist */
|
|
|
|
|
|
|
|
di_other.data = hashmap_get(back->dependencies[k], other);
|
|
|
|
if (!di_other.data)
|
|
|
|
continue; /* dependency isn't set, let's try the next one */
|
|
|
|
|
|
|
|
di_u.data = hashmap_get(back->dependencies[k], u);
|
|
|
|
|
2020-05-28 15:25:22 +02:00
|
|
|
UnitDependencyInfo di_merged = {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
.origin_mask = di_u.origin_mask | di_other.origin_mask,
|
|
|
|
.destination_mask = di_u.destination_mask | di_other.destination_mask,
|
|
|
|
};
|
|
|
|
|
|
|
|
r = hashmap_remove_and_replace(back->dependencies[k], other, u, di_merged.data);
|
|
|
|
if (r < 0)
|
|
|
|
log_warning_errno(r, "Failed to remove/replace: back=%s other=%s u=%s: %m", back->id, other_id, u->id);
|
|
|
|
assert(r >= 0);
|
|
|
|
|
|
|
|
/* assert_se(hashmap_remove_and_replace(back->dependencies[k], other, u, di_merged.data) >= 0); */
|
2014-08-08 02:42:58 +02:00
|
|
|
}
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2014-08-08 02:42:58 +02:00
|
|
|
/* Also do not move dependencies on u to itself */
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
back = hashmap_remove(other->dependencies[d], u);
|
2014-08-08 02:46:49 +02:00
|
|
|
if (back)
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
maybe_warn_about_dependency(u, other_id, d);
|
2014-08-08 02:42:58 +02:00
|
|
|
|
2014-10-15 00:23:21 +02:00
|
|
|
/* The move cannot fail. The caller must have performed a reservation. */
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
assert_se(hashmap_complete_move(&u->dependencies[d], &other->dependencies[d]) == 0);
|
2010-04-06 02:43:58 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
other->dependencies[d] = hashmap_free(other->dependencies[d]);
|
2010-04-06 02:43:58 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
int unit_merge(Unit *u, Unit *other) {
|
2014-08-08 02:46:49 +02:00
|
|
|
const char *other_id = NULL;
|
2014-10-15 00:00:30 +02:00
|
|
|
int r;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(other);
|
2012-01-15 12:04:08 +01:00
|
|
|
assert(u->manager == other->manager);
|
|
|
|
assert(u->type != _UNIT_TYPE_INVALID);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2010-04-08 04:34:42 +02:00
|
|
|
other = unit_follow_merge(other);
|
|
|
|
|
2010-04-06 02:43:58 +02:00
|
|
|
if (other == u)
|
|
|
|
return 0;
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->type != other->type)
|
2010-04-15 03:11:11 +02:00
|
|
|
return -EINVAL;
|
|
|
|
|
2016-04-30 22:21:41 +02:00
|
|
|
if (!unit_type_may_alias(u->type)) /* Merging only applies to unit names that support aliases */
|
2016-04-29 17:31:02 +02:00
|
|
|
return -EEXIST;
|
|
|
|
|
2017-09-29 09:58:22 +02:00
|
|
|
if (!IN_SET(other->load_state, UNIT_STUB, UNIT_NOT_FOUND))
|
2010-04-06 02:43:58 +02:00
|
|
|
return -EEXIST;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2020-05-27 16:39:35 +02:00
|
|
|
if (!streq_ptr(u->instance, other->instance))
|
|
|
|
return -EINVAL;
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (other->job)
|
2010-04-10 04:42:36 +02:00
|
|
|
return -EEXIST;
|
|
|
|
|
core: add NOP jobs, job type collapsing
Two of our current job types are special:
JOB_TRY_RESTART, JOB_RELOAD_OR_START.
They differ from other job types by being sensitive to the unit active state.
They perform some action when the unit is active and some other action
otherwise. This raises a question: when exactly should the unit state be
checked to make the decision?
Currently the unit state is checked when the job becomes runnable. It's more
sensible to check the state immediately when the job is added by the user.
When the user types "systemctl try-restart foo.service", he really intends
to restart the service if it's running right now. If it isn't running right
now, the restart is pointless.
Consider the example (from Bugzilla[1]):
sleep.service takes some time to start.
hello.service has After=sleep.service.
Both services get started. Two jobs will appear:
hello.service/start waiting
sleep.service/start running
Then someone runs "systemctl try-restart hello.service".
Currently the try-restart operation will block and wait for
sleep.service/start to complete.
The correct result is to complete the try-restart operation immediately
with success, because hello.service is not running. The two original
jobs must not be disturbed by this.
To fix this we introduce two new concepts:
- a new job type: JOB_NOP
A JOB_NOP job does not do anything to the unit. It does not pull in any
dependencies. It is always immediately runnable. When installed to a unit,
it sits in a special slot (u->nop_job) where it never conflicts with
the installed job (u->job) of a different type. It never merges with jobs
of other types, but it can merge into an already installed JOB_NOP job.
- "collapsing" of job types
When a job of one of the two special types is added, the state of the unit
is checked immediately and the job type changes:
JOB_TRY_RESTART -> JOB_RESTART or JOB_NOP
JOB_RELOAD_OR_START -> JOB_RELOAD or JOB_START
Should a job type JOB_RELOAD_OR_START appear later during job merging, it
collapses immediately afterwards.
Collapsing actually makes some things simpler, because there are now fewer
job types that are allowed in the transaction.
[1] Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=753586
2012-04-25 11:58:27 +02:00
|
|
|
if (other->nop_job)
|
|
|
|
return -EEXIST;
|
|
|
|
|
2010-08-31 00:23:34 +02:00
|
|
|
if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
|
2010-04-10 04:42:36 +02:00
|
|
|
return -EEXIST;
|
|
|
|
|
2014-08-08 02:46:49 +02:00
|
|
|
if (other->id)
|
|
|
|
other_id = strdupa(other->id);
|
|
|
|
|
2014-10-15 00:00:30 +02:00
|
|
|
/* Make reservations to ensure merge_dependencies() won't fail */
|
2020-05-28 15:25:22 +02:00
|
|
|
for (UnitDependency d = 0; d < _UNIT_DEPENDENCY_MAX; d++) {
|
2014-10-15 00:00:30 +02:00
|
|
|
r = reserve_dependencies(u, other, d);
|
|
|
|
/*
|
|
|
|
* We don't rollback reservations if we fail. We don't have
|
|
|
|
* a way to undo reservations. A reservation is not a leak.
|
|
|
|
*/
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
/* Merge names */
|
2014-10-15 00:23:21 +02:00
|
|
|
r = merge_names(u, other);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2012-01-06 23:08:54 +01:00
|
|
|
/* Redirect all references */
|
2018-02-13 13:12:43 +01:00
|
|
|
while (other->refs_by_target)
|
|
|
|
unit_ref_set(other->refs_by_target, other->refs_by_target->source, u);
|
2012-01-06 23:08:54 +01:00
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
/* Merge dependencies */
|
2020-05-28 15:25:22 +02:00
|
|
|
for (UnitDependency d = 0; d < _UNIT_DEPENDENCY_MAX; d++)
|
2014-08-08 02:46:49 +02:00
|
|
|
merge_dependencies(u, other, other_id, d);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
other->load_state = UNIT_MERGED;
|
|
|
|
other->merged_into = u;
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2010-04-10 04:43:21 +02:00
|
|
|
/* If there is still some data attached to the other node, we
|
|
|
|
* don't need it anymore, and can free it. */
|
2012-01-15 12:04:08 +01:00
|
|
|
if (other->load_state != UNIT_STUB)
|
2010-04-10 04:43:21 +02:00
|
|
|
if (UNIT_VTABLE(other)->done)
|
|
|
|
UNIT_VTABLE(other)->done(other);
|
|
|
|
|
|
|
|
unit_add_to_dbus_queue(u);
|
2010-04-06 02:43:58 +02:00
|
|
|
unit_add_to_cleanup_queue(other);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_merge_by_name(Unit *u, const char *name) {
|
2016-04-29 17:31:02 +02:00
|
|
|
_cleanup_free_ char *s = NULL;
|
2010-04-06 02:43:58 +02:00
|
|
|
Unit *other;
|
2010-04-15 03:11:11 +02:00
|
|
|
int r;
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2019-07-18 13:11:28 +02:00
|
|
|
/* Either add name to u, or if a unit with name already exists, merge it with u.
|
|
|
|
* If name is a template, do the same for name@instance, where instance is u's instance. */
|
|
|
|
|
2010-04-06 02:43:58 +02:00
|
|
|
assert(u);
|
|
|
|
assert(name);
|
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
|
2012-01-15 12:04:08 +01:00
|
|
|
if (!u->instance)
|
2010-04-15 03:11:11 +02:00
|
|
|
return -EINVAL;
|
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
r = unit_name_replace_instance(name, u->instance, &s);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2010-04-15 03:11:11 +02:00
|
|
|
|
|
|
|
name = s;
|
|
|
|
}
|
|
|
|
|
2013-06-28 04:12:58 +02:00
|
|
|
other = manager_get_unit(u->manager, name);
|
2015-04-30 20:21:00 +02:00
|
|
|
if (other)
|
|
|
|
return unit_merge(u, other);
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
return unit_add_name(u, name);
|
2010-04-06 02:43:58 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
Unit* unit_follow_merge(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
while (u->load_state == UNIT_MERGED)
|
|
|
|
assert_se(u = u->merged_into);
|
2010-04-06 02:43:58 +02:00
|
|
|
|
|
|
|
return u;
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_add_exec_dependencies(Unit *u, ExecContext *c) {
|
2017-08-31 11:19:35 +02:00
|
|
|
ExecDirectoryType dt;
|
|
|
|
char **dp;
|
2010-04-06 02:43:58 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(c);
|
|
|
|
|
2018-10-24 13:22:01 +02:00
|
|
|
if (c->working_directory && !c->working_directory_missing_ok) {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
r = unit_require_mounts_for(u, c->working_directory, UNIT_DEPENDENCY_FILE);
|
2014-01-27 07:23:20 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (c->root_directory) {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
r = unit_require_mounts_for(u, c->root_directory, UNIT_DEPENDENCY_FILE);
|
2014-01-27 07:23:20 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2016-12-23 14:26:05 +01:00
|
|
|
if (c->root_image) {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
r = unit_require_mounts_for(u, c->root_image, UNIT_DEPENDENCY_FILE);
|
2016-12-23 14:26:05 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2017-09-28 16:58:43 +02:00
|
|
|
for (dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) {
|
2017-08-31 11:19:35 +02:00
|
|
|
if (!u->manager->prefix[dt])
|
|
|
|
continue;
|
|
|
|
|
|
|
|
STRV_FOREACH(dp, c->directories[dt].paths) {
|
|
|
|
_cleanup_free_ char *p;
|
|
|
|
|
2019-06-20 20:07:01 +02:00
|
|
|
p = path_join(u->manager->prefix[dt], *dp);
|
2017-08-31 11:19:35 +02:00
|
|
|
if (!p)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
r = unit_require_mounts_for(u, p, UNIT_DEPENDENCY_FILE);
|
2017-08-31 11:19:35 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-02-24 21:24:23 +01:00
|
|
|
if (!MANAGER_IS_SYSTEM(u->manager))
|
2014-03-19 16:55:43 +01:00
|
|
|
return 0;
|
|
|
|
|
2020-04-08 16:08:35 +02:00
|
|
|
/* For the following three directory types we need write access, and /var/ is possibly on the root
|
|
|
|
* fs. Hence order after systemd-remount-fs.service, to ensure things are writable. */
|
|
|
|
if (!strv_isempty(c->directories[EXEC_DIRECTORY_STATE].paths) ||
|
|
|
|
!strv_isempty(c->directories[EXEC_DIRECTORY_CACHE].paths) ||
|
|
|
|
!strv_isempty(c->directories[EXEC_DIRECTORY_LOGS].paths)) {
|
|
|
|
r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_REMOUNT_FS_SERVICE, true, UNIT_DEPENDENCY_FILE);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2014-03-19 16:55:43 +01:00
|
|
|
if (c->private_tmp) {
|
2016-12-27 23:25:24 +01:00
|
|
|
const char *p;
|
|
|
|
|
|
|
|
FOREACH_STRING(p, "/tmp", "/var/tmp") {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
r = unit_require_mounts_for(u, p, UNIT_DEPENDENCY_FILE);
|
2016-12-27 23:25:24 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
2014-03-19 16:55:43 +01:00
|
|
|
|
2018-09-15 19:57:52 +02:00
|
|
|
r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_TMPFILES_SETUP_SERVICE, true, UNIT_DEPENDENCY_FILE);
|
2014-03-19 16:55:43 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2020-04-21 13:04:39 +02:00
|
|
|
if (c->root_image) {
|
|
|
|
/* We need to wait for /dev/loopX to appear when doing RootImage=, hence let's add an
|
|
|
|
* implicit dependency on udev */
|
|
|
|
|
|
|
|
r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_UDEVD_SERVICE, true, UNIT_DEPENDENCY_FILE);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2016-10-18 02:05:49 +02:00
|
|
|
if (!IN_SET(c->std_output,
|
|
|
|
EXEC_OUTPUT_JOURNAL, EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
|
2020-05-14 12:20:39 +02:00
|
|
|
EXEC_OUTPUT_KMSG, EXEC_OUTPUT_KMSG_AND_CONSOLE) &&
|
2016-10-18 02:05:49 +02:00
|
|
|
!IN_SET(c->std_error,
|
|
|
|
EXEC_OUTPUT_JOURNAL, EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
|
2020-05-14 12:20:39 +02:00
|
|
|
EXEC_OUTPUT_KMSG, EXEC_OUTPUT_KMSG_AND_CONSOLE) &&
|
2019-11-25 16:22:45 +01:00
|
|
|
!c->log_namespace)
|
2010-04-06 02:43:58 +02:00
|
|
|
return 0;
|
|
|
|
|
2019-11-25 16:22:45 +01:00
|
|
|
/* If syslog or kernel logging is requested (or log namespacing is), make sure our own logging daemon
|
|
|
|
* is run first. */
|
|
|
|
|
|
|
|
if (c->log_namespace) {
|
2019-11-27 14:47:37 +01:00
|
|
|
_cleanup_free_ char *socket_unit = NULL, *varlink_socket_unit = NULL;
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2019-11-25 16:22:45 +01:00
|
|
|
r = unit_name_build_from_type("systemd-journald", c->log_namespace, UNIT_SOCKET, &socket_unit);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_REQUIRES, socket_unit, true, UNIT_DEPENDENCY_FILE);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2019-11-27 14:47:37 +01:00
|
|
|
|
|
|
|
r = unit_name_build_from_type("systemd-journald-varlink", c->log_namespace, UNIT_SOCKET, &varlink_socket_unit);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_REQUIRES, varlink_socket_unit, true, UNIT_DEPENDENCY_FILE);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2019-11-25 16:22:45 +01:00
|
|
|
} else
|
|
|
|
r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_JOURNALD_SOCKET, true, UNIT_DEPENDENCY_FILE);
|
2014-03-19 16:55:43 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
const char *unit_description(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->description)
|
|
|
|
return u->description;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
return strna(u->id);
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
2019-06-06 17:33:59 +02:00
|
|
|
const char *unit_status_string(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (u->manager->status_unit_format == STATUS_UNIT_FORMAT_NAME && u->id)
|
|
|
|
return u->id;
|
|
|
|
|
|
|
|
return unit_description(u);
|
|
|
|
}
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
static void print_unit_dependency_mask(FILE *f, const char *kind, UnitDependencyMask mask, bool *space) {
|
|
|
|
const struct {
|
|
|
|
UnitDependencyMask mask;
|
|
|
|
const char *name;
|
|
|
|
} table[] = {
|
|
|
|
{ UNIT_DEPENDENCY_FILE, "file" },
|
|
|
|
{ UNIT_DEPENDENCY_IMPLICIT, "implicit" },
|
|
|
|
{ UNIT_DEPENDENCY_DEFAULT, "default" },
|
|
|
|
{ UNIT_DEPENDENCY_UDEV, "udev" },
|
|
|
|
{ UNIT_DEPENDENCY_PATH, "path" },
|
|
|
|
{ UNIT_DEPENDENCY_MOUNTINFO_IMPLICIT, "mountinfo-implicit" },
|
|
|
|
{ UNIT_DEPENDENCY_MOUNTINFO_DEFAULT, "mountinfo-default" },
|
|
|
|
{ UNIT_DEPENDENCY_PROC_SWAP, "proc-swap" },
|
|
|
|
};
|
|
|
|
size_t i;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(kind);
|
|
|
|
assert(space);
|
|
|
|
|
|
|
|
for (i = 0; i < ELEMENTSOF(table); i++) {
|
|
|
|
|
|
|
|
if (mask == 0)
|
|
|
|
break;
|
|
|
|
|
2018-04-20 15:36:20 +02:00
|
|
|
if (FLAGS_SET(mask, table[i].mask)) {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
if (*space)
|
|
|
|
fputc(' ', f);
|
|
|
|
else
|
|
|
|
*space = true;
|
|
|
|
|
|
|
|
fputs(kind, f);
|
|
|
|
fputs("-", f);
|
|
|
|
fputs(table[i].name, f);
|
|
|
|
|
|
|
|
mask &= ~table[i].mask;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(mask == 0);
|
|
|
|
}
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
void unit_dump(Unit *u, FILE *f, const char *prefix) {
|
2012-05-21 15:12:18 +02:00
|
|
|
char *t, **j;
|
2010-01-26 21:39:06 +01:00
|
|
|
Iterator i;
|
2010-02-03 14:21:48 +01:00
|
|
|
const char *prefix2;
|
2019-07-05 13:44:31 +02:00
|
|
|
char timestamp[5][FORMAT_TIMESTAMP_MAX], timespan[FORMAT_TIMESPAN_MAX];
|
2010-07-21 05:00:29 +02:00
|
|
|
Unit *following;
|
2013-11-25 21:16:37 +01:00
|
|
|
_cleanup_set_free_ Set *following_set = NULL;
|
2016-08-15 18:12:01 +02:00
|
|
|
const char *n;
|
2017-11-09 15:29:34 +01:00
|
|
|
CGroupMask m;
|
|
|
|
int r;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
assert(u);
|
2012-01-15 12:04:08 +01:00
|
|
|
assert(u->type >= 0);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2014-08-21 16:15:49 +02:00
|
|
|
prefix = strempty(prefix);
|
2015-02-03 02:05:59 +01:00
|
|
|
prefix2 = strjoina(prefix, "\t");
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
fprintf(f,
|
2019-07-05 13:34:54 +02:00
|
|
|
"%s-> Unit %s:\n",
|
|
|
|
prefix, u->id);
|
|
|
|
|
2020-05-27 15:49:17 +02:00
|
|
|
SET_FOREACH(t, u->aliases, i)
|
|
|
|
fprintf(f, "%s\tAlias: %s\n", prefix, t);
|
2019-07-05 13:34:54 +02:00
|
|
|
|
|
|
|
fprintf(f,
|
2010-01-26 21:39:06 +01:00
|
|
|
"%s\tDescription: %s\n"
|
2010-04-15 03:11:11 +02:00
|
|
|
"%s\tInstance: %s\n"
|
2010-01-26 21:39:06 +01:00
|
|
|
"%s\tUnit Load State: %s\n"
|
2010-04-10 04:43:57 +02:00
|
|
|
"%s\tUnit Active State: %s\n"
|
2016-02-12 21:33:39 +01:00
|
|
|
"%s\tState Change Timestamp: %s\n"
|
2010-05-14 03:05:38 +02:00
|
|
|
"%s\tInactive Exit Timestamp: %s\n"
|
2010-04-10 04:43:57 +02:00
|
|
|
"%s\tActive Enter Timestamp: %s\n"
|
2010-04-21 06:01:13 +02:00
|
|
|
"%s\tActive Exit Timestamp: %s\n"
|
2010-05-14 03:05:38 +02:00
|
|
|
"%s\tInactive Enter Timestamp: %s\n"
|
2018-02-13 10:50:13 +01:00
|
|
|
"%s\tMay GC: %s\n"
|
2013-06-20 03:45:08 +02:00
|
|
|
"%s\tNeed Daemon Reload: %s\n"
|
2013-06-28 04:12:58 +02:00
|
|
|
"%s\tTransient: %s\n"
|
2016-10-24 21:41:54 +02:00
|
|
|
"%s\tPerpetual: %s\n"
|
2017-11-13 17:14:07 +01:00
|
|
|
"%s\tGarbage Collection Mode: %s\n"
|
2013-06-27 04:14:27 +02:00
|
|
|
"%s\tSlice: %s\n"
|
|
|
|
"%s\tCGroup: %s\n"
|
2017-05-02 09:59:17 +02:00
|
|
|
"%s\tCGroup realized: %s\n",
|
2010-01-26 21:39:06 +01:00
|
|
|
prefix, unit_description(u),
|
2012-01-15 12:04:08 +01:00
|
|
|
prefix, strna(u->instance),
|
|
|
|
prefix, unit_load_state_to_string(u->load_state),
|
2010-04-10 04:43:57 +02:00
|
|
|
prefix, unit_active_state_to_string(unit_active_state(u)),
|
2019-07-05 13:44:31 +02:00
|
|
|
prefix, strna(format_timestamp(timestamp[0], sizeof(timestamp[0]), u->state_change_timestamp.realtime)),
|
|
|
|
prefix, strna(format_timestamp(timestamp[1], sizeof(timestamp[1]), u->inactive_exit_timestamp.realtime)),
|
|
|
|
prefix, strna(format_timestamp(timestamp[2], sizeof(timestamp[2]), u->active_enter_timestamp.realtime)),
|
|
|
|
prefix, strna(format_timestamp(timestamp[3], sizeof(timestamp[3]), u->active_exit_timestamp.realtime)),
|
|
|
|
prefix, strna(format_timestamp(timestamp[4], sizeof(timestamp[4]), u->inactive_enter_timestamp.realtime)),
|
2018-02-13 10:50:13 +01:00
|
|
|
prefix, yes_no(unit_may_gc(u)),
|
2013-06-20 03:45:08 +02:00
|
|
|
prefix, yes_no(unit_need_daemon_reload(u)),
|
2013-06-28 04:12:58 +02:00
|
|
|
prefix, yes_no(u->transient),
|
2016-10-24 21:41:54 +02:00
|
|
|
prefix, yes_no(u->perpetual),
|
2017-11-13 17:14:07 +01:00
|
|
|
prefix, collect_mode_to_string(u->collect_mode),
|
2013-06-27 04:14:27 +02:00
|
|
|
prefix, strna(unit_slice_name(u)),
|
|
|
|
prefix, strna(u->cgroup_path),
|
2017-05-02 09:59:17 +02:00
|
|
|
prefix, yes_no(u->cgroup_realized));
|
|
|
|
|
|
|
|
if (u->cgroup_realized_mask != 0) {
|
|
|
|
_cleanup_free_ char *s = NULL;
|
|
|
|
(void) cg_mask_to_string(u->cgroup_realized_mask, &s);
|
2017-11-09 15:29:34 +01:00
|
|
|
fprintf(f, "%s\tCGroup realized mask: %s\n", prefix, strnull(s));
|
|
|
|
}
|
2018-11-21 17:48:41 +01:00
|
|
|
|
2017-11-09 15:29:34 +01:00
|
|
|
if (u->cgroup_enabled_mask != 0) {
|
|
|
|
_cleanup_free_ char *s = NULL;
|
|
|
|
(void) cg_mask_to_string(u->cgroup_enabled_mask, &s);
|
|
|
|
fprintf(f, "%s\tCGroup enabled mask: %s\n", prefix, strnull(s));
|
|
|
|
}
|
2018-11-21 17:48:41 +01:00
|
|
|
|
2017-11-09 15:29:34 +01:00
|
|
|
m = unit_get_own_mask(u);
|
|
|
|
if (m != 0) {
|
|
|
|
_cleanup_free_ char *s = NULL;
|
|
|
|
(void) cg_mask_to_string(m, &s);
|
|
|
|
fprintf(f, "%s\tCGroup own mask: %s\n", prefix, strnull(s));
|
2017-05-02 09:59:17 +02:00
|
|
|
}
|
2018-11-21 17:48:41 +01:00
|
|
|
|
2017-11-09 15:29:34 +01:00
|
|
|
m = unit_get_members_mask(u);
|
|
|
|
if (m != 0) {
|
2017-05-02 09:59:17 +02:00
|
|
|
_cleanup_free_ char *s = NULL;
|
2017-11-09 15:29:34 +01:00
|
|
|
(void) cg_mask_to_string(m, &s);
|
2017-05-02 09:59:17 +02:00
|
|
|
fprintf(f, "%s\tCGroup members mask: %s\n", prefix, strnull(s));
|
|
|
|
}
|
2010-01-27 00:15:56 +01:00
|
|
|
|
2018-11-21 17:48:41 +01:00
|
|
|
m = unit_get_delegate_mask(u);
|
|
|
|
if (m != 0) {
|
|
|
|
_cleanup_free_ char *s = NULL;
|
|
|
|
(void) cg_mask_to_string(m, &s);
|
|
|
|
fprintf(f, "%s\tCGroup delegate mask: %s\n", prefix, strnull(s));
|
|
|
|
}
|
|
|
|
|
2016-08-30 23:18:46 +02:00
|
|
|
if (!sd_id128_is_null(u->invocation_id))
|
|
|
|
fprintf(f, "%s\tInvocation ID: " SD_ID128_FORMAT_STR "\n",
|
|
|
|
prefix, SD_ID128_FORMAT_VAL(u->invocation_id));
|
|
|
|
|
2012-05-21 15:12:18 +02:00
|
|
|
STRV_FOREACH(j, u->documentation)
|
|
|
|
fprintf(f, "%s\tDocumentation: %s\n", prefix, *j);
|
|
|
|
|
2013-11-25 21:16:37 +01:00
|
|
|
following = unit_following(u);
|
|
|
|
if (following)
|
2012-01-15 12:04:08 +01:00
|
|
|
fprintf(f, "%s\tFollowing: %s\n", prefix, following->id);
|
2010-07-20 20:33:19 +02:00
|
|
|
|
2013-11-25 21:16:37 +01:00
|
|
|
r = unit_following_set(u, &following_set);
|
|
|
|
if (r >= 0) {
|
|
|
|
Unit *other;
|
|
|
|
|
|
|
|
SET_FOREACH(other, following_set, i)
|
|
|
|
fprintf(f, "%s\tFollowing Set Member: %s\n", prefix, other->id);
|
|
|
|
}
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->fragment_path)
|
|
|
|
fprintf(f, "%s\tFragment Path: %s\n", prefix, u->fragment_path);
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2012-05-22 23:08:24 +02:00
|
|
|
if (u->source_path)
|
|
|
|
fprintf(f, "%s\tSource Path: %s\n", prefix, u->source_path);
|
|
|
|
|
2013-04-01 12:32:35 +02:00
|
|
|
STRV_FOREACH(j, u->dropin_paths)
|
2013-04-02 16:24:45 +02:00
|
|
|
fprintf(f, "%s\tDropIn Path: %s\n", prefix, *j);
|
2013-04-01 12:32:35 +02:00
|
|
|
|
2017-11-16 15:18:01 +01:00
|
|
|
if (u->failure_action != EMERGENCY_ACTION_NONE)
|
|
|
|
fprintf(f, "%s\tFailure Action: %s\n", prefix, emergency_action_to_string(u->failure_action));
|
2018-11-16 11:41:18 +01:00
|
|
|
if (u->failure_action_exit_status >= 0)
|
|
|
|
fprintf(f, "%s\tFailure Action Exit Status: %i\n", prefix, u->failure_action_exit_status);
|
2017-11-16 15:18:01 +01:00
|
|
|
if (u->success_action != EMERGENCY_ACTION_NONE)
|
|
|
|
fprintf(f, "%s\tSuccess Action: %s\n", prefix, emergency_action_to_string(u->success_action));
|
2018-11-16 11:41:18 +01:00
|
|
|
if (u->success_action_exit_status >= 0)
|
|
|
|
fprintf(f, "%s\tSuccess Action Exit Status: %i\n", prefix, u->success_action_exit_status);
|
2017-11-16 15:18:01 +01:00
|
|
|
|
core: rework unit timeout handling, and add new setting RuntimeMaxSec=
This clean-ups timeout handling in PID 1. Specifically, instead of storing 0 in internal timeout variables as
indication for a disabled timeout, use USEC_INFINITY which is in-line with how we do this in the rest of our code
(following the logic that 0 means "no", and USEC_INFINITY means "never").
This also replace all usec_t additions with invocations to usec_add(), so that USEC_INFINITY is properly propagated,
and sd-event considers it has indication for turning off the event source.
This also alters the deserialization of the units to restart timeouts from the time they were originally started from.
Before this patch timeouts would be restarted beginning with the time of the deserialization, which could lead to
artificially prolonged timeouts if a daemon reload took place.
Finally, a new RuntimeMaxSec= setting is introduced for service units, that specifies a maximum runtime after which a
specific service is forcibly terminated. This is useful to put time limits on time-intensive processing jobs.
This also simplifies the various xyz_spawn() calls of the various types in that explicit distruction of the timers is
removed, as that is done anyway by the state change handlers, and a state change is always done when the xyz_spawn()
calls fail.
Fixes: #2249
2016-02-01 21:48:10 +01:00
|
|
|
if (u->job_timeout != USEC_INFINITY)
|
2013-04-04 02:56:56 +02:00
|
|
|
fprintf(f, "%s\tJob Timeout: %s\n", prefix, format_timespan(timespan, sizeof(timespan), u->job_timeout, 0));
|
2010-07-17 04:09:28 +02:00
|
|
|
|
2016-10-20 15:27:37 +02:00
|
|
|
if (u->job_timeout_action != EMERGENCY_ACTION_NONE)
|
|
|
|
fprintf(f, "%s\tJob Timeout Action: %s\n", prefix, emergency_action_to_string(u->job_timeout_action));
|
2014-10-28 01:49:07 +01:00
|
|
|
|
|
|
|
if (u->job_timeout_reboot_arg)
|
|
|
|
fprintf(f, "%s\tJob Timeout Reboot Argument: %s\n", prefix, u->job_timeout_reboot_arg);
|
|
|
|
|
2014-11-06 13:43:45 +01:00
|
|
|
condition_dump_list(u->conditions, f, prefix, condition_type_to_string);
|
|
|
|
condition_dump_list(u->asserts, f, prefix, assert_type_to_string);
|
2010-10-13 02:15:41 +02:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (dual_timestamp_is_set(&u->condition_timestamp))
|
2011-03-17 04:36:19 +01:00
|
|
|
fprintf(f,
|
|
|
|
"%s\tCondition Timestamp: %s\n"
|
|
|
|
"%s\tCondition Result: %s\n",
|
2019-07-05 13:44:31 +02:00
|
|
|
prefix, strna(format_timestamp(timestamp[0], sizeof(timestamp[0]), u->condition_timestamp.realtime)),
|
2012-01-15 12:04:08 +01:00
|
|
|
prefix, yes_no(u->condition_result));
|
2011-03-17 04:36:19 +01:00
|
|
|
|
2014-11-06 13:43:45 +01:00
|
|
|
if (dual_timestamp_is_set(&u->assert_timestamp))
|
|
|
|
fprintf(f,
|
|
|
|
"%s\tAssert Timestamp: %s\n"
|
|
|
|
"%s\tAssert Result: %s\n",
|
2019-07-05 13:44:31 +02:00
|
|
|
prefix, strna(format_timestamp(timestamp[0], sizeof(timestamp[0]), u->assert_timestamp.realtime)),
|
2014-11-06 13:43:45 +01:00
|
|
|
prefix, yes_no(u->assert_result));
|
|
|
|
|
2020-05-28 15:25:22 +02:00
|
|
|
for (UnitDependency d = 0; d < _UNIT_DEPENDENCY_MAX; d++) {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
UnitDependencyInfo di;
|
2010-01-26 21:39:06 +01:00
|
|
|
Unit *other;
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(di.data, other, u->dependencies[d], i) {
|
|
|
|
bool space = false;
|
|
|
|
|
|
|
|
fprintf(f, "%s\t%s: %s (", prefix, unit_dependency_to_string(d), other->id);
|
|
|
|
|
|
|
|
print_unit_dependency_mask(f, "origin", di.origin_mask, &space);
|
|
|
|
print_unit_dependency_mask(f, "destination", di.destination_mask, &space);
|
|
|
|
|
|
|
|
fputs(")\n", f);
|
|
|
|
}
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
if (!hashmap_isempty(u->requires_mounts_for)) {
|
|
|
|
UnitDependencyInfo di;
|
|
|
|
const char *path;
|
|
|
|
|
|
|
|
HASHMAP_FOREACH_KEY(di.data, path, u->requires_mounts_for, i) {
|
|
|
|
bool space = false;
|
2012-04-29 14:26:07 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
fprintf(f, "%s\tRequiresMountsFor: %s (", prefix, path);
|
2012-04-29 14:26:07 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
print_unit_dependency_mask(f, "origin", di.origin_mask, &space);
|
|
|
|
print_unit_dependency_mask(f, "destination", di.destination_mask, &space);
|
|
|
|
|
|
|
|
fputs(")\n", f);
|
|
|
|
}
|
2012-04-29 14:26:07 +02:00
|
|
|
}
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->load_state == UNIT_LOADED) {
|
2011-08-20 00:20:41 +02:00
|
|
|
|
2010-04-10 04:44:24 +02:00
|
|
|
fprintf(f,
|
2010-07-03 19:48:33 +02:00
|
|
|
"%s\tStopWhenUnneeded: %s\n"
|
2010-08-10 20:57:21 +02:00
|
|
|
"%s\tRefuseManualStart: %s\n"
|
|
|
|
"%s\tRefuseManualStop: %s\n"
|
2011-04-07 04:11:31 +02:00
|
|
|
"%s\tDefaultDependencies: %s\n"
|
2013-11-26 01:39:53 +01:00
|
|
|
"%s\tOnFailureJobMode: %s\n"
|
Remove snapshot unit type
Snapshots were never useful or used for anything. Many systemd
developers that I spoke to at systemd.conf2015, didn't even know they
existed, so it is fairly safe to assume that this type can be deleted
without harm.
The fundamental problem with snapshots is that the state of the system
is dynamic, devices come and go, users log in and out, timers fire...
and restoring all units to some state from the past would "undo"
those changes, which isn't really possible.
Tested by creating a snapshot, running the new binary, and checking
that the transition did not cause errors, and the snapshot is gone,
and snapshots cannot be created anymore.
New systemctl says:
Unknown operation snapshot.
Old systemctl says:
Failed to create snapshot: Support for snapshots has been removed.
IgnoreOnSnaphost settings are warned about and ignored:
Support for option IgnoreOnSnapshot= has been removed and it is ignored
http://lists.freedesktop.org/archives/systemd-devel/2015-November/034872.html
2015-11-08 14:12:54 +01:00
|
|
|
"%s\tIgnoreOnIsolate: %s\n",
|
2012-01-15 12:04:08 +01:00
|
|
|
prefix, yes_no(u->stop_when_unneeded),
|
|
|
|
prefix, yes_no(u->refuse_manual_start),
|
|
|
|
prefix, yes_no(u->refuse_manual_stop),
|
|
|
|
prefix, yes_no(u->default_dependencies),
|
2013-11-26 01:39:53 +01:00
|
|
|
prefix, job_mode_to_string(u->on_failure_job_mode),
|
Remove snapshot unit type
Snapshots were never useful or used for anything. Many systemd
developers that I spoke to at systemd.conf2015, didn't even know they
existed, so it is fairly safe to assume that this type can be deleted
without harm.
The fundamental problem with snapshots is that the state of the system
is dynamic, devices come and go, users log in and out, timers fire...
and restoring all units to some state from the past would "undo"
those changes, which isn't really possible.
Tested by creating a snapshot, running the new binary, and checking
that the transition did not cause errors, and the snapshot is gone,
and snapshots cannot be created anymore.
New systemctl says:
Unknown operation snapshot.
Old systemctl says:
Failed to create snapshot: Support for snapshots has been removed.
IgnoreOnSnaphost settings are warned about and ignored:
Support for option IgnoreOnSnapshot= has been removed and it is ignored
http://lists.freedesktop.org/archives/systemd-devel/2015-November/034872.html
2015-11-08 14:12:54 +01:00
|
|
|
prefix, yes_no(u->ignore_on_isolate));
|
2012-01-15 12:04:08 +01:00
|
|
|
|
2010-04-06 02:43:58 +02:00
|
|
|
if (UNIT_VTABLE(u)->dump)
|
|
|
|
UNIT_VTABLE(u)->dump(u, f, prefix2);
|
2010-04-10 04:44:24 +02:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
} else if (u->load_state == UNIT_MERGED)
|
2010-04-10 04:44:24 +02:00
|
|
|
fprintf(f,
|
|
|
|
"%s\tMerged into: %s\n",
|
2012-01-15 12:04:08 +01:00
|
|
|
prefix, u->merged_into->id);
|
|
|
|
else if (u->load_state == UNIT_ERROR)
|
2019-07-03 16:56:17 +02:00
|
|
|
fprintf(f, "%s\tLoad Error Code: %s\n", prefix, strerror_safe(u->load_error));
|
2010-08-12 01:05:35 +02:00
|
|
|
|
2016-08-15 18:12:01 +02:00
|
|
|
for (n = sd_bus_track_first(u->bus_track); n; n = sd_bus_track_next(u->bus_track))
|
|
|
|
fprintf(f, "%s\tBus Ref: %s\n", prefix, n);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->job)
|
|
|
|
job_dump(u->job, f, prefix2);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
core: add NOP jobs, job type collapsing
Two of our current job types are special:
JOB_TRY_RESTART, JOB_RELOAD_OR_START.
They differ from other job types by being sensitive to the unit active state.
They perform some action when the unit is active and some other action
otherwise. This raises a question: when exactly should the unit state be
checked to make the decision?
Currently the unit state is checked when the job becomes runnable. It's more
sensible to check the state immediately when the job is added by the user.
When the user types "systemctl try-restart foo.service", he really intends
to restart the service if it's running right now. If it isn't running right
now, the restart is pointless.
Consider the example (from Bugzilla[1]):
sleep.service takes some time to start.
hello.service has After=sleep.service.
Both services get started. Two jobs will appear:
hello.service/start waiting
sleep.service/start running
Then someone runs "systemctl try-restart hello.service".
Currently the try-restart operation will block and wait for
sleep.service/start to complete.
The correct result is to complete the try-restart operation immediately
with success, because hello.service is not running. The two original
jobs must not be disturbed by this.
To fix this we introduce two new concepts:
- a new job type: JOB_NOP
A JOB_NOP job does not do anything to the unit. It does not pull in any
dependencies. It is always immediately runnable. When installed to a unit,
it sits in a special slot (u->nop_job) where it never conflicts with
the installed job (u->job) of a different type. It never merges with jobs
of other types, but it can merge into an already installed JOB_NOP job.
- "collapsing" of job types
When a job of one of the two special types is added, the state of the unit
is checked immediately and the job type changes:
JOB_TRY_RESTART -> JOB_RESTART or JOB_NOP
JOB_RELOAD_OR_START -> JOB_RELOAD or JOB_START
Should a job type JOB_RELOAD_OR_START appear later during job merging, it
collapses immediately afterwards.
Collapsing actually makes some things simpler, because there are now fewer
job types that are allowed in the transaction.
[1] Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=753586
2012-04-25 11:58:27 +02:00
|
|
|
if (u->nop_job)
|
|
|
|
job_dump(u->nop_job, f, prefix2);
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Common implementation for multiple backends */
|
2019-10-11 10:41:44 +02:00
|
|
|
int unit_load_fragment_and_dropin(Unit *u, bool fragment_required) {
|
2010-04-06 02:43:58 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
2013-10-22 01:54:10 +02:00
|
|
|
/* Load a .{service,socket,...} file */
|
2013-06-27 04:14:27 +02:00
|
|
|
r = unit_load_fragment(u);
|
|
|
|
if (r < 0)
|
2010-04-06 02:43:58 +02:00
|
|
|
return r;
|
|
|
|
|
2019-10-11 10:41:44 +02:00
|
|
|
if (u->load_state == UNIT_STUB) {
|
|
|
|
if (fragment_required)
|
|
|
|
return -ENOENT;
|
|
|
|
|
|
|
|
u->load_state = UNIT_LOADED;
|
|
|
|
}
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2017-07-22 14:39:49 +02:00
|
|
|
/* Load drop-in directory data. If u is an alias, we might be reloading the
|
|
|
|
* target unit needlessly. But we cannot be sure which drops-ins have already
|
|
|
|
* been loaded and which not, at least without doing complicated book-keeping,
|
|
|
|
* so let's always reread all drop-ins. */
|
2020-05-31 14:35:40 +02:00
|
|
|
r = unit_load_dropin(unit_follow_merge(u));
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
if (u->source_path) {
|
|
|
|
struct stat st;
|
|
|
|
|
|
|
|
if (stat(u->source_path, &st) >= 0)
|
|
|
|
u->source_mtime = timespec_load(&st.st_mtim);
|
|
|
|
else
|
|
|
|
u->source_mtime = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
2010-04-06 02:43:58 +02:00
|
|
|
}
|
|
|
|
|
2018-03-23 15:28:06 +01:00
|
|
|
void unit_add_to_target_deps_queue(Unit *u) {
|
|
|
|
Manager *m = u->manager;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (u->in_target_deps_queue)
|
|
|
|
return;
|
|
|
|
|
|
|
|
LIST_PREPEND(target_deps_queue, m->target_deps_queue, u);
|
|
|
|
u->in_target_deps_queue = true;
|
|
|
|
}
|
|
|
|
|
2010-09-14 01:51:30 +02:00
|
|
|
int unit_add_default_target_dependency(Unit *u, Unit *target) {
|
2010-09-13 12:06:49 +02:00
|
|
|
assert(u);
|
|
|
|
assert(target);
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (target->type != UNIT_TARGET)
|
2010-09-13 12:06:49 +02:00
|
|
|
return 0;
|
|
|
|
|
2011-02-21 15:32:17 +01:00
|
|
|
/* Only add the dependency if both units are loaded, so that
|
2010-09-14 01:51:30 +02:00
|
|
|
* that loop check below is reliable */
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->load_state != UNIT_LOADED ||
|
|
|
|
target->load_state != UNIT_LOADED)
|
2010-09-14 01:51:30 +02:00
|
|
|
return 0;
|
|
|
|
|
2011-03-08 03:24:42 +01:00
|
|
|
/* If either side wants no automatic dependencies, then let's
|
|
|
|
* skip this */
|
2012-01-15 12:04:08 +01:00
|
|
|
if (!u->default_dependencies ||
|
|
|
|
!target->default_dependencies)
|
2011-03-08 03:24:42 +01:00
|
|
|
return 0;
|
|
|
|
|
2010-09-13 12:06:49 +02:00
|
|
|
/* Don't create loops */
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
if (hashmap_get(target->dependencies[UNIT_BEFORE], u))
|
2010-09-13 12:06:49 +02:00
|
|
|
return 0;
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
return unit_add_dependency(target, UNIT_AFTER, u, true, UNIT_DEPENDENCY_DEFAULT);
|
2010-09-13 12:06:49 +02:00
|
|
|
}
|
|
|
|
|
2014-02-17 01:19:08 +01:00
|
|
|
static int unit_add_slice_dependencies(Unit *u) {
|
|
|
|
assert(u);
|
2010-10-28 23:18:47 +02:00
|
|
|
|
2015-08-28 17:14:59 +02:00
|
|
|
if (!UNIT_HAS_CGROUP_CONTEXT(u))
|
2014-02-17 01:19:08 +01:00
|
|
|
return 0;
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
/* Slice units are implicitly ordered against their parent slices (as this relationship is encoded in the
|
|
|
|
name), while all other units are ordered based on configuration (as in their case Slice= configures the
|
|
|
|
relationship). */
|
2020-05-28 15:25:22 +02:00
|
|
|
UnitDependencyMask mask = u->type == UNIT_SLICE ? UNIT_DEPENDENCY_IMPLICIT : UNIT_DEPENDENCY_FILE;
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
|
2014-02-17 01:19:08 +01:00
|
|
|
if (UNIT_ISSET(u->slice))
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
return unit_add_two_dependencies(u, UNIT_AFTER, UNIT_REQUIRES, UNIT_DEREF(u->slice), true, mask);
|
2014-02-17 01:19:08 +01:00
|
|
|
|
2015-09-29 13:06:28 +02:00
|
|
|
if (unit_has_name(u, SPECIAL_ROOT_SLICE))
|
2014-08-08 02:46:49 +02:00
|
|
|
return 0;
|
|
|
|
|
2018-09-15 20:02:00 +02:00
|
|
|
return unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_REQUIRES, SPECIAL_ROOT_SLICE, true, mask);
|
2010-09-13 12:06:49 +02:00
|
|
|
}
|
|
|
|
|
2014-02-17 01:19:08 +01:00
|
|
|
static int unit_add_mount_dependencies(Unit *u) {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
UnitDependencyInfo di;
|
|
|
|
const char *path;
|
|
|
|
Iterator i;
|
2013-11-08 18:11:09 +01:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(di.data, path, u->requires_mounts_for, i) {
|
|
|
|
char prefix[strlen(path) + 1];
|
2013-11-08 18:11:09 +01:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
PATH_FOREACH_PREFIX_MORE(prefix, path) {
|
2015-10-08 19:06:06 +02:00
|
|
|
_cleanup_free_ char *p = NULL;
|
2013-11-08 18:11:09 +01:00
|
|
|
Unit *m;
|
|
|
|
|
2015-10-08 19:06:06 +02:00
|
|
|
r = unit_name_from_path(prefix, ".mount", &p);
|
2013-11-08 18:11:09 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2015-10-08 19:06:06 +02:00
|
|
|
|
|
|
|
m = manager_get_unit(u->manager, p);
|
|
|
|
if (!m) {
|
|
|
|
/* Make sure to load the mount unit if
|
|
|
|
* it exists. If so the dependencies
|
|
|
|
* on this unit will be added later
|
|
|
|
* during the loading of the mount
|
|
|
|
* unit. */
|
|
|
|
(void) manager_load_unit_prepare(u->manager, p, NULL, NULL, &m);
|
2013-11-08 18:11:09 +01:00
|
|
|
continue;
|
2015-10-08 19:06:06 +02:00
|
|
|
}
|
2013-11-08 18:11:09 +01:00
|
|
|
if (m == u)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (m->load_state != UNIT_LOADED)
|
|
|
|
continue;
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
r = unit_add_dependency(u, UNIT_AFTER, m, true, di.origin_mask);
|
2013-11-08 18:11:09 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
if (m->fragment_path) {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
r = unit_add_dependency(u, UNIT_REQUIRES, m, true, di.origin_mask);
|
2013-11-08 18:11:09 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2014-05-15 17:09:34 +02:00
|
|
|
static int unit_add_startup_units(Unit *u) {
|
|
|
|
CGroupContext *c;
|
|
|
|
|
|
|
|
c = unit_get_cgroup_context(u);
|
2014-05-22 00:06:16 +02:00
|
|
|
if (!c)
|
|
|
|
return 0;
|
|
|
|
|
2015-09-11 16:48:24 +02:00
|
|
|
if (c->startup_cpu_shares == CGROUP_CPU_SHARES_INVALID &&
|
2016-05-05 22:42:55 +02:00
|
|
|
c->startup_io_weight == CGROUP_WEIGHT_INVALID &&
|
2015-09-11 16:48:24 +02:00
|
|
|
c->startup_blockio_weight == CGROUP_BLKIO_WEIGHT_INVALID)
|
2014-05-22 00:06:16 +02:00
|
|
|
return 0;
|
|
|
|
|
2020-06-05 15:12:29 +02:00
|
|
|
return set_ensure_put(&u->manager->startup_units, NULL, u);
|
2014-05-15 17:09:34 +02:00
|
|
|
}
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
int unit_load(Unit *u) {
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->in_load_queue) {
|
2013-10-14 06:10:14 +02:00
|
|
|
LIST_REMOVE(load_queue, u->manager->load_queue, u);
|
2012-01-15 12:04:08 +01:00
|
|
|
u->in_load_queue = false;
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->type == _UNIT_TYPE_INVALID)
|
2010-04-10 17:53:17 +02:00
|
|
|
return -EINVAL;
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->load_state != UNIT_STUB)
|
2010-01-26 21:39:06 +01:00
|
|
|
return 0;
|
|
|
|
|
2016-04-07 15:43:59 +02:00
|
|
|
if (u->transient_file) {
|
2018-11-22 22:25:27 +01:00
|
|
|
/* Finalize transient file: if this is a transient unit file, as soon as we reach unit_load() the setup
|
|
|
|
* is complete, hence let's synchronize the unit file we just wrote to disk. */
|
|
|
|
|
2016-04-07 15:43:59 +02:00
|
|
|
r = fflush_and_check(u->transient_file);
|
|
|
|
if (r < 0)
|
|
|
|
goto fail;
|
|
|
|
|
2017-11-28 21:24:20 +01:00
|
|
|
u->transient_file = safe_fclose(u->transient_file);
|
2016-05-06 18:22:22 +02:00
|
|
|
u->fragment_mtime = now(CLOCK_REALTIME);
|
2016-04-07 15:43:59 +02:00
|
|
|
}
|
|
|
|
|
2019-10-11 11:10:50 +02:00
|
|
|
r = UNIT_VTABLE(u)->load(u);
|
|
|
|
if (r < 0)
|
2010-04-06 02:43:58 +02:00
|
|
|
goto fail;
|
2019-10-11 11:10:50 +02:00
|
|
|
|
|
|
|
assert(u->load_state != UNIT_STUB);
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2012-04-29 14:26:07 +02:00
|
|
|
if (u->load_state == UNIT_LOADED) {
|
2018-03-23 15:28:06 +01:00
|
|
|
unit_add_to_target_deps_queue(u);
|
2014-02-17 01:19:08 +01:00
|
|
|
|
|
|
|
r = unit_add_slice_dependencies(u);
|
|
|
|
if (r < 0)
|
|
|
|
goto fail;
|
2013-06-28 04:12:58 +02:00
|
|
|
|
2014-02-17 01:19:08 +01:00
|
|
|
r = unit_add_mount_dependencies(u);
|
2012-04-29 14:26:07 +02:00
|
|
|
if (r < 0)
|
2013-06-28 04:12:58 +02:00
|
|
|
goto fail;
|
2012-04-29 14:26:07 +02:00
|
|
|
|
2014-05-15 17:09:34 +02:00
|
|
|
r = unit_add_startup_units(u);
|
|
|
|
if (r < 0)
|
|
|
|
goto fail;
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
if (u->on_failure_job_mode == JOB_ISOLATE && hashmap_size(u->dependencies[UNIT_ON_FAILURE]) > 1) {
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
log_unit_error(u, "More than one OnFailure= dependencies specified but OnFailureJobMode=isolate set. Refusing.");
|
2018-06-01 18:06:54 +02:00
|
|
|
r = -ENOEXEC;
|
2013-06-28 04:12:58 +02:00
|
|
|
goto fail;
|
|
|
|
}
|
2014-02-14 19:11:07 +01:00
|
|
|
|
2017-02-17 17:47:20 +01:00
|
|
|
if (u->job_running_timeout != USEC_INFINITY && u->job_running_timeout > u->job_timeout)
|
|
|
|
log_unit_warning(u, "JobRunningTimeoutSec= is greater than JobTimeoutSec=, it has no effect.");
|
|
|
|
|
cgroup: drastically simplify caching of cgroups members mask
Previously we tried to be smart: when a new unit appeared and it only
added controllers to the cgroup mask we'd update the cached members mask
in all parents by ORing in the controller flags in their cached values.
Unfortunately this was quite broken, as we missed some conditions when
this cache had to be reset (for example, when a unit got unloaded),
moreover the optimization doesn't work when a controller is removed
anyway (as in that case there's no other way for the parent to iterate
though all children if any other, remaining child unit still needs it).
Hence, let's simplify the logic substantially: instead of updating the
cache on the right events (which we didn't get right), let's simply
invalidate the cache, and generate it lazily when we encounter it later.
This should actually result in better behaviour as we don't have to
calculate the new members mask for a whole subtree whever we have the
suspicion something changed, but can delay it to the point where we
actually need the members mask.
This allows us to simplify things quite a bit, which is good, since
validating this cache for correctness is hard enough.
Fixes: #9512
2018-11-23 01:07:34 +01:00
|
|
|
/* We finished loading, let's ensure our parents recalculate the members mask */
|
|
|
|
unit_invalidate_cgroup_members_masks(u);
|
2011-04-07 18:47:35 +02:00
|
|
|
}
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
assert((u->load_state != UNIT_MERGED) == !u->merged_into);
|
2010-04-06 02:43:58 +02:00
|
|
|
|
|
|
|
unit_add_to_dbus_queue(unit_follow_merge(u));
|
2010-04-21 06:01:13 +02:00
|
|
|
unit_add_to_gc_queue(u);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
fail:
|
2018-06-01 17:46:01 +02:00
|
|
|
/* We convert ENOEXEC errors to the UNIT_BAD_SETTING load state here. Configuration parsing code should hence
|
|
|
|
* return ENOEXEC to ensure units are placed in this state after loading */
|
|
|
|
|
|
|
|
u->load_state = u->load_state == UNIT_STUB ? UNIT_NOT_FOUND :
|
|
|
|
r == -ENOEXEC ? UNIT_BAD_SETTING :
|
|
|
|
UNIT_ERROR;
|
2012-01-15 12:04:08 +01:00
|
|
|
u->load_error = r;
|
2018-06-01 17:46:01 +02:00
|
|
|
|
2020-06-16 19:46:55 +02:00
|
|
|
/* Record the last time we tried to load the unit, so that if the cache gets updated between now
|
|
|
|
* and the next time an attempt is made to load this unit, we know we need to check again */
|
|
|
|
if (u->load_state == UNIT_NOT_FOUND)
|
|
|
|
u->fragment_loadtime = now(CLOCK_REALTIME);
|
|
|
|
|
2010-02-05 00:38:41 +01:00
|
|
|
unit_add_to_dbus_queue(u);
|
2011-12-06 00:47:28 +01:00
|
|
|
unit_add_to_gc_queue(u);
|
2010-04-06 02:43:58 +02:00
|
|
|
|
2018-06-01 17:46:01 +02:00
|
|
|
return log_unit_debug_errno(u, r, "Failed to load configuration: %m");
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
2019-03-08 06:23:03 +01:00
|
|
|
_printf_(7, 8)
|
|
|
|
static int log_unit_internal(void *userdata, int level, int error, const char *file, int line, const char *func, const char *format, ...) {
|
|
|
|
Unit *u = userdata;
|
|
|
|
va_list ap;
|
|
|
|
int r;
|
2014-11-06 14:09:51 +01:00
|
|
|
|
2019-03-08 06:23:03 +01:00
|
|
|
va_start(ap, format);
|
|
|
|
if (u)
|
|
|
|
r = log_object_internalv(level, error, file, line, func,
|
|
|
|
u->manager->unit_log_field,
|
|
|
|
u->id,
|
|
|
|
u->manager->invocation_log_field,
|
|
|
|
u->invocation_id_string,
|
|
|
|
format, ap);
|
|
|
|
else
|
|
|
|
r = log_internalv(level, error, file, line, func, format, ap);
|
|
|
|
va_end(ap);
|
2014-11-06 14:09:51 +01:00
|
|
|
|
2019-03-08 06:23:03 +01:00
|
|
|
return r;
|
2014-11-06 14:09:51 +01:00
|
|
|
}
|
|
|
|
|
2019-03-18 12:21:27 +01:00
|
|
|
static bool unit_test_condition(Unit *u) {
|
2020-05-14 19:13:03 +02:00
|
|
|
_cleanup_strv_free_ char **env = NULL;
|
|
|
|
int r;
|
|
|
|
|
2011-03-09 23:58:17 +01:00
|
|
|
assert(u);
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
dual_timestamp_get(&u->condition_timestamp);
|
2011-03-09 23:58:17 +01:00
|
|
|
|
2020-05-14 19:13:03 +02:00
|
|
|
r = manager_get_effective_environment(u->manager, &env);
|
|
|
|
if (r < 0) {
|
|
|
|
log_unit_error_errno(u, r, "Failed to determine effective environment: %m");
|
|
|
|
u->condition_result = CONDITION_ERROR;
|
|
|
|
} else
|
|
|
|
u->condition_result = condition_test_list(
|
|
|
|
u->conditions,
|
|
|
|
env,
|
|
|
|
condition_type_to_string,
|
|
|
|
log_unit_internal,
|
|
|
|
u);
|
2018-11-29 16:34:59 +01:00
|
|
|
|
2020-05-14 19:13:03 +02:00
|
|
|
unit_add_to_dbus_queue(u);
|
2012-01-15 12:04:08 +01:00
|
|
|
return u->condition_result;
|
2011-03-09 23:58:17 +01:00
|
|
|
}
|
|
|
|
|
2019-03-18 12:21:27 +01:00
|
|
|
static bool unit_test_assert(Unit *u) {
|
2020-05-14 19:13:03 +02:00
|
|
|
_cleanup_strv_free_ char **env = NULL;
|
|
|
|
int r;
|
|
|
|
|
2014-11-06 13:43:45 +01:00
|
|
|
assert(u);
|
|
|
|
|
|
|
|
dual_timestamp_get(&u->assert_timestamp);
|
|
|
|
|
2020-05-14 19:13:03 +02:00
|
|
|
r = manager_get_effective_environment(u->manager, &env);
|
|
|
|
if (r < 0) {
|
|
|
|
log_unit_error_errno(u, r, "Failed to determine effective environment: %m");
|
|
|
|
u->assert_result = CONDITION_ERROR;
|
|
|
|
} else
|
|
|
|
u->assert_result = condition_test_list(
|
|
|
|
u->asserts,
|
|
|
|
env,
|
|
|
|
assert_type_to_string,
|
|
|
|
log_unit_internal,
|
|
|
|
u);
|
2018-11-29 16:34:59 +01:00
|
|
|
|
2020-05-14 19:13:03 +02:00
|
|
|
unit_add_to_dbus_queue(u);
|
2014-11-06 13:43:45 +01:00
|
|
|
return u->assert_result;
|
|
|
|
}
|
|
|
|
|
2020-02-29 17:19:46 +01:00
|
|
|
void unit_status_printf(Unit *u, StatusType status_type, const char *status, const char *unit_status_msg_format) {
|
2018-11-23 17:46:25 +01:00
|
|
|
const char *d;
|
|
|
|
|
2019-06-06 17:33:59 +02:00
|
|
|
d = unit_status_string(u);
|
2018-11-23 17:46:25 +01:00
|
|
|
if (log_get_show_color())
|
|
|
|
d = strjoina(ANSI_HIGHLIGHT, d, ANSI_NORMAL);
|
|
|
|
|
2015-11-17 17:11:44 +01:00
|
|
|
DISABLE_WARNING_FORMAT_NONLITERAL;
|
2020-02-29 17:19:46 +01:00
|
|
|
manager_status_printf(u->manager, status_type, status, unit_status_msg_format, d);
|
2015-11-17 17:11:44 +01:00
|
|
|
REENABLE_WARNING;
|
|
|
|
}
|
|
|
|
|
2019-03-18 12:21:27 +01:00
|
|
|
int unit_test_start_limit(Unit *u) {
|
2018-11-08 13:15:25 +01:00
|
|
|
const char *reason;
|
|
|
|
|
core: make the StartLimitXYZ= settings generic and apply to any kind of unit, not just services
This moves the StartLimitBurst=, StartLimitInterval=, StartLimitAction=, RebootArgument= from the [Service] section
into the [Unit] section of unit files, and thus support it in all unit types, not just in services.
This way we can enforce the start limit much earlier, in particular before testing the unit conditions, so that
repeated start-up failure due to failed conditions is also considered for the start limit logic.
For compatibility the four options may also be configured in the [Service] section still, but we only document them in
their new section [Unit].
This also renamed the socket unit failure code "service-failed-permanent" into "service-start-limit-hit" to express
more clearly what it is about, after all it's only triggered through the start limit being hit.
Finally, the code in busname_trigger_notify() and socket_trigger_notify() is altered to become more alike.
Fixes: #2467
2016-02-09 18:38:03 +01:00
|
|
|
assert(u);
|
|
|
|
|
2019-09-19 17:45:41 +02:00
|
|
|
if (ratelimit_below(&u->start_ratelimit)) {
|
core: make the StartLimitXYZ= settings generic and apply to any kind of unit, not just services
This moves the StartLimitBurst=, StartLimitInterval=, StartLimitAction=, RebootArgument= from the [Service] section
into the [Unit] section of unit files, and thus support it in all unit types, not just in services.
This way we can enforce the start limit much earlier, in particular before testing the unit conditions, so that
repeated start-up failure due to failed conditions is also considered for the start limit logic.
For compatibility the four options may also be configured in the [Service] section still, but we only document them in
their new section [Unit].
This also renamed the socket unit failure code "service-failed-permanent" into "service-start-limit-hit" to express
more clearly what it is about, after all it's only triggered through the start limit being hit.
Finally, the code in busname_trigger_notify() and socket_trigger_notify() is altered to become more alike.
Fixes: #2467
2016-02-09 18:38:03 +01:00
|
|
|
u->start_limit_hit = false;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
log_unit_warning(u, "Start request repeated too quickly.");
|
|
|
|
u->start_limit_hit = true;
|
|
|
|
|
2018-11-08 13:15:25 +01:00
|
|
|
reason = strjoina("unit ", u->id, " failed");
|
|
|
|
|
2019-03-18 13:20:54 +01:00
|
|
|
emergency_action(u->manager, u->start_limit_action,
|
|
|
|
EMERGENCY_ACTION_IS_WATCHDOG|EMERGENCY_ACTION_WARN,
|
|
|
|
u->reboot_arg, -1, reason);
|
|
|
|
|
|
|
|
return -ECANCELED;
|
core: make the StartLimitXYZ= settings generic and apply to any kind of unit, not just services
This moves the StartLimitBurst=, StartLimitInterval=, StartLimitAction=, RebootArgument= from the [Service] section
into the [Unit] section of unit files, and thus support it in all unit types, not just in services.
This way we can enforce the start limit much earlier, in particular before testing the unit conditions, so that
repeated start-up failure due to failed conditions is also considered for the start limit logic.
For compatibility the four options may also be configured in the [Service] section still, but we only document them in
their new section [Unit].
This also renamed the socket unit failure code "service-failed-permanent" into "service-start-limit-hit" to express
more clearly what it is about, after all it's only triggered through the start limit being hit.
Finally, the code in busname_trigger_notify() and socket_trigger_notify() is altered to become more alike.
Fixes: #2467
2016-02-09 18:38:03 +01:00
|
|
|
}
|
|
|
|
|
2016-11-14 17:37:40 +01:00
|
|
|
bool unit_shall_confirm_spawn(Unit *u) {
|
2016-11-24 18:47:48 +01:00
|
|
|
assert(u);
|
2016-11-14 17:37:40 +01:00
|
|
|
|
|
|
|
if (manager_is_confirm_spawn_disabled(u->manager))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/* For some reasons units remaining in the same process group
|
|
|
|
* as PID 1 fail to acquire the console even if it's not used
|
|
|
|
* by any process. So skip the confirmation question for them. */
|
|
|
|
return !unit_get_exec_context(u)->same_pgrp;
|
|
|
|
}
|
|
|
|
|
2016-11-24 18:47:48 +01:00
|
|
|
static bool unit_verify_deps(Unit *u) {
|
|
|
|
Unit *other;
|
|
|
|
Iterator j;
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
void *v;
|
2016-11-24 18:47:48 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Checks whether all BindsTo= dependencies of this unit are fulfilled — if they are also combined with
|
|
|
|
* After=. We do not check Requires= or Requisite= here as they only should have an effect on the job
|
|
|
|
* processing, but do not have any effect afterwards. We don't check BindsTo= dependencies that are not used in
|
|
|
|
* conjunction with After= as for them any such check would make things entirely racy. */
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BINDS_TO], j) {
|
2016-11-24 18:47:48 +01:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
if (!hashmap_contains(u->dependencies[UNIT_AFTER], other))
|
2016-11-24 18:47:48 +01:00
|
|
|
continue;
|
|
|
|
|
|
|
|
if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(other))) {
|
|
|
|
log_unit_notice(u, "Bound to unit %s, but unit isn't active.", other->id);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2019-03-18 13:28:59 +01:00
|
|
|
/* Errors that aren't really errors:
|
core: make the StartLimitXYZ= settings generic and apply to any kind of unit, not just services
This moves the StartLimitBurst=, StartLimitInterval=, StartLimitAction=, RebootArgument= from the [Service] section
into the [Unit] section of unit files, and thus support it in all unit types, not just in services.
This way we can enforce the start limit much earlier, in particular before testing the unit conditions, so that
repeated start-up failure due to failed conditions is also considered for the start limit logic.
For compatibility the four options may also be configured in the [Service] section still, but we only document them in
their new section [Unit].
This also renamed the socket unit failure code "service-failed-permanent" into "service-start-limit-hit" to express
more clearly what it is about, after all it's only triggered through the start limit being hit.
Finally, the code in busname_trigger_notify() and socket_trigger_notify() is altered to become more alike.
Fixes: #2467
2016-02-09 18:38:03 +01:00
|
|
|
* -EALREADY: Unit is already started.
|
2019-03-18 13:28:59 +01:00
|
|
|
* -ECOMM: Condition failed
|
core: make the StartLimitXYZ= settings generic and apply to any kind of unit, not just services
This moves the StartLimitBurst=, StartLimitInterval=, StartLimitAction=, RebootArgument= from the [Service] section
into the [Unit] section of unit files, and thus support it in all unit types, not just in services.
This way we can enforce the start limit much earlier, in particular before testing the unit conditions, so that
repeated start-up failure due to failed conditions is also considered for the start limit logic.
For compatibility the four options may also be configured in the [Service] section still, but we only document them in
their new section [Unit].
This also renamed the socket unit failure code "service-failed-permanent" into "service-start-limit-hit" to express
more clearly what it is about, after all it's only triggered through the start limit being hit.
Finally, the code in busname_trigger_notify() and socket_trigger_notify() is altered to become more alike.
Fixes: #2467
2016-02-09 18:38:03 +01:00
|
|
|
* -EAGAIN: An operation is already in progress. Retry later.
|
2019-03-18 13:28:59 +01:00
|
|
|
*
|
|
|
|
* Errors that are real errors:
|
|
|
|
* -EBADR: This unit type does not support starting.
|
2019-03-18 13:14:19 +01:00
|
|
|
* -ECANCELED: Start limit hit, too many requests for now
|
core: make the StartLimitXYZ= settings generic and apply to any kind of unit, not just services
This moves the StartLimitBurst=, StartLimitInterval=, StartLimitAction=, RebootArgument= from the [Service] section
into the [Unit] section of unit files, and thus support it in all unit types, not just in services.
This way we can enforce the start limit much earlier, in particular before testing the unit conditions, so that
repeated start-up failure due to failed conditions is also considered for the start limit logic.
For compatibility the four options may also be configured in the [Service] section still, but we only document them in
their new section [Unit].
This also renamed the socket unit failure code "service-failed-permanent" into "service-start-limit-hit" to express
more clearly what it is about, after all it's only triggered through the start limit being hit.
Finally, the code in busname_trigger_notify() and socket_trigger_notify() is altered to become more alike.
Fixes: #2467
2016-02-09 18:38:03 +01:00
|
|
|
* -EPROTO: Assert failed
|
|
|
|
* -EINVAL: Unit not loaded
|
|
|
|
* -EOPNOTSUPP: Unit type not supported
|
2016-11-24 18:47:48 +01:00
|
|
|
* -ENOLINK: The necessary dependencies are not fulfilled.
|
2018-04-27 20:35:10 +02:00
|
|
|
* -ESTALE: This unit has been started before and can't be started a second time
|
2019-03-18 12:29:08 +01:00
|
|
|
* -ENOENT: This is a triggering unit and unit to trigger is not loaded
|
2010-01-26 21:39:06 +01:00
|
|
|
*/
|
|
|
|
int unit_start(Unit *u) {
|
|
|
|
UnitActiveState state;
|
2010-11-14 23:26:53 +01:00
|
|
|
Unit *following;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
2019-03-18 12:36:03 +01:00
|
|
|
/* If this is already started, then this will succeed. Note that this will even succeed if this unit
|
|
|
|
* is not startable by the user. This is relied on to detect when we need to wait for units and when
|
|
|
|
* waiting is finished. */
|
2010-01-26 21:39:06 +01:00
|
|
|
state = unit_active_state(u);
|
|
|
|
if (UNIT_IS_ACTIVE_OR_RELOADING(state))
|
|
|
|
return -EALREADY;
|
2019-06-25 11:31:28 +02:00
|
|
|
if (state == UNIT_MAINTENANCE)
|
|
|
|
return -EAGAIN;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
core: make the StartLimitXYZ= settings generic and apply to any kind of unit, not just services
This moves the StartLimitBurst=, StartLimitInterval=, StartLimitAction=, RebootArgument= from the [Service] section
into the [Unit] section of unit files, and thus support it in all unit types, not just in services.
This way we can enforce the start limit much earlier, in particular before testing the unit conditions, so that
repeated start-up failure due to failed conditions is also considered for the start limit logic.
For compatibility the four options may also be configured in the [Service] section still, but we only document them in
their new section [Unit].
This also renamed the socket unit failure code "service-failed-permanent" into "service-start-limit-hit" to express
more clearly what it is about, after all it's only triggered through the start limit being hit.
Finally, the code in busname_trigger_notify() and socket_trigger_notify() is altered to become more alike.
Fixes: #2467
2016-02-09 18:38:03 +01:00
|
|
|
/* Units that aren't loaded cannot be started */
|
|
|
|
if (u->load_state != UNIT_LOADED)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2018-04-27 20:35:10 +02:00
|
|
|
/* Refuse starting scope units more than once */
|
|
|
|
if (UNIT_VTABLE(u)->once_only && dual_timestamp_is_set(&u->inactive_enter_timestamp))
|
|
|
|
return -ESTALE;
|
|
|
|
|
2019-03-18 12:36:03 +01:00
|
|
|
/* If the conditions failed, don't do anything at all. If we already are activating this call might
|
|
|
|
* still be useful to speed up activation in case there is some hold-off time, but we don't want to
|
|
|
|
* recheck the condition in that case. */
|
2011-09-22 21:32:18 +02:00
|
|
|
if (state != UNIT_ACTIVATING &&
|
2019-08-30 18:21:05 +02:00
|
|
|
!unit_test_condition(u))
|
2019-03-18 12:36:03 +01:00
|
|
|
return log_unit_debug_errno(u, SYNTHETIC_ERRNO(ECOMM), "Starting requested but condition failed. Not starting unit.");
|
2010-10-13 02:15:41 +02:00
|
|
|
|
2014-11-06 13:43:45 +01:00
|
|
|
/* If the asserts failed, fail the entire job */
|
|
|
|
if (state != UNIT_ACTIVATING &&
|
2019-03-18 12:36:03 +01:00
|
|
|
!unit_test_assert(u))
|
|
|
|
return log_unit_notice_errno(u, SYNTHETIC_ERRNO(EPROTO), "Starting requested but asserts failed.");
|
2014-11-06 13:43:45 +01:00
|
|
|
|
2019-03-18 12:36:03 +01:00
|
|
|
/* Units of types that aren't supported cannot be started. Note that we do this test only after the
|
|
|
|
* condition checks, so that we rather return condition check errors (which are usually not
|
|
|
|
* considered a true failure) than "not supported" errors (which are considered a failure).
|
2015-09-07 14:08:24 +02:00
|
|
|
*/
|
2019-07-08 17:43:14 +02:00
|
|
|
if (!unit_type_supported(u->type))
|
2015-09-07 14:08:24 +02:00
|
|
|
return -EOPNOTSUPP;
|
|
|
|
|
2019-03-18 12:36:03 +01:00
|
|
|
/* Let's make sure that the deps really are in order before we start this. Normally the job engine
|
|
|
|
* should have taken care of this already, but let's check this here again. After all, our
|
|
|
|
* dependencies might not be in effect anymore, due to a reload or due to a failed condition. */
|
2016-11-24 18:47:48 +01:00
|
|
|
if (!unit_verify_deps(u))
|
|
|
|
return -ENOLINK;
|
|
|
|
|
2010-11-14 23:26:53 +01:00
|
|
|
/* Forward to the main object, if we aren't it. */
|
systemd,systemctl: export condition status and show failing condition
$ systemctl --user status hoohoo
hoohoo.service
Loaded: loaded (/home/zbyszek/.config/systemd/user/hoohoo.service; static)
Active: inactive (dead)
start condition failed at Tue 2013-06-25 18:08:42 EDT; 1s ago
ConditionPathExists=/tmp/hoo was not met
Full information is exported over D-Bus:
[(condition, trigger, negate, param, state),...]
where state is one of "failed" (<0), "untested" (0), "OK" (>0).
I've decided to use 0 for "untested", because it might be useful to
differentiate different types of failure later on, without breaking
compatibility.
systemctl shows the failing condition, if there was a non-trigger
failing condition, or says "none of the trigger conditions were met",
because there're often many trigger conditions, and they must all
fail for the condition to fail, so printing them all would consume
a lot of space, and bring unnecessary attention to something that is
quite low-level.
2013-06-25 22:09:07 +02:00
|
|
|
following = unit_following(u);
|
|
|
|
if (following) {
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
log_unit_debug(u, "Redirecting start request from %s to %s.", u->id, following->id);
|
2010-11-14 23:26:53 +01:00
|
|
|
return unit_start(following);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* If it is stopped, but we cannot start it, then fail */
|
|
|
|
if (!UNIT_VTABLE(u)->start)
|
|
|
|
return -EBADR;
|
|
|
|
|
2019-03-18 12:36:03 +01:00
|
|
|
/* We don't suppress calls to ->start() here when we are already starting, to allow this request to
|
|
|
|
* be used as a "hurry up" call, for example when the unit is in some "auto restart" state where it
|
|
|
|
* waits for a holdoff timer to elapse before it will start again. */
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2010-02-05 00:38:41 +01:00
|
|
|
unit_add_to_dbus_queue(u);
|
2020-04-29 17:53:43 +02:00
|
|
|
unit_cgroup_freezer_action(u, FREEZER_THAW);
|
2010-07-07 00:00:59 +02:00
|
|
|
|
2015-07-16 20:08:30 +02:00
|
|
|
return UNIT_VTABLE(u)->start(u);
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
bool unit_can_start(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
2015-09-05 20:21:46 +02:00
|
|
|
if (u->load_state != UNIT_LOADED)
|
|
|
|
return false;
|
|
|
|
|
2019-07-08 17:43:14 +02:00
|
|
|
if (!unit_type_supported(u->type))
|
2015-09-05 20:21:46 +02:00
|
|
|
return false;
|
|
|
|
|
2018-04-27 20:35:10 +02:00
|
|
|
/* Scope units may be started only once */
|
|
|
|
if (UNIT_VTABLE(u)->once_only && dual_timestamp_is_set(&u->inactive_exit_timestamp))
|
|
|
|
return false;
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
return !!UNIT_VTABLE(u)->start;
|
|
|
|
}
|
|
|
|
|
2010-08-30 22:45:46 +02:00
|
|
|
bool unit_can_isolate(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
return unit_can_start(u) &&
|
2012-01-15 12:04:08 +01:00
|
|
|
u->allow_isolate;
|
2010-08-30 22:45:46 +02:00
|
|
|
}
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
/* Errors:
|
|
|
|
* -EBADR: This unit type does not support stopping.
|
|
|
|
* -EALREADY: Unit is already stopped.
|
|
|
|
* -EAGAIN: An operation is already in progress. Retry later.
|
|
|
|
*/
|
|
|
|
int unit_stop(Unit *u) {
|
|
|
|
UnitActiveState state;
|
2010-11-14 23:26:53 +01:00
|
|
|
Unit *following;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
state = unit_active_state(u);
|
2010-08-31 00:23:34 +02:00
|
|
|
if (UNIT_IS_INACTIVE_OR_FAILED(state))
|
2010-01-26 21:39:06 +01:00
|
|
|
return -EALREADY;
|
|
|
|
|
2014-12-12 21:05:32 +01:00
|
|
|
following = unit_following(u);
|
|
|
|
if (following) {
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
log_unit_debug(u, "Redirecting stop request from %s to %s.", u->id, following->id);
|
2010-11-14 23:26:53 +01:00
|
|
|
return unit_stop(following);
|
|
|
|
}
|
|
|
|
|
2010-02-12 02:40:28 +01:00
|
|
|
if (!UNIT_VTABLE(u)->stop)
|
|
|
|
return -EBADR;
|
|
|
|
|
2010-02-05 00:38:41 +01:00
|
|
|
unit_add_to_dbus_queue(u);
|
2020-04-29 17:53:43 +02:00
|
|
|
unit_cgroup_freezer_action(u, FREEZER_THAW);
|
2010-07-07 00:00:59 +02:00
|
|
|
|
2015-07-16 20:08:30 +02:00
|
|
|
return UNIT_VTABLE(u)->stop(u);
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
2016-10-24 21:41:54 +02:00
|
|
|
bool unit_can_stop(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
2019-07-08 17:43:14 +02:00
|
|
|
if (!unit_type_supported(u->type))
|
2016-10-24 21:41:54 +02:00
|
|
|
return false;
|
|
|
|
|
|
|
|
if (u->perpetual)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
return !!UNIT_VTABLE(u)->stop;
|
|
|
|
}
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
/* Errors:
|
|
|
|
* -EBADR: This unit type does not support reloading.
|
|
|
|
* -ENOEXEC: Unit is not started.
|
|
|
|
* -EAGAIN: An operation is already in progress. Retry later.
|
|
|
|
*/
|
|
|
|
int unit_reload(Unit *u) {
|
|
|
|
UnitActiveState state;
|
2010-11-14 23:26:53 +01:00
|
|
|
Unit *following;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->load_state != UNIT_LOADED)
|
2010-07-01 00:31:53 +02:00
|
|
|
return -EINVAL;
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
if (!unit_can_reload(u))
|
|
|
|
return -EBADR;
|
|
|
|
|
|
|
|
state = unit_active_state(u);
|
2010-08-11 22:04:22 +02:00
|
|
|
if (state == UNIT_RELOADING)
|
2019-01-20 23:12:24 +01:00
|
|
|
return -EAGAIN;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2013-12-03 03:52:51 +01:00
|
|
|
if (state != UNIT_ACTIVE) {
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
log_unit_warning(u, "Unit cannot be reloaded because it is inactive.");
|
2010-01-26 21:39:06 +01:00
|
|
|
return -ENOEXEC;
|
2013-12-03 03:52:51 +01:00
|
|
|
}
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2013-10-22 01:54:10 +02:00
|
|
|
following = unit_following(u);
|
|
|
|
if (following) {
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
log_unit_debug(u, "Redirecting reload request from %s to %s.", u->id, following->id);
|
2010-11-14 23:26:53 +01:00
|
|
|
return unit_reload(following);
|
|
|
|
}
|
|
|
|
|
2010-02-05 00:38:41 +01:00
|
|
|
unit_add_to_dbus_queue(u);
|
2015-01-28 15:07:13 +01:00
|
|
|
|
2017-07-22 17:30:57 +02:00
|
|
|
if (!UNIT_VTABLE(u)->reload) {
|
|
|
|
/* Unit doesn't have a reload function, but we need to propagate the reload anyway */
|
2018-06-01 19:06:19 +02:00
|
|
|
unit_notify(u, unit_active_state(u), unit_active_state(u), 0);
|
2017-07-22 17:30:57 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2020-04-29 17:53:43 +02:00
|
|
|
unit_cgroup_freezer_action(u, FREEZER_THAW);
|
|
|
|
|
2015-07-16 20:08:30 +02:00
|
|
|
return UNIT_VTABLE(u)->reload(u);
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
bool unit_can_reload(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
2017-07-22 17:30:57 +02:00
|
|
|
if (UNIT_VTABLE(u)->can_reload)
|
|
|
|
return UNIT_VTABLE(u)->can_reload(u);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
if (!hashmap_isempty(u->dependencies[UNIT_PROPAGATES_RELOAD_TO]))
|
2010-01-26 21:39:06 +01:00
|
|
|
return true;
|
|
|
|
|
2017-07-22 17:30:57 +02:00
|
|
|
return UNIT_VTABLE(u)->reload;
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
2018-08-09 16:26:27 +02:00
|
|
|
bool unit_is_unneeded(Unit *u) {
|
|
|
|
static const UnitDependency deps[] = {
|
2015-05-19 01:24:28 +02:00
|
|
|
UNIT_REQUIRED_BY,
|
2015-06-23 11:13:13 +02:00
|
|
|
UNIT_REQUISITE_OF,
|
2015-05-19 01:24:28 +02:00
|
|
|
UNIT_WANTED_BY,
|
|
|
|
UNIT_BOUND_BY,
|
|
|
|
};
|
2018-08-09 16:26:27 +02:00
|
|
|
size_t j;
|
2010-01-29 20:47:09 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (!u->stop_when_unneeded)
|
2018-08-09 16:26:27 +02:00
|
|
|
return false;
|
2010-01-29 20:47:09 +01:00
|
|
|
|
2018-08-09 16:26:27 +02:00
|
|
|
/* Don't clean up while the unit is transitioning or is even inactive. */
|
|
|
|
if (!UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
|
|
|
|
return false;
|
|
|
|
if (u->job)
|
|
|
|
return false;
|
2010-01-29 20:47:09 +01:00
|
|
|
|
2018-08-09 16:26:27 +02:00
|
|
|
for (j = 0; j < ELEMENTSOF(deps); j++) {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
Unit *other;
|
|
|
|
Iterator i;
|
|
|
|
void *v;
|
|
|
|
|
2018-08-20 10:43:31 +02:00
|
|
|
/* If a dependent unit has a job queued, is active or transitioning, or is marked for
|
2018-08-09 16:26:27 +02:00
|
|
|
* restart, then don't clean this one up. */
|
2010-10-28 23:18:47 +02:00
|
|
|
|
2018-08-09 16:26:27 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[deps[j]], i) {
|
2018-10-09 22:23:14 +02:00
|
|
|
if (other->job)
|
2018-08-09 16:26:27 +02:00
|
|
|
return false;
|
|
|
|
|
|
|
|
if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
if (unit_will_restart(other))
|
|
|
|
return false;
|
|
|
|
}
|
2015-05-19 16:00:24 +02:00
|
|
|
}
|
|
|
|
|
2018-08-09 16:26:27 +02:00
|
|
|
return true;
|
|
|
|
}
|
2010-01-29 20:47:09 +01:00
|
|
|
|
2018-08-09 16:26:27 +02:00
|
|
|
static void check_unneeded_dependencies(Unit *u) {
|
|
|
|
|
|
|
|
static const UnitDependency deps[] = {
|
|
|
|
UNIT_REQUIRES,
|
|
|
|
UNIT_REQUISITE,
|
|
|
|
UNIT_WANTS,
|
|
|
|
UNIT_BINDS_TO,
|
|
|
|
};
|
|
|
|
size_t j;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Add all units this unit depends on to the queue that processes StopWhenUnneeded= behaviour. */
|
|
|
|
|
|
|
|
for (j = 0; j < ELEMENTSOF(deps); j++) {
|
|
|
|
Unit *other;
|
|
|
|
Iterator i;
|
|
|
|
void *v;
|
|
|
|
|
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[deps[j]], i)
|
2018-08-20 10:43:31 +02:00
|
|
|
unit_submit_to_stop_when_unneeded_queue(other);
|
2018-08-09 16:26:27 +02:00
|
|
|
}
|
2010-01-29 20:47:09 +01:00
|
|
|
}
|
|
|
|
|
2014-08-18 22:21:42 +02:00
|
|
|
static void unit_check_binds_to(Unit *u) {
|
tree-wide: expose "p"-suffix unref calls in public APIs to make gcc cleanup easy
GLIB has recently started to officially support the gcc cleanup
attribute in its public API, hence let's do the same for our APIs.
With this patch we'll define an xyz_unrefp() call for each public
xyz_unref() call, to make it easy to use inside a
__attribute__((cleanup())) expression. Then, all code is ported over to
make use of this.
The new calls are also documented in the man pages, with examples how to
use them (well, I only added docs where the _unref() call itself already
had docs, and the examples, only cover sd_bus_unrefp() and
sd_event_unrefp()).
This also renames sd_lldp_free() to sd_lldp_unref(), since that's how we
tend to call our destructors these days.
Note that this defines no public macro that wraps gcc's attribute and
makes it easier to use. While I think it's our duty in the library to
make our stuff easy to use, I figure it's not our duty to make gcc's own
features easy to use on its own. Most likely, client code which wants to
make use of this should define its own:
#define _cleanup_(function) __attribute__((cleanup(function)))
Or similar, to make the gcc feature easier to use.
Making this logic public has the benefit that we can remove three header
files whose only purpose was to define these functions internally.
See #2008.
2015-11-27 19:13:45 +01:00
|
|
|
_cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
|
2014-08-18 22:21:42 +02:00
|
|
|
bool stop = false;
|
|
|
|
Unit *other;
|
|
|
|
Iterator i;
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
void *v;
|
2015-05-19 16:23:14 +02:00
|
|
|
int r;
|
2014-08-18 22:21:42 +02:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (u->job)
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (unit_active_state(u) != UNIT_ACTIVE)
|
|
|
|
return;
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BINDS_TO], i) {
|
2014-08-18 22:21:42 +02:00
|
|
|
if (other->job)
|
|
|
|
continue;
|
2017-07-11 10:45:03 +02:00
|
|
|
|
|
|
|
if (!other->coldplugged)
|
|
|
|
/* We might yet create a job for the other unit… */
|
|
|
|
continue;
|
2014-08-18 22:21:42 +02:00
|
|
|
|
|
|
|
if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
stop = true;
|
2015-02-17 19:47:34 +01:00
|
|
|
break;
|
2014-08-18 22:21:42 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (!stop)
|
|
|
|
return;
|
|
|
|
|
2016-07-12 12:52:11 +02:00
|
|
|
/* If stopping a unit fails continuously we might enter a stop
|
2015-05-19 16:23:14 +02:00
|
|
|
* loop here, hence stop acting on the service being
|
|
|
|
* unnecessary after a while. */
|
2018-05-11 11:16:52 +02:00
|
|
|
if (!ratelimit_below(&u->auto_stop_ratelimit)) {
|
2015-05-19 16:23:14 +02:00
|
|
|
log_unit_warning(u, "Unit is bound to inactive unit %s, but not stopping since we tried this too often recently.", other->id);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2015-02-17 19:47:34 +01:00
|
|
|
assert(other);
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
log_unit_info(u, "Unit is bound to inactive unit %s. Stopping, too.", other->id);
|
2014-08-18 22:21:42 +02:00
|
|
|
|
|
|
|
/* A unit we need to run is gone. Sniff. Let's stop this. */
|
2019-03-22 20:57:30 +01:00
|
|
|
r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, NULL, &error, NULL);
|
2015-05-19 16:23:14 +02:00
|
|
|
if (r < 0)
|
2015-11-12 19:52:31 +01:00
|
|
|
log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
|
2014-08-18 22:21:42 +02:00
|
|
|
}
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
static void retroactively_start_dependencies(Unit *u) {
|
|
|
|
Iterator i;
|
|
|
|
Unit *other;
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
void *v;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)));
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_REQUIRES], i)
|
|
|
|
if (!hashmap_get(u->dependencies[UNIT_AFTER], other) &&
|
2010-10-28 23:18:47 +02:00
|
|
|
!UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
|
2019-03-22 20:57:30 +01:00
|
|
|
manager_add_job(u->manager, JOB_START, other, JOB_REPLACE, NULL, NULL, NULL);
|
2010-10-28 23:18:47 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BINDS_TO], i)
|
|
|
|
if (!hashmap_get(u->dependencies[UNIT_AFTER], other) &&
|
2010-10-28 23:18:47 +02:00
|
|
|
!UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
|
2019-03-22 20:57:30 +01:00
|
|
|
manager_add_job(u->manager, JOB_START, other, JOB_REPLACE, NULL, NULL, NULL);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_WANTS], i)
|
|
|
|
if (!hashmap_get(u->dependencies[UNIT_AFTER], other) &&
|
2010-10-28 23:18:47 +02:00
|
|
|
!UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
|
2019-03-22 20:57:30 +01:00
|
|
|
manager_add_job(u->manager, JOB_START, other, JOB_FAIL, NULL, NULL, NULL);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_CONFLICTS], i)
|
2010-10-28 23:18:47 +02:00
|
|
|
if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
|
2019-03-22 20:57:30 +01:00
|
|
|
manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL, NULL);
|
2010-08-09 22:32:30 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_CONFLICTED_BY], i)
|
2010-10-28 23:18:47 +02:00
|
|
|
if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
|
2019-03-22 20:57:30 +01:00
|
|
|
manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL, NULL);
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
static void retroactively_stop_dependencies(Unit *u) {
|
|
|
|
Unit *other;
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
Iterator i;
|
|
|
|
void *v;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)));
|
|
|
|
|
2010-10-28 23:18:47 +02:00
|
|
|
/* Pull down units which are bound to us recursively if enabled */
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BOUND_BY], i)
|
2010-10-28 23:18:47 +02:00
|
|
|
if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
|
2019-03-22 20:57:30 +01:00
|
|
|
manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL, NULL);
|
2011-12-09 15:25:29 +01:00
|
|
|
}
|
|
|
|
|
2013-04-23 20:53:16 +02:00
|
|
|
void unit_start_on_failure(Unit *u) {
|
2011-02-24 03:24:23 +01:00
|
|
|
Unit *other;
|
|
|
|
Iterator i;
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
void *v;
|
2018-06-01 19:04:37 +02:00
|
|
|
int r;
|
2011-02-24 03:24:23 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
if (hashmap_size(u->dependencies[UNIT_ON_FAILURE]) <= 0)
|
2011-04-07 04:11:31 +02:00
|
|
|
return;
|
|
|
|
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
log_unit_info(u, "Triggering OnFailure= dependencies.");
|
2011-04-07 04:11:31 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_ON_FAILURE], i) {
|
2018-06-01 19:04:37 +02:00
|
|
|
_cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
|
2011-04-07 04:11:31 +02:00
|
|
|
|
2019-03-22 20:57:30 +01:00
|
|
|
r = manager_add_job(u->manager, JOB_START, other, u->on_failure_job_mode, NULL, &error, NULL);
|
2013-04-15 07:40:44 +02:00
|
|
|
if (r < 0)
|
2018-06-01 19:04:37 +02:00
|
|
|
log_unit_warning_errno(u, r, "Failed to enqueue OnFailure= job, ignoring: %s", bus_error_message(&error, r));
|
2011-04-07 04:11:31 +02:00
|
|
|
}
|
2011-02-24 03:24:23 +01:00
|
|
|
}
|
|
|
|
|
2013-04-23 20:53:16 +02:00
|
|
|
void unit_trigger_notify(Unit *u) {
|
|
|
|
Unit *other;
|
|
|
|
Iterator i;
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
void *v;
|
2013-04-23 20:53:16 +02:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_TRIGGERED_BY], i)
|
2013-04-23 20:53:16 +02:00
|
|
|
if (UNIT_VTABLE(other)->trigger_notify)
|
|
|
|
UNIT_VTABLE(other)->trigger_notify(other, u);
|
|
|
|
}
|
|
|
|
|
2019-07-27 23:40:24 +02:00
|
|
|
static int raise_level(int log_level, bool condition_info, bool condition_notice) {
|
|
|
|
if (condition_notice && log_level > LOG_NOTICE)
|
|
|
|
return LOG_NOTICE;
|
|
|
|
if (condition_info && log_level > LOG_INFO)
|
|
|
|
return LOG_INFO;
|
|
|
|
return log_level;
|
|
|
|
}
|
|
|
|
|
2017-09-21 14:05:35 +02:00
|
|
|
static int unit_log_resources(Unit *u) {
|
2019-03-22 12:18:00 +01:00
|
|
|
struct iovec iovec[1 + _CGROUP_IP_ACCOUNTING_METRIC_MAX + _CGROUP_IO_ACCOUNTING_METRIC_MAX + 4];
|
|
|
|
bool any_traffic = false, have_ip_accounting = false, any_io = false, have_io_accounting = false;
|
|
|
|
_cleanup_free_ char *igress = NULL, *egress = NULL, *rr = NULL, *wr = NULL;
|
2020-04-16 17:50:21 +02:00
|
|
|
int log_level = LOG_DEBUG; /* May be raised if resources consumed over a threshold */
|
2017-09-21 14:05:35 +02:00
|
|
|
size_t n_message_parts = 0, n_iovec = 0;
|
2019-03-22 12:18:00 +01:00
|
|
|
char* message_parts[1 + 2 + 2 + 1], *t;
|
2017-09-21 14:05:35 +02:00
|
|
|
nsec_t nsec = NSEC_INFINITY;
|
|
|
|
CGroupIPAccountingMetric m;
|
|
|
|
size_t i;
|
|
|
|
int r;
|
|
|
|
const char* const ip_fields[_CGROUP_IP_ACCOUNTING_METRIC_MAX] = {
|
|
|
|
[CGROUP_IP_INGRESS_BYTES] = "IP_METRIC_INGRESS_BYTES",
|
|
|
|
[CGROUP_IP_INGRESS_PACKETS] = "IP_METRIC_INGRESS_PACKETS",
|
|
|
|
[CGROUP_IP_EGRESS_BYTES] = "IP_METRIC_EGRESS_BYTES",
|
|
|
|
[CGROUP_IP_EGRESS_PACKETS] = "IP_METRIC_EGRESS_PACKETS",
|
|
|
|
};
|
2019-03-22 12:18:00 +01:00
|
|
|
const char* const io_fields[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {
|
|
|
|
[CGROUP_IO_READ_BYTES] = "IO_METRIC_READ_BYTES",
|
|
|
|
[CGROUP_IO_WRITE_BYTES] = "IO_METRIC_WRITE_BYTES",
|
|
|
|
[CGROUP_IO_READ_OPERATIONS] = "IO_METRIC_READ_OPERATIONS",
|
|
|
|
[CGROUP_IO_WRITE_OPERATIONS] = "IO_METRIC_WRITE_OPERATIONS",
|
|
|
|
};
|
2017-09-21 14:05:35 +02:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Invoked whenever a unit enters failed or dead state. Logs information about consumed resources if resource
|
|
|
|
* accounting was enabled for a unit. It does this in two ways: a friendly human readable string with reduced
|
|
|
|
* information and the complete data in structured fields. */
|
|
|
|
|
|
|
|
(void) unit_get_cpu_usage(u, &nsec);
|
|
|
|
if (nsec != NSEC_INFINITY) {
|
|
|
|
char buf[FORMAT_TIMESPAN_MAX] = "";
|
|
|
|
|
|
|
|
/* Format the CPU time for inclusion in the structured log message */
|
|
|
|
if (asprintf(&t, "CPU_USAGE_NSEC=%" PRIu64, nsec) < 0) {
|
|
|
|
r = log_oom();
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
iovec[n_iovec++] = IOVEC_MAKE_STRING(t);
|
|
|
|
|
|
|
|
/* Format the CPU time for inclusion in the human language message string */
|
|
|
|
format_timespan(buf, sizeof(buf), nsec / NSEC_PER_USEC, USEC_PER_MSEC);
|
2018-10-19 02:19:24 +02:00
|
|
|
t = strjoin("consumed ", buf, " CPU time");
|
2017-09-21 14:05:35 +02:00
|
|
|
if (!t) {
|
|
|
|
r = log_oom();
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
|
|
|
|
message_parts[n_message_parts++] = t;
|
2019-07-27 23:40:24 +02:00
|
|
|
|
|
|
|
log_level = raise_level(log_level,
|
|
|
|
nsec > NOTICEWORTHY_CPU_NSEC,
|
|
|
|
nsec > MENTIONWORTHY_CPU_NSEC);
|
2017-09-21 14:05:35 +02:00
|
|
|
}
|
|
|
|
|
2019-03-22 12:18:00 +01:00
|
|
|
for (CGroupIOAccountingMetric k = 0; k < _CGROUP_IO_ACCOUNTING_METRIC_MAX; k++) {
|
|
|
|
char buf[FORMAT_BYTES_MAX] = "";
|
|
|
|
uint64_t value = UINT64_MAX;
|
|
|
|
|
|
|
|
assert(io_fields[k]);
|
|
|
|
|
|
|
|
(void) unit_get_io_accounting(u, k, k > 0, &value);
|
|
|
|
if (value == UINT64_MAX)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
have_io_accounting = true;
|
|
|
|
if (value > 0)
|
|
|
|
any_io = true;
|
|
|
|
|
|
|
|
/* Format IO accounting data for inclusion in the structured log message */
|
|
|
|
if (asprintf(&t, "%s=%" PRIu64, io_fields[k], value) < 0) {
|
|
|
|
r = log_oom();
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
iovec[n_iovec++] = IOVEC_MAKE_STRING(t);
|
|
|
|
|
|
|
|
/* Format the IO accounting data for inclusion in the human language message string, but only
|
|
|
|
* for the bytes counters (and not for the operations counters) */
|
|
|
|
if (k == CGROUP_IO_READ_BYTES) {
|
|
|
|
assert(!rr);
|
|
|
|
rr = strjoin("read ", format_bytes(buf, sizeof(buf), value), " from disk");
|
|
|
|
if (!rr) {
|
|
|
|
r = log_oom();
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
} else if (k == CGROUP_IO_WRITE_BYTES) {
|
|
|
|
assert(!wr);
|
|
|
|
wr = strjoin("written ", format_bytes(buf, sizeof(buf), value), " to disk");
|
|
|
|
if (!wr) {
|
|
|
|
r = log_oom();
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
}
|
2019-07-27 23:40:24 +02:00
|
|
|
|
|
|
|
if (IN_SET(k, CGROUP_IO_READ_BYTES, CGROUP_IO_WRITE_BYTES))
|
|
|
|
log_level = raise_level(log_level,
|
|
|
|
value > MENTIONWORTHY_IO_BYTES,
|
|
|
|
value > NOTICEWORTHY_IO_BYTES);
|
2019-03-22 12:18:00 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
if (have_io_accounting) {
|
|
|
|
if (any_io) {
|
|
|
|
if (rr)
|
|
|
|
message_parts[n_message_parts++] = TAKE_PTR(rr);
|
|
|
|
if (wr)
|
|
|
|
message_parts[n_message_parts++] = TAKE_PTR(wr);
|
|
|
|
|
|
|
|
} else {
|
|
|
|
char *k;
|
|
|
|
|
|
|
|
k = strdup("no IO");
|
|
|
|
if (!k) {
|
|
|
|
r = log_oom();
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
|
|
|
|
message_parts[n_message_parts++] = k;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-09-21 14:05:35 +02:00
|
|
|
for (m = 0; m < _CGROUP_IP_ACCOUNTING_METRIC_MAX; m++) {
|
|
|
|
char buf[FORMAT_BYTES_MAX] = "";
|
|
|
|
uint64_t value = UINT64_MAX;
|
|
|
|
|
|
|
|
assert(ip_fields[m]);
|
|
|
|
|
|
|
|
(void) unit_get_ip_accounting(u, m, &value);
|
|
|
|
if (value == UINT64_MAX)
|
|
|
|
continue;
|
2018-11-13 20:51:31 +01:00
|
|
|
|
|
|
|
have_ip_accounting = true;
|
2018-10-19 02:04:12 +02:00
|
|
|
if (value > 0)
|
|
|
|
any_traffic = true;
|
2017-09-21 14:05:35 +02:00
|
|
|
|
|
|
|
/* Format IP accounting data for inclusion in the structured log message */
|
|
|
|
if (asprintf(&t, "%s=%" PRIu64, ip_fields[m], value) < 0) {
|
|
|
|
r = log_oom();
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
iovec[n_iovec++] = IOVEC_MAKE_STRING(t);
|
|
|
|
|
|
|
|
/* Format the IP accounting data for inclusion in the human language message string, but only for the
|
|
|
|
* bytes counters (and not for the packets counters) */
|
2018-10-19 02:04:12 +02:00
|
|
|
if (m == CGROUP_IP_INGRESS_BYTES) {
|
|
|
|
assert(!igress);
|
2018-10-19 02:19:24 +02:00
|
|
|
igress = strjoin("received ", format_bytes(buf, sizeof(buf), value), " IP traffic");
|
2018-10-19 02:04:12 +02:00
|
|
|
if (!igress) {
|
|
|
|
r = log_oom();
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
} else if (m == CGROUP_IP_EGRESS_BYTES) {
|
|
|
|
assert(!egress);
|
2018-10-19 02:19:24 +02:00
|
|
|
egress = strjoin("sent ", format_bytes(buf, sizeof(buf), value), " IP traffic");
|
2018-10-19 02:04:12 +02:00
|
|
|
if (!egress) {
|
|
|
|
r = log_oom();
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
}
|
2019-07-27 23:40:24 +02:00
|
|
|
|
|
|
|
if (IN_SET(m, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES))
|
|
|
|
log_level = raise_level(log_level,
|
|
|
|
value > MENTIONWORTHY_IP_BYTES,
|
|
|
|
value > NOTICEWORTHY_IP_BYTES);
|
2018-10-19 02:04:12 +02:00
|
|
|
}
|
|
|
|
|
2018-11-13 20:51:31 +01:00
|
|
|
if (have_ip_accounting) {
|
|
|
|
if (any_traffic) {
|
|
|
|
if (igress)
|
|
|
|
message_parts[n_message_parts++] = TAKE_PTR(igress);
|
|
|
|
if (egress)
|
|
|
|
message_parts[n_message_parts++] = TAKE_PTR(egress);
|
2018-10-19 02:04:12 +02:00
|
|
|
|
2018-11-13 20:51:31 +01:00
|
|
|
} else {
|
|
|
|
char *k;
|
2017-09-21 14:05:35 +02:00
|
|
|
|
2018-11-13 20:51:31 +01:00
|
|
|
k = strdup("no IP traffic");
|
|
|
|
if (!k) {
|
|
|
|
r = log_oom();
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
|
|
|
|
message_parts[n_message_parts++] = k;
|
|
|
|
}
|
2017-09-21 14:05:35 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Is there any accounting data available at all? */
|
|
|
|
if (n_iovec == 0) {
|
|
|
|
r = 0;
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (n_message_parts == 0)
|
2018-10-19 02:04:12 +02:00
|
|
|
t = strjoina("MESSAGE=", u->id, ": Completed.");
|
2017-09-21 14:05:35 +02:00
|
|
|
else {
|
|
|
|
_cleanup_free_ char *joined;
|
|
|
|
|
|
|
|
message_parts[n_message_parts] = NULL;
|
|
|
|
|
|
|
|
joined = strv_join(message_parts, ", ");
|
|
|
|
if (!joined) {
|
|
|
|
r = log_oom();
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
|
2018-10-19 02:19:24 +02:00
|
|
|
joined[0] = ascii_toupper(joined[0]);
|
2018-10-19 02:04:12 +02:00
|
|
|
t = strjoina("MESSAGE=", u->id, ": ", joined, ".");
|
2017-09-21 14:05:35 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/* The following four fields we allocate on the stack or are static strings, we hence don't want to free them,
|
|
|
|
* and hence don't increase n_iovec for them */
|
|
|
|
iovec[n_iovec] = IOVEC_MAKE_STRING(t);
|
|
|
|
iovec[n_iovec + 1] = IOVEC_MAKE_STRING("MESSAGE_ID=" SD_MESSAGE_UNIT_RESOURCES_STR);
|
|
|
|
|
|
|
|
t = strjoina(u->manager->unit_log_field, u->id);
|
|
|
|
iovec[n_iovec + 2] = IOVEC_MAKE_STRING(t);
|
|
|
|
|
|
|
|
t = strjoina(u->manager->invocation_log_field, u->invocation_id_string);
|
|
|
|
iovec[n_iovec + 3] = IOVEC_MAKE_STRING(t);
|
|
|
|
|
2019-07-27 23:40:24 +02:00
|
|
|
log_struct_iovec(log_level, iovec, n_iovec + 4);
|
2017-09-21 14:05:35 +02:00
|
|
|
r = 0;
|
|
|
|
|
|
|
|
finish:
|
|
|
|
for (i = 0; i < n_message_parts; i++)
|
|
|
|
free(message_parts[i]);
|
|
|
|
|
|
|
|
for (i = 0; i < n_iovec; i++)
|
|
|
|
free(iovec[i].iov_base);
|
|
|
|
|
|
|
|
return r;
|
|
|
|
|
|
|
|
}
|
|
|
|
|
2018-01-24 19:59:55 +01:00
|
|
|
static void unit_update_on_console(Unit *u) {
|
|
|
|
bool b;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
b = unit_needs_console(u);
|
|
|
|
if (u->on_console == b)
|
|
|
|
return;
|
|
|
|
|
|
|
|
u->on_console = b;
|
|
|
|
if (b)
|
|
|
|
manager_ref_console(u->manager);
|
|
|
|
else
|
|
|
|
manager_unref_console(u->manager);
|
|
|
|
}
|
|
|
|
|
2018-11-13 20:59:20 +01:00
|
|
|
static void unit_emit_audit_start(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (u->type != UNIT_SERVICE)
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* Write audit record if we have just finished starting up */
|
|
|
|
manager_send_unit_audit(u->manager, u, AUDIT_SERVICE_START, true);
|
|
|
|
u->in_audit = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void unit_emit_audit_stop(Unit *u, UnitActiveState state) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (u->type != UNIT_SERVICE)
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (u->in_audit) {
|
|
|
|
/* Write audit record if we have just finished shutting down */
|
|
|
|
manager_send_unit_audit(u->manager, u, AUDIT_SERVICE_STOP, state == UNIT_INACTIVE);
|
|
|
|
u->in_audit = false;
|
|
|
|
} else {
|
|
|
|
/* Hmm, if there was no start record written write it now, so that we always have a nice pair */
|
|
|
|
manager_send_unit_audit(u->manager, u, AUDIT_SERVICE_START, state == UNIT_INACTIVE);
|
|
|
|
|
|
|
|
if (state == UNIT_INACTIVE)
|
|
|
|
manager_send_unit_audit(u->manager, u, AUDIT_SERVICE_STOP, true);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-12-10 20:56:57 +01:00
|
|
|
static bool unit_process_job(Job *j, UnitActiveState ns, UnitNotifyFlags flags) {
|
|
|
|
bool unexpected = false;
|
2019-06-29 02:02:30 +02:00
|
|
|
JobResult result;
|
2018-12-10 20:56:57 +01:00
|
|
|
|
|
|
|
assert(j);
|
|
|
|
|
|
|
|
if (j->state == JOB_WAITING)
|
|
|
|
|
|
|
|
/* So we reached a different state for this job. Let's see if we can run it now if it failed previously
|
|
|
|
* due to EAGAIN. */
|
|
|
|
job_add_to_run_queue(j);
|
|
|
|
|
|
|
|
/* Let's check whether the unit's new state constitutes a finished job, or maybe contradicts a running job and
|
|
|
|
* hence needs to invalidate jobs. */
|
|
|
|
|
|
|
|
switch (j->type) {
|
|
|
|
|
|
|
|
case JOB_START:
|
|
|
|
case JOB_VERIFY_ACTIVE:
|
|
|
|
|
|
|
|
if (UNIT_IS_ACTIVE_OR_RELOADING(ns))
|
|
|
|
job_finish_and_invalidate(j, JOB_DONE, true, false);
|
|
|
|
else if (j->state == JOB_RUNNING && ns != UNIT_ACTIVATING) {
|
|
|
|
unexpected = true;
|
|
|
|
|
2019-06-29 02:02:30 +02:00
|
|
|
if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
|
|
|
|
if (ns == UNIT_FAILED)
|
|
|
|
result = JOB_FAILED;
|
|
|
|
else if (FLAGS_SET(flags, UNIT_NOTIFY_SKIP_CONDITION))
|
|
|
|
result = JOB_SKIPPED;
|
|
|
|
else
|
|
|
|
result = JOB_DONE;
|
|
|
|
|
|
|
|
job_finish_and_invalidate(j, result, true, false);
|
|
|
|
}
|
2018-12-10 20:56:57 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
case JOB_RELOAD:
|
|
|
|
case JOB_RELOAD_OR_START:
|
|
|
|
case JOB_TRY_RELOAD:
|
|
|
|
|
|
|
|
if (j->state == JOB_RUNNING) {
|
|
|
|
if (ns == UNIT_ACTIVE)
|
|
|
|
job_finish_and_invalidate(j, (flags & UNIT_NOTIFY_RELOAD_FAILURE) ? JOB_FAILED : JOB_DONE, true, false);
|
|
|
|
else if (!IN_SET(ns, UNIT_ACTIVATING, UNIT_RELOADING)) {
|
|
|
|
unexpected = true;
|
|
|
|
|
|
|
|
if (UNIT_IS_INACTIVE_OR_FAILED(ns))
|
|
|
|
job_finish_and_invalidate(j, ns == UNIT_FAILED ? JOB_FAILED : JOB_DONE, true, false);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
case JOB_STOP:
|
|
|
|
case JOB_RESTART:
|
|
|
|
case JOB_TRY_RESTART:
|
|
|
|
|
|
|
|
if (UNIT_IS_INACTIVE_OR_FAILED(ns))
|
|
|
|
job_finish_and_invalidate(j, JOB_DONE, true, false);
|
|
|
|
else if (j->state == JOB_RUNNING && ns != UNIT_DEACTIVATING) {
|
|
|
|
unexpected = true;
|
|
|
|
job_finish_and_invalidate(j, JOB_FAILED, true, false);
|
|
|
|
}
|
|
|
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
default:
|
|
|
|
assert_not_reached("Job type unknown");
|
|
|
|
}
|
|
|
|
|
|
|
|
return unexpected;
|
|
|
|
}
|
|
|
|
|
2018-06-01 19:06:19 +02:00
|
|
|
void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, UnitNotifyFlags flags) {
|
2018-11-08 13:15:25 +01:00
|
|
|
const char *reason;
|
core: rework how we connect to the bus
This removes the current bus_init() call, as it had multiple problems:
it munged handling of the three bus connections we care about (private,
"api" and system) into one, even though the conditions when which was
ready are very different. It also added redundant logging, as the
individual calls it called all logged on their own anyway.
The three calls bus_init_api(), bus_init_private() and bus_init_system()
are now made public. A new call manager_dbus_is_running() is added that
works much like manager_journal_is_running() and is a lot more careful
when checking whether dbus is around. Optionally it checks the unit's
deserialized_state rather than state, in order to accomodate for cases
where we cant to connect to the bus before deserializing the
"subscribed" list, before coldplugging the units.
manager_recheck_dbus() is added, that works a lot like
manager_recheck_journal() and is invoked in unit_notify(), i.e. when
units change state.
All in all this should make handling a bit more alike to journal
handling, and it also fixes one major bug: when running in user mode
we'll now connect to the system bus early on, without conditionalizing
this in anyway.
2018-02-07 14:52:22 +01:00
|
|
|
Manager *m;
|
2010-04-08 01:22:51 +02:00
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
assert(u);
|
|
|
|
assert(os < _UNIT_ACTIVE_STATE_MAX);
|
|
|
|
assert(ns < _UNIT_ACTIVE_STATE_MAX);
|
|
|
|
|
core: rework how we connect to the bus
This removes the current bus_init() call, as it had multiple problems:
it munged handling of the three bus connections we care about (private,
"api" and system) into one, even though the conditions when which was
ready are very different. It also added redundant logging, as the
individual calls it called all logged on their own anyway.
The three calls bus_init_api(), bus_init_private() and bus_init_system()
are now made public. A new call manager_dbus_is_running() is added that
works much like manager_journal_is_running() and is a lot more careful
when checking whether dbus is around. Optionally it checks the unit's
deserialized_state rather than state, in order to accomodate for cases
where we cant to connect to the bus before deserializing the
"subscribed" list, before coldplugging the units.
manager_recheck_dbus() is added, that works a lot like
manager_recheck_journal() and is invoked in unit_notify(), i.e. when
units change state.
All in all this should make handling a bit more alike to journal
handling, and it also fixes one major bug: when running in user mode
we'll now connect to the system bus early on, without conditionalizing
this in anyway.
2018-02-07 14:52:22 +01:00
|
|
|
/* Note that this is called for all low-level state changes, even if they might map to the same high-level
|
|
|
|
* UnitActiveState! That means that ns == os is an expected behavior here. For example: if a mount point is
|
|
|
|
* remounted this function will be called too! */
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2013-02-27 23:58:10 +01:00
|
|
|
m = u->manager;
|
|
|
|
|
2018-11-29 18:48:52 +01:00
|
|
|
/* Let's enqueue the change signal early. In case this unit has a job associated we want that this unit is in
|
|
|
|
* the bus queue, so that any job change signal queued will force out the unit change signal first. */
|
|
|
|
unit_add_to_dbus_queue(u);
|
|
|
|
|
2014-03-12 20:55:13 +01:00
|
|
|
/* Update timestamps for state changes */
|
2016-02-24 21:36:09 +01:00
|
|
|
if (!MANAGER_IS_RELOADING(m)) {
|
2016-02-01 16:01:25 +01:00
|
|
|
dual_timestamp_get(&u->state_change_timestamp);
|
2010-05-14 03:05:38 +02:00
|
|
|
|
2011-03-30 20:04:20 +02:00
|
|
|
if (UNIT_IS_INACTIVE_OR_FAILED(os) && !UNIT_IS_INACTIVE_OR_FAILED(ns))
|
2016-02-01 16:01:25 +01:00
|
|
|
u->inactive_exit_timestamp = u->state_change_timestamp;
|
2011-03-30 20:04:20 +02:00
|
|
|
else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_INACTIVE_OR_FAILED(ns))
|
2016-02-01 16:01:25 +01:00
|
|
|
u->inactive_enter_timestamp = u->state_change_timestamp;
|
2011-03-30 20:04:20 +02:00
|
|
|
|
|
|
|
if (!UNIT_IS_ACTIVE_OR_RELOADING(os) && UNIT_IS_ACTIVE_OR_RELOADING(ns))
|
2016-02-01 16:01:25 +01:00
|
|
|
u->active_enter_timestamp = u->state_change_timestamp;
|
2011-03-30 20:04:20 +02:00
|
|
|
else if (UNIT_IS_ACTIVE_OR_RELOADING(os) && !UNIT_IS_ACTIVE_OR_RELOADING(ns))
|
2016-02-01 16:01:25 +01:00
|
|
|
u->active_exit_timestamp = u->state_change_timestamp;
|
2011-03-30 20:04:20 +02:00
|
|
|
}
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2014-05-23 20:56:42 +02:00
|
|
|
/* Keep track of failed units */
|
2018-11-13 05:44:20 +01:00
|
|
|
(void) manager_update_failed_units(m, u, ns == UNIT_FAILED);
|
2014-03-12 20:55:13 +01:00
|
|
|
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
/* Make sure the cgroup and state files are always removed when we become inactive */
|
|
|
|
if (UNIT_IS_INACTIVE_OR_FAILED(ns)) {
|
core: unified cgroup hierarchy support
This patch set adds full support the new unified cgroup hierarchy logic
of modern kernels.
A new kernel command line option "systemd.unified_cgroup_hierarchy=1" is
added. If specified the unified hierarchy is mounted to /sys/fs/cgroup
instead of a tmpfs. No further hierarchies are mounted. The kernel
command line option defaults to off. We can turn it on by default as
soon as the kernel's APIs regarding this are stabilized (but even then
downstream distros might want to turn this off, as this will break any
tools that access cgroupfs directly).
It is possibly to choose for each boot individually whether the unified
or the legacy hierarchy is used. nspawn will by default provide the
legacy hierarchy to containers if the host is using it, and the unified
otherwise. However it is possible to run containers with the unified
hierarchy on a legacy host and vice versa, by setting the
$UNIFIED_CGROUP_HIERARCHY environment variable for nspawn to 1 or 0,
respectively.
The unified hierarchy provides reliable cgroup empty notifications for
the first time, via inotify. To make use of this we maintain one
manager-wide inotify fd, and each cgroup to it.
This patch also removes cg_delete() which is unused now.
On kernel 4.2 only the "memory" controller is compatible with the
unified hierarchy, hence that's the only controller systemd exposes when
booted in unified heirarchy mode.
This introduces a new enum for enumerating supported controllers, plus a
related enum for the mask bits mapping to it. The core is changed to
make use of this everywhere.
This moves PID 1 into a new "init.scope" implicit scope unit in the root
slice. This is necessary since on the unified hierarchy cgroups may
either contain subgroups or processes but not both. PID 1 hence has to
move out of the root cgroup (strictly speaking the root cgroup is the
only one where processes and subgroups are still allowed, but in order
to support containers nicey, we move PID 1 into the new scope in all
cases.) This new unit is also used on legacy hierarchy setups. It's
actually pretty useful on all systems, as it can then be used to filter
journal messages coming from PID 1, and so on.
The root slice ("-.slice") is now implicitly created and started (and
does not require a unit file on disk anymore), since
that's where "init.scope" is located and the slice needs to be started
before the scope can.
To check whether we are in unified or legacy hierarchy mode we use
statfs() on /sys/fs/cgroup. If the .f_type field reports tmpfs we are in
legacy mode, if it reports cgroupfs we are in unified mode.
This patch set carefuly makes sure that cgls and cgtop continue to work
as desired.
When invoking nspawn as a service it will implicitly create two
subcgroups in the cgroup it is using, one to move the nspawn process
into, the other to move the actual container processes into. This is
done because of the requirement that cgroups may either contain
processes or other subgroups.
2015-09-01 19:22:36 +02:00
|
|
|
unit_prune_cgroup(u);
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
unit_unlink_state_files(u);
|
|
|
|
}
|
2010-07-10 17:34:42 +02:00
|
|
|
|
2018-01-24 19:59:55 +01:00
|
|
|
unit_update_on_console(u);
|
2013-02-28 00:01:10 +01:00
|
|
|
|
2016-02-24 21:36:09 +01:00
|
|
|
if (!MANAGER_IS_RELOADING(m)) {
|
2018-12-11 11:59:39 +01:00
|
|
|
bool unexpected;
|
|
|
|
|
|
|
|
/* Let's propagate state changes to the job */
|
|
|
|
if (u->job)
|
|
|
|
unexpected = unit_process_job(u->job, ns, flags);
|
|
|
|
else
|
|
|
|
unexpected = true;
|
2010-01-29 20:47:09 +01:00
|
|
|
|
2018-12-11 11:59:39 +01:00
|
|
|
/* If this state change happened without being requested by a job, then let's retroactively start or
|
|
|
|
* stop dependencies. We skip that step when deserializing, since we don't want to create any
|
|
|
|
* additional jobs just because something is already activated. */
|
2011-03-30 20:04:20 +02:00
|
|
|
|
|
|
|
if (unexpected) {
|
|
|
|
if (UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_ACTIVE_OR_ACTIVATING(ns))
|
|
|
|
retroactively_start_dependencies(u);
|
|
|
|
else if (UNIT_IS_ACTIVE_OR_ACTIVATING(os) && UNIT_IS_INACTIVE_OR_DEACTIVATING(ns))
|
|
|
|
retroactively_stop_dependencies(u);
|
|
|
|
}
|
2010-07-17 00:58:47 +02:00
|
|
|
|
2011-12-09 15:25:29 +01:00
|
|
|
/* stop unneeded units regardless if going down was expected or not */
|
2018-08-09 16:26:27 +02:00
|
|
|
if (UNIT_IS_INACTIVE_OR_FAILED(ns))
|
2011-12-09 15:25:29 +01:00
|
|
|
check_unneeded_dependencies(u);
|
|
|
|
|
2011-03-30 20:04:20 +02:00
|
|
|
if (ns != os && ns == UNIT_FAILED) {
|
2017-09-26 23:35:58 +02:00
|
|
|
log_unit_debug(u, "Unit entered failed state.");
|
2018-06-01 19:06:19 +02:00
|
|
|
|
|
|
|
if (!(flags & UNIT_NOTIFY_WILL_AUTO_RESTART))
|
|
|
|
unit_start_on_failure(u);
|
2010-08-11 04:38:55 +02:00
|
|
|
}
|
2010-04-10 17:53:17 +02:00
|
|
|
|
2018-10-09 10:05:44 +02:00
|
|
|
if (UNIT_IS_ACTIVE_OR_RELOADING(ns) && !UNIT_IS_ACTIVE_OR_RELOADING(os)) {
|
|
|
|
/* This unit just finished starting up */
|
2011-03-30 20:16:07 +02:00
|
|
|
|
2018-11-13 20:59:20 +01:00
|
|
|
unit_emit_audit_start(u);
|
2013-02-27 23:58:10 +01:00
|
|
|
manager_send_unit_plymouth(m, u);
|
2018-10-09 10:05:44 +02:00
|
|
|
}
|
2011-03-30 20:04:20 +02:00
|
|
|
|
2018-10-09 10:05:44 +02:00
|
|
|
if (UNIT_IS_INACTIVE_OR_FAILED(ns) && !UNIT_IS_INACTIVE_OR_FAILED(os)) {
|
2017-09-21 14:05:35 +02:00
|
|
|
/* This unit just stopped/failed. */
|
2018-10-09 10:05:44 +02:00
|
|
|
|
2018-11-13 20:59:20 +01:00
|
|
|
unit_emit_audit_stop(u, ns);
|
2017-09-21 14:05:35 +02:00
|
|
|
unit_log_resources(u);
|
2010-08-11 04:38:55 +02:00
|
|
|
}
|
2010-04-06 16:32:07 +02:00
|
|
|
}
|
|
|
|
|
2018-03-21 12:03:45 +01:00
|
|
|
manager_recheck_journal(m);
|
|
|
|
manager_recheck_dbus(m);
|
2018-03-16 23:01:05 +01:00
|
|
|
|
2013-04-23 20:53:16 +02:00
|
|
|
unit_trigger_notify(u);
|
2011-03-30 20:16:07 +02:00
|
|
|
|
2018-11-13 05:44:20 +01:00
|
|
|
if (!MANAGER_IS_RELOADING(m)) {
|
core: rework how we connect to the bus
This removes the current bus_init() call, as it had multiple problems:
it munged handling of the three bus connections we care about (private,
"api" and system) into one, even though the conditions when which was
ready are very different. It also added redundant logging, as the
individual calls it called all logged on their own anyway.
The three calls bus_init_api(), bus_init_private() and bus_init_system()
are now made public. A new call manager_dbus_is_running() is added that
works much like manager_journal_is_running() and is a lot more careful
when checking whether dbus is around. Optionally it checks the unit's
deserialized_state rather than state, in order to accomodate for cases
where we cant to connect to the bus before deserializing the
"subscribed" list, before coldplugging the units.
manager_recheck_dbus() is added, that works a lot like
manager_recheck_journal() and is invoked in unit_notify(), i.e. when
units change state.
All in all this should make handling a bit more alike to journal
handling, and it also fixes one major bug: when running in user mode
we'll now connect to the system bus early on, without conditionalizing
this in anyway.
2018-02-07 14:52:22 +01:00
|
|
|
/* Maybe we finished startup and are now ready for being stopped because unneeded? */
|
2018-08-20 10:43:31 +02:00
|
|
|
unit_submit_to_stop_when_unneeded_queue(u);
|
2010-02-05 00:38:41 +01:00
|
|
|
|
core: rework how we connect to the bus
This removes the current bus_init() call, as it had multiple problems:
it munged handling of the three bus connections we care about (private,
"api" and system) into one, even though the conditions when which was
ready are very different. It also added redundant logging, as the
individual calls it called all logged on their own anyway.
The three calls bus_init_api(), bus_init_private() and bus_init_system()
are now made public. A new call manager_dbus_is_running() is added that
works much like manager_journal_is_running() and is a lot more careful
when checking whether dbus is around. Optionally it checks the unit's
deserialized_state rather than state, in order to accomodate for cases
where we cant to connect to the bus before deserializing the
"subscribed" list, before coldplugging the units.
manager_recheck_dbus() is added, that works a lot like
manager_recheck_journal() and is invoked in unit_notify(), i.e. when
units change state.
All in all this should make handling a bit more alike to journal
handling, and it also fixes one major bug: when running in user mode
we'll now connect to the system bus early on, without conditionalizing
this in anyway.
2018-02-07 14:52:22 +01:00
|
|
|
/* Maybe we finished startup, but something we needed has vanished? Let's die then. (This happens when
|
|
|
|
* something BindsTo= to a Type=oneshot unit, as these units go directly from starting to inactive,
|
2014-08-18 22:21:42 +02:00
|
|
|
* without ever entering started.) */
|
|
|
|
unit_check_binds_to(u);
|
2017-11-16 15:02:56 +01:00
|
|
|
|
2018-11-08 13:15:25 +01:00
|
|
|
if (os != UNIT_FAILED && ns == UNIT_FAILED) {
|
|
|
|
reason = strjoina("unit ", u->id, " failed");
|
2019-03-18 13:20:54 +01:00
|
|
|
emergency_action(m, u->failure_action, 0, u->reboot_arg, unit_failure_action_exit_status(u), reason);
|
2018-11-08 13:15:25 +01:00
|
|
|
} else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && ns == UNIT_INACTIVE) {
|
|
|
|
reason = strjoina("unit ", u->id, " succeeded");
|
2019-03-18 13:20:54 +01:00
|
|
|
emergency_action(m, u->success_action, 0, u->reboot_arg, unit_success_action_exit_status(u), reason);
|
2018-11-08 13:15:25 +01:00
|
|
|
}
|
2014-08-18 22:21:42 +02:00
|
|
|
}
|
|
|
|
|
2010-04-21 06:01:13 +02:00
|
|
|
unit_add_to_gc_queue(u);
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
2019-03-18 20:59:36 +01:00
|
|
|
int unit_watch_pid(Unit *u, pid_t pid, bool exclusive) {
|
core: rework how we track which PIDs to watch for a unit
Previously, we'd maintain two hashmaps keyed by PIDs, pointing to Unit
interested in SIGCHLD events for them. This scheme allowed a specific
PID to be watched by exactly 0, 1 or 2 units.
With this rework this is replaced by a single hashmap which is primarily
keyed by the PID and points to a Unit interested in it. However, it
optionally also keyed by the negated PID, in which case it points to a
NULL terminated array of additional Unit objects also interested. This
scheme means arbitrary numbers of Units may now watch the same PID.
Runtime and memory behaviour should not be impact by this change, as for
the common case (i.e. each PID only watched by a single unit) behaviour
stays the same, but for the uncommon case (a PID watched by more than
one unit) we only pay with a single additional memory allocation for the
array.
Why this all? Primarily, because allowing exactly two units to watch a
specific PID is not sufficient for some niche cases, as processes can
belong to more than one unit these days:
1. sd_notify() with MAINPID= can be used to attach a process from a
different cgroup to multiple units.
2. Similar, the PIDFile= setting in unit files can be used for similar
setups,
3. By creating a scope unit a main process of a service may join a
different unit, too.
4. On cgroupsv1 we frequently end up watching all processes remaining in
a scope, and if a process opens lots of scopes one after the other it
might thus end up being watch by many of them.
This patch hence removes the 2-unit-per-PID limit. It also makes a
couple of other changes, some of them quite relevant:
- manager_get_unit_by_pid() (and the bus call wrapping it) when there's
ambiguity will prefer returning the Unit the process belongs to based on
cgroup membership, and only check the watch-pids hashmap if that
fails. This change in logic is probably more in line with what people
expect and makes things more stable as each process can belong to
exactly one cgroup only.
- Every SIGCHLD event is now dispatched to all units interested in its
PID. Previously, there was some magic conditionalization: the SIGCHLD
would only be dispatched to the unit if it was only interested in a
single PID only, or the PID belonged to the control or main PID or we
didn't dispatch a signle SIGCHLD to the unit in the current event loop
iteration yet. These rules were quite arbitrary and also redundant as
the the per-unit handlers would filter the PIDs anyway a second time.
With this change we'll hence relax the rules: all we do now is
dispatch every SIGCHLD event exactly once to each unit interested in
it, and it's up to the unit to then use or ignore this. We use a
generation counter in the unit to ensure that we only invoke the unit
handler once for each event, protecting us from confusion if a unit is
both associated with a specific PID through cgroup membership and
through the "watch_pids" logic. It also protects us from being
confused if the "watch_pids" hashmap is altered while we are
dispatching to it (which is a very likely case).
- sd_notify() message dispatching has been reworked to be very similar
to SIGCHLD handling now. A generation counter is used for dispatching
as well.
This also adds a new test that validates that "watch_pid" registration
and unregstration works correctly.
2018-01-12 13:41:05 +01:00
|
|
|
int r;
|
2014-02-06 17:17:51 +01:00
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
assert(u);
|
core: rework how we track which PIDs to watch for a unit
Previously, we'd maintain two hashmaps keyed by PIDs, pointing to Unit
interested in SIGCHLD events for them. This scheme allowed a specific
PID to be watched by exactly 0, 1 or 2 units.
With this rework this is replaced by a single hashmap which is primarily
keyed by the PID and points to a Unit interested in it. However, it
optionally also keyed by the negated PID, in which case it points to a
NULL terminated array of additional Unit objects also interested. This
scheme means arbitrary numbers of Units may now watch the same PID.
Runtime and memory behaviour should not be impact by this change, as for
the common case (i.e. each PID only watched by a single unit) behaviour
stays the same, but for the uncommon case (a PID watched by more than
one unit) we only pay with a single additional memory allocation for the
array.
Why this all? Primarily, because allowing exactly two units to watch a
specific PID is not sufficient for some niche cases, as processes can
belong to more than one unit these days:
1. sd_notify() with MAINPID= can be used to attach a process from a
different cgroup to multiple units.
2. Similar, the PIDFile= setting in unit files can be used for similar
setups,
3. By creating a scope unit a main process of a service may join a
different unit, too.
4. On cgroupsv1 we frequently end up watching all processes remaining in
a scope, and if a process opens lots of scopes one after the other it
might thus end up being watch by many of them.
This patch hence removes the 2-unit-per-PID limit. It also makes a
couple of other changes, some of them quite relevant:
- manager_get_unit_by_pid() (and the bus call wrapping it) when there's
ambiguity will prefer returning the Unit the process belongs to based on
cgroup membership, and only check the watch-pids hashmap if that
fails. This change in logic is probably more in line with what people
expect and makes things more stable as each process can belong to
exactly one cgroup only.
- Every SIGCHLD event is now dispatched to all units interested in its
PID. Previously, there was some magic conditionalization: the SIGCHLD
would only be dispatched to the unit if it was only interested in a
single PID only, or the PID belonged to the control or main PID or we
didn't dispatch a signle SIGCHLD to the unit in the current event loop
iteration yet. These rules were quite arbitrary and also redundant as
the the per-unit handlers would filter the PIDs anyway a second time.
With this change we'll hence relax the rules: all we do now is
dispatch every SIGCHLD event exactly once to each unit interested in
it, and it's up to the unit to then use or ignore this. We use a
generation counter in the unit to ensure that we only invoke the unit
handler once for each event, protecting us from confusion if a unit is
both associated with a specific PID through cgroup membership and
through the "watch_pids" logic. It also protects us from being
confused if the "watch_pids" hashmap is altered while we are
dispatching to it (which is a very likely case).
- sd_notify() message dispatching has been reworked to be very similar
to SIGCHLD handling now. A generation counter is used for dispatching
as well.
This also adds a new test that validates that "watch_pid" registration
and unregstration works correctly.
2018-01-12 13:41:05 +01:00
|
|
|
assert(pid_is_valid(pid));
|
2010-01-26 21:39:06 +01:00
|
|
|
|
core: rework how we track which PIDs to watch for a unit
Previously, we'd maintain two hashmaps keyed by PIDs, pointing to Unit
interested in SIGCHLD events for them. This scheme allowed a specific
PID to be watched by exactly 0, 1 or 2 units.
With this rework this is replaced by a single hashmap which is primarily
keyed by the PID and points to a Unit interested in it. However, it
optionally also keyed by the negated PID, in which case it points to a
NULL terminated array of additional Unit objects also interested. This
scheme means arbitrary numbers of Units may now watch the same PID.
Runtime and memory behaviour should not be impact by this change, as for
the common case (i.e. each PID only watched by a single unit) behaviour
stays the same, but for the uncommon case (a PID watched by more than
one unit) we only pay with a single additional memory allocation for the
array.
Why this all? Primarily, because allowing exactly two units to watch a
specific PID is not sufficient for some niche cases, as processes can
belong to more than one unit these days:
1. sd_notify() with MAINPID= can be used to attach a process from a
different cgroup to multiple units.
2. Similar, the PIDFile= setting in unit files can be used for similar
setups,
3. By creating a scope unit a main process of a service may join a
different unit, too.
4. On cgroupsv1 we frequently end up watching all processes remaining in
a scope, and if a process opens lots of scopes one after the other it
might thus end up being watch by many of them.
This patch hence removes the 2-unit-per-PID limit. It also makes a
couple of other changes, some of them quite relevant:
- manager_get_unit_by_pid() (and the bus call wrapping it) when there's
ambiguity will prefer returning the Unit the process belongs to based on
cgroup membership, and only check the watch-pids hashmap if that
fails. This change in logic is probably more in line with what people
expect and makes things more stable as each process can belong to
exactly one cgroup only.
- Every SIGCHLD event is now dispatched to all units interested in its
PID. Previously, there was some magic conditionalization: the SIGCHLD
would only be dispatched to the unit if it was only interested in a
single PID only, or the PID belonged to the control or main PID or we
didn't dispatch a signle SIGCHLD to the unit in the current event loop
iteration yet. These rules were quite arbitrary and also redundant as
the the per-unit handlers would filter the PIDs anyway a second time.
With this change we'll hence relax the rules: all we do now is
dispatch every SIGCHLD event exactly once to each unit interested in
it, and it's up to the unit to then use or ignore this. We use a
generation counter in the unit to ensure that we only invoke the unit
handler once for each event, protecting us from confusion if a unit is
both associated with a specific PID through cgroup membership and
through the "watch_pids" logic. It also protects us from being
confused if the "watch_pids" hashmap is altered while we are
dispatching to it (which is a very likely case).
- sd_notify() message dispatching has been reworked to be very similar
to SIGCHLD handling now. A generation counter is used for dispatching
as well.
This also adds a new test that validates that "watch_pid" registration
and unregstration works correctly.
2018-01-12 13:41:05 +01:00
|
|
|
/* Watch a specific PID */
|
2014-02-07 11:58:25 +01:00
|
|
|
|
2019-03-18 20:59:36 +01:00
|
|
|
/* Caller might be sure that this PID belongs to this unit only. Let's take this
|
|
|
|
* opportunity to remove any stalled references to this PID as they can be created
|
|
|
|
* easily (when watching a process which is not our direct child). */
|
|
|
|
if (exclusive)
|
|
|
|
manager_unwatch_pid(u->manager, pid);
|
|
|
|
|
2014-08-13 01:00:18 +02:00
|
|
|
r = set_ensure_allocated(&u->pids, NULL);
|
2014-02-07 11:58:25 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
core: rework how we track which PIDs to watch for a unit
Previously, we'd maintain two hashmaps keyed by PIDs, pointing to Unit
interested in SIGCHLD events for them. This scheme allowed a specific
PID to be watched by exactly 0, 1 or 2 units.
With this rework this is replaced by a single hashmap which is primarily
keyed by the PID and points to a Unit interested in it. However, it
optionally also keyed by the negated PID, in which case it points to a
NULL terminated array of additional Unit objects also interested. This
scheme means arbitrary numbers of Units may now watch the same PID.
Runtime and memory behaviour should not be impact by this change, as for
the common case (i.e. each PID only watched by a single unit) behaviour
stays the same, but for the uncommon case (a PID watched by more than
one unit) we only pay with a single additional memory allocation for the
array.
Why this all? Primarily, because allowing exactly two units to watch a
specific PID is not sufficient for some niche cases, as processes can
belong to more than one unit these days:
1. sd_notify() with MAINPID= can be used to attach a process from a
different cgroup to multiple units.
2. Similar, the PIDFile= setting in unit files can be used for similar
setups,
3. By creating a scope unit a main process of a service may join a
different unit, too.
4. On cgroupsv1 we frequently end up watching all processes remaining in
a scope, and if a process opens lots of scopes one after the other it
might thus end up being watch by many of them.
This patch hence removes the 2-unit-per-PID limit. It also makes a
couple of other changes, some of them quite relevant:
- manager_get_unit_by_pid() (and the bus call wrapping it) when there's
ambiguity will prefer returning the Unit the process belongs to based on
cgroup membership, and only check the watch-pids hashmap if that
fails. This change in logic is probably more in line with what people
expect and makes things more stable as each process can belong to
exactly one cgroup only.
- Every SIGCHLD event is now dispatched to all units interested in its
PID. Previously, there was some magic conditionalization: the SIGCHLD
would only be dispatched to the unit if it was only interested in a
single PID only, or the PID belonged to the control or main PID or we
didn't dispatch a signle SIGCHLD to the unit in the current event loop
iteration yet. These rules were quite arbitrary and also redundant as
the the per-unit handlers would filter the PIDs anyway a second time.
With this change we'll hence relax the rules: all we do now is
dispatch every SIGCHLD event exactly once to each unit interested in
it, and it's up to the unit to then use or ignore this. We use a
generation counter in the unit to ensure that we only invoke the unit
handler once for each event, protecting us from confusion if a unit is
both associated with a specific PID through cgroup membership and
through the "watch_pids" logic. It also protects us from being
confused if the "watch_pids" hashmap is altered while we are
dispatching to it (which is a very likely case).
- sd_notify() message dispatching has been reworked to be very similar
to SIGCHLD handling now. A generation counter is used for dispatching
as well.
This also adds a new test that validates that "watch_pid" registration
and unregstration works correctly.
2018-01-12 13:41:05 +01:00
|
|
|
r = hashmap_ensure_allocated(&u->manager->watch_pids, NULL);
|
2014-02-06 17:17:51 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
core: rework how we track which PIDs to watch for a unit
Previously, we'd maintain two hashmaps keyed by PIDs, pointing to Unit
interested in SIGCHLD events for them. This scheme allowed a specific
PID to be watched by exactly 0, 1 or 2 units.
With this rework this is replaced by a single hashmap which is primarily
keyed by the PID and points to a Unit interested in it. However, it
optionally also keyed by the negated PID, in which case it points to a
NULL terminated array of additional Unit objects also interested. This
scheme means arbitrary numbers of Units may now watch the same PID.
Runtime and memory behaviour should not be impact by this change, as for
the common case (i.e. each PID only watched by a single unit) behaviour
stays the same, but for the uncommon case (a PID watched by more than
one unit) we only pay with a single additional memory allocation for the
array.
Why this all? Primarily, because allowing exactly two units to watch a
specific PID is not sufficient for some niche cases, as processes can
belong to more than one unit these days:
1. sd_notify() with MAINPID= can be used to attach a process from a
different cgroup to multiple units.
2. Similar, the PIDFile= setting in unit files can be used for similar
setups,
3. By creating a scope unit a main process of a service may join a
different unit, too.
4. On cgroupsv1 we frequently end up watching all processes remaining in
a scope, and if a process opens lots of scopes one after the other it
might thus end up being watch by many of them.
This patch hence removes the 2-unit-per-PID limit. It also makes a
couple of other changes, some of them quite relevant:
- manager_get_unit_by_pid() (and the bus call wrapping it) when there's
ambiguity will prefer returning the Unit the process belongs to based on
cgroup membership, and only check the watch-pids hashmap if that
fails. This change in logic is probably more in line with what people
expect and makes things more stable as each process can belong to
exactly one cgroup only.
- Every SIGCHLD event is now dispatched to all units interested in its
PID. Previously, there was some magic conditionalization: the SIGCHLD
would only be dispatched to the unit if it was only interested in a
single PID only, or the PID belonged to the control or main PID or we
didn't dispatch a signle SIGCHLD to the unit in the current event loop
iteration yet. These rules were quite arbitrary and also redundant as
the the per-unit handlers would filter the PIDs anyway a second time.
With this change we'll hence relax the rules: all we do now is
dispatch every SIGCHLD event exactly once to each unit interested in
it, and it's up to the unit to then use or ignore this. We use a
generation counter in the unit to ensure that we only invoke the unit
handler once for each event, protecting us from confusion if a unit is
both associated with a specific PID through cgroup membership and
through the "watch_pids" logic. It also protects us from being
confused if the "watch_pids" hashmap is altered while we are
dispatching to it (which is a very likely case).
- sd_notify() message dispatching has been reworked to be very similar
to SIGCHLD handling now. A generation counter is used for dispatching
as well.
This also adds a new test that validates that "watch_pid" registration
and unregstration works correctly.
2018-01-12 13:41:05 +01:00
|
|
|
/* First try, let's add the unit keyed by "pid". */
|
|
|
|
r = hashmap_put(u->manager->watch_pids, PID_TO_PTR(pid), u);
|
|
|
|
if (r == -EEXIST) {
|
|
|
|
Unit **array;
|
|
|
|
bool found = false;
|
|
|
|
size_t n = 0;
|
2010-04-15 23:16:16 +02:00
|
|
|
|
core: rework how we track which PIDs to watch for a unit
Previously, we'd maintain two hashmaps keyed by PIDs, pointing to Unit
interested in SIGCHLD events for them. This scheme allowed a specific
PID to be watched by exactly 0, 1 or 2 units.
With this rework this is replaced by a single hashmap which is primarily
keyed by the PID and points to a Unit interested in it. However, it
optionally also keyed by the negated PID, in which case it points to a
NULL terminated array of additional Unit objects also interested. This
scheme means arbitrary numbers of Units may now watch the same PID.
Runtime and memory behaviour should not be impact by this change, as for
the common case (i.e. each PID only watched by a single unit) behaviour
stays the same, but for the uncommon case (a PID watched by more than
one unit) we only pay with a single additional memory allocation for the
array.
Why this all? Primarily, because allowing exactly two units to watch a
specific PID is not sufficient for some niche cases, as processes can
belong to more than one unit these days:
1. sd_notify() with MAINPID= can be used to attach a process from a
different cgroup to multiple units.
2. Similar, the PIDFile= setting in unit files can be used for similar
setups,
3. By creating a scope unit a main process of a service may join a
different unit, too.
4. On cgroupsv1 we frequently end up watching all processes remaining in
a scope, and if a process opens lots of scopes one after the other it
might thus end up being watch by many of them.
This patch hence removes the 2-unit-per-PID limit. It also makes a
couple of other changes, some of them quite relevant:
- manager_get_unit_by_pid() (and the bus call wrapping it) when there's
ambiguity will prefer returning the Unit the process belongs to based on
cgroup membership, and only check the watch-pids hashmap if that
fails. This change in logic is probably more in line with what people
expect and makes things more stable as each process can belong to
exactly one cgroup only.
- Every SIGCHLD event is now dispatched to all units interested in its
PID. Previously, there was some magic conditionalization: the SIGCHLD
would only be dispatched to the unit if it was only interested in a
single PID only, or the PID belonged to the control or main PID or we
didn't dispatch a signle SIGCHLD to the unit in the current event loop
iteration yet. These rules were quite arbitrary and also redundant as
the the per-unit handlers would filter the PIDs anyway a second time.
With this change we'll hence relax the rules: all we do now is
dispatch every SIGCHLD event exactly once to each unit interested in
it, and it's up to the unit to then use or ignore this. We use a
generation counter in the unit to ensure that we only invoke the unit
handler once for each event, protecting us from confusion if a unit is
both associated with a specific PID through cgroup membership and
through the "watch_pids" logic. It also protects us from being
confused if the "watch_pids" hashmap is altered while we are
dispatching to it (which is a very likely case).
- sd_notify() message dispatching has been reworked to be very similar
to SIGCHLD handling now. A generation counter is used for dispatching
as well.
This also adds a new test that validates that "watch_pid" registration
and unregstration works correctly.
2018-01-12 13:41:05 +01:00
|
|
|
/* OK, the "pid" key is already assigned to a different unit. Let's see if the "-pid" key (which points
|
|
|
|
* to an array of Units rather than just a Unit), lists us already. */
|
2014-02-06 17:17:51 +01:00
|
|
|
|
core: rework how we track which PIDs to watch for a unit
Previously, we'd maintain two hashmaps keyed by PIDs, pointing to Unit
interested in SIGCHLD events for them. This scheme allowed a specific
PID to be watched by exactly 0, 1 or 2 units.
With this rework this is replaced by a single hashmap which is primarily
keyed by the PID and points to a Unit interested in it. However, it
optionally also keyed by the negated PID, in which case it points to a
NULL terminated array of additional Unit objects also interested. This
scheme means arbitrary numbers of Units may now watch the same PID.
Runtime and memory behaviour should not be impact by this change, as for
the common case (i.e. each PID only watched by a single unit) behaviour
stays the same, but for the uncommon case (a PID watched by more than
one unit) we only pay with a single additional memory allocation for the
array.
Why this all? Primarily, because allowing exactly two units to watch a
specific PID is not sufficient for some niche cases, as processes can
belong to more than one unit these days:
1. sd_notify() with MAINPID= can be used to attach a process from a
different cgroup to multiple units.
2. Similar, the PIDFile= setting in unit files can be used for similar
setups,
3. By creating a scope unit a main process of a service may join a
different unit, too.
4. On cgroupsv1 we frequently end up watching all processes remaining in
a scope, and if a process opens lots of scopes one after the other it
might thus end up being watch by many of them.
This patch hence removes the 2-unit-per-PID limit. It also makes a
couple of other changes, some of them quite relevant:
- manager_get_unit_by_pid() (and the bus call wrapping it) when there's
ambiguity will prefer returning the Unit the process belongs to based on
cgroup membership, and only check the watch-pids hashmap if that
fails. This change in logic is probably more in line with what people
expect and makes things more stable as each process can belong to
exactly one cgroup only.
- Every SIGCHLD event is now dispatched to all units interested in its
PID. Previously, there was some magic conditionalization: the SIGCHLD
would only be dispatched to the unit if it was only interested in a
single PID only, or the PID belonged to the control or main PID or we
didn't dispatch a signle SIGCHLD to the unit in the current event loop
iteration yet. These rules were quite arbitrary and also redundant as
the the per-unit handlers would filter the PIDs anyway a second time.
With this change we'll hence relax the rules: all we do now is
dispatch every SIGCHLD event exactly once to each unit interested in
it, and it's up to the unit to then use or ignore this. We use a
generation counter in the unit to ensure that we only invoke the unit
handler once for each event, protecting us from confusion if a unit is
both associated with a specific PID through cgroup membership and
through the "watch_pids" logic. It also protects us from being
confused if the "watch_pids" hashmap is altered while we are
dispatching to it (which is a very likely case).
- sd_notify() message dispatching has been reworked to be very similar
to SIGCHLD handling now. A generation counter is used for dispatching
as well.
This also adds a new test that validates that "watch_pid" registration
and unregstration works correctly.
2018-01-12 13:41:05 +01:00
|
|
|
array = hashmap_get(u->manager->watch_pids, PID_TO_PTR(-pid));
|
|
|
|
if (array)
|
|
|
|
for (; array[n]; n++)
|
|
|
|
if (array[n] == u)
|
|
|
|
found = true;
|
2014-02-06 17:17:51 +01:00
|
|
|
|
core: rework how we track which PIDs to watch for a unit
Previously, we'd maintain two hashmaps keyed by PIDs, pointing to Unit
interested in SIGCHLD events for them. This scheme allowed a specific
PID to be watched by exactly 0, 1 or 2 units.
With this rework this is replaced by a single hashmap which is primarily
keyed by the PID and points to a Unit interested in it. However, it
optionally also keyed by the negated PID, in which case it points to a
NULL terminated array of additional Unit objects also interested. This
scheme means arbitrary numbers of Units may now watch the same PID.
Runtime and memory behaviour should not be impact by this change, as for
the common case (i.e. each PID only watched by a single unit) behaviour
stays the same, but for the uncommon case (a PID watched by more than
one unit) we only pay with a single additional memory allocation for the
array.
Why this all? Primarily, because allowing exactly two units to watch a
specific PID is not sufficient for some niche cases, as processes can
belong to more than one unit these days:
1. sd_notify() with MAINPID= can be used to attach a process from a
different cgroup to multiple units.
2. Similar, the PIDFile= setting in unit files can be used for similar
setups,
3. By creating a scope unit a main process of a service may join a
different unit, too.
4. On cgroupsv1 we frequently end up watching all processes remaining in
a scope, and if a process opens lots of scopes one after the other it
might thus end up being watch by many of them.
This patch hence removes the 2-unit-per-PID limit. It also makes a
couple of other changes, some of them quite relevant:
- manager_get_unit_by_pid() (and the bus call wrapping it) when there's
ambiguity will prefer returning the Unit the process belongs to based on
cgroup membership, and only check the watch-pids hashmap if that
fails. This change in logic is probably more in line with what people
expect and makes things more stable as each process can belong to
exactly one cgroup only.
- Every SIGCHLD event is now dispatched to all units interested in its
PID. Previously, there was some magic conditionalization: the SIGCHLD
would only be dispatched to the unit if it was only interested in a
single PID only, or the PID belonged to the control or main PID or we
didn't dispatch a signle SIGCHLD to the unit in the current event loop
iteration yet. These rules were quite arbitrary and also redundant as
the the per-unit handlers would filter the PIDs anyway a second time.
With this change we'll hence relax the rules: all we do now is
dispatch every SIGCHLD event exactly once to each unit interested in
it, and it's up to the unit to then use or ignore this. We use a
generation counter in the unit to ensure that we only invoke the unit
handler once for each event, protecting us from confusion if a unit is
both associated with a specific PID through cgroup membership and
through the "watch_pids" logic. It also protects us from being
confused if the "watch_pids" hashmap is altered while we are
dispatching to it (which is a very likely case).
- sd_notify() message dispatching has been reworked to be very similar
to SIGCHLD handling now. A generation counter is used for dispatching
as well.
This also adds a new test that validates that "watch_pid" registration
and unregstration works correctly.
2018-01-12 13:41:05 +01:00
|
|
|
if (found) /* Found it already? if so, do nothing */
|
|
|
|
r = 0;
|
|
|
|
else {
|
|
|
|
Unit **new_array;
|
|
|
|
|
|
|
|
/* Allocate a new array */
|
|
|
|
new_array = new(Unit*, n + 2);
|
|
|
|
if (!new_array)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
memcpy_safe(new_array, array, sizeof(Unit*) * n);
|
|
|
|
new_array[n] = u;
|
|
|
|
new_array[n+1] = NULL;
|
|
|
|
|
|
|
|
/* Add or replace the old array */
|
|
|
|
r = hashmap_replace(u->manager->watch_pids, PID_TO_PTR(-pid), new_array);
|
|
|
|
if (r < 0) {
|
|
|
|
free(new_array);
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
free(array);
|
|
|
|
}
|
|
|
|
} else if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
r = set_put(u->pids, PID_TO_PTR(pid));
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
return 0;
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
void unit_unwatch_pid(Unit *u, pid_t pid) {
|
core: rework how we track which PIDs to watch for a unit
Previously, we'd maintain two hashmaps keyed by PIDs, pointing to Unit
interested in SIGCHLD events for them. This scheme allowed a specific
PID to be watched by exactly 0, 1 or 2 units.
With this rework this is replaced by a single hashmap which is primarily
keyed by the PID and points to a Unit interested in it. However, it
optionally also keyed by the negated PID, in which case it points to a
NULL terminated array of additional Unit objects also interested. This
scheme means arbitrary numbers of Units may now watch the same PID.
Runtime and memory behaviour should not be impact by this change, as for
the common case (i.e. each PID only watched by a single unit) behaviour
stays the same, but for the uncommon case (a PID watched by more than
one unit) we only pay with a single additional memory allocation for the
array.
Why this all? Primarily, because allowing exactly two units to watch a
specific PID is not sufficient for some niche cases, as processes can
belong to more than one unit these days:
1. sd_notify() with MAINPID= can be used to attach a process from a
different cgroup to multiple units.
2. Similar, the PIDFile= setting in unit files can be used for similar
setups,
3. By creating a scope unit a main process of a service may join a
different unit, too.
4. On cgroupsv1 we frequently end up watching all processes remaining in
a scope, and if a process opens lots of scopes one after the other it
might thus end up being watch by many of them.
This patch hence removes the 2-unit-per-PID limit. It also makes a
couple of other changes, some of them quite relevant:
- manager_get_unit_by_pid() (and the bus call wrapping it) when there's
ambiguity will prefer returning the Unit the process belongs to based on
cgroup membership, and only check the watch-pids hashmap if that
fails. This change in logic is probably more in line with what people
expect and makes things more stable as each process can belong to
exactly one cgroup only.
- Every SIGCHLD event is now dispatched to all units interested in its
PID. Previously, there was some magic conditionalization: the SIGCHLD
would only be dispatched to the unit if it was only interested in a
single PID only, or the PID belonged to the control or main PID or we
didn't dispatch a signle SIGCHLD to the unit in the current event loop
iteration yet. These rules were quite arbitrary and also redundant as
the the per-unit handlers would filter the PIDs anyway a second time.
With this change we'll hence relax the rules: all we do now is
dispatch every SIGCHLD event exactly once to each unit interested in
it, and it's up to the unit to then use or ignore this. We use a
generation counter in the unit to ensure that we only invoke the unit
handler once for each event, protecting us from confusion if a unit is
both associated with a specific PID through cgroup membership and
through the "watch_pids" logic. It also protects us from being
confused if the "watch_pids" hashmap is altered while we are
dispatching to it (which is a very likely case).
- sd_notify() message dispatching has been reworked to be very similar
to SIGCHLD handling now. A generation counter is used for dispatching
as well.
This also adds a new test that validates that "watch_pid" registration
and unregstration works correctly.
2018-01-12 13:41:05 +01:00
|
|
|
Unit **array;
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
assert(u);
|
core: rework how we track which PIDs to watch for a unit
Previously, we'd maintain two hashmaps keyed by PIDs, pointing to Unit
interested in SIGCHLD events for them. This scheme allowed a specific
PID to be watched by exactly 0, 1 or 2 units.
With this rework this is replaced by a single hashmap which is primarily
keyed by the PID and points to a Unit interested in it. However, it
optionally also keyed by the negated PID, in which case it points to a
NULL terminated array of additional Unit objects also interested. This
scheme means arbitrary numbers of Units may now watch the same PID.
Runtime and memory behaviour should not be impact by this change, as for
the common case (i.e. each PID only watched by a single unit) behaviour
stays the same, but for the uncommon case (a PID watched by more than
one unit) we only pay with a single additional memory allocation for the
array.
Why this all? Primarily, because allowing exactly two units to watch a
specific PID is not sufficient for some niche cases, as processes can
belong to more than one unit these days:
1. sd_notify() with MAINPID= can be used to attach a process from a
different cgroup to multiple units.
2. Similar, the PIDFile= setting in unit files can be used for similar
setups,
3. By creating a scope unit a main process of a service may join a
different unit, too.
4. On cgroupsv1 we frequently end up watching all processes remaining in
a scope, and if a process opens lots of scopes one after the other it
might thus end up being watch by many of them.
This patch hence removes the 2-unit-per-PID limit. It also makes a
couple of other changes, some of them quite relevant:
- manager_get_unit_by_pid() (and the bus call wrapping it) when there's
ambiguity will prefer returning the Unit the process belongs to based on
cgroup membership, and only check the watch-pids hashmap if that
fails. This change in logic is probably more in line with what people
expect and makes things more stable as each process can belong to
exactly one cgroup only.
- Every SIGCHLD event is now dispatched to all units interested in its
PID. Previously, there was some magic conditionalization: the SIGCHLD
would only be dispatched to the unit if it was only interested in a
single PID only, or the PID belonged to the control or main PID or we
didn't dispatch a signle SIGCHLD to the unit in the current event loop
iteration yet. These rules were quite arbitrary and also redundant as
the the per-unit handlers would filter the PIDs anyway a second time.
With this change we'll hence relax the rules: all we do now is
dispatch every SIGCHLD event exactly once to each unit interested in
it, and it's up to the unit to then use or ignore this. We use a
generation counter in the unit to ensure that we only invoke the unit
handler once for each event, protecting us from confusion if a unit is
both associated with a specific PID through cgroup membership and
through the "watch_pids" logic. It also protects us from being
confused if the "watch_pids" hashmap is altered while we are
dispatching to it (which is a very likely case).
- sd_notify() message dispatching has been reworked to be very similar
to SIGCHLD handling now. A generation counter is used for dispatching
as well.
This also adds a new test that validates that "watch_pid" registration
and unregstration works correctly.
2018-01-12 13:41:05 +01:00
|
|
|
assert(pid_is_valid(pid));
|
|
|
|
|
|
|
|
/* First let's drop the unit in case it's keyed as "pid". */
|
|
|
|
(void) hashmap_remove_value(u->manager->watch_pids, PID_TO_PTR(pid), u);
|
|
|
|
|
|
|
|
/* Then, let's also drop the unit, in case it's in the array keyed by -pid */
|
|
|
|
array = hashmap_get(u->manager->watch_pids, PID_TO_PTR(-pid));
|
|
|
|
if (array) {
|
|
|
|
size_t n, m = 0;
|
|
|
|
|
|
|
|
/* Let's iterate through the array, dropping our own entry */
|
|
|
|
for (n = 0; array[n]; n++)
|
|
|
|
if (array[n] != u)
|
|
|
|
array[m++] = array[n];
|
|
|
|
array[m] = NULL;
|
|
|
|
|
|
|
|
if (m == 0) {
|
|
|
|
/* The array is now empty, remove the entire entry */
|
2020-05-08 13:32:11 +02:00
|
|
|
assert_se(hashmap_remove(u->manager->watch_pids, PID_TO_PTR(-pid)) == array);
|
core: rework how we track which PIDs to watch for a unit
Previously, we'd maintain two hashmaps keyed by PIDs, pointing to Unit
interested in SIGCHLD events for them. This scheme allowed a specific
PID to be watched by exactly 0, 1 or 2 units.
With this rework this is replaced by a single hashmap which is primarily
keyed by the PID and points to a Unit interested in it. However, it
optionally also keyed by the negated PID, in which case it points to a
NULL terminated array of additional Unit objects also interested. This
scheme means arbitrary numbers of Units may now watch the same PID.
Runtime and memory behaviour should not be impact by this change, as for
the common case (i.e. each PID only watched by a single unit) behaviour
stays the same, but for the uncommon case (a PID watched by more than
one unit) we only pay with a single additional memory allocation for the
array.
Why this all? Primarily, because allowing exactly two units to watch a
specific PID is not sufficient for some niche cases, as processes can
belong to more than one unit these days:
1. sd_notify() with MAINPID= can be used to attach a process from a
different cgroup to multiple units.
2. Similar, the PIDFile= setting in unit files can be used for similar
setups,
3. By creating a scope unit a main process of a service may join a
different unit, too.
4. On cgroupsv1 we frequently end up watching all processes remaining in
a scope, and if a process opens lots of scopes one after the other it
might thus end up being watch by many of them.
This patch hence removes the 2-unit-per-PID limit. It also makes a
couple of other changes, some of them quite relevant:
- manager_get_unit_by_pid() (and the bus call wrapping it) when there's
ambiguity will prefer returning the Unit the process belongs to based on
cgroup membership, and only check the watch-pids hashmap if that
fails. This change in logic is probably more in line with what people
expect and makes things more stable as each process can belong to
exactly one cgroup only.
- Every SIGCHLD event is now dispatched to all units interested in its
PID. Previously, there was some magic conditionalization: the SIGCHLD
would only be dispatched to the unit if it was only interested in a
single PID only, or the PID belonged to the control or main PID or we
didn't dispatch a signle SIGCHLD to the unit in the current event loop
iteration yet. These rules were quite arbitrary and also redundant as
the the per-unit handlers would filter the PIDs anyway a second time.
With this change we'll hence relax the rules: all we do now is
dispatch every SIGCHLD event exactly once to each unit interested in
it, and it's up to the unit to then use or ignore this. We use a
generation counter in the unit to ensure that we only invoke the unit
handler once for each event, protecting us from confusion if a unit is
both associated with a specific PID through cgroup membership and
through the "watch_pids" logic. It also protects us from being
confused if the "watch_pids" hashmap is altered while we are
dispatching to it (which is a very likely case).
- sd_notify() message dispatching has been reworked to be very similar
to SIGCHLD handling now. A generation counter is used for dispatching
as well.
This also adds a new test that validates that "watch_pid" registration
and unregstration works correctly.
2018-01-12 13:41:05 +01:00
|
|
|
free(array);
|
|
|
|
}
|
|
|
|
}
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2015-09-03 13:22:51 +02:00
|
|
|
(void) set_remove(u->pids, PID_TO_PTR(pid));
|
2014-02-06 17:17:51 +01:00
|
|
|
}
|
|
|
|
|
2014-03-06 02:19:42 +01:00
|
|
|
void unit_unwatch_all_pids(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
while (!set_isempty(u->pids))
|
2015-09-03 13:22:51 +02:00
|
|
|
unit_unwatch_pid(u, PTR_TO_PID(set_first(u->pids)));
|
2014-03-06 02:19:42 +01:00
|
|
|
|
core: unified cgroup hierarchy support
This patch set adds full support the new unified cgroup hierarchy logic
of modern kernels.
A new kernel command line option "systemd.unified_cgroup_hierarchy=1" is
added. If specified the unified hierarchy is mounted to /sys/fs/cgroup
instead of a tmpfs. No further hierarchies are mounted. The kernel
command line option defaults to off. We can turn it on by default as
soon as the kernel's APIs regarding this are stabilized (but even then
downstream distros might want to turn this off, as this will break any
tools that access cgroupfs directly).
It is possibly to choose for each boot individually whether the unified
or the legacy hierarchy is used. nspawn will by default provide the
legacy hierarchy to containers if the host is using it, and the unified
otherwise. However it is possible to run containers with the unified
hierarchy on a legacy host and vice versa, by setting the
$UNIFIED_CGROUP_HIERARCHY environment variable for nspawn to 1 or 0,
respectively.
The unified hierarchy provides reliable cgroup empty notifications for
the first time, via inotify. To make use of this we maintain one
manager-wide inotify fd, and each cgroup to it.
This patch also removes cg_delete() which is unused now.
On kernel 4.2 only the "memory" controller is compatible with the
unified hierarchy, hence that's the only controller systemd exposes when
booted in unified heirarchy mode.
This introduces a new enum for enumerating supported controllers, plus a
related enum for the mask bits mapping to it. The core is changed to
make use of this everywhere.
This moves PID 1 into a new "init.scope" implicit scope unit in the root
slice. This is necessary since on the unified hierarchy cgroups may
either contain subgroups or processes but not both. PID 1 hence has to
move out of the root cgroup (strictly speaking the root cgroup is the
only one where processes and subgroups are still allowed, but in order
to support containers nicey, we move PID 1 into the new scope in all
cases.) This new unit is also used on legacy hierarchy setups. It's
actually pretty useful on all systems, as it can then be used to filter
journal messages coming from PID 1, and so on.
The root slice ("-.slice") is now implicitly created and started (and
does not require a unit file on disk anymore), since
that's where "init.scope" is located and the slice needs to be started
before the scope can.
To check whether we are in unified or legacy hierarchy mode we use
statfs() on /sys/fs/cgroup. If the .f_type field reports tmpfs we are in
legacy mode, if it reports cgroupfs we are in unified mode.
This patch set carefuly makes sure that cgls and cgtop continue to work
as desired.
When invoking nspawn as a service it will implicitly create two
subcgroups in the cgroup it is using, one to move the nspawn process
into, the other to move the actual container processes into. This is
done because of the requirement that cgroups may either contain
processes or other subgroups.
2015-09-01 19:22:36 +02:00
|
|
|
u->pids = set_free(u->pids);
|
2014-02-06 17:17:51 +01:00
|
|
|
}
|
|
|
|
|
core: rework how we track service and scope PIDs
This reworks how systemd tracks processes on cgroupv1 systems where
cgroup notification is not reliable. Previously, whenever we had reason
to believe that new processes showed up or got removed we'd scan the
cgroup of the scope or service unit for new processes, and would tidy up
the list of PIDs previously watched. This scanning is relatively slow,
and does not scale well. With this change behaviour is changed: instead
of scanning for new/removed processes right away we do this work in a
per-unit deferred event loop job. This event source is scheduled at a
very low priority, so that it is executed when we have time but does not
starve other event sources. This has two benefits: this expensive work is
coalesced, if events happen in quick succession, and we won't delay
SIGCHLD handling for too long.
This patch basically replaces all direct invocation of
unit_watch_all_pids() in scope.c and service.c with invocations of the
new unit_enqueue_rewatch_pids() call which just enqueues a request of
watching/tidying up the PID sets (with one exception: in
scope_enter_signal() and service_enter_signal() we'll still do
unit_watch_all_pids() synchronously first, since we really want to know
all processes we are about to kill so that we can track them properly.
Moreover, all direct invocations of unit_tidy_watch_pids() and
unit_synthesize_cgroup_empty_event() are removed too, when the
unit_enqueue_rewatch_pids() call is invoked, as the queued job will run
those operations too.
All of this is done on cgroupsv1 systems only, and is disabled on
cgroupsv2 systems as cgroup-empty notifications are reliable there, and
we do not need SIGCHLD events to track processes there.
Fixes: #9138
2018-05-31 15:41:59 +02:00
|
|
|
static void unit_tidy_watch_pids(Unit *u) {
|
|
|
|
pid_t except1, except2;
|
2014-02-06 17:17:51 +01:00
|
|
|
Iterator i;
|
|
|
|
void *e;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Cleans dead PIDs from our list */
|
|
|
|
|
core: rework how we track service and scope PIDs
This reworks how systemd tracks processes on cgroupv1 systems where
cgroup notification is not reliable. Previously, whenever we had reason
to believe that new processes showed up or got removed we'd scan the
cgroup of the scope or service unit for new processes, and would tidy up
the list of PIDs previously watched. This scanning is relatively slow,
and does not scale well. With this change behaviour is changed: instead
of scanning for new/removed processes right away we do this work in a
per-unit deferred event loop job. This event source is scheduled at a
very low priority, so that it is executed when we have time but does not
starve other event sources. This has two benefits: this expensive work is
coalesced, if events happen in quick succession, and we won't delay
SIGCHLD handling for too long.
This patch basically replaces all direct invocation of
unit_watch_all_pids() in scope.c and service.c with invocations of the
new unit_enqueue_rewatch_pids() call which just enqueues a request of
watching/tidying up the PID sets (with one exception: in
scope_enter_signal() and service_enter_signal() we'll still do
unit_watch_all_pids() synchronously first, since we really want to know
all processes we are about to kill so that we can track them properly.
Moreover, all direct invocations of unit_tidy_watch_pids() and
unit_synthesize_cgroup_empty_event() are removed too, when the
unit_enqueue_rewatch_pids() call is invoked, as the queued job will run
those operations too.
All of this is done on cgroupsv1 systems only, and is disabled on
cgroupsv2 systems as cgroup-empty notifications are reliable there, and
we do not need SIGCHLD events to track processes there.
Fixes: #9138
2018-05-31 15:41:59 +02:00
|
|
|
except1 = unit_main_pid(u);
|
|
|
|
except2 = unit_control_pid(u);
|
|
|
|
|
2014-02-06 17:17:51 +01:00
|
|
|
SET_FOREACH(e, u->pids, i) {
|
2015-09-03 13:22:51 +02:00
|
|
|
pid_t pid = PTR_TO_PID(e);
|
2014-02-06 17:17:51 +01:00
|
|
|
|
|
|
|
if (pid == except1 || pid == except2)
|
|
|
|
continue;
|
|
|
|
|
2014-02-17 18:28:53 +01:00
|
|
|
if (!pid_is_unwaited(pid))
|
2014-03-06 02:19:42 +01:00
|
|
|
unit_unwatch_pid(u, pid);
|
2014-02-06 17:17:51 +01:00
|
|
|
}
|
2010-01-26 21:39:06 +01:00
|
|
|
}
|
|
|
|
|
core: rework how we track service and scope PIDs
This reworks how systemd tracks processes on cgroupv1 systems where
cgroup notification is not reliable. Previously, whenever we had reason
to believe that new processes showed up or got removed we'd scan the
cgroup of the scope or service unit for new processes, and would tidy up
the list of PIDs previously watched. This scanning is relatively slow,
and does not scale well. With this change behaviour is changed: instead
of scanning for new/removed processes right away we do this work in a
per-unit deferred event loop job. This event source is scheduled at a
very low priority, so that it is executed when we have time but does not
starve other event sources. This has two benefits: this expensive work is
coalesced, if events happen in quick succession, and we won't delay
SIGCHLD handling for too long.
This patch basically replaces all direct invocation of
unit_watch_all_pids() in scope.c and service.c with invocations of the
new unit_enqueue_rewatch_pids() call which just enqueues a request of
watching/tidying up the PID sets (with one exception: in
scope_enter_signal() and service_enter_signal() we'll still do
unit_watch_all_pids() synchronously first, since we really want to know
all processes we are about to kill so that we can track them properly.
Moreover, all direct invocations of unit_tidy_watch_pids() and
unit_synthesize_cgroup_empty_event() are removed too, when the
unit_enqueue_rewatch_pids() call is invoked, as the queued job will run
those operations too.
All of this is done on cgroupsv1 systems only, and is disabled on
cgroupsv2 systems as cgroup-empty notifications are reliable there, and
we do not need SIGCHLD events to track processes there.
Fixes: #9138
2018-05-31 15:41:59 +02:00
|
|
|
static int on_rewatch_pids_event(sd_event_source *s, void *userdata) {
|
|
|
|
Unit *u = userdata;
|
|
|
|
|
|
|
|
assert(s);
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
unit_tidy_watch_pids(u);
|
|
|
|
unit_watch_all_pids(u);
|
|
|
|
|
|
|
|
/* If the PID set is empty now, then let's finish this off. */
|
|
|
|
unit_synthesize_cgroup_empty_event(u);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_enqueue_rewatch_pids(Unit *u) {
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (!u->cgroup_path)
|
|
|
|
return -ENOENT;
|
|
|
|
|
|
|
|
r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
if (r > 0) /* On unified we can use proper notifications */
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* Enqueues a low-priority job that will clean up dead PIDs from our list of PIDs to watch and subscribe to new
|
|
|
|
* PIDs that might have appeared. We do this in a delayed job because the work might be quite slow, as it
|
|
|
|
* involves issuing kill(pid, 0) on all processes we watch. */
|
|
|
|
|
|
|
|
if (!u->rewatch_pids_event_source) {
|
|
|
|
_cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL;
|
|
|
|
|
|
|
|
r = sd_event_add_defer(u->manager->event, &s, on_rewatch_pids_event, u);
|
|
|
|
if (r < 0)
|
|
|
|
return log_error_errno(r, "Failed to allocate event source for tidying watched PIDs: %m");
|
|
|
|
|
|
|
|
r = sd_event_source_set_priority(s, SD_EVENT_PRIORITY_IDLE);
|
|
|
|
if (r < 0)
|
2019-09-25 00:31:27 +02:00
|
|
|
return log_error_errno(r, "Failed to adjust priority of event source for tidying watched PIDs: %m");
|
core: rework how we track service and scope PIDs
This reworks how systemd tracks processes on cgroupv1 systems where
cgroup notification is not reliable. Previously, whenever we had reason
to believe that new processes showed up or got removed we'd scan the
cgroup of the scope or service unit for new processes, and would tidy up
the list of PIDs previously watched. This scanning is relatively slow,
and does not scale well. With this change behaviour is changed: instead
of scanning for new/removed processes right away we do this work in a
per-unit deferred event loop job. This event source is scheduled at a
very low priority, so that it is executed when we have time but does not
starve other event sources. This has two benefits: this expensive work is
coalesced, if events happen in quick succession, and we won't delay
SIGCHLD handling for too long.
This patch basically replaces all direct invocation of
unit_watch_all_pids() in scope.c and service.c with invocations of the
new unit_enqueue_rewatch_pids() call which just enqueues a request of
watching/tidying up the PID sets (with one exception: in
scope_enter_signal() and service_enter_signal() we'll still do
unit_watch_all_pids() synchronously first, since we really want to know
all processes we are about to kill so that we can track them properly.
Moreover, all direct invocations of unit_tidy_watch_pids() and
unit_synthesize_cgroup_empty_event() are removed too, when the
unit_enqueue_rewatch_pids() call is invoked, as the queued job will run
those operations too.
All of this is done on cgroupsv1 systems only, and is disabled on
cgroupsv2 systems as cgroup-empty notifications are reliable there, and
we do not need SIGCHLD events to track processes there.
Fixes: #9138
2018-05-31 15:41:59 +02:00
|
|
|
|
|
|
|
(void) sd_event_source_set_description(s, "tidy-watch-pids");
|
|
|
|
|
|
|
|
u->rewatch_pids_event_source = TAKE_PTR(s);
|
|
|
|
}
|
|
|
|
|
|
|
|
r = sd_event_source_set_enabled(u->rewatch_pids_event_source, SD_EVENT_ONESHOT);
|
|
|
|
if (r < 0)
|
|
|
|
return log_error_errno(r, "Failed to enable event source for tidying watched PIDs: %m");
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
void unit_dequeue_rewatch_pids(Unit *u) {
|
|
|
|
int r;
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (!u->rewatch_pids_event_source)
|
|
|
|
return;
|
|
|
|
|
|
|
|
r = sd_event_source_set_enabled(u->rewatch_pids_event_source, SD_EVENT_OFF);
|
|
|
|
if (r < 0)
|
|
|
|
log_warning_errno(r, "Failed to disable event source for tidying watched PIDs, ignoring: %m");
|
|
|
|
|
|
|
|
u->rewatch_pids_event_source = sd_event_source_unref(u->rewatch_pids_event_source);
|
|
|
|
}
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
bool unit_job_is_applicable(Unit *u, JobType j) {
|
|
|
|
assert(u);
|
|
|
|
assert(j >= 0 && j < _JOB_TYPE_MAX);
|
|
|
|
|
|
|
|
switch (j) {
|
|
|
|
|
|
|
|
case JOB_VERIFY_ACTIVE:
|
|
|
|
case JOB_START:
|
core: add NOP jobs, job type collapsing
Two of our current job types are special:
JOB_TRY_RESTART, JOB_RELOAD_OR_START.
They differ from other job types by being sensitive to the unit active state.
They perform some action when the unit is active and some other action
otherwise. This raises a question: when exactly should the unit state be
checked to make the decision?
Currently the unit state is checked when the job becomes runnable. It's more
sensible to check the state immediately when the job is added by the user.
When the user types "systemctl try-restart foo.service", he really intends
to restart the service if it's running right now. If it isn't running right
now, the restart is pointless.
Consider the example (from Bugzilla[1]):
sleep.service takes some time to start.
hello.service has After=sleep.service.
Both services get started. Two jobs will appear:
hello.service/start waiting
sleep.service/start running
Then someone runs "systemctl try-restart hello.service".
Currently the try-restart operation will block and wait for
sleep.service/start to complete.
The correct result is to complete the try-restart operation immediately
with success, because hello.service is not running. The two original
jobs must not be disturbed by this.
To fix this we introduce two new concepts:
- a new job type: JOB_NOP
A JOB_NOP job does not do anything to the unit. It does not pull in any
dependencies. It is always immediately runnable. When installed to a unit,
it sits in a special slot (u->nop_job) where it never conflicts with
the installed job (u->job) of a different type. It never merges with jobs
of other types, but it can merge into an already installed JOB_NOP job.
- "collapsing" of job types
When a job of one of the two special types is added, the state of the unit
is checked immediately and the job type changes:
JOB_TRY_RESTART -> JOB_RESTART or JOB_NOP
JOB_RELOAD_OR_START -> JOB_RELOAD or JOB_START
Should a job type JOB_RELOAD_OR_START appear later during job merging, it
collapses immediately afterwards.
Collapsing actually makes some things simpler, because there are now fewer
job types that are allowed in the transaction.
[1] Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=753586
2012-04-25 11:58:27 +02:00
|
|
|
case JOB_NOP:
|
2016-10-24 21:41:54 +02:00
|
|
|
/* Note that we don't check unit_can_start() here. That's because .device units and suchlike are not
|
2020-04-21 20:46:53 +02:00
|
|
|
* startable by us but may appear due to external events, and it thus makes sense to permit enqueuing
|
2016-10-24 21:41:54 +02:00
|
|
|
* jobs for it. */
|
2010-01-26 21:39:06 +01:00
|
|
|
return true;
|
|
|
|
|
2016-10-24 21:41:54 +02:00
|
|
|
case JOB_STOP:
|
|
|
|
/* Similar as above. However, perpetual units can never be stopped (neither explicitly nor due to
|
2020-04-21 20:46:53 +02:00
|
|
|
* external events), hence it makes no sense to permit enqueuing such a request either. */
|
2016-10-24 21:41:54 +02:00
|
|
|
return !u->perpetual;
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
case JOB_RESTART:
|
|
|
|
case JOB_TRY_RESTART:
|
2016-10-24 21:41:54 +02:00
|
|
|
return unit_can_stop(u) && unit_can_start(u);
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
case JOB_RELOAD:
|
2016-01-28 18:48:42 +01:00
|
|
|
case JOB_TRY_RELOAD:
|
2010-01-26 21:39:06 +01:00
|
|
|
return unit_can_reload(u);
|
|
|
|
|
|
|
|
case JOB_RELOAD_OR_START:
|
|
|
|
return unit_can_reload(u) && unit_can_start(u);
|
|
|
|
|
|
|
|
default:
|
|
|
|
assert_not_reached("Invalid job type");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
static void maybe_warn_about_dependency(Unit *u, const char *other, UnitDependency dependency) {
|
|
|
|
assert(u);
|
2014-08-08 02:46:49 +02:00
|
|
|
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
/* Only warn about some unit types */
|
|
|
|
if (!IN_SET(dependency, UNIT_CONFLICTS, UNIT_CONFLICTED_BY, UNIT_BEFORE, UNIT_AFTER, UNIT_ON_FAILURE, UNIT_TRIGGERS, UNIT_TRIGGERED_BY))
|
|
|
|
return;
|
2014-08-18 22:25:24 +02:00
|
|
|
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
if (streq_ptr(u->id, other))
|
|
|
|
log_unit_warning(u, "Dependency %s=%s dropped", unit_dependency_to_string(dependency), u->id);
|
|
|
|
else
|
|
|
|
log_unit_warning(u, "Dependency %s=%s dropped, merged into %s", unit_dependency_to_string(dependency), strna(other), u->id);
|
2014-08-08 02:46:49 +02:00
|
|
|
}
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
static int unit_add_dependency_hashmap(
|
|
|
|
Hashmap **h,
|
|
|
|
Unit *other,
|
|
|
|
UnitDependencyMask origin_mask,
|
|
|
|
UnitDependencyMask destination_mask) {
|
|
|
|
|
|
|
|
UnitDependencyInfo info;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(h);
|
|
|
|
assert(other);
|
|
|
|
assert(origin_mask < _UNIT_DEPENDENCY_MASK_FULL);
|
|
|
|
assert(destination_mask < _UNIT_DEPENDENCY_MASK_FULL);
|
|
|
|
assert(origin_mask > 0 || destination_mask > 0);
|
|
|
|
|
|
|
|
r = hashmap_ensure_allocated(h, NULL);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
assert_cc(sizeof(void*) == sizeof(info));
|
|
|
|
|
|
|
|
info.data = hashmap_get(*h, other);
|
|
|
|
if (info.data) {
|
|
|
|
/* Entry already exists. Add in our mask. */
|
|
|
|
|
2018-04-20 15:36:20 +02:00
|
|
|
if (FLAGS_SET(origin_mask, info.origin_mask) &&
|
|
|
|
FLAGS_SET(destination_mask, info.destination_mask))
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
return 0; /* NOP */
|
|
|
|
|
|
|
|
info.origin_mask |= origin_mask;
|
|
|
|
info.destination_mask |= destination_mask;
|
|
|
|
|
|
|
|
r = hashmap_update(*h, other, info.data);
|
|
|
|
} else {
|
|
|
|
info = (UnitDependencyInfo) {
|
|
|
|
.origin_mask = origin_mask,
|
|
|
|
.destination_mask = destination_mask,
|
|
|
|
};
|
|
|
|
|
|
|
|
r = hashmap_put(*h, other, info.data);
|
|
|
|
}
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_add_dependency(
|
|
|
|
Unit *u,
|
|
|
|
UnitDependency d,
|
|
|
|
Unit *other,
|
|
|
|
bool add_reference,
|
|
|
|
UnitDependencyMask mask) {
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
static const UnitDependency inverse_table[_UNIT_DEPENDENCY_MAX] = {
|
|
|
|
[UNIT_REQUIRES] = UNIT_REQUIRED_BY,
|
|
|
|
[UNIT_WANTS] = UNIT_WANTED_BY,
|
2015-05-19 01:24:28 +02:00
|
|
|
[UNIT_REQUISITE] = UNIT_REQUISITE_OF,
|
2012-07-13 23:34:40 +02:00
|
|
|
[UNIT_BINDS_TO] = UNIT_BOUND_BY,
|
2012-07-26 09:42:26 +02:00
|
|
|
[UNIT_PART_OF] = UNIT_CONSISTS_OF,
|
2015-05-19 01:24:28 +02:00
|
|
|
[UNIT_REQUIRED_BY] = UNIT_REQUIRES,
|
|
|
|
[UNIT_REQUISITE_OF] = UNIT_REQUISITE,
|
|
|
|
[UNIT_WANTED_BY] = UNIT_WANTS,
|
2012-07-13 23:34:40 +02:00
|
|
|
[UNIT_BOUND_BY] = UNIT_BINDS_TO,
|
2012-07-26 09:42:26 +02:00
|
|
|
[UNIT_CONSISTS_OF] = UNIT_PART_OF,
|
2010-08-09 22:32:30 +02:00
|
|
|
[UNIT_CONFLICTS] = UNIT_CONFLICTED_BY,
|
|
|
|
[UNIT_CONFLICTED_BY] = UNIT_CONFLICTS,
|
2010-01-26 21:39:06 +01:00
|
|
|
[UNIT_BEFORE] = UNIT_AFTER,
|
2010-04-21 06:01:13 +02:00
|
|
|
[UNIT_AFTER] = UNIT_BEFORE,
|
2010-07-17 00:58:47 +02:00
|
|
|
[UNIT_ON_FAILURE] = _UNIT_DEPENDENCY_INVALID,
|
2010-04-21 06:01:13 +02:00
|
|
|
[UNIT_REFERENCES] = UNIT_REFERENCED_BY,
|
2012-01-06 23:08:54 +01:00
|
|
|
[UNIT_REFERENCED_BY] = UNIT_REFERENCES,
|
|
|
|
[UNIT_TRIGGERS] = UNIT_TRIGGERED_BY,
|
2012-01-11 02:47:14 +01:00
|
|
|
[UNIT_TRIGGERED_BY] = UNIT_TRIGGERS,
|
2012-07-13 23:34:40 +02:00
|
|
|
[UNIT_PROPAGATES_RELOAD_TO] = UNIT_RELOAD_PROPAGATED_FROM,
|
2012-07-20 15:55:01 +02:00
|
|
|
[UNIT_RELOAD_PROPAGATED_FROM] = UNIT_PROPAGATES_RELOAD_TO,
|
2013-11-27 20:23:18 +01:00
|
|
|
[UNIT_JOINS_NAMESPACE_OF] = UNIT_JOINS_NAMESPACE_OF,
|
2010-01-26 21:39:06 +01:00
|
|
|
};
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
Unit *original_u = u, *original_other = other;
|
|
|
|
int r;
|
2010-01-26 21:39:06 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(d >= 0 && d < _UNIT_DEPENDENCY_MAX);
|
|
|
|
assert(other);
|
|
|
|
|
2010-09-27 23:24:17 +02:00
|
|
|
u = unit_follow_merge(u);
|
|
|
|
other = unit_follow_merge(other);
|
|
|
|
|
2010-01-26 21:39:06 +01:00
|
|
|
/* We won't allow dependencies on ourselves. We will not
|
|
|
|
* consider them an error however. */
|
2014-08-08 02:46:49 +02:00
|
|
|
if (u == other) {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
maybe_warn_about_dependency(original_u, original_other->id, d);
|
2010-01-26 21:39:06 +01:00
|
|
|
return 0;
|
2014-08-08 02:46:49 +02:00
|
|
|
}
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2020-03-30 10:49:29 +02:00
|
|
|
/* Note that ordering a device unit after a unit is permitted since it
|
|
|
|
* allows to start its job running timeout at a specific time. */
|
|
|
|
if (d == UNIT_BEFORE && other->type == UNIT_DEVICE) {
|
|
|
|
log_unit_warning(u, "Dependency Before=%s ignored (.device units cannot be delayed)", other->id);
|
2020-01-07 11:48:57 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (d == UNIT_ON_FAILURE && !UNIT_VTABLE(u)->can_fail) {
|
|
|
|
log_unit_warning(u, "Requested dependency OnFailure=%s ignored (%s units cannot fail).", other->id, unit_type_to_string(u->type));
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (d == UNIT_TRIGGERS && !UNIT_VTABLE(u)->can_trigger)
|
|
|
|
return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
|
|
|
|
"Requested dependency Triggers=%s refused (%s units cannot trigger other units).", other->id, unit_type_to_string(u->type));
|
|
|
|
if (d == UNIT_TRIGGERED_BY && !UNIT_VTABLE(other)->can_trigger)
|
|
|
|
return log_unit_error_errno(u, SYNTHETIC_ERRNO(EINVAL),
|
|
|
|
"Requested dependency TriggeredBy=%s refused (%s units cannot trigger other units).", other->id, unit_type_to_string(other->type));
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
r = unit_add_dependency_hashmap(u->dependencies + d, other, mask, 0);
|
2013-11-27 20:23:18 +01:00
|
|
|
if (r < 0)
|
2010-01-26 21:39:06 +01:00
|
|
|
return r;
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
if (inverse_table[d] != _UNIT_DEPENDENCY_INVALID && inverse_table[d] != d) {
|
|
|
|
r = unit_add_dependency_hashmap(other->dependencies + inverse_table[d], u, 0, mask);
|
2013-11-27 20:23:18 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (add_reference) {
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
r = unit_add_dependency_hashmap(u->dependencies + UNIT_REFERENCES, other, mask, 0);
|
2013-11-27 20:23:18 +01:00
|
|
|
if (r < 0)
|
2010-07-17 00:58:47 +02:00
|
|
|
return r;
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
r = unit_add_dependency_hashmap(other->dependencies + UNIT_REFERENCED_BY, u, 0, mask);
|
2013-11-27 20:23:18 +01:00
|
|
|
if (r < 0)
|
2010-04-21 06:01:13 +02:00
|
|
|
return r;
|
2013-11-27 20:23:18 +01:00
|
|
|
}
|
2010-01-26 21:39:06 +01:00
|
|
|
|
2010-02-05 00:38:41 +01:00
|
|
|
unit_add_to_dbus_queue(u);
|
2010-01-26 21:39:06 +01:00
|
|
|
return 0;
|
|
|
|
}
|
2010-01-27 00:15:56 +01:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
int unit_add_two_dependencies(Unit *u, UnitDependency d, UnitDependency e, Unit *other, bool add_reference, UnitDependencyMask mask) {
|
2010-07-03 19:46:38 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
r = unit_add_dependency(u, d, other, add_reference, mask);
|
2014-08-18 22:25:24 +02:00
|
|
|
if (r < 0)
|
2010-07-03 19:46:38 +02:00
|
|
|
return r;
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
return unit_add_dependency(u, e, other, add_reference, mask);
|
2010-07-03 19:46:38 +02:00
|
|
|
}
|
|
|
|
|
2018-09-15 20:03:32 +02:00
|
|
|
static int resolve_template(Unit *u, const char *name, char **buf, const char **ret) {
|
2015-04-30 20:21:00 +02:00
|
|
|
int r;
|
2010-04-15 03:11:11 +02:00
|
|
|
|
|
|
|
assert(u);
|
2018-09-15 20:03:32 +02:00
|
|
|
assert(name);
|
2015-04-30 20:21:00 +02:00
|
|
|
assert(buf);
|
|
|
|
assert(ret);
|
2010-04-15 03:11:11 +02:00
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
if (!unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
|
|
|
|
*buf = NULL;
|
|
|
|
*ret = name;
|
|
|
|
return 0;
|
2010-04-15 03:11:11 +02:00
|
|
|
}
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->instance)
|
2015-04-30 20:21:00 +02:00
|
|
|
r = unit_name_replace_instance(name, u->instance, buf);
|
2010-04-15 03:11:11 +02:00
|
|
|
else {
|
2013-04-23 04:10:13 +02:00
|
|
|
_cleanup_free_ char *i = NULL;
|
2010-04-15 03:11:11 +02:00
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
r = unit_name_to_prefix(u->id, &i);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2010-04-15 03:11:11 +02:00
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
r = unit_name_replace_instance(name, i, buf);
|
2010-04-15 03:11:11 +02:00
|
|
|
}
|
2015-04-30 20:21:00 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2010-04-15 03:11:11 +02:00
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
*ret = *buf;
|
|
|
|
return 0;
|
2010-04-15 03:11:11 +02:00
|
|
|
}
|
|
|
|
|
2018-09-15 19:57:52 +02:00
|
|
|
int unit_add_dependency_by_name(Unit *u, UnitDependency d, const char *name, bool add_reference, UnitDependencyMask mask) {
|
2015-04-30 20:21:00 +02:00
|
|
|
_cleanup_free_ char *buf = NULL;
|
2010-01-28 06:43:23 +01:00
|
|
|
Unit *other;
|
|
|
|
int r;
|
|
|
|
|
2010-04-15 03:11:11 +02:00
|
|
|
assert(u);
|
2018-09-15 19:57:52 +02:00
|
|
|
assert(name);
|
2010-01-28 06:43:23 +01:00
|
|
|
|
2018-09-15 20:03:32 +02:00
|
|
|
r = resolve_template(u, name, &buf, &name);
|
2015-04-30 20:21:00 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2010-01-28 06:43:23 +01:00
|
|
|
|
2018-09-15 19:57:52 +02:00
|
|
|
r = manager_load_unit(u->manager, name, NULL, NULL, &other);
|
2013-01-11 00:21:06 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2010-04-15 03:11:11 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
return unit_add_dependency(u, d, other, add_reference, mask);
|
2010-01-28 06:43:23 +01:00
|
|
|
}
|
|
|
|
|
2018-09-15 20:02:00 +02:00
|
|
|
int unit_add_two_dependencies_by_name(Unit *u, UnitDependency d, UnitDependency e, const char *name, bool add_reference, UnitDependencyMask mask) {
|
2015-04-30 20:21:00 +02:00
|
|
|
_cleanup_free_ char *buf = NULL;
|
2010-07-03 19:46:38 +02:00
|
|
|
Unit *other;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
2018-09-15 20:02:00 +02:00
|
|
|
assert(name);
|
2010-07-03 19:46:38 +02:00
|
|
|
|
2018-09-15 20:03:32 +02:00
|
|
|
r = resolve_template(u, name, &buf, &name);
|
2015-04-30 20:21:00 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2010-07-03 19:46:38 +02:00
|
|
|
|
2018-09-15 20:02:00 +02:00
|
|
|
r = manager_load_unit(u->manager, name, NULL, NULL, &other);
|
2014-08-18 22:25:24 +02:00
|
|
|
if (r < 0)
|
2013-07-25 17:36:01 +02:00
|
|
|
return r;
|
2010-07-03 19:46:38 +02:00
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
return unit_add_two_dependencies(u, d, e, other, add_reference, mask);
|
2010-07-03 19:46:38 +02:00
|
|
|
}
|
|
|
|
|
2010-01-27 00:15:56 +01:00
|
|
|
int set_unit_path(const char *p) {
|
|
|
|
/* This is mostly for debug purposes */
|
2015-10-31 12:14:37 +01:00
|
|
|
if (setenv("SYSTEMD_UNIT_PATH", p, 1) < 0)
|
2013-02-27 18:50:41 +01:00
|
|
|
return -errno;
|
2010-01-27 00:15:56 +01:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
2010-01-28 06:44:30 +01:00
|
|
|
|
2010-02-01 03:33:24 +01:00
|
|
|
char *unit_dbus_path(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (!u->id)
|
2010-08-09 17:02:09 +02:00
|
|
|
return NULL;
|
|
|
|
|
2012-06-13 18:22:08 +02:00
|
|
|
return unit_dbus_path_from_name(u->id);
|
2010-02-01 03:33:24 +01:00
|
|
|
}
|
|
|
|
|
2016-08-30 23:18:46 +02:00
|
|
|
char *unit_dbus_path_invocation_id(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (sd_id128_is_null(u->invocation_id))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return unit_dbus_path_from_name(u->invocation_id_string);
|
|
|
|
}
|
|
|
|
|
2020-05-28 14:09:43 +02:00
|
|
|
static int unit_set_invocation_id(Unit *u, sd_id128_t id) {
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Set the invocation ID for this unit. If we cannot, this will not roll back, but reset the whole thing. */
|
|
|
|
|
|
|
|
if (sd_id128_equal(u->invocation_id, id))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (!sd_id128_is_null(u->invocation_id))
|
|
|
|
(void) hashmap_remove_value(u->manager->units_by_invocation_id, &u->invocation_id, u);
|
|
|
|
|
|
|
|
if (sd_id128_is_null(id)) {
|
|
|
|
r = 0;
|
|
|
|
goto reset;
|
|
|
|
}
|
|
|
|
|
|
|
|
r = hashmap_ensure_allocated(&u->manager->units_by_invocation_id, &id128_hash_ops);
|
|
|
|
if (r < 0)
|
|
|
|
goto reset;
|
|
|
|
|
|
|
|
u->invocation_id = id;
|
|
|
|
sd_id128_to_string(id, u->invocation_id_string);
|
|
|
|
|
|
|
|
r = hashmap_put(u->manager->units_by_invocation_id, &u->invocation_id, u);
|
|
|
|
if (r < 0)
|
|
|
|
goto reset;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
reset:
|
|
|
|
u->invocation_id = SD_ID128_NULL;
|
|
|
|
u->invocation_id_string[0] = 0;
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2015-08-28 17:36:39 +02:00
|
|
|
int unit_set_slice(Unit *u, Unit *slice) {
|
|
|
|
assert(u);
|
|
|
|
assert(slice);
|
|
|
|
|
|
|
|
/* Sets the unit slice if it has not been set before. Is extra
|
|
|
|
* careful, to only allow this for units that actually have a
|
|
|
|
* cgroup context. Also, we don't allow to set this for slices
|
|
|
|
* (since the parent slice is derived from the name). Make
|
|
|
|
* sure the unit we set is actually a slice. */
|
|
|
|
|
|
|
|
if (!UNIT_HAS_CGROUP_CONTEXT(u))
|
|
|
|
return -EOPNOTSUPP;
|
|
|
|
|
|
|
|
if (u->type == UNIT_SLICE)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2015-09-01 18:53:29 +02:00
|
|
|
if (unit_active_state(u) != UNIT_INACTIVE)
|
|
|
|
return -EBUSY;
|
|
|
|
|
2015-08-28 17:36:39 +02:00
|
|
|
if (slice->type != UNIT_SLICE)
|
|
|
|
return -EINVAL;
|
|
|
|
|
core: unified cgroup hierarchy support
This patch set adds full support the new unified cgroup hierarchy logic
of modern kernels.
A new kernel command line option "systemd.unified_cgroup_hierarchy=1" is
added. If specified the unified hierarchy is mounted to /sys/fs/cgroup
instead of a tmpfs. No further hierarchies are mounted. The kernel
command line option defaults to off. We can turn it on by default as
soon as the kernel's APIs regarding this are stabilized (but even then
downstream distros might want to turn this off, as this will break any
tools that access cgroupfs directly).
It is possibly to choose for each boot individually whether the unified
or the legacy hierarchy is used. nspawn will by default provide the
legacy hierarchy to containers if the host is using it, and the unified
otherwise. However it is possible to run containers with the unified
hierarchy on a legacy host and vice versa, by setting the
$UNIFIED_CGROUP_HIERARCHY environment variable for nspawn to 1 or 0,
respectively.
The unified hierarchy provides reliable cgroup empty notifications for
the first time, via inotify. To make use of this we maintain one
manager-wide inotify fd, and each cgroup to it.
This patch also removes cg_delete() which is unused now.
On kernel 4.2 only the "memory" controller is compatible with the
unified hierarchy, hence that's the only controller systemd exposes when
booted in unified heirarchy mode.
This introduces a new enum for enumerating supported controllers, plus a
related enum for the mask bits mapping to it. The core is changed to
make use of this everywhere.
This moves PID 1 into a new "init.scope" implicit scope unit in the root
slice. This is necessary since on the unified hierarchy cgroups may
either contain subgroups or processes but not both. PID 1 hence has to
move out of the root cgroup (strictly speaking the root cgroup is the
only one where processes and subgroups are still allowed, but in order
to support containers nicey, we move PID 1 into the new scope in all
cases.) This new unit is also used on legacy hierarchy setups. It's
actually pretty useful on all systems, as it can then be used to filter
journal messages coming from PID 1, and so on.
The root slice ("-.slice") is now implicitly created and started (and
does not require a unit file on disk anymore), since
that's where "init.scope" is located and the slice needs to be started
before the scope can.
To check whether we are in unified or legacy hierarchy mode we use
statfs() on /sys/fs/cgroup. If the .f_type field reports tmpfs we are in
legacy mode, if it reports cgroupfs we are in unified mode.
This patch set carefuly makes sure that cgls and cgtop continue to work
as desired.
When invoking nspawn as a service it will implicitly create two
subcgroups in the cgroup it is using, one to move the nspawn process
into, the other to move the actual container processes into. This is
done because of the requirement that cgroups may either contain
processes or other subgroups.
2015-09-01 19:22:36 +02:00
|
|
|
if (unit_has_name(u, SPECIAL_INIT_SCOPE) &&
|
|
|
|
!unit_has_name(slice, SPECIAL_ROOT_SLICE))
|
|
|
|
return -EPERM;
|
|
|
|
|
2015-08-28 17:36:39 +02:00
|
|
|
if (UNIT_DEREF(u->slice) == slice)
|
|
|
|
return 0;
|
|
|
|
|
2016-05-14 21:56:53 +02:00
|
|
|
/* Disallow slice changes if @u is already bound to cgroups */
|
|
|
|
if (UNIT_ISSET(u->slice) && u->cgroup_realized)
|
2015-08-28 17:36:39 +02:00
|
|
|
return -EBUSY;
|
|
|
|
|
2018-02-13 13:12:43 +01:00
|
|
|
unit_ref_set(&u->slice, u, slice);
|
2015-08-28 17:36:39 +02:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_set_default_slice(Unit *u) {
|
2013-07-11 03:52:43 +02:00
|
|
|
const char *slice_name;
|
2013-06-17 21:33:26 +02:00
|
|
|
Unit *slice;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
2013-06-20 03:45:08 +02:00
|
|
|
if (UNIT_ISSET(u->slice))
|
2013-06-17 21:33:26 +02:00
|
|
|
return 0;
|
|
|
|
|
2013-07-11 03:52:43 +02:00
|
|
|
if (u->instance) {
|
|
|
|
_cleanup_free_ char *prefix = NULL, *escaped = NULL;
|
2013-07-25 17:36:01 +02:00
|
|
|
|
2013-07-11 03:52:43 +02:00
|
|
|
/* Implicitly place all instantiated units in their
|
|
|
|
* own per-template slice */
|
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
r = unit_name_to_prefix(u->id, &prefix);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2013-07-11 03:52:43 +02:00
|
|
|
|
|
|
|
/* The prefix is already escaped, but it might include
|
|
|
|
* "-" which has a special meaning for slice units,
|
|
|
|
* hence escape it here extra. */
|
2015-04-30 20:21:00 +02:00
|
|
|
escaped = unit_name_escape(prefix);
|
2013-07-11 03:52:43 +02:00
|
|
|
if (!escaped)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2016-02-24 21:24:23 +01:00
|
|
|
if (MANAGER_IS_SYSTEM(u->manager))
|
2018-11-21 16:43:40 +01:00
|
|
|
slice_name = strjoina("system-", escaped, ".slice");
|
2013-07-11 03:52:43 +02:00
|
|
|
else
|
2018-11-21 16:43:40 +01:00
|
|
|
slice_name = strjoina(escaped, ".slice");
|
2013-07-11 03:52:43 +02:00
|
|
|
} else
|
|
|
|
slice_name =
|
2016-02-24 21:24:23 +01:00
|
|
|
MANAGER_IS_SYSTEM(u->manager) && !unit_has_name(u, SPECIAL_INIT_SCOPE)
|
2013-07-11 03:52:43 +02:00
|
|
|
? SPECIAL_SYSTEM_SLICE
|
|
|
|
: SPECIAL_ROOT_SLICE;
|
|
|
|
|
|
|
|
r = manager_load_unit(u->manager, slice_name, NULL, NULL, &slice);
|
2013-06-17 21:33:26 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2015-08-28 17:36:39 +02:00
|
|
|
return unit_set_slice(u, slice);
|
2013-06-17 21:33:26 +02:00
|
|
|
}
|
|
|
|
|
2013-06-20 03:45:08 +02:00
|
|
|
const char *unit_slice_name(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (!UNIT_ISSET(u->slice))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return UNIT_DEREF(u->slice)->id;
|
|
|
|
}
|
|
|
|
|
2010-04-10 04:48:33 +02:00
|
|
|
int unit_load_related_unit(Unit *u, const char *type, Unit **_found) {
|
2013-04-25 00:01:29 +02:00
|
|
|
_cleanup_free_ char *t = NULL;
|
2010-04-10 04:48:33 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(type);
|
|
|
|
assert(_found);
|
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
r = unit_name_change_suffix(u->id, type, &t);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
if (unit_has_name(u, t))
|
|
|
|
return -EINVAL;
|
2010-04-10 04:48:33 +02:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
r = manager_load_unit(u->manager, t, NULL, NULL, _found);
|
2010-04-15 03:11:11 +02:00
|
|
|
assert(r < 0 || *_found != u);
|
2010-04-10 04:48:33 +02:00
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2015-08-05 17:47:45 +02:00
|
|
|
static int signal_name_owner_changed(sd_bus_message *message, void *userdata, sd_bus_error *error) {
|
core: drop initial ListNames() bus call from PID 1
Previously, when first connecting to the bus after connecting to it we'd
issue a ListNames() bus call to the driver to figure out which bus names
are currently active. This information was then used to initialize the
initial state for services that use BusName=.
This change removes the whole code for this and replaces it with
something vastly simpler.
First of all, the ListNames() call was issues synchronosuly, which meant
if dbus was for some reason synchronously calling into PID1 for some
reason we'd deadlock. As it turns out there's now a good chance it does:
the nss-systemd userdb hookup means that any user dbus-daemon resolves
might result in a varlink call into PID 1, and dbus resolves quite a lot
of users while parsing its policy. My original goal was to fix this
deadlock.
But as it turns out we don't need the ListNames() call at all anymore,
since #12957 has been merged. That PR was supposed to fix a race where
asynchronous installation of bus matches would cause us missing the
initial owner of a bus name when a service is first started. It fixed it
(correctly) by enquiring with GetOwnerName() who currently owns the
name, right after installing the match. But this means whenever we start watching a bus name we anyway
issue a GetOwnerName() for it, and that means also when first connecting
to the bus we don't need to issue ListNames() anymore since that just
tells us the same info: which names are currently owned.
hence, let's drop ListNames() and instead make better use of the
GetOwnerName() result: if it failed the name is not owned.
Also, while we are at it, let's simplify the unit's owner_name_changed()
callback(): let's drop the "old_owner" argument. We never used that
besides logging, and it's hard to synthesize from just the return of a
GetOwnerName(), hence don't bother.
2019-12-23 17:31:34 +01:00
|
|
|
const char *new_owner;
|
2015-08-05 17:47:45 +02:00
|
|
|
Unit *u = userdata;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(message);
|
|
|
|
assert(u);
|
|
|
|
|
core: drop initial ListNames() bus call from PID 1
Previously, when first connecting to the bus after connecting to it we'd
issue a ListNames() bus call to the driver to figure out which bus names
are currently active. This information was then used to initialize the
initial state for services that use BusName=.
This change removes the whole code for this and replaces it with
something vastly simpler.
First of all, the ListNames() call was issues synchronosuly, which meant
if dbus was for some reason synchronously calling into PID1 for some
reason we'd deadlock. As it turns out there's now a good chance it does:
the nss-systemd userdb hookup means that any user dbus-daemon resolves
might result in a varlink call into PID 1, and dbus resolves quite a lot
of users while parsing its policy. My original goal was to fix this
deadlock.
But as it turns out we don't need the ListNames() call at all anymore,
since #12957 has been merged. That PR was supposed to fix a race where
asynchronous installation of bus matches would cause us missing the
initial owner of a bus name when a service is first started. It fixed it
(correctly) by enquiring with GetOwnerName() who currently owns the
name, right after installing the match. But this means whenever we start watching a bus name we anyway
issue a GetOwnerName() for it, and that means also when first connecting
to the bus we don't need to issue ListNames() anymore since that just
tells us the same info: which names are currently owned.
hence, let's drop ListNames() and instead make better use of the
GetOwnerName() result: if it failed the name is not owned.
Also, while we are at it, let's simplify the unit's owner_name_changed()
callback(): let's drop the "old_owner" argument. We never used that
besides logging, and it's hard to synthesize from just the return of a
GetOwnerName(), hence don't bother.
2019-12-23 17:31:34 +01:00
|
|
|
r = sd_bus_message_read(message, "sss", NULL, NULL, &new_owner);
|
2015-08-05 17:47:45 +02:00
|
|
|
if (r < 0) {
|
|
|
|
bus_log_parse_error(r);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (UNIT_VTABLE(u)->bus_name_owner_change)
|
core: drop initial ListNames() bus call from PID 1
Previously, when first connecting to the bus after connecting to it we'd
issue a ListNames() bus call to the driver to figure out which bus names
are currently active. This information was then used to initialize the
initial state for services that use BusName=.
This change removes the whole code for this and replaces it with
something vastly simpler.
First of all, the ListNames() call was issues synchronosuly, which meant
if dbus was for some reason synchronously calling into PID1 for some
reason we'd deadlock. As it turns out there's now a good chance it does:
the nss-systemd userdb hookup means that any user dbus-daemon resolves
might result in a varlink call into PID 1, and dbus resolves quite a lot
of users while parsing its policy. My original goal was to fix this
deadlock.
But as it turns out we don't need the ListNames() call at all anymore,
since #12957 has been merged. That PR was supposed to fix a race where
asynchronous installation of bus matches would cause us missing the
initial owner of a bus name when a service is first started. It fixed it
(correctly) by enquiring with GetOwnerName() who currently owns the
name, right after installing the match. But this means whenever we start watching a bus name we anyway
issue a GetOwnerName() for it, and that means also when first connecting
to the bus we don't need to issue ListNames() anymore since that just
tells us the same info: which names are currently owned.
hence, let's drop ListNames() and instead make better use of the
GetOwnerName() result: if it failed the name is not owned.
Also, while we are at it, let's simplify the unit's owner_name_changed()
callback(): let's drop the "old_owner" argument. We never used that
besides logging, and it's hard to synthesize from just the return of a
GetOwnerName(), hence don't bother.
2019-12-23 17:31:34 +01:00
|
|
|
UNIT_VTABLE(u)->bus_name_owner_change(u, empty_to_null(new_owner));
|
2019-07-11 18:13:46 +02:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int get_name_owner_handler(sd_bus_message *message, void *userdata, sd_bus_error *error) {
|
|
|
|
const sd_bus_error *e;
|
|
|
|
const char *new_owner;
|
|
|
|
Unit *u = userdata;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(message);
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
|
|
|
|
|
|
|
|
e = sd_bus_message_get_error(message);
|
|
|
|
if (e) {
|
2019-12-23 16:35:15 +01:00
|
|
|
if (!sd_bus_error_has_name(e, "org.freedesktop.DBus.Error.NameHasNoOwner"))
|
|
|
|
log_unit_error(u, "Unexpected error response from GetNameOwner(): %s", e->message);
|
2019-07-11 18:13:46 +02:00
|
|
|
|
core: drop initial ListNames() bus call from PID 1
Previously, when first connecting to the bus after connecting to it we'd
issue a ListNames() bus call to the driver to figure out which bus names
are currently active. This information was then used to initialize the
initial state for services that use BusName=.
This change removes the whole code for this and replaces it with
something vastly simpler.
First of all, the ListNames() call was issues synchronosuly, which meant
if dbus was for some reason synchronously calling into PID1 for some
reason we'd deadlock. As it turns out there's now a good chance it does:
the nss-systemd userdb hookup means that any user dbus-daemon resolves
might result in a varlink call into PID 1, and dbus resolves quite a lot
of users while parsing its policy. My original goal was to fix this
deadlock.
But as it turns out we don't need the ListNames() call at all anymore,
since #12957 has been merged. That PR was supposed to fix a race where
asynchronous installation of bus matches would cause us missing the
initial owner of a bus name when a service is first started. It fixed it
(correctly) by enquiring with GetOwnerName() who currently owns the
name, right after installing the match. But this means whenever we start watching a bus name we anyway
issue a GetOwnerName() for it, and that means also when first connecting
to the bus we don't need to issue ListNames() anymore since that just
tells us the same info: which names are currently owned.
hence, let's drop ListNames() and instead make better use of the
GetOwnerName() result: if it failed the name is not owned.
Also, while we are at it, let's simplify the unit's owner_name_changed()
callback(): let's drop the "old_owner" argument. We never used that
besides logging, and it's hard to synthesize from just the return of a
GetOwnerName(), hence don't bother.
2019-12-23 17:31:34 +01:00
|
|
|
new_owner = NULL;
|
|
|
|
} else {
|
|
|
|
r = sd_bus_message_read(message, "s", &new_owner);
|
|
|
|
if (r < 0)
|
|
|
|
return bus_log_parse_error(r);
|
2019-07-11 18:13:46 +02:00
|
|
|
|
core: drop initial ListNames() bus call from PID 1
Previously, when first connecting to the bus after connecting to it we'd
issue a ListNames() bus call to the driver to figure out which bus names
are currently active. This information was then used to initialize the
initial state for services that use BusName=.
This change removes the whole code for this and replaces it with
something vastly simpler.
First of all, the ListNames() call was issues synchronosuly, which meant
if dbus was for some reason synchronously calling into PID1 for some
reason we'd deadlock. As it turns out there's now a good chance it does:
the nss-systemd userdb hookup means that any user dbus-daemon resolves
might result in a varlink call into PID 1, and dbus resolves quite a lot
of users while parsing its policy. My original goal was to fix this
deadlock.
But as it turns out we don't need the ListNames() call at all anymore,
since #12957 has been merged. That PR was supposed to fix a race where
asynchronous installation of bus matches would cause us missing the
initial owner of a bus name when a service is first started. It fixed it
(correctly) by enquiring with GetOwnerName() who currently owns the
name, right after installing the match. But this means whenever we start watching a bus name we anyway
issue a GetOwnerName() for it, and that means also when first connecting
to the bus we don't need to issue ListNames() anymore since that just
tells us the same info: which names are currently owned.
hence, let's drop ListNames() and instead make better use of the
GetOwnerName() result: if it failed the name is not owned.
Also, while we are at it, let's simplify the unit's owner_name_changed()
callback(): let's drop the "old_owner" argument. We never used that
besides logging, and it's hard to synthesize from just the return of a
GetOwnerName(), hence don't bother.
2019-12-23 17:31:34 +01:00
|
|
|
assert(!isempty(new_owner));
|
|
|
|
}
|
2019-07-11 18:13:46 +02:00
|
|
|
|
|
|
|
if (UNIT_VTABLE(u)->bus_name_owner_change)
|
core: drop initial ListNames() bus call from PID 1
Previously, when first connecting to the bus after connecting to it we'd
issue a ListNames() bus call to the driver to figure out which bus names
are currently active. This information was then used to initialize the
initial state for services that use BusName=.
This change removes the whole code for this and replaces it with
something vastly simpler.
First of all, the ListNames() call was issues synchronosuly, which meant
if dbus was for some reason synchronously calling into PID1 for some
reason we'd deadlock. As it turns out there's now a good chance it does:
the nss-systemd userdb hookup means that any user dbus-daemon resolves
might result in a varlink call into PID 1, and dbus resolves quite a lot
of users while parsing its policy. My original goal was to fix this
deadlock.
But as it turns out we don't need the ListNames() call at all anymore,
since #12957 has been merged. That PR was supposed to fix a race where
asynchronous installation of bus matches would cause us missing the
initial owner of a bus name when a service is first started. It fixed it
(correctly) by enquiring with GetOwnerName() who currently owns the
name, right after installing the match. But this means whenever we start watching a bus name we anyway
issue a GetOwnerName() for it, and that means also when first connecting
to the bus we don't need to issue ListNames() anymore since that just
tells us the same info: which names are currently owned.
hence, let's drop ListNames() and instead make better use of the
GetOwnerName() result: if it failed the name is not owned.
Also, while we are at it, let's simplify the unit's owner_name_changed()
callback(): let's drop the "old_owner" argument. We never used that
besides logging, and it's hard to synthesize from just the return of a
GetOwnerName(), hence don't bother.
2019-12-23 17:31:34 +01:00
|
|
|
UNIT_VTABLE(u)->bus_name_owner_change(u, new_owner);
|
2015-08-05 17:47:45 +02:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2015-10-17 16:25:10 +02:00
|
|
|
int unit_install_bus_match(Unit *u, sd_bus *bus, const char *name) {
|
|
|
|
const char *match;
|
2019-12-23 16:48:18 +01:00
|
|
|
int r;
|
2015-08-05 17:47:45 +02:00
|
|
|
|
2015-10-17 16:25:10 +02:00
|
|
|
assert(u);
|
|
|
|
assert(bus);
|
|
|
|
assert(name);
|
2015-08-05 17:47:45 +02:00
|
|
|
|
2019-12-23 16:48:18 +01:00
|
|
|
if (u->match_bus_slot || u->get_name_owner_slot)
|
2015-08-05 17:47:45 +02:00
|
|
|
return -EBUSY;
|
|
|
|
|
2015-10-17 16:25:10 +02:00
|
|
|
match = strjoina("type='signal',"
|
2016-04-09 20:04:09 +02:00
|
|
|
"sender='org.freedesktop.DBus',"
|
|
|
|
"path='/org/freedesktop/DBus',"
|
|
|
|
"interface='org.freedesktop.DBus',"
|
|
|
|
"member='NameOwnerChanged',"
|
|
|
|
"arg0='", name, "'");
|
2015-08-05 17:47:45 +02:00
|
|
|
|
2019-12-23 16:48:18 +01:00
|
|
|
r = sd_bus_add_match_async(bus, &u->match_bus_slot, match, signal_name_owner_changed, NULL, u);
|
2019-07-11 18:13:46 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2019-12-23 16:48:18 +01:00
|
|
|
r = sd_bus_call_method_async(
|
|
|
|
bus,
|
|
|
|
&u->get_name_owner_slot,
|
|
|
|
"org.freedesktop.DBus",
|
|
|
|
"/org/freedesktop/DBus",
|
|
|
|
"org.freedesktop.DBus",
|
|
|
|
"GetNameOwner",
|
|
|
|
get_name_owner_handler,
|
|
|
|
u,
|
|
|
|
"s", name);
|
|
|
|
if (r < 0) {
|
|
|
|
u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
log_unit_debug(u, "Watching D-Bus name '%s'.", name);
|
|
|
|
return 0;
|
2015-08-05 17:47:45 +02:00
|
|
|
}
|
|
|
|
|
2010-04-15 23:16:16 +02:00
|
|
|
int unit_watch_bus_name(Unit *u, const char *name) {
|
2015-08-05 17:47:45 +02:00
|
|
|
int r;
|
|
|
|
|
2010-04-15 23:16:16 +02:00
|
|
|
assert(u);
|
|
|
|
assert(name);
|
|
|
|
|
|
|
|
/* Watch a specific name on the bus. We only support one unit
|
|
|
|
* watching each name for now. */
|
|
|
|
|
2015-08-05 17:47:45 +02:00
|
|
|
if (u->manager->api_bus) {
|
|
|
|
/* If the bus is already available, install the match directly.
|
|
|
|
* Otherwise, just put the name in the list. bus_setup_api() will take care later. */
|
2015-10-17 16:25:10 +02:00
|
|
|
r = unit_install_bus_match(u, u->manager->api_bus, name);
|
2015-08-05 17:47:45 +02:00
|
|
|
if (r < 0)
|
2015-10-15 17:39:14 +02:00
|
|
|
return log_warning_errno(r, "Failed to subscribe to NameOwnerChanged signal for '%s': %m", name);
|
2015-08-05 17:47:45 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
r = hashmap_put(u->manager->watch_bus, name, u);
|
|
|
|
if (r < 0) {
|
|
|
|
u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
|
2019-12-23 16:48:18 +01:00
|
|
|
u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
|
2015-08-05 17:47:45 +02:00
|
|
|
return log_warning_errno(r, "Failed to put bus name to hashmap: %m");
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
2010-04-15 23:16:16 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
void unit_unwatch_bus_name(Unit *u, const char *name) {
|
|
|
|
assert(u);
|
|
|
|
assert(name);
|
|
|
|
|
2017-02-10 03:54:48 +01:00
|
|
|
(void) hashmap_remove_value(u->manager->watch_bus, name, u);
|
2015-08-05 17:47:45 +02:00
|
|
|
u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
|
2019-07-11 18:13:46 +02:00
|
|
|
u->get_name_owner_slot = sd_bus_slot_unref(u->get_name_owner_slot);
|
2010-04-15 23:16:16 +02:00
|
|
|
}
|
|
|
|
|
2010-04-21 03:27:44 +02:00
|
|
|
bool unit_can_serialize(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
return UNIT_VTABLE(u)->serialize && UNIT_VTABLE(u)->deserialize_item;
|
|
|
|
}
|
|
|
|
|
2018-10-17 20:40:09 +02:00
|
|
|
static int serialize_cgroup_mask(FILE *f, const char *key, CGroupMask mask) {
|
2017-03-27 18:00:54 +02:00
|
|
|
_cleanup_free_ char *s = NULL;
|
2018-10-17 20:40:09 +02:00
|
|
|
int r;
|
2017-03-27 18:00:54 +02:00
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(key);
|
|
|
|
|
2018-10-17 20:40:09 +02:00
|
|
|
if (mask == 0)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
r = cg_mask_to_string(mask, &s);
|
|
|
|
if (r < 0)
|
|
|
|
return log_error_errno(r, "Failed to format cgroup mask: %m");
|
|
|
|
|
|
|
|
return serialize_item(f, key, s);
|
2017-03-27 18:00:54 +02:00
|
|
|
}
|
|
|
|
|
2019-03-22 19:19:32 +01:00
|
|
|
static const char *const ip_accounting_metric_field[_CGROUP_IP_ACCOUNTING_METRIC_MAX] = {
|
2017-09-07 14:07:13 +02:00
|
|
|
[CGROUP_IP_INGRESS_BYTES] = "ip-accounting-ingress-bytes",
|
|
|
|
[CGROUP_IP_INGRESS_PACKETS] = "ip-accounting-ingress-packets",
|
|
|
|
[CGROUP_IP_EGRESS_BYTES] = "ip-accounting-egress-bytes",
|
|
|
|
[CGROUP_IP_EGRESS_PACKETS] = "ip-accounting-egress-packets",
|
|
|
|
};
|
|
|
|
|
2019-03-22 12:16:03 +01:00
|
|
|
static const char *const io_accounting_metric_field_base[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {
|
|
|
|
[CGROUP_IO_READ_BYTES] = "io-accounting-read-bytes-base",
|
|
|
|
[CGROUP_IO_WRITE_BYTES] = "io-accounting-write-bytes-base",
|
|
|
|
[CGROUP_IO_READ_OPERATIONS] = "io-accounting-read-operations-base",
|
|
|
|
[CGROUP_IO_WRITE_OPERATIONS] = "io-accounting-write-operations-base",
|
|
|
|
};
|
|
|
|
|
|
|
|
static const char *const io_accounting_metric_field_last[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {
|
|
|
|
[CGROUP_IO_READ_BYTES] = "io-accounting-read-bytes-last",
|
|
|
|
[CGROUP_IO_WRITE_BYTES] = "io-accounting-write-bytes-last",
|
|
|
|
[CGROUP_IO_READ_OPERATIONS] = "io-accounting-read-operations-last",
|
|
|
|
[CGROUP_IO_WRITE_OPERATIONS] = "io-accounting-write-operations-last",
|
|
|
|
};
|
|
|
|
|
2012-07-18 01:46:52 +02:00
|
|
|
int unit_serialize(Unit *u, FILE *f, FDSet *fds, bool serialize_jobs) {
|
2017-09-07 14:07:13 +02:00
|
|
|
CGroupIPAccountingMetric m;
|
2010-04-21 03:27:44 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(f);
|
|
|
|
assert(fds);
|
|
|
|
|
2014-05-16 01:15:03 +02:00
|
|
|
if (unit_can_serialize(u)) {
|
|
|
|
r = UNIT_VTABLE(u)->serialize(u, f, fds);
|
2013-11-27 20:23:18 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
core: add NOP jobs, job type collapsing
Two of our current job types are special:
JOB_TRY_RESTART, JOB_RELOAD_OR_START.
They differ from other job types by being sensitive to the unit active state.
They perform some action when the unit is active and some other action
otherwise. This raises a question: when exactly should the unit state be
checked to make the decision?
Currently the unit state is checked when the job becomes runnable. It's more
sensible to check the state immediately when the job is added by the user.
When the user types "systemctl try-restart foo.service", he really intends
to restart the service if it's running right now. If it isn't running right
now, the restart is pointless.
Consider the example (from Bugzilla[1]):
sleep.service takes some time to start.
hello.service has After=sleep.service.
Both services get started. Two jobs will appear:
hello.service/start waiting
sleep.service/start running
Then someone runs "systemctl try-restart hello.service".
Currently the try-restart operation will block and wait for
sleep.service/start to complete.
The correct result is to complete the try-restart operation immediately
with success, because hello.service is not running. The two original
jobs must not be disturbed by this.
To fix this we introduce two new concepts:
- a new job type: JOB_NOP
A JOB_NOP job does not do anything to the unit. It does not pull in any
dependencies. It is always immediately runnable. When installed to a unit,
it sits in a special slot (u->nop_job) where it never conflicts with
the installed job (u->job) of a different type. It never merges with jobs
of other types, but it can merge into an already installed JOB_NOP job.
- "collapsing" of job types
When a job of one of the two special types is added, the state of the unit
is checked immediately and the job type changes:
JOB_TRY_RESTART -> JOB_RESTART or JOB_NOP
JOB_RELOAD_OR_START -> JOB_RELOAD or JOB_START
Should a job type JOB_RELOAD_OR_START appear later during job merging, it
collapses immediately afterwards.
Collapsing actually makes some things simpler, because there are now fewer
job types that are allowed in the transaction.
[1] Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=753586
2012-04-25 11:58:27 +02:00
|
|
|
}
|
|
|
|
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_dual_timestamp(f, "state-change-timestamp", &u->state_change_timestamp);
|
2016-02-01 16:01:25 +01:00
|
|
|
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_dual_timestamp(f, "inactive-exit-timestamp", &u->inactive_exit_timestamp);
|
|
|
|
(void) serialize_dual_timestamp(f, "active-enter-timestamp", &u->active_enter_timestamp);
|
|
|
|
(void) serialize_dual_timestamp(f, "active-exit-timestamp", &u->active_exit_timestamp);
|
|
|
|
(void) serialize_dual_timestamp(f, "inactive-enter-timestamp", &u->inactive_enter_timestamp);
|
2016-02-01 16:01:25 +01:00
|
|
|
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_dual_timestamp(f, "condition-timestamp", &u->condition_timestamp);
|
|
|
|
(void) serialize_dual_timestamp(f, "assert-timestamp", &u->assert_timestamp);
|
2011-03-17 04:36:19 +01:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (dual_timestamp_is_set(&u->condition_timestamp))
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_bool(f, "condition-result", u->condition_result);
|
2010-10-27 00:01:12 +02:00
|
|
|
|
2014-11-06 13:43:45 +01:00
|
|
|
if (dual_timestamp_is_set(&u->assert_timestamp))
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_bool(f, "assert-result", u->assert_result);
|
2014-11-06 13:43:45 +01:00
|
|
|
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_bool(f, "transient", u->transient);
|
|
|
|
(void) serialize_bool(f, "in-audit", u->in_audit);
|
2016-08-18 20:58:10 +02:00
|
|
|
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_bool(f, "exported-invocation-id", u->exported_invocation_id);
|
|
|
|
(void) serialize_bool(f, "exported-log-level-max", u->exported_log_level_max);
|
|
|
|
(void) serialize_bool(f, "exported-log-extra-fields", u->exported_log_extra_fields);
|
2019-09-19 17:49:14 +02:00
|
|
|
(void) serialize_bool(f, "exported-log-rate-limit-interval", u->exported_log_ratelimit_interval);
|
|
|
|
(void) serialize_bool(f, "exported-log-rate-limit-burst", u->exported_log_ratelimit_burst);
|
2018-10-09 10:08:44 +02:00
|
|
|
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_item_format(f, "cpu-usage-base", "%" PRIu64, u->cpu_usage_base);
|
2016-08-18 20:58:10 +02:00
|
|
|
if (u->cpu_usage_last != NSEC_INFINITY)
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_item_format(f, "cpu-usage-last", "%" PRIu64, u->cpu_usage_last);
|
2013-06-28 04:12:58 +02:00
|
|
|
|
2019-03-19 19:05:19 +01:00
|
|
|
if (u->oom_kill_last > 0)
|
|
|
|
(void) serialize_item_format(f, "oom-kill-last", "%" PRIu64, u->oom_kill_last);
|
|
|
|
|
2019-03-22 12:16:03 +01:00
|
|
|
for (CGroupIOAccountingMetric im = 0; im < _CGROUP_IO_ACCOUNTING_METRIC_MAX; im++) {
|
|
|
|
(void) serialize_item_format(f, io_accounting_metric_field_base[im], "%" PRIu64, u->io_accounting_base[im]);
|
|
|
|
|
|
|
|
if (u->io_accounting_last[im] != UINT64_MAX)
|
|
|
|
(void) serialize_item_format(f, io_accounting_metric_field_last[im], "%" PRIu64, u->io_accounting_last[im]);
|
|
|
|
}
|
|
|
|
|
2013-06-28 04:12:58 +02:00
|
|
|
if (u->cgroup_path)
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_item(f, "cgroup", u->cgroup_path);
|
|
|
|
|
|
|
|
(void) serialize_bool(f, "cgroup-realized", u->cgroup_realized);
|
|
|
|
(void) serialize_cgroup_mask(f, "cgroup-realized-mask", u->cgroup_realized_mask);
|
|
|
|
(void) serialize_cgroup_mask(f, "cgroup-enabled-mask", u->cgroup_enabled_mask);
|
|
|
|
(void) serialize_cgroup_mask(f, "cgroup-invalidated-mask", u->cgroup_invalidated_mask);
|
2013-06-28 04:12:58 +02:00
|
|
|
|
2016-08-01 19:24:40 +02:00
|
|
|
if (uid_is_valid(u->ref_uid))
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_item_format(f, "ref-uid", UID_FMT, u->ref_uid);
|
2016-08-01 19:24:40 +02:00
|
|
|
if (gid_is_valid(u->ref_gid))
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_item_format(f, "ref-gid", GID_FMT, u->ref_gid);
|
2016-08-01 19:24:40 +02:00
|
|
|
|
2016-08-30 23:18:46 +02:00
|
|
|
if (!sd_id128_is_null(u->invocation_id))
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_item_format(f, "invocation-id", SD_ID128_FORMAT_STR, SD_ID128_FORMAT_VAL(u->invocation_id));
|
2016-08-30 23:18:46 +02:00
|
|
|
|
2020-04-29 17:53:43 +02:00
|
|
|
(void) serialize_item_format(f, "freezer-state", "%s", freezer_state_to_string(unit_freezer_state(u)));
|
|
|
|
|
2016-08-15 18:12:01 +02:00
|
|
|
bus_track_serialize(u->bus_track, f, "ref");
|
|
|
|
|
2017-09-07 14:07:13 +02:00
|
|
|
for (m = 0; m < _CGROUP_IP_ACCOUNTING_METRIC_MAX; m++) {
|
|
|
|
uint64_t v;
|
|
|
|
|
|
|
|
r = unit_get_ip_accounting(u, m, &v);
|
|
|
|
if (r >= 0)
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) serialize_item_format(f, ip_accounting_metric_field[m], "%" PRIu64, v);
|
2017-09-07 14:07:13 +02:00
|
|
|
}
|
|
|
|
|
2013-11-27 20:23:18 +01:00
|
|
|
if (serialize_jobs) {
|
|
|
|
if (u->job) {
|
2018-10-17 20:40:09 +02:00
|
|
|
fputs("job\n", f);
|
2016-08-15 18:12:01 +02:00
|
|
|
job_serialize(u->job, f);
|
2013-11-27 20:23:18 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
if (u->nop_job) {
|
2018-10-17 20:40:09 +02:00
|
|
|
fputs("job\n", f);
|
2016-08-15 18:12:01 +02:00
|
|
|
job_serialize(u->nop_job, f);
|
2013-11-27 20:23:18 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-04-21 03:27:44 +02:00
|
|
|
/* End marker */
|
|
|
|
fputc('\n', f);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2018-12-10 19:40:37 +01:00
|
|
|
static int unit_deserialize_job(Unit *u, FILE *f) {
|
|
|
|
_cleanup_(job_freep) Job *j = NULL;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(f);
|
|
|
|
|
|
|
|
j = job_new_raw(u);
|
|
|
|
if (!j)
|
|
|
|
return log_oom();
|
|
|
|
|
|
|
|
r = job_deserialize(j, f);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
r = job_install_deserialized(j);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
TAKE_PTR(j);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2010-04-21 03:27:44 +02:00
|
|
|
int unit_deserialize(Unit *u, FILE *f, FDSet *fds) {
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(f);
|
|
|
|
assert(fds);
|
|
|
|
|
|
|
|
for (;;) {
|
2018-10-17 18:36:24 +02:00
|
|
|
_cleanup_free_ char *line = NULL;
|
|
|
|
char *l, *v;
|
2019-03-22 11:47:29 +01:00
|
|
|
ssize_t m;
|
2010-04-21 03:27:44 +02:00
|
|
|
size_t k;
|
|
|
|
|
2018-10-17 18:36:24 +02:00
|
|
|
r = read_line(f, LONG_LINE_MAX, &line);
|
|
|
|
if (r < 0)
|
|
|
|
return log_error_errno(r, "Failed to read serialization line: %m");
|
|
|
|
if (r == 0) /* eof */
|
|
|
|
break;
|
2010-04-21 03:27:44 +02:00
|
|
|
|
|
|
|
l = strstrip(line);
|
2018-10-17 18:36:24 +02:00
|
|
|
if (isempty(l)) /* End marker */
|
2016-02-01 16:01:25 +01:00
|
|
|
break;
|
2010-04-21 03:27:44 +02:00
|
|
|
|
|
|
|
k = strcspn(l, "=");
|
|
|
|
|
|
|
|
if (l[k] == '=') {
|
|
|
|
l[k] = 0;
|
|
|
|
v = l+k+1;
|
|
|
|
} else
|
|
|
|
v = l+k;
|
|
|
|
|
2010-06-03 14:26:50 +02:00
|
|
|
if (streq(l, "job")) {
|
2012-04-23 01:24:04 +02:00
|
|
|
if (v[0] == '\0') {
|
2018-12-10 19:40:37 +01:00
|
|
|
/* New-style serialized job */
|
|
|
|
r = unit_deserialize_job(u, f);
|
|
|
|
if (r < 0)
|
core: add NOP jobs, job type collapsing
Two of our current job types are special:
JOB_TRY_RESTART, JOB_RELOAD_OR_START.
They differ from other job types by being sensitive to the unit active state.
They perform some action when the unit is active and some other action
otherwise. This raises a question: when exactly should the unit state be
checked to make the decision?
Currently the unit state is checked when the job becomes runnable. It's more
sensible to check the state immediately when the job is added by the user.
When the user types "systemctl try-restart foo.service", he really intends
to restart the service if it's running right now. If it isn't running right
now, the restart is pointless.
Consider the example (from Bugzilla[1]):
sleep.service takes some time to start.
hello.service has After=sleep.service.
Both services get started. Two jobs will appear:
hello.service/start waiting
sleep.service/start running
Then someone runs "systemctl try-restart hello.service".
Currently the try-restart operation will block and wait for
sleep.service/start to complete.
The correct result is to complete the try-restart operation immediately
with success, because hello.service is not running. The two original
jobs must not be disturbed by this.
To fix this we introduce two new concepts:
- a new job type: JOB_NOP
A JOB_NOP job does not do anything to the unit. It does not pull in any
dependencies. It is always immediately runnable. When installed to a unit,
it sits in a special slot (u->nop_job) where it never conflicts with
the installed job (u->job) of a different type. It never merges with jobs
of other types, but it can merge into an already installed JOB_NOP job.
- "collapsing" of job types
When a job of one of the two special types is added, the state of the unit
is checked immediately and the job type changes:
JOB_TRY_RESTART -> JOB_RESTART or JOB_NOP
JOB_RELOAD_OR_START -> JOB_RELOAD or JOB_START
Should a job type JOB_RELOAD_OR_START appear later during job merging, it
collapses immediately afterwards.
Collapsing actually makes some things simpler, because there are now fewer
job types that are allowed in the transaction.
[1] Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=753586
2012-04-25 11:58:27 +02:00
|
|
|
return r;
|
2018-12-10 19:40:37 +01:00
|
|
|
} else /* Legacy for pre-44 */
|
2015-05-19 16:41:14 +02:00
|
|
|
log_unit_warning(u, "Update from too old systemd versions are unsupported, cannot deserialize job: %s", v);
|
2010-06-03 14:26:50 +02:00
|
|
|
continue;
|
2016-02-01 16:01:25 +01:00
|
|
|
} else if (streq(l, "state-change-timestamp")) {
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) deserialize_dual_timestamp(v, &u->state_change_timestamp);
|
2016-02-01 16:01:25 +01:00
|
|
|
continue;
|
2010-10-28 02:19:21 +02:00
|
|
|
} else if (streq(l, "inactive-exit-timestamp")) {
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) deserialize_dual_timestamp(v, &u->inactive_exit_timestamp);
|
2010-10-28 02:19:21 +02:00
|
|
|
continue;
|
|
|
|
} else if (streq(l, "active-enter-timestamp")) {
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) deserialize_dual_timestamp(v, &u->active_enter_timestamp);
|
2010-10-28 02:19:21 +02:00
|
|
|
continue;
|
|
|
|
} else if (streq(l, "active-exit-timestamp")) {
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) deserialize_dual_timestamp(v, &u->active_exit_timestamp);
|
2010-10-28 02:19:21 +02:00
|
|
|
continue;
|
|
|
|
} else if (streq(l, "inactive-enter-timestamp")) {
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) deserialize_dual_timestamp(v, &u->inactive_enter_timestamp);
|
2010-10-28 02:19:21 +02:00
|
|
|
continue;
|
2011-03-17 04:36:19 +01:00
|
|
|
} else if (streq(l, "condition-timestamp")) {
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) deserialize_dual_timestamp(v, &u->condition_timestamp);
|
2011-03-17 04:36:19 +01:00
|
|
|
continue;
|
2014-11-06 13:43:45 +01:00
|
|
|
} else if (streq(l, "assert-timestamp")) {
|
2018-10-17 20:40:09 +02:00
|
|
|
(void) deserialize_dual_timestamp(v, &u->assert_timestamp);
|
2014-11-06 13:43:45 +01:00
|
|
|
continue;
|
2011-03-17 04:36:19 +01:00
|
|
|
} else if (streq(l, "condition-result")) {
|
|
|
|
|
2015-04-21 20:22:51 +02:00
|
|
|
r = parse_boolean(v);
|
|
|
|
if (r < 0)
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
log_unit_debug(u, "Failed to parse condition result value %s, ignoring.", v);
|
2011-03-17 04:36:19 +01:00
|
|
|
else
|
2015-04-21 20:22:51 +02:00
|
|
|
u->condition_result = r;
|
2011-03-29 23:32:31 +02:00
|
|
|
|
|
|
|
continue;
|
2013-06-28 04:12:58 +02:00
|
|
|
|
2014-11-06 13:43:45 +01:00
|
|
|
} else if (streq(l, "assert-result")) {
|
|
|
|
|
2015-04-21 20:22:51 +02:00
|
|
|
r = parse_boolean(v);
|
|
|
|
if (r < 0)
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
log_unit_debug(u, "Failed to parse assert result value %s, ignoring.", v);
|
2014-11-06 13:43:45 +01:00
|
|
|
else
|
2015-04-21 20:22:51 +02:00
|
|
|
u->assert_result = r;
|
2014-11-06 13:43:45 +01:00
|
|
|
|
|
|
|
continue;
|
|
|
|
|
2013-06-28 04:12:58 +02:00
|
|
|
} else if (streq(l, "transient")) {
|
|
|
|
|
2015-04-21 20:22:51 +02:00
|
|
|
r = parse_boolean(v);
|
|
|
|
if (r < 0)
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
log_unit_debug(u, "Failed to parse transient bool %s, ignoring.", v);
|
2013-06-28 04:12:58 +02:00
|
|
|
else
|
2015-04-21 20:22:51 +02:00
|
|
|
u->transient = r;
|
2013-06-28 04:12:58 +02:00
|
|
|
|
|
|
|
continue;
|
2015-04-21 20:22:51 +02:00
|
|
|
|
2018-10-09 10:08:44 +02:00
|
|
|
} else if (streq(l, "in-audit")) {
|
|
|
|
|
|
|
|
r = parse_boolean(v);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse in-audit bool %s, ignoring.", v);
|
|
|
|
else
|
|
|
|
u->in_audit = r;
|
|
|
|
|
|
|
|
continue;
|
|
|
|
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
} else if (streq(l, "exported-invocation-id")) {
|
|
|
|
|
|
|
|
r = parse_boolean(v);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse exported invocation ID bool %s, ignoring.", v);
|
|
|
|
else
|
|
|
|
u->exported_invocation_id = r;
|
|
|
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
} else if (streq(l, "exported-log-level-max")) {
|
|
|
|
|
|
|
|
r = parse_boolean(v);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse exported log level max bool %s, ignoring.", v);
|
|
|
|
else
|
|
|
|
u->exported_log_level_max = r;
|
|
|
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
} else if (streq(l, "exported-log-extra-fields")) {
|
|
|
|
|
|
|
|
r = parse_boolean(v);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse exported log extra fields bool %s, ignoring.", v);
|
|
|
|
else
|
|
|
|
u->exported_log_extra_fields = r;
|
|
|
|
|
|
|
|
continue;
|
|
|
|
|
2018-10-08 05:28:36 +02:00
|
|
|
} else if (streq(l, "exported-log-rate-limit-interval")) {
|
|
|
|
|
|
|
|
r = parse_boolean(v);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse exported log rate limit interval %s, ignoring.", v);
|
|
|
|
else
|
2019-09-19 17:49:14 +02:00
|
|
|
u->exported_log_ratelimit_interval = r;
|
2018-10-08 05:28:36 +02:00
|
|
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
} else if (streq(l, "exported-log-rate-limit-burst")) {
|
|
|
|
|
|
|
|
r = parse_boolean(v);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse exported log rate limit burst %s, ignoring.", v);
|
|
|
|
else
|
2019-09-19 17:49:14 +02:00
|
|
|
u->exported_log_ratelimit_burst = r;
|
2018-10-08 05:28:36 +02:00
|
|
|
|
|
|
|
continue;
|
|
|
|
|
2016-08-18 20:58:10 +02:00
|
|
|
} else if (STR_IN_SET(l, "cpu-usage-base", "cpuacct-usage-base")) {
|
2015-03-01 16:24:19 +01:00
|
|
|
|
2016-08-07 15:45:39 +02:00
|
|
|
r = safe_atou64(v, &u->cpu_usage_base);
|
2015-03-01 16:24:19 +01:00
|
|
|
if (r < 0)
|
2016-08-18 20:58:10 +02:00
|
|
|
log_unit_debug(u, "Failed to parse CPU usage base %s, ignoring.", v);
|
|
|
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
} else if (streq(l, "cpu-usage-last")) {
|
|
|
|
|
|
|
|
r = safe_atou64(v, &u->cpu_usage_last);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to read CPU usage last %s, ignoring.", v);
|
2015-03-01 16:24:19 +01:00
|
|
|
|
2015-04-16 14:10:33 +02:00
|
|
|
continue;
|
2014-04-18 04:12:25 +02:00
|
|
|
|
2019-03-19 19:05:19 +01:00
|
|
|
} else if (streq(l, "oom-kill-last")) {
|
|
|
|
|
|
|
|
r = safe_atou64(v, &u->oom_kill_last);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to read OOM kill last %s, ignoring.", v);
|
|
|
|
|
|
|
|
continue;
|
|
|
|
|
2015-04-21 20:22:51 +02:00
|
|
|
} else if (streq(l, "cgroup")) {
|
2013-07-10 21:17:37 +02:00
|
|
|
|
2015-04-21 20:22:51 +02:00
|
|
|
r = unit_set_cgroup_path(u, v);
|
|
|
|
if (r < 0)
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
log_unit_debug_errno(u, r, "Failed to set cgroup path %s, ignoring: %m", v);
|
2014-04-18 04:12:25 +02:00
|
|
|
|
core: unified cgroup hierarchy support
This patch set adds full support the new unified cgroup hierarchy logic
of modern kernels.
A new kernel command line option "systemd.unified_cgroup_hierarchy=1" is
added. If specified the unified hierarchy is mounted to /sys/fs/cgroup
instead of a tmpfs. No further hierarchies are mounted. The kernel
command line option defaults to off. We can turn it on by default as
soon as the kernel's APIs regarding this are stabilized (but even then
downstream distros might want to turn this off, as this will break any
tools that access cgroupfs directly).
It is possibly to choose for each boot individually whether the unified
or the legacy hierarchy is used. nspawn will by default provide the
legacy hierarchy to containers if the host is using it, and the unified
otherwise. However it is possible to run containers with the unified
hierarchy on a legacy host and vice versa, by setting the
$UNIFIED_CGROUP_HIERARCHY environment variable for nspawn to 1 or 0,
respectively.
The unified hierarchy provides reliable cgroup empty notifications for
the first time, via inotify. To make use of this we maintain one
manager-wide inotify fd, and each cgroup to it.
This patch also removes cg_delete() which is unused now.
On kernel 4.2 only the "memory" controller is compatible with the
unified hierarchy, hence that's the only controller systemd exposes when
booted in unified heirarchy mode.
This introduces a new enum for enumerating supported controllers, plus a
related enum for the mask bits mapping to it. The core is changed to
make use of this everywhere.
This moves PID 1 into a new "init.scope" implicit scope unit in the root
slice. This is necessary since on the unified hierarchy cgroups may
either contain subgroups or processes but not both. PID 1 hence has to
move out of the root cgroup (strictly speaking the root cgroup is the
only one where processes and subgroups are still allowed, but in order
to support containers nicey, we move PID 1 into the new scope in all
cases.) This new unit is also used on legacy hierarchy setups. It's
actually pretty useful on all systems, as it can then be used to filter
journal messages coming from PID 1, and so on.
The root slice ("-.slice") is now implicitly created and started (and
does not require a unit file on disk anymore), since
that's where "init.scope" is located and the slice needs to be started
before the scope can.
To check whether we are in unified or legacy hierarchy mode we use
statfs() on /sys/fs/cgroup. If the .f_type field reports tmpfs we are in
legacy mode, if it reports cgroupfs we are in unified mode.
This patch set carefuly makes sure that cgls and cgtop continue to work
as desired.
When invoking nspawn as a service it will implicitly create two
subcgroups in the cgroup it is using, one to move the nspawn process
into, the other to move the actual container processes into. This is
done because of the requirement that cgroups may either contain
processes or other subgroups.
2015-09-01 19:22:36 +02:00
|
|
|
(void) unit_watch_cgroup(u);
|
2019-03-19 19:05:19 +01:00
|
|
|
(void) unit_watch_cgroup_memory(u);
|
core: unified cgroup hierarchy support
This patch set adds full support the new unified cgroup hierarchy logic
of modern kernels.
A new kernel command line option "systemd.unified_cgroup_hierarchy=1" is
added. If specified the unified hierarchy is mounted to /sys/fs/cgroup
instead of a tmpfs. No further hierarchies are mounted. The kernel
command line option defaults to off. We can turn it on by default as
soon as the kernel's APIs regarding this are stabilized (but even then
downstream distros might want to turn this off, as this will break any
tools that access cgroupfs directly).
It is possibly to choose for each boot individually whether the unified
or the legacy hierarchy is used. nspawn will by default provide the
legacy hierarchy to containers if the host is using it, and the unified
otherwise. However it is possible to run containers with the unified
hierarchy on a legacy host and vice versa, by setting the
$UNIFIED_CGROUP_HIERARCHY environment variable for nspawn to 1 or 0,
respectively.
The unified hierarchy provides reliable cgroup empty notifications for
the first time, via inotify. To make use of this we maintain one
manager-wide inotify fd, and each cgroup to it.
This patch also removes cg_delete() which is unused now.
On kernel 4.2 only the "memory" controller is compatible with the
unified hierarchy, hence that's the only controller systemd exposes when
booted in unified heirarchy mode.
This introduces a new enum for enumerating supported controllers, plus a
related enum for the mask bits mapping to it. The core is changed to
make use of this everywhere.
This moves PID 1 into a new "init.scope" implicit scope unit in the root
slice. This is necessary since on the unified hierarchy cgroups may
either contain subgroups or processes but not both. PID 1 hence has to
move out of the root cgroup (strictly speaking the root cgroup is the
only one where processes and subgroups are still allowed, but in order
to support containers nicey, we move PID 1 into the new scope in all
cases.) This new unit is also used on legacy hierarchy setups. It's
actually pretty useful on all systems, as it can then be used to filter
journal messages coming from PID 1, and so on.
The root slice ("-.slice") is now implicitly created and started (and
does not require a unit file on disk anymore), since
that's where "init.scope" is located and the slice needs to be started
before the scope can.
To check whether we are in unified or legacy hierarchy mode we use
statfs() on /sys/fs/cgroup. If the .f_type field reports tmpfs we are in
legacy mode, if it reports cgroupfs we are in unified mode.
This patch set carefuly makes sure that cgls and cgtop continue to work
as desired.
When invoking nspawn as a service it will implicitly create two
subcgroups in the cgroup it is using, one to move the nspawn process
into, the other to move the actual container processes into. This is
done because of the requirement that cgroups may either contain
processes or other subgroups.
2015-09-01 19:22:36 +02:00
|
|
|
|
2015-06-10 14:36:50 +02:00
|
|
|
continue;
|
|
|
|
} else if (streq(l, "cgroup-realized")) {
|
|
|
|
int b;
|
|
|
|
|
|
|
|
b = parse_boolean(v);
|
|
|
|
if (b < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse cgroup-realized bool %s, ignoring.", v);
|
|
|
|
else
|
|
|
|
u->cgroup_realized = b;
|
|
|
|
|
2013-06-28 04:12:58 +02:00
|
|
|
continue;
|
2016-08-01 19:24:40 +02:00
|
|
|
|
2017-03-27 18:00:54 +02:00
|
|
|
} else if (streq(l, "cgroup-realized-mask")) {
|
|
|
|
|
|
|
|
r = cg_mask_from_string(v, &u->cgroup_realized_mask);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse cgroup-realized-mask %s, ignoring.", v);
|
|
|
|
continue;
|
|
|
|
|
|
|
|
} else if (streq(l, "cgroup-enabled-mask")) {
|
|
|
|
|
|
|
|
r = cg_mask_from_string(v, &u->cgroup_enabled_mask);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse cgroup-enabled-mask %s, ignoring.", v);
|
|
|
|
continue;
|
|
|
|
|
2018-09-30 12:33:16 +02:00
|
|
|
} else if (streq(l, "cgroup-invalidated-mask")) {
|
2017-09-05 19:27:53 +02:00
|
|
|
|
2018-09-30 12:33:16 +02:00
|
|
|
r = cg_mask_from_string(v, &u->cgroup_invalidated_mask);
|
2017-09-05 19:27:53 +02:00
|
|
|
if (r < 0)
|
2018-09-30 12:33:16 +02:00
|
|
|
log_unit_debug(u, "Failed to parse cgroup-invalidated-mask %s, ignoring.", v);
|
2017-09-05 19:27:53 +02:00
|
|
|
continue;
|
|
|
|
|
2016-08-01 19:24:40 +02:00
|
|
|
} else if (streq(l, "ref-uid")) {
|
|
|
|
uid_t uid;
|
|
|
|
|
|
|
|
r = parse_uid(v, &uid);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse referenced UID %s, ignoring.", v);
|
|
|
|
else
|
|
|
|
unit_ref_uid_gid(u, uid, GID_INVALID);
|
|
|
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
} else if (streq(l, "ref-gid")) {
|
|
|
|
gid_t gid;
|
|
|
|
|
|
|
|
r = parse_gid(v, &gid);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse referenced GID %s, ignoring.", v);
|
|
|
|
else
|
|
|
|
unit_ref_uid_gid(u, UID_INVALID, gid);
|
|
|
|
|
2018-10-09 15:33:13 +02:00
|
|
|
continue;
|
|
|
|
|
2016-08-15 18:12:01 +02:00
|
|
|
} else if (streq(l, "ref")) {
|
|
|
|
|
|
|
|
r = strv_extend(&u->deserialized_refs, v);
|
|
|
|
if (r < 0)
|
2018-10-17 20:40:09 +02:00
|
|
|
return log_oom();
|
2016-08-15 18:12:01 +02:00
|
|
|
|
2016-08-30 23:18:46 +02:00
|
|
|
continue;
|
|
|
|
} else if (streq(l, "invocation-id")) {
|
|
|
|
sd_id128_t id;
|
|
|
|
|
|
|
|
r = sd_id128_from_string(v, &id);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse invocation id %s, ignoring.", v);
|
|
|
|
else {
|
|
|
|
r = unit_set_invocation_id(u, id);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_warning_errno(u, r, "Failed to set invocation ID for unit: %m");
|
|
|
|
}
|
|
|
|
|
2020-04-29 17:53:43 +02:00
|
|
|
continue;
|
|
|
|
} else if (streq(l, "freezer-state")) {
|
|
|
|
FreezerState s;
|
|
|
|
|
|
|
|
s = freezer_state_from_string(v);
|
|
|
|
if (s < 0)
|
|
|
|
log_unit_debug(u, "Failed to deserialize freezer-state '%s', ignoring.", v);
|
|
|
|
else
|
|
|
|
u->freezer_state = s;
|
|
|
|
|
2016-08-01 19:24:40 +02:00
|
|
|
continue;
|
2010-10-28 02:19:21 +02:00
|
|
|
}
|
2010-06-03 14:26:50 +02:00
|
|
|
|
2017-09-07 14:07:13 +02:00
|
|
|
/* Check if this is an IP accounting metric serialization field */
|
2019-03-22 11:47:29 +01:00
|
|
|
m = string_table_lookup(ip_accounting_metric_field, ELEMENTSOF(ip_accounting_metric_field), l);
|
|
|
|
if (m >= 0) {
|
2017-09-07 14:07:13 +02:00
|
|
|
uint64_t c;
|
|
|
|
|
|
|
|
r = safe_atou64(v, &c);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse IP accounting value %s, ignoring.", v);
|
|
|
|
else
|
|
|
|
u->ip_accounting_extra[m] = c;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2019-03-22 12:16:03 +01:00
|
|
|
m = string_table_lookup(io_accounting_metric_field_base, ELEMENTSOF(io_accounting_metric_field_base), l);
|
|
|
|
if (m >= 0) {
|
|
|
|
uint64_t c;
|
|
|
|
|
|
|
|
r = safe_atou64(v, &c);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse IO accounting base value %s, ignoring.", v);
|
|
|
|
else
|
|
|
|
u->io_accounting_base[m] = c;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
m = string_table_lookup(io_accounting_metric_field_last, ELEMENTSOF(io_accounting_metric_field_last), l);
|
|
|
|
if (m >= 0) {
|
|
|
|
uint64_t c;
|
|
|
|
|
|
|
|
r = safe_atou64(v, &c);
|
|
|
|
if (r < 0)
|
|
|
|
log_unit_debug(u, "Failed to parse IO accounting last value %s, ignoring.", v);
|
|
|
|
else
|
|
|
|
u->io_accounting_last[m] = c;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2014-05-16 01:15:03 +02:00
|
|
|
if (unit_can_serialize(u)) {
|
2018-02-06 08:00:34 +01:00
|
|
|
r = exec_runtime_deserialize_compat(u, l, v, fds);
|
|
|
|
if (r < 0) {
|
|
|
|
log_unit_warning(u, "Failed to deserialize runtime parameter '%s', ignoring.", l);
|
|
|
|
continue;
|
2014-05-16 01:15:03 +02:00
|
|
|
}
|
|
|
|
|
2018-02-06 08:00:34 +01:00
|
|
|
/* Returns positive if key was handled by the call */
|
|
|
|
if (r > 0)
|
|
|
|
continue;
|
|
|
|
|
2014-05-16 01:15:03 +02:00
|
|
|
r = UNIT_VTABLE(u)->deserialize_item(u, l, v, fds);
|
2013-11-27 20:23:18 +01:00
|
|
|
if (r < 0)
|
core,network: major per-object logging rework
This changes log_unit_info() (and friends) to take a real Unit* object
insted of just a unit name as parameter. The call will now prefix all
logged messages with the unit name, thus allowing the unit name to be
dropped from the various passed romat strings, simplifying invocations
drastically, and unifying log output across messages. Also, UNIT= vs.
USER_UNIT= is now derived from the Manager object attached to the Unit
object, instead of getpid(). This has the benefit of correcting the
field for --test runs.
Also contains a couple of other logging improvements:
- Drops a couple of strerror() invocations in favour of using %m.
- Not only .mount units now warn if a symlinks exist for the mount
point already, .automount units do that too, now.
- A few invocations of log_struct() that didn't actually pass any
additional structured data have been replaced by simpler invocations
of log_unit_info() and friends.
- For structured data a new LOG_UNIT_MESSAGE() macro has been added,
that works like LOG_MESSAGE() but prefixes the message with the unit
name. Similar, there's now LOG_LINK_MESSAGE() and
LOG_NETDEV_MESSAGE().
- For structured data new LOG_UNIT_ID(), LOG_LINK_INTERFACE(),
LOG_NETDEV_INTERFACE() macros have been added that generate the
necessary per object fields. The old log_unit_struct() call has been
removed in favour of these new macros used in raw log_struct()
invocations. In addition to removing one more function call this
allows generated structured log messages that contain two object
fields, as necessary for example for network interfaces that are
joined into another network interface, and whose messages shall be
indexed by both.
- The LOG_ERRNO() macro has been removed, in favour of
log_struct_errno(). The latter has the benefit of ensuring that %m in
format strings is properly resolved to the specified error number.
- A number of logging messages have been converted to use
log_unit_info() instead of log_info()
- The client code in sysv-generator no longer #includes core code from
src/core/.
- log_unit_full_errno() has been removed, log_unit_full() instead takes
an errno now, too.
- log_unit_info(), log_link_info(), log_netdev_info() and friends, now
avoid double evaluation of their parameters
2015-05-11 20:38:21 +02:00
|
|
|
log_unit_warning(u, "Failed to deserialize unit parameter '%s', ignoring.", l);
|
2013-11-27 20:23:18 +01:00
|
|
|
}
|
2010-04-21 03:27:44 +02:00
|
|
|
}
|
2016-02-01 16:01:25 +01:00
|
|
|
|
|
|
|
/* Versions before 228 did not carry a state change timestamp. In this case, take the current time. This is
|
|
|
|
* useful, so that timeouts based on this timestamp don't trigger too early, and is in-line with the logic from
|
2016-01-13 14:52:51 +01:00
|
|
|
* before 228 where the base for timeouts was not persistent across reboots. */
|
2016-02-01 16:01:25 +01:00
|
|
|
|
|
|
|
if (!dual_timestamp_is_set(&u->state_change_timestamp))
|
|
|
|
dual_timestamp_get(&u->state_change_timestamp);
|
|
|
|
|
2017-09-07 14:32:33 +02:00
|
|
|
/* Let's make sure that everything that is deserialized also gets any potential new cgroup settings applied
|
|
|
|
* after we are done. For that we invalidate anything already realized, so that we can realize it again. */
|
|
|
|
unit_invalidate_cgroup(u, _CGROUP_MASK_ALL);
|
|
|
|
unit_invalidate_cgroup_bpf(u);
|
|
|
|
|
2016-02-01 16:01:25 +01:00
|
|
|
return 0;
|
2010-04-21 03:27:44 +02:00
|
|
|
}
|
|
|
|
|
2018-10-17 18:36:24 +02:00
|
|
|
int unit_deserialize_skip(FILE *f) {
|
|
|
|
int r;
|
2017-07-31 08:05:35 +02:00
|
|
|
assert(f);
|
|
|
|
|
|
|
|
/* Skip serialized data for this unit. We don't know what it is. */
|
|
|
|
|
|
|
|
for (;;) {
|
2018-10-17 18:36:24 +02:00
|
|
|
_cleanup_free_ char *line = NULL;
|
|
|
|
char *l;
|
2017-07-31 08:05:35 +02:00
|
|
|
|
2018-10-17 18:36:24 +02:00
|
|
|
r = read_line(f, LONG_LINE_MAX, &line);
|
|
|
|
if (r < 0)
|
|
|
|
return log_error_errno(r, "Failed to read serialization line: %m");
|
|
|
|
if (r == 0)
|
|
|
|
return 0;
|
2017-07-31 08:05:35 +02:00
|
|
|
|
|
|
|
l = strstrip(line);
|
|
|
|
|
|
|
|
/* End marker */
|
|
|
|
if (isempty(l))
|
2018-10-17 18:36:24 +02:00
|
|
|
return 1;
|
2017-07-31 08:05:35 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-10-28 18:50:43 +01:00
|
|
|
int unit_add_node_dependency(Unit *u, const char *what, UnitDependency dep, UnitDependencyMask mask) {
|
2013-07-25 17:36:01 +02:00
|
|
|
_cleanup_free_ char *e = NULL;
|
2020-01-21 18:19:08 +01:00
|
|
|
Unit *device;
|
2010-05-13 03:07:16 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Adds in links to the device node that this unit is based on */
|
2015-04-24 17:28:06 +02:00
|
|
|
if (isempty(what))
|
|
|
|
return 0;
|
2010-05-13 03:07:16 +02:00
|
|
|
|
2010-05-16 18:13:58 +02:00
|
|
|
if (!is_device_path(what))
|
2010-05-13 03:07:16 +02:00
|
|
|
return 0;
|
|
|
|
|
2020-01-21 18:19:08 +01:00
|
|
|
/* When device units aren't supported (such as in a container), don't create dependencies on them. */
|
2015-04-30 01:29:00 +02:00
|
|
|
if (!unit_type_supported(UNIT_DEVICE))
|
2015-04-24 17:28:06 +02:00
|
|
|
return 0;
|
|
|
|
|
2015-04-30 20:21:00 +02:00
|
|
|
r = unit_name_from_path(what, ".device", &e);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2010-05-13 03:07:16 +02:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
r = manager_load_unit(u->manager, e, NULL, NULL, &device);
|
2010-05-13 03:07:16 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2016-12-16 17:13:58 +01:00
|
|
|
if (dep == UNIT_REQUIRES && device_shall_be_bound_by(device, u))
|
|
|
|
dep = UNIT_BINDS_TO;
|
|
|
|
|
2019-10-28 18:50:43 +01:00
|
|
|
return unit_add_two_dependencies(u, UNIT_AFTER,
|
|
|
|
MANAGER_IS_SYSTEM(u->manager) ? dep : UNIT_WANTS,
|
|
|
|
device, true, mask);
|
2010-05-13 03:07:16 +02:00
|
|
|
}
|
2010-04-21 03:27:44 +02:00
|
|
|
|
2020-01-21 18:19:08 +01:00
|
|
|
int unit_add_blockdev_dependency(Unit *u, const char *what, UnitDependencyMask mask) {
|
|
|
|
_cleanup_free_ char *escaped = NULL, *target = NULL;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (isempty(what))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (!path_startswith(what, "/dev/"))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* If we don't support devices, then also don't bother with blockdev@.target */
|
|
|
|
if (!unit_type_supported(UNIT_DEVICE))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
r = unit_name_path_escape(what, &escaped);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
r = unit_name_build("blockdev", escaped, ".target", &target);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
return unit_add_dependency_by_name(u, UNIT_AFTER, target, true, mask);
|
|
|
|
}
|
|
|
|
|
2015-04-24 15:27:19 +02:00
|
|
|
int unit_coldplug(Unit *u) {
|
2016-08-15 18:12:01 +02:00
|
|
|
int r = 0, q;
|
|
|
|
char **i;
|
2019-07-19 12:28:04 +02:00
|
|
|
Job *uj;
|
2010-06-03 14:26:50 +02:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
2018-06-05 16:53:22 +02:00
|
|
|
/* Make sure we don't enter a loop, when coldplugging recursively. */
|
2015-04-24 16:04:50 +02:00
|
|
|
if (u->coldplugged)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
u->coldplugged = true;
|
|
|
|
|
2016-08-15 18:12:01 +02:00
|
|
|
STRV_FOREACH(i, u->deserialized_refs) {
|
|
|
|
q = bus_unit_track_add_name(u, *i);
|
|
|
|
if (q < 0 && r >= 0)
|
|
|
|
r = q;
|
|
|
|
}
|
|
|
|
u->deserialized_refs = strv_free(u->deserialized_refs);
|
2010-06-03 14:26:50 +02:00
|
|
|
|
2016-08-15 18:12:01 +02:00
|
|
|
if (UNIT_VTABLE(u)->coldplug) {
|
|
|
|
q = UNIT_VTABLE(u)->coldplug(u);
|
|
|
|
if (q < 0 && r >= 0)
|
|
|
|
r = q;
|
|
|
|
}
|
2015-11-10 20:38:30 +01:00
|
|
|
|
2019-07-19 12:28:04 +02:00
|
|
|
uj = u->job ?: u->nop_job;
|
|
|
|
if (uj) {
|
|
|
|
q = job_coldplug(uj);
|
2016-08-15 18:12:01 +02:00
|
|
|
if (q < 0 && r >= 0)
|
|
|
|
r = q;
|
|
|
|
}
|
2010-06-03 14:26:50 +02:00
|
|
|
|
2016-08-15 18:12:01 +02:00
|
|
|
return r;
|
2010-06-03 14:26:50 +02:00
|
|
|
}
|
|
|
|
|
2018-06-05 16:53:22 +02:00
|
|
|
void unit_catchup(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (UNIT_VTABLE(u)->catchup)
|
|
|
|
UNIT_VTABLE(u)->catchup(u);
|
|
|
|
}
|
|
|
|
|
2016-10-17 07:15:03 +02:00
|
|
|
static bool fragment_mtime_newer(const char *path, usec_t mtime, bool path_masked) {
|
2016-03-31 06:08:34 +02:00
|
|
|
struct stat st;
|
|
|
|
|
|
|
|
if (!path)
|
|
|
|
return false;
|
|
|
|
|
2017-03-01 16:25:08 +01:00
|
|
|
/* If the source is some virtual kernel file system, then we assume we watch it anyway, and hence pretend we
|
|
|
|
* are never out-of-date. */
|
|
|
|
if (PATH_STARTSWITH_SET(path, "/proc", "/sys"))
|
|
|
|
return false;
|
|
|
|
|
2016-03-31 06:08:34 +02:00
|
|
|
if (stat(path, &st) < 0)
|
|
|
|
/* What, cannot access this anymore? */
|
|
|
|
return true;
|
|
|
|
|
2016-10-17 07:15:03 +02:00
|
|
|
if (path_masked)
|
|
|
|
/* For masked files check if they are still so */
|
|
|
|
return !null_or_empty(&st);
|
|
|
|
else
|
2016-03-31 06:22:33 +02:00
|
|
|
/* For non-empty files check the mtime */
|
2016-05-02 14:50:27 +02:00
|
|
|
return timespec_load(&st.st_mtim) > mtime;
|
2016-03-31 06:08:34 +02:00
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2010-07-17 00:57:51 +02:00
|
|
|
bool unit_need_daemon_reload(Unit *u) {
|
2013-04-01 12:32:35 +02:00
|
|
|
_cleanup_strv_free_ char **t = NULL;
|
|
|
|
char **path;
|
2012-05-22 23:08:24 +02:00
|
|
|
|
2010-07-17 00:57:51 +02:00
|
|
|
assert(u);
|
|
|
|
|
2016-10-17 07:15:03 +02:00
|
|
|
/* For unit files, we allow masking… */
|
|
|
|
if (fragment_mtime_newer(u->fragment_path, u->fragment_mtime,
|
|
|
|
u->load_state == UNIT_MASKED))
|
2016-03-31 06:08:34 +02:00
|
|
|
return true;
|
2011-06-15 15:34:19 +02:00
|
|
|
|
2016-10-17 07:15:03 +02:00
|
|
|
/* Source paths should not be masked… */
|
|
|
|
if (fragment_mtime_newer(u->source_path, u->source_mtime, false))
|
2016-05-02 15:07:40 +02:00
|
|
|
return true;
|
2013-04-01 12:32:35 +02:00
|
|
|
|
2017-10-18 08:38:50 +02:00
|
|
|
if (u->load_state == UNIT_LOADED)
|
|
|
|
(void) unit_find_dropin_paths(u, &t);
|
2016-05-02 15:07:40 +02:00
|
|
|
if (!strv_equal(u->dropin_paths, t))
|
|
|
|
return true;
|
2016-04-12 11:10:57 +02:00
|
|
|
|
2016-10-17 07:15:03 +02:00
|
|
|
/* … any drop-ins that are masked are simply omitted from the list. */
|
2016-05-02 15:07:40 +02:00
|
|
|
STRV_FOREACH(path, u->dropin_paths)
|
2016-10-17 07:15:03 +02:00
|
|
|
if (fragment_mtime_newer(*path, u->dropin_mtime, false))
|
2016-05-02 15:07:40 +02:00
|
|
|
return true;
|
2016-03-31 06:08:34 +02:00
|
|
|
|
2016-05-02 15:07:40 +02:00
|
|
|
return false;
|
2010-07-17 00:57:51 +02:00
|
|
|
}
|
|
|
|
|
2010-08-31 00:23:34 +02:00
|
|
|
void unit_reset_failed(Unit *u) {
|
2010-07-18 04:58:01 +02:00
|
|
|
assert(u);
|
|
|
|
|
2010-08-31 00:23:34 +02:00
|
|
|
if (UNIT_VTABLE(u)->reset_failed)
|
|
|
|
UNIT_VTABLE(u)->reset_failed(u);
|
core: make the StartLimitXYZ= settings generic and apply to any kind of unit, not just services
This moves the StartLimitBurst=, StartLimitInterval=, StartLimitAction=, RebootArgument= from the [Service] section
into the [Unit] section of unit files, and thus support it in all unit types, not just in services.
This way we can enforce the start limit much earlier, in particular before testing the unit conditions, so that
repeated start-up failure due to failed conditions is also considered for the start limit logic.
For compatibility the four options may also be configured in the [Service] section still, but we only document them in
their new section [Unit].
This also renamed the socket unit failure code "service-failed-permanent" into "service-start-limit-hit" to express
more clearly what it is about, after all it's only triggered through the start limit being hit.
Finally, the code in busname_trigger_notify() and socket_trigger_notify() is altered to become more alike.
Fixes: #2467
2016-02-09 18:38:03 +01:00
|
|
|
|
2019-09-19 17:45:41 +02:00
|
|
|
ratelimit_reset(&u->start_ratelimit);
|
core: make the StartLimitXYZ= settings generic and apply to any kind of unit, not just services
This moves the StartLimitBurst=, StartLimitInterval=, StartLimitAction=, RebootArgument= from the [Service] section
into the [Unit] section of unit files, and thus support it in all unit types, not just in services.
This way we can enforce the start limit much earlier, in particular before testing the unit conditions, so that
repeated start-up failure due to failed conditions is also considered for the start limit logic.
For compatibility the four options may also be configured in the [Service] section still, but we only document them in
their new section [Unit].
This also renamed the socket unit failure code "service-failed-permanent" into "service-start-limit-hit" to express
more clearly what it is about, after all it's only triggered through the start limit being hit.
Finally, the code in busname_trigger_notify() and socket_trigger_notify() is altered to become more alike.
Fixes: #2467
2016-02-09 18:38:03 +01:00
|
|
|
u->start_limit_hit = false;
|
2010-07-18 04:58:01 +02:00
|
|
|
}
|
|
|
|
|
2010-07-21 05:00:29 +02:00
|
|
|
Unit *unit_following(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (UNIT_VTABLE(u)->following)
|
|
|
|
return UNIT_VTABLE(u)->following(u);
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2013-04-26 02:57:41 +02:00
|
|
|
bool unit_stop_pending(Unit *u) {
|
2010-09-01 03:35:04 +02:00
|
|
|
assert(u);
|
|
|
|
|
2013-04-26 02:57:41 +02:00
|
|
|
/* This call does check the current state of the unit. It's
|
|
|
|
* hence useful to be called from state change calls of the
|
|
|
|
* unit itself, where the state isn't updated yet. This is
|
|
|
|
* different from unit_inactive_or_pending() which checks both
|
|
|
|
* the current state and for a queued job. */
|
2010-09-01 03:35:04 +02:00
|
|
|
|
2019-10-01 14:58:55 +02:00
|
|
|
return unit_has_job_type(u, JOB_STOP);
|
2013-04-26 02:57:41 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
bool unit_inactive_or_pending(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Returns true if the unit is inactive or going down */
|
2010-09-01 03:35:04 +02:00
|
|
|
|
2013-05-06 22:28:39 +02:00
|
|
|
if (UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)))
|
|
|
|
return true;
|
|
|
|
|
2013-04-26 02:57:41 +02:00
|
|
|
if (unit_stop_pending(u))
|
2010-09-01 03:35:04 +02:00
|
|
|
return true;
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2013-04-26 02:57:41 +02:00
|
|
|
bool unit_active_or_pending(Unit *u) {
|
2010-10-05 19:49:15 +02:00
|
|
|
assert(u);
|
|
|
|
|
2011-12-09 15:24:04 +01:00
|
|
|
/* Returns true if the unit is active or going up */
|
2010-10-05 19:49:15 +02:00
|
|
|
|
|
|
|
if (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)))
|
|
|
|
return true;
|
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
if (u->job &&
|
2017-09-29 00:37:23 +02:00
|
|
|
IN_SET(u->job->type, JOB_START, JOB_RELOAD_OR_START, JOB_RESTART))
|
2010-10-05 19:49:15 +02:00
|
|
|
return true;
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2019-08-22 17:08:16 +02:00
|
|
|
bool unit_will_restart_default(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
2019-10-01 14:58:55 +02:00
|
|
|
return unit_has_job_type(u, JOB_START);
|
2019-08-22 17:08:16 +02:00
|
|
|
}
|
|
|
|
|
2017-12-05 16:51:19 +01:00
|
|
|
bool unit_will_restart(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (!UNIT_VTABLE(u)->will_restart)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
return UNIT_VTABLE(u)->will_restart(u);
|
|
|
|
}
|
|
|
|
|
2013-11-19 21:12:59 +01:00
|
|
|
int unit_kill(Unit *u, KillWho w, int signo, sd_bus_error *error) {
|
2010-10-22 16:11:50 +02:00
|
|
|
assert(u);
|
|
|
|
assert(w >= 0 && w < _KILL_WHO_MAX);
|
2016-04-08 11:27:28 +02:00
|
|
|
assert(SIGNAL_VALID(signo));
|
2010-10-22 16:11:50 +02:00
|
|
|
|
|
|
|
if (!UNIT_VTABLE(u)->kill)
|
2015-03-13 14:08:00 +01:00
|
|
|
return -EOPNOTSUPP;
|
2010-10-22 16:11:50 +02:00
|
|
|
|
2012-07-20 00:00:04 +02:00
|
|
|
return UNIT_VTABLE(u)->kill(u, w, signo, error);
|
2010-10-22 16:11:50 +02:00
|
|
|
}
|
|
|
|
|
2013-07-30 01:54:59 +02:00
|
|
|
static Set *unit_pid_set(pid_t main_pid, pid_t control_pid) {
|
2018-05-14 07:13:57 +02:00
|
|
|
_cleanup_set_free_ Set *pid_set = NULL;
|
2013-07-30 01:54:59 +02:00
|
|
|
int r;
|
|
|
|
|
2014-08-13 01:00:18 +02:00
|
|
|
pid_set = set_new(NULL);
|
2013-07-30 01:54:59 +02:00
|
|
|
if (!pid_set)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/* Exclude the main/control pids from being killed via the cgroup */
|
|
|
|
if (main_pid > 0) {
|
2015-09-03 13:22:51 +02:00
|
|
|
r = set_put(pid_set, PID_TO_PTR(main_pid));
|
2013-07-30 01:54:59 +02:00
|
|
|
if (r < 0)
|
2018-05-11 18:43:40 +02:00
|
|
|
return NULL;
|
2013-07-30 01:54:59 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (control_pid > 0) {
|
2015-09-03 13:22:51 +02:00
|
|
|
r = set_put(pid_set, PID_TO_PTR(control_pid));
|
2013-07-30 01:54:59 +02:00
|
|
|
if (r < 0)
|
2018-05-11 18:43:40 +02:00
|
|
|
return NULL;
|
2013-07-30 01:54:59 +02:00
|
|
|
}
|
|
|
|
|
2018-05-11 18:43:40 +02:00
|
|
|
return TAKE_PTR(pid_set);
|
2013-07-30 01:54:59 +02:00
|
|
|
}
|
|
|
|
|
2013-03-22 23:25:54 +01:00
|
|
|
int unit_kill_common(
|
|
|
|
Unit *u,
|
|
|
|
KillWho who,
|
|
|
|
int signo,
|
|
|
|
pid_t main_pid,
|
|
|
|
pid_t control_pid,
|
2013-11-19 21:12:59 +01:00
|
|
|
sd_bus_error *error) {
|
2013-03-22 23:25:54 +01:00
|
|
|
|
2013-03-02 22:31:09 +01:00
|
|
|
int r = 0;
|
2015-09-18 11:56:53 +02:00
|
|
|
bool killed = false;
|
2013-03-02 22:31:09 +01:00
|
|
|
|
2015-09-18 11:56:53 +02:00
|
|
|
if (IN_SET(who, KILL_MAIN, KILL_MAIN_FAIL)) {
|
2013-03-02 22:31:09 +01:00
|
|
|
if (main_pid < 0)
|
2014-10-31 01:32:17 +01:00
|
|
|
return sd_bus_error_setf(error, BUS_ERROR_NO_SUCH_PROCESS, "%s units have no main processes", unit_type_to_string(u->type));
|
2015-08-28 18:29:02 +02:00
|
|
|
else if (main_pid == 0)
|
2014-10-31 01:32:17 +01:00
|
|
|
return sd_bus_error_set_const(error, BUS_ERROR_NO_SUCH_PROCESS, "No main process to kill");
|
2013-03-02 22:31:09 +01:00
|
|
|
}
|
|
|
|
|
2015-09-18 11:56:53 +02:00
|
|
|
if (IN_SET(who, KILL_CONTROL, KILL_CONTROL_FAIL)) {
|
2013-03-02 22:31:09 +01:00
|
|
|
if (control_pid < 0)
|
2014-10-31 01:32:17 +01:00
|
|
|
return sd_bus_error_setf(error, BUS_ERROR_NO_SUCH_PROCESS, "%s units have no control processes", unit_type_to_string(u->type));
|
2015-08-28 18:29:02 +02:00
|
|
|
else if (control_pid == 0)
|
2014-10-31 01:32:17 +01:00
|
|
|
return sd_bus_error_set_const(error, BUS_ERROR_NO_SUCH_PROCESS, "No control process to kill");
|
2013-03-02 22:31:09 +01:00
|
|
|
}
|
|
|
|
|
2015-09-18 11:56:53 +02:00
|
|
|
if (IN_SET(who, KILL_CONTROL, KILL_CONTROL_FAIL, KILL_ALL, KILL_ALL_FAIL))
|
|
|
|
if (control_pid > 0) {
|
2013-03-02 22:31:09 +01:00
|
|
|
if (kill(control_pid, signo) < 0)
|
|
|
|
r = -errno;
|
2015-09-18 11:56:53 +02:00
|
|
|
else
|
|
|
|
killed = true;
|
|
|
|
}
|
2013-03-02 22:31:09 +01:00
|
|
|
|
2015-09-18 11:56:53 +02:00
|
|
|
if (IN_SET(who, KILL_MAIN, KILL_MAIN_FAIL, KILL_ALL, KILL_ALL_FAIL))
|
|
|
|
if (main_pid > 0) {
|
2013-03-02 22:31:09 +01:00
|
|
|
if (kill(main_pid, signo) < 0)
|
|
|
|
r = -errno;
|
2015-09-18 11:56:53 +02:00
|
|
|
else
|
|
|
|
killed = true;
|
|
|
|
}
|
2013-03-02 22:31:09 +01:00
|
|
|
|
2015-09-18 11:56:53 +02:00
|
|
|
if (IN_SET(who, KILL_ALL, KILL_ALL_FAIL) && u->cgroup_path) {
|
2013-03-02 22:31:09 +01:00
|
|
|
_cleanup_set_free_ Set *pid_set = NULL;
|
|
|
|
int q;
|
|
|
|
|
2013-07-30 01:54:59 +02:00
|
|
|
/* Exclude the main/control pids from being killed via the cgroup */
|
|
|
|
pid_set = unit_pid_set(main_pid, control_pid);
|
2013-03-02 22:31:09 +01:00
|
|
|
if (!pid_set)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2016-07-20 11:16:05 +02:00
|
|
|
q = cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, signo, 0, pid_set, NULL, NULL);
|
2017-10-04 16:01:32 +02:00
|
|
|
if (q < 0 && !IN_SET(q, -EAGAIN, -ESRCH, -ENOENT))
|
2013-03-02 22:31:09 +01:00
|
|
|
r = q;
|
2015-09-18 11:56:53 +02:00
|
|
|
else
|
|
|
|
killed = true;
|
2013-03-02 22:31:09 +01:00
|
|
|
}
|
|
|
|
|
2016-01-10 18:10:08 +01:00
|
|
|
if (r == 0 && !killed && IN_SET(who, KILL_ALL_FAIL, KILL_CONTROL_FAIL))
|
2015-09-18 11:56:53 +02:00
|
|
|
return -ESRCH;
|
|
|
|
|
2013-03-02 22:31:09 +01:00
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2010-11-14 23:47:53 +01:00
|
|
|
int unit_following_set(Unit *u, Set **s) {
|
|
|
|
assert(u);
|
|
|
|
assert(s);
|
|
|
|
|
|
|
|
if (UNIT_VTABLE(u)->following_set)
|
|
|
|
return UNIT_VTABLE(u)->following_set(u, s);
|
|
|
|
|
|
|
|
*s = NULL;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-07-31 18:28:02 +02:00
|
|
|
UnitFileState unit_get_unit_file_state(Unit *u) {
|
install: follow unit file symlinks in /usr, but not /etc when looking for [Install] data
Some distributions use alias unit files via symlinks in /usr to cover
for legacy service names. With this change we'll allow "systemctl
enable" on such aliases.
Previously, our rule was that symlinks are user configuration that
"systemctl enable" + "systemctl disable" creates and removes, while unit
files is where the instructions to do so are store. As a result of the
rule we'd never read install information through symlinks, since that
would mix enablement state with installation instructions.
Now, the new rule is that only symlinks inside of /etc are
configuration. Unit files, and symlinks in /usr are now valid for
installation instructions.
This patch is quite a rework of the whole install logic, and makes the
following addional changes:
- Adds a complete test "test-instal-root" that tests the install logic
pretty comprehensively.
- Never uses canonicalize_file_name(), because that's incompatible with
operation relative to a specific root directory.
- unit_file_get_state() is reworked to return a proper error, and
returns the state in a call-by-ref parameter. This cleans up confusion
between the enum type and errno-like errors.
- The new logic puts a limit on how long to follow unit file symlinks:
it will do so only for 64 steps at max.
- The InstallContext object's fields are renamed to will_process and
has_processed (will_install and has_installed) since they are also
used for deinstallation and all kinds of other operations.
- The root directory is always verified before use.
- install.c is reordered to place the exported functions together.
- Stricter rules are followed when traversing symlinks: the unit suffix
must say identical, and it's not allowed to link between regular units
and templated units.
- Various modernizations
- The "invalid" unit file state has been renamed to "bad", in order to
avoid confusion between UNIT_FILE_INVALID and
_UNIT_FILE_STATE_INVALID. Given that the state should normally not be
seen and is not documented this should not be a problematic change.
The new name is now documented however.
Fixes #1375, #1718, #1706
2015-10-08 22:31:56 +02:00
|
|
|
int r;
|
|
|
|
|
2011-07-31 18:28:02 +02:00
|
|
|
assert(u);
|
|
|
|
|
install: follow unit file symlinks in /usr, but not /etc when looking for [Install] data
Some distributions use alias unit files via symlinks in /usr to cover
for legacy service names. With this change we'll allow "systemctl
enable" on such aliases.
Previously, our rule was that symlinks are user configuration that
"systemctl enable" + "systemctl disable" creates and removes, while unit
files is where the instructions to do so are store. As a result of the
rule we'd never read install information through symlinks, since that
would mix enablement state with installation instructions.
Now, the new rule is that only symlinks inside of /etc are
configuration. Unit files, and symlinks in /usr are now valid for
installation instructions.
This patch is quite a rework of the whole install logic, and makes the
following addional changes:
- Adds a complete test "test-instal-root" that tests the install logic
pretty comprehensively.
- Never uses canonicalize_file_name(), because that's incompatible with
operation relative to a specific root directory.
- unit_file_get_state() is reworked to return a proper error, and
returns the state in a call-by-ref parameter. This cleans up confusion
between the enum type and errno-like errors.
- The new logic puts a limit on how long to follow unit file symlinks:
it will do so only for 64 steps at max.
- The InstallContext object's fields are renamed to will_process and
has_processed (will_install and has_installed) since they are also
used for deinstallation and all kinds of other operations.
- The root directory is always verified before use.
- install.c is reordered to place the exported functions together.
- Stricter rules are followed when traversing symlinks: the unit suffix
must say identical, and it's not allowed to link between regular units
and templated units.
- Various modernizations
- The "invalid" unit file state has been renamed to "bad", in order to
avoid confusion between UNIT_FILE_INVALID and
_UNIT_FILE_STATE_INVALID. Given that the state should normally not be
seen and is not documented this should not be a problematic change.
The new name is now documented however.
Fixes #1375, #1718, #1706
2015-10-08 22:31:56 +02:00
|
|
|
if (u->unit_file_state < 0 && u->fragment_path) {
|
|
|
|
r = unit_file_get_state(
|
2016-02-24 21:24:23 +01:00
|
|
|
u->manager->unit_file_scope,
|
install: follow unit file symlinks in /usr, but not /etc when looking for [Install] data
Some distributions use alias unit files via symlinks in /usr to cover
for legacy service names. With this change we'll allow "systemctl
enable" on such aliases.
Previously, our rule was that symlinks are user configuration that
"systemctl enable" + "systemctl disable" creates and removes, while unit
files is where the instructions to do so are store. As a result of the
rule we'd never read install information through symlinks, since that
would mix enablement state with installation instructions.
Now, the new rule is that only symlinks inside of /etc are
configuration. Unit files, and symlinks in /usr are now valid for
installation instructions.
This patch is quite a rework of the whole install logic, and makes the
following addional changes:
- Adds a complete test "test-instal-root" that tests the install logic
pretty comprehensively.
- Never uses canonicalize_file_name(), because that's incompatible with
operation relative to a specific root directory.
- unit_file_get_state() is reworked to return a proper error, and
returns the state in a call-by-ref parameter. This cleans up confusion
between the enum type and errno-like errors.
- The new logic puts a limit on how long to follow unit file symlinks:
it will do so only for 64 steps at max.
- The InstallContext object's fields are renamed to will_process and
has_processed (will_install and has_installed) since they are also
used for deinstallation and all kinds of other operations.
- The root directory is always verified before use.
- install.c is reordered to place the exported functions together.
- Stricter rules are followed when traversing symlinks: the unit suffix
must say identical, and it's not allowed to link between regular units
and templated units.
- Various modernizations
- The "invalid" unit file state has been renamed to "bad", in order to
avoid confusion between UNIT_FILE_INVALID and
_UNIT_FILE_STATE_INVALID. Given that the state should normally not be
seen and is not documented this should not be a problematic change.
The new name is now documented however.
Fixes #1375, #1718, #1706
2015-10-08 22:31:56 +02:00
|
|
|
NULL,
|
2018-02-07 14:08:02 +01:00
|
|
|
u->id,
|
install: follow unit file symlinks in /usr, but not /etc when looking for [Install] data
Some distributions use alias unit files via symlinks in /usr to cover
for legacy service names. With this change we'll allow "systemctl
enable" on such aliases.
Previously, our rule was that symlinks are user configuration that
"systemctl enable" + "systemctl disable" creates and removes, while unit
files is where the instructions to do so are store. As a result of the
rule we'd never read install information through symlinks, since that
would mix enablement state with installation instructions.
Now, the new rule is that only symlinks inside of /etc are
configuration. Unit files, and symlinks in /usr are now valid for
installation instructions.
This patch is quite a rework of the whole install logic, and makes the
following addional changes:
- Adds a complete test "test-instal-root" that tests the install logic
pretty comprehensively.
- Never uses canonicalize_file_name(), because that's incompatible with
operation relative to a specific root directory.
- unit_file_get_state() is reworked to return a proper error, and
returns the state in a call-by-ref parameter. This cleans up confusion
between the enum type and errno-like errors.
- The new logic puts a limit on how long to follow unit file symlinks:
it will do so only for 64 steps at max.
- The InstallContext object's fields are renamed to will_process and
has_processed (will_install and has_installed) since they are also
used for deinstallation and all kinds of other operations.
- The root directory is always verified before use.
- install.c is reordered to place the exported functions together.
- Stricter rules are followed when traversing symlinks: the unit suffix
must say identical, and it's not allowed to link between regular units
and templated units.
- Various modernizations
- The "invalid" unit file state has been renamed to "bad", in order to
avoid confusion between UNIT_FILE_INVALID and
_UNIT_FILE_STATE_INVALID. Given that the state should normally not be
seen and is not documented this should not be a problematic change.
The new name is now documented however.
Fixes #1375, #1718, #1706
2015-10-08 22:31:56 +02:00
|
|
|
&u->unit_file_state);
|
|
|
|
if (r < 0)
|
|
|
|
u->unit_file_state = UNIT_FILE_BAD;
|
|
|
|
}
|
2011-07-31 18:28:02 +02:00
|
|
|
|
2012-01-15 12:04:08 +01:00
|
|
|
return u->unit_file_state;
|
2011-07-31 18:28:02 +02:00
|
|
|
}
|
|
|
|
|
2014-12-02 02:38:18 +01:00
|
|
|
int unit_get_unit_file_preset(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (u->unit_file_preset < 0 && u->fragment_path)
|
|
|
|
u->unit_file_preset = unit_file_query_preset(
|
2016-02-24 21:24:23 +01:00
|
|
|
u->manager->unit_file_scope,
|
install: follow unit file symlinks in /usr, but not /etc when looking for [Install] data
Some distributions use alias unit files via symlinks in /usr to cover
for legacy service names. With this change we'll allow "systemctl
enable" on such aliases.
Previously, our rule was that symlinks are user configuration that
"systemctl enable" + "systemctl disable" creates and removes, while unit
files is where the instructions to do so are store. As a result of the
rule we'd never read install information through symlinks, since that
would mix enablement state with installation instructions.
Now, the new rule is that only symlinks inside of /etc are
configuration. Unit files, and symlinks in /usr are now valid for
installation instructions.
This patch is quite a rework of the whole install logic, and makes the
following addional changes:
- Adds a complete test "test-instal-root" that tests the install logic
pretty comprehensively.
- Never uses canonicalize_file_name(), because that's incompatible with
operation relative to a specific root directory.
- unit_file_get_state() is reworked to return a proper error, and
returns the state in a call-by-ref parameter. This cleans up confusion
between the enum type and errno-like errors.
- The new logic puts a limit on how long to follow unit file symlinks:
it will do so only for 64 steps at max.
- The InstallContext object's fields are renamed to will_process and
has_processed (will_install and has_installed) since they are also
used for deinstallation and all kinds of other operations.
- The root directory is always verified before use.
- install.c is reordered to place the exported functions together.
- Stricter rules are followed when traversing symlinks: the unit suffix
must say identical, and it's not allowed to link between regular units
and templated units.
- Various modernizations
- The "invalid" unit file state has been renamed to "bad", in order to
avoid confusion between UNIT_FILE_INVALID and
_UNIT_FILE_STATE_INVALID. Given that the state should normally not be
seen and is not documented this should not be a problematic change.
The new name is now documented however.
Fixes #1375, #1718, #1706
2015-10-08 22:31:56 +02:00
|
|
|
NULL,
|
2020-04-30 22:37:34 +02:00
|
|
|
basename(u->fragment_path),
|
|
|
|
NULL);
|
2014-12-02 02:38:18 +01:00
|
|
|
|
|
|
|
return u->unit_file_preset;
|
|
|
|
}
|
|
|
|
|
2018-02-13 13:12:43 +01:00
|
|
|
Unit* unit_ref_set(UnitRef *ref, Unit *source, Unit *target) {
|
2012-01-06 23:08:54 +01:00
|
|
|
assert(ref);
|
2018-02-13 13:12:43 +01:00
|
|
|
assert(source);
|
|
|
|
assert(target);
|
2012-01-06 23:08:54 +01:00
|
|
|
|
2018-02-13 13:12:43 +01:00
|
|
|
if (ref->target)
|
2012-01-06 23:08:54 +01:00
|
|
|
unit_ref_unset(ref);
|
|
|
|
|
2018-02-13 13:12:43 +01:00
|
|
|
ref->source = source;
|
|
|
|
ref->target = target;
|
|
|
|
LIST_PREPEND(refs_by_target, target->refs_by_target, ref);
|
|
|
|
return target;
|
2012-01-06 23:08:54 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
void unit_ref_unset(UnitRef *ref) {
|
|
|
|
assert(ref);
|
|
|
|
|
2018-02-13 13:12:43 +01:00
|
|
|
if (!ref->target)
|
2012-01-06 23:08:54 +01:00
|
|
|
return;
|
|
|
|
|
2016-04-29 11:18:53 +02:00
|
|
|
/* We are about to drop a reference to the unit, make sure the garbage collection has a look at it as it might
|
|
|
|
* be unreferenced now. */
|
2018-02-13 13:12:43 +01:00
|
|
|
unit_add_to_gc_queue(ref->target);
|
2016-04-29 11:18:53 +02:00
|
|
|
|
2018-02-13 13:12:43 +01:00
|
|
|
LIST_REMOVE(refs_by_target, ref->target->refs_by_target, ref);
|
|
|
|
ref->source = ref->target = NULL;
|
2012-01-06 23:08:54 +01:00
|
|
|
}
|
|
|
|
|
2016-07-14 12:37:28 +02:00
|
|
|
static int user_from_unit_name(Unit *u, char **ret) {
|
|
|
|
|
|
|
|
static const uint8_t hash_key[] = {
|
|
|
|
0x58, 0x1a, 0xaf, 0xe6, 0x28, 0x58, 0x4e, 0x96,
|
|
|
|
0xb4, 0x4e, 0xf5, 0x3b, 0x8c, 0x92, 0x07, 0xec
|
|
|
|
};
|
|
|
|
|
|
|
|
_cleanup_free_ char *n = NULL;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
r = unit_name_to_prefix(u->id, &n);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2020-04-04 12:23:02 +02:00
|
|
|
if (valid_user_group_name(n, 0)) {
|
2018-03-22 16:53:26 +01:00
|
|
|
*ret = TAKE_PTR(n);
|
2016-07-14 12:37:28 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* If we can't use the unit name as a user name, then let's hash it and use that */
|
|
|
|
if (asprintf(ret, "_du%016" PRIx64, siphash24(n, strlen(n), hash_key)) < 0)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
int unit_patch_contexts(Unit *u) {
|
|
|
|
CGroupContext *cc;
|
|
|
|
ExecContext *ec;
|
2012-07-20 00:09:35 +02:00
|
|
|
unsigned i;
|
|
|
|
int r;
|
|
|
|
|
2012-07-16 12:44:42 +02:00
|
|
|
assert(u);
|
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
/* Patch in the manager defaults into the exec and cgroup
|
|
|
|
* contexts, _after_ the rest of the settings have been
|
|
|
|
* initialized */
|
2014-02-24 23:50:10 +01:00
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
ec = unit_get_exec_context(u);
|
|
|
|
if (ec) {
|
|
|
|
/* This only copies in the ones that need memory */
|
|
|
|
for (i = 0; i < _RLIMIT_MAX; i++)
|
|
|
|
if (u->manager->rlimit[i] && !ec->rlimit[i]) {
|
|
|
|
ec->rlimit[i] = newdup(struct rlimit, u->manager->rlimit[i], 1);
|
|
|
|
if (!ec->rlimit[i])
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
|
2016-02-24 21:24:23 +01:00
|
|
|
if (MANAGER_IS_USER(u->manager) &&
|
2014-03-19 20:40:05 +01:00
|
|
|
!ec->working_directory) {
|
|
|
|
|
|
|
|
r = get_home_dir(&ec->working_directory);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2015-02-12 12:21:16 +01:00
|
|
|
|
|
|
|
/* Allow user services to run, even if the
|
|
|
|
* home directory is missing */
|
|
|
|
ec->working_directory_missing_ok = true;
|
2012-07-20 00:09:35 +02:00
|
|
|
}
|
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
if (ec->private_devices)
|
2016-10-07 20:38:05 +02:00
|
|
|
ec->capability_bounding_set &= ~((UINT64_C(1) << CAP_MKNOD) | (UINT64_C(1) << CAP_SYS_RAWIO));
|
2016-10-12 13:31:21 +02:00
|
|
|
|
|
|
|
if (ec->protect_kernel_modules)
|
|
|
|
ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYS_MODULE);
|
2016-07-14 12:37:28 +02:00
|
|
|
|
2019-11-05 02:18:42 +01:00
|
|
|
if (ec->protect_kernel_logs)
|
|
|
|
ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYSLOG);
|
|
|
|
|
2020-01-26 21:23:33 +01:00
|
|
|
if (ec->protect_clock)
|
|
|
|
ec->capability_bounding_set &= ~((UINT64_C(1) << CAP_SYS_TIME) | (UINT64_C(1) << CAP_WAKE_ALARM));
|
|
|
|
|
2016-07-14 12:37:28 +02:00
|
|
|
if (ec->dynamic_user) {
|
|
|
|
if (!ec->user) {
|
|
|
|
r = user_from_unit_name(u, &ec->user);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!ec->group) {
|
|
|
|
ec->group = strdup(ec->user);
|
|
|
|
if (!ec->group)
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
|
2019-03-20 20:19:38 +01:00
|
|
|
/* If the dynamic user option is on, let's make sure that the unit can't leave its
|
|
|
|
* UID/GID around in the file system or on IPC objects. Hence enforce a strict
|
|
|
|
* sandbox. */
|
2016-08-25 16:12:46 +02:00
|
|
|
|
2016-07-14 12:37:28 +02:00
|
|
|
ec->private_tmp = true;
|
2016-08-01 19:24:40 +02:00
|
|
|
ec->remove_ipc = true;
|
2016-08-25 16:12:46 +02:00
|
|
|
ec->protect_system = PROTECT_SYSTEM_STRICT;
|
|
|
|
if (ec->protect_home == PROTECT_HOME_NO)
|
|
|
|
ec->protect_home = PROTECT_HOME_READ_ONLY;
|
2019-03-20 20:19:38 +01:00
|
|
|
|
|
|
|
/* Make sure this service can neither benefit from SUID/SGID binaries nor create
|
|
|
|
* them. */
|
|
|
|
ec->no_new_privileges = true;
|
|
|
|
ec->restrict_suid_sgid = true;
|
2016-07-14 12:37:28 +02:00
|
|
|
}
|
2012-07-20 00:09:35 +02:00
|
|
|
}
|
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
cc = unit_get_cgroup_context(u);
|
2018-08-06 07:02:28 +02:00
|
|
|
if (cc && ec) {
|
2014-02-26 02:28:52 +01:00
|
|
|
|
2018-08-06 07:02:28 +02:00
|
|
|
if (ec->private_devices &&
|
2019-11-08 15:12:23 +01:00
|
|
|
cc->device_policy == CGROUP_DEVICE_POLICY_AUTO)
|
|
|
|
cc->device_policy = CGROUP_DEVICE_POLICY_CLOSED;
|
2018-08-06 07:02:28 +02:00
|
|
|
|
2020-07-14 17:18:41 +02:00
|
|
|
if ((ec->root_image || !LIST_IS_EMPTY(ec->mount_images)) &&
|
2019-11-08 15:12:23 +01:00
|
|
|
(cc->device_policy != CGROUP_DEVICE_POLICY_AUTO || cc->device_allow)) {
|
2020-06-26 13:19:48 +02:00
|
|
|
const char *p;
|
2018-08-06 07:02:28 +02:00
|
|
|
|
2020-07-14 17:18:41 +02:00
|
|
|
/* When RootImage= or MountImages= is specified, the following devices are touched. */
|
2020-06-26 13:19:48 +02:00
|
|
|
FOREACH_STRING(p, "/dev/loop-control", "/dev/mapper/control") {
|
|
|
|
r = cgroup_add_device_allow(cc, p, "rw");
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
FOREACH_STRING(p, "block-loop", "block-blkext", "block-device-mapper") {
|
|
|
|
r = cgroup_add_device_allow(cc, p, "rwm");
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
2020-01-07 18:53:31 +01:00
|
|
|
|
2020-06-26 13:19:48 +02:00
|
|
|
/* Make sure "block-loop" can be resolved, i.e. make sure "loop" shows up in /proc/devices.
|
|
|
|
* Same for mapper and verity. */
|
|
|
|
FOREACH_STRING(p, "modprobe@loop.service", "modprobe@dm_mod.service", "modprobe@dm_verity.service") {
|
|
|
|
r = unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_WANTS, p, true, UNIT_DEPENDENCY_FILE);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
2018-08-06 07:02:28 +02:00
|
|
|
}
|
2020-01-26 21:23:33 +01:00
|
|
|
|
|
|
|
if (ec->protect_clock) {
|
|
|
|
r = cgroup_add_device_allow(cc, "char-rtc", "r");
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
2014-03-19 20:40:05 +01:00
|
|
|
}
|
2014-03-18 17:58:19 +01:00
|
|
|
|
2012-07-20 00:09:35 +02:00
|
|
|
return 0;
|
2012-07-16 12:44:42 +02:00
|
|
|
}
|
|
|
|
|
2012-09-18 11:40:01 +02:00
|
|
|
ExecContext *unit_get_exec_context(Unit *u) {
|
|
|
|
size_t offset;
|
|
|
|
assert(u);
|
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
if (u->type < 0)
|
|
|
|
return NULL;
|
|
|
|
|
2012-09-18 11:40:01 +02:00
|
|
|
offset = UNIT_VTABLE(u)->exec_context_offset;
|
|
|
|
if (offset <= 0)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return (ExecContext*) ((uint8_t*) u + offset);
|
|
|
|
}
|
|
|
|
|
2013-11-19 21:12:59 +01:00
|
|
|
KillContext *unit_get_kill_context(Unit *u) {
|
|
|
|
size_t offset;
|
|
|
|
assert(u);
|
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
if (u->type < 0)
|
|
|
|
return NULL;
|
|
|
|
|
2013-11-19 21:12:59 +01:00
|
|
|
offset = UNIT_VTABLE(u)->kill_context_offset;
|
|
|
|
if (offset <= 0)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return (KillContext*) ((uint8_t*) u + offset);
|
|
|
|
}
|
|
|
|
|
2013-06-27 04:14:27 +02:00
|
|
|
CGroupContext *unit_get_cgroup_context(Unit *u) {
|
|
|
|
size_t offset;
|
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
if (u->type < 0)
|
|
|
|
return NULL;
|
|
|
|
|
2013-06-27 04:14:27 +02:00
|
|
|
offset = UNIT_VTABLE(u)->cgroup_context_offset;
|
|
|
|
if (offset <= 0)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return (CGroupContext*) ((uint8_t*) u + offset);
|
|
|
|
}
|
|
|
|
|
2013-11-27 20:23:18 +01:00
|
|
|
ExecRuntime *unit_get_exec_runtime(Unit *u) {
|
|
|
|
size_t offset;
|
|
|
|
|
2014-03-19 20:40:05 +01:00
|
|
|
if (u->type < 0)
|
|
|
|
return NULL;
|
|
|
|
|
2013-11-27 20:23:18 +01:00
|
|
|
offset = UNIT_VTABLE(u)->exec_runtime_offset;
|
|
|
|
if (offset <= 0)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return *(ExecRuntime**) ((uint8_t*) u + offset);
|
|
|
|
}
|
|
|
|
|
2017-11-22 15:03:51 +01:00
|
|
|
static const char* unit_drop_in_dir(Unit *u, UnitWriteFlags flags) {
|
2015-08-28 16:05:32 +02:00
|
|
|
assert(u);
|
|
|
|
|
2017-11-22 15:03:51 +01:00
|
|
|
if (UNIT_WRITE_FLAGS_NOOP(flags))
|
2016-04-07 15:43:59 +02:00
|
|
|
return NULL;
|
|
|
|
|
2016-02-25 01:13:57 +01:00
|
|
|
if (u->transient) /* Redirect drop-ins for transient units always into the transient directory. */
|
|
|
|
return u->manager->lookup_paths.transient;
|
2013-02-27 18:50:41 +01:00
|
|
|
|
2017-11-22 15:03:51 +01:00
|
|
|
if (flags & UNIT_PERSISTENT)
|
2016-04-07 15:43:59 +02:00
|
|
|
return u->manager->lookup_paths.persistent_control;
|
2013-02-27 18:50:41 +01:00
|
|
|
|
2017-11-22 15:03:51 +01:00
|
|
|
if (flags & UNIT_RUNTIME)
|
|
|
|
return u->manager->lookup_paths.runtime_control;
|
|
|
|
|
2016-02-25 01:13:57 +01:00
|
|
|
return NULL;
|
2013-01-19 01:01:41 +01:00
|
|
|
}
|
|
|
|
|
2017-11-22 15:03:51 +01:00
|
|
|
char* unit_escape_setting(const char *s, UnitWriteFlags flags, char **buf) {
|
|
|
|
char *ret = NULL;
|
|
|
|
|
|
|
|
if (!s)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/* Escapes the input string as requested. Returns the escaped string. If 'buf' is specified then the allocated
|
|
|
|
* return buffer pointer is also written to *buf, except if no escaping was necessary, in which case *buf is
|
|
|
|
* set to NULL, and the input pointer is returned as-is. This means the return value always contains a properly
|
|
|
|
* escaped version, but *buf when passed only contains a pointer if an allocation was necessary. If *buf is
|
|
|
|
* not specified, then the return value always needs to be freed. Callers can use this to optimize memory
|
|
|
|
* allocations. */
|
|
|
|
|
|
|
|
if (flags & UNIT_ESCAPE_SPECIFIERS) {
|
|
|
|
ret = specifier_escape(s);
|
|
|
|
if (!ret)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
s = ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (flags & UNIT_ESCAPE_C) {
|
|
|
|
char *a;
|
|
|
|
|
|
|
|
a = cescape(s);
|
|
|
|
free(ret);
|
|
|
|
if (!a)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
ret = a;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (buf) {
|
|
|
|
*buf = ret;
|
|
|
|
return ret ?: (char*) s;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret ?: strdup(s);
|
|
|
|
}
|
|
|
|
|
|
|
|
char* unit_concat_strv(char **l, UnitWriteFlags flags) {
|
|
|
|
_cleanup_free_ char *result = NULL;
|
|
|
|
size_t n = 0, allocated = 0;
|
2018-03-22 16:53:26 +01:00
|
|
|
char **i;
|
2017-11-22 15:03:51 +01:00
|
|
|
|
|
|
|
/* Takes a list of strings, escapes them, and concatenates them. This may be used to format command lines in a
|
|
|
|
* way suitable for ExecStart= stanzas */
|
|
|
|
|
|
|
|
STRV_FOREACH(i, l) {
|
|
|
|
_cleanup_free_ char *buf = NULL;
|
|
|
|
const char *p;
|
|
|
|
size_t a;
|
|
|
|
char *q;
|
|
|
|
|
|
|
|
p = unit_escape_setting(*i, flags, &buf);
|
|
|
|
if (!p)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
a = (n > 0) + 1 + strlen(p) + 1; /* separating space + " + entry + " */
|
|
|
|
if (!GREEDY_REALLOC(result, allocated, n + a + 1))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
q = result + n;
|
|
|
|
if (n > 0)
|
|
|
|
*(q++) = ' ';
|
|
|
|
|
|
|
|
*(q++) = '"';
|
|
|
|
q = stpcpy(q, p);
|
|
|
|
*(q++) = '"';
|
|
|
|
|
|
|
|
n += a;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!GREEDY_REALLOC(result, allocated, n + 1))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
result[n] = 0;
|
|
|
|
|
2018-03-22 16:53:26 +01:00
|
|
|
return TAKE_PTR(result);
|
2017-11-22 15:03:51 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
int unit_write_setting(Unit *u, UnitWriteFlags flags, const char *name, const char *data) {
|
|
|
|
_cleanup_free_ char *p = NULL, *q = NULL, *escaped = NULL;
|
2016-05-28 22:06:44 +02:00
|
|
|
const char *dir, *wrapped;
|
2013-02-27 18:50:41 +01:00
|
|
|
int r;
|
2013-01-19 01:01:41 +01:00
|
|
|
|
|
|
|
assert(u);
|
2017-11-22 15:03:51 +01:00
|
|
|
assert(name);
|
|
|
|
assert(data);
|
|
|
|
|
|
|
|
if (UNIT_WRITE_FLAGS_NOOP(flags))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
data = unit_escape_setting(data, flags, &escaped);
|
|
|
|
if (!data)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
/* Prefix the section header. If we are writing this out as transient file, then let's suppress this if the
|
|
|
|
* previous section header is the same */
|
|
|
|
|
|
|
|
if (flags & UNIT_PRIVATE) {
|
|
|
|
if (!UNIT_VTABLE(u)->private_section)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (!u->transient_file || u->last_section_private < 0)
|
|
|
|
data = strjoina("[", UNIT_VTABLE(u)->private_section, "]\n", data);
|
|
|
|
else if (u->last_section_private == 0)
|
|
|
|
data = strjoina("\n[", UNIT_VTABLE(u)->private_section, "]\n", data);
|
|
|
|
} else {
|
|
|
|
if (!u->transient_file || u->last_section_private < 0)
|
|
|
|
data = strjoina("[Unit]\n", data);
|
|
|
|
else if (u->last_section_private > 0)
|
|
|
|
data = strjoina("\n[Unit]\n", data);
|
|
|
|
}
|
2013-01-19 01:01:41 +01:00
|
|
|
|
2016-04-07 15:43:59 +02:00
|
|
|
if (u->transient_file) {
|
|
|
|
/* When this is a transient unit file in creation, then let's not create a new drop-in but instead
|
|
|
|
* write to the transient unit file. */
|
|
|
|
fputs(data, u->transient_file);
|
|
|
|
|
2017-11-22 15:03:51 +01:00
|
|
|
if (!endswith(data, "\n"))
|
|
|
|
fputc('\n', u->transient_file);
|
|
|
|
|
|
|
|
/* Remember which section we wrote this entry to */
|
|
|
|
u->last_section_private = !!(flags & UNIT_PRIVATE);
|
2013-06-27 21:14:56 +02:00
|
|
|
return 0;
|
2017-11-22 15:03:51 +01:00
|
|
|
}
|
2013-06-27 21:14:56 +02:00
|
|
|
|
2017-11-22 15:03:51 +01:00
|
|
|
dir = unit_drop_in_dir(u, flags);
|
2016-02-25 01:13:57 +01:00
|
|
|
if (!dir)
|
|
|
|
return -EINVAL;
|
2013-01-19 01:01:41 +01:00
|
|
|
|
2016-05-28 22:06:44 +02:00
|
|
|
wrapped = strjoina("# This is a drop-in unit file extension, created via \"systemctl set-property\"\n"
|
2016-06-09 19:53:45 +02:00
|
|
|
"# or an equivalent operation. Do not edit.\n",
|
2016-05-28 22:06:44 +02:00
|
|
|
data,
|
|
|
|
"\n");
|
2016-04-08 18:10:32 +02:00
|
|
|
|
2016-04-08 18:13:02 +02:00
|
|
|
r = drop_in_file(dir, u->id, 50, name, &p, &q);
|
2014-12-09 13:46:30 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2017-11-23 17:45:58 +01:00
|
|
|
(void) mkdir_p_label(p, 0755);
|
2019-08-21 09:14:52 +02:00
|
|
|
|
|
|
|
/* Make sure the drop-in dir is registered in our path cache. This way we don't need to stupidly
|
|
|
|
* recreate the cache after every drop-in we write. */
|
|
|
|
if (u->manager->unit_path_cache) {
|
2020-04-29 08:47:51 +02:00
|
|
|
r = set_put_strdup(&u->manager->unit_path_cache, p);
|
2019-08-21 09:14:52 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2016-05-28 22:06:44 +02:00
|
|
|
r = write_string_file_atomic_label(q, wrapped);
|
2014-12-09 13:46:30 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2016-04-08 18:13:02 +02:00
|
|
|
r = strv_push(&u->dropin_paths, q);
|
2014-12-09 13:46:30 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2016-04-08 18:13:02 +02:00
|
|
|
q = NULL;
|
2014-12-09 13:46:30 +01:00
|
|
|
|
|
|
|
strv_uniq(u->dropin_paths);
|
|
|
|
|
|
|
|
u->dropin_mtime = now(CLOCK_REALTIME);
|
|
|
|
|
|
|
|
return 0;
|
2013-02-27 18:50:41 +01:00
|
|
|
}
|
2013-01-19 01:01:41 +01:00
|
|
|
|
2017-11-22 15:03:51 +01:00
|
|
|
int unit_write_settingf(Unit *u, UnitWriteFlags flags, const char *name, const char *format, ...) {
|
2013-07-11 21:29:33 +02:00
|
|
|
_cleanup_free_ char *p = NULL;
|
|
|
|
va_list ap;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(name);
|
|
|
|
assert(format);
|
|
|
|
|
2017-11-22 15:03:51 +01:00
|
|
|
if (UNIT_WRITE_FLAGS_NOOP(flags))
|
2013-07-11 21:29:33 +02:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
va_start(ap, format);
|
|
|
|
r = vasprintf(&p, format, ap);
|
|
|
|
va_end(ap);
|
|
|
|
|
|
|
|
if (r < 0)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2017-11-22 15:03:51 +01:00
|
|
|
return unit_write_setting(u, flags, name, p);
|
2013-07-11 21:29:33 +02:00
|
|
|
}
|
2013-01-19 01:01:41 +01:00
|
|
|
|
2013-06-28 04:12:58 +02:00
|
|
|
int unit_make_transient(Unit *u) {
|
2017-11-23 17:39:25 +01:00
|
|
|
_cleanup_free_ char *path = NULL;
|
2016-04-07 15:43:59 +02:00
|
|
|
FILE *f;
|
|
|
|
|
2013-06-28 04:12:58 +02:00
|
|
|
assert(u);
|
|
|
|
|
2015-08-28 16:05:32 +02:00
|
|
|
if (!UNIT_VTABLE(u)->can_transient)
|
|
|
|
return -EOPNOTSUPP;
|
|
|
|
|
2017-11-23 17:45:58 +01:00
|
|
|
(void) mkdir_p_label(u->manager->lookup_paths.transient, 0755);
|
|
|
|
|
2019-06-20 20:07:01 +02:00
|
|
|
path = path_join(u->manager->lookup_paths.transient, u->id);
|
2016-04-07 15:43:59 +02:00
|
|
|
if (!path)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
/* Let's open the file we'll write the transient settings into. This file is kept open as long as we are
|
|
|
|
* creating the transient, and is closed in unit_load(), as soon as we start loading the file. */
|
|
|
|
|
2016-04-09 04:20:22 +02:00
|
|
|
RUN_WITH_UMASK(0022) {
|
2016-04-07 15:43:59 +02:00
|
|
|
f = fopen(path, "we");
|
2017-11-23 17:39:25 +01:00
|
|
|
if (!f)
|
2016-04-09 04:20:22 +02:00
|
|
|
return -errno;
|
2016-04-07 15:43:59 +02:00
|
|
|
}
|
|
|
|
|
2017-11-23 17:39:25 +01:00
|
|
|
safe_fclose(u->transient_file);
|
2016-04-07 15:43:59 +02:00
|
|
|
u->transient_file = f;
|
|
|
|
|
2017-11-23 17:39:25 +01:00
|
|
|
free_and_replace(u->fragment_path, path);
|
2015-11-17 14:09:16 +01:00
|
|
|
|
|
|
|
u->source_path = mfree(u->source_path);
|
|
|
|
u->dropin_paths = strv_free(u->dropin_paths);
|
|
|
|
u->fragment_mtime = u->source_mtime = u->dropin_mtime = 0;
|
|
|
|
|
2016-04-07 15:43:59 +02:00
|
|
|
u->load_state = UNIT_STUB;
|
|
|
|
u->load_error = 0;
|
|
|
|
u->transient = true;
|
|
|
|
|
2015-11-17 14:09:16 +01:00
|
|
|
unit_add_to_dbus_queue(u);
|
|
|
|
unit_add_to_gc_queue(u);
|
2013-06-28 04:12:58 +02:00
|
|
|
|
2016-04-07 15:43:59 +02:00
|
|
|
fputs("# This is a transient unit file, created programmatically via the systemd API. Do not edit.\n",
|
|
|
|
u->transient_file);
|
|
|
|
|
2015-08-28 16:05:32 +02:00
|
|
|
return 0;
|
2013-06-28 04:12:58 +02:00
|
|
|
}
|
|
|
|
|
2019-01-16 11:20:18 +01:00
|
|
|
static int log_kill(pid_t pid, int sig, void *userdata) {
|
2016-07-20 11:16:05 +02:00
|
|
|
_cleanup_free_ char *comm = NULL;
|
|
|
|
|
|
|
|
(void) get_process_comm(pid, &comm);
|
|
|
|
|
|
|
|
/* Don't log about processes marked with brackets, under the assumption that these are temporary processes
|
|
|
|
only, like for example systemd's own PAM stub process. */
|
|
|
|
if (comm && comm[0] == '(')
|
2019-01-16 11:20:18 +01:00
|
|
|
return 0;
|
2016-07-20 11:16:05 +02:00
|
|
|
|
|
|
|
log_unit_notice(userdata,
|
|
|
|
"Killing process " PID_FMT " (%s) with signal SIG%s.",
|
|
|
|
pid,
|
|
|
|
strna(comm),
|
|
|
|
signal_to_string(sig));
|
2019-01-16 11:20:18 +01:00
|
|
|
|
|
|
|
return 1;
|
2016-07-20 11:16:05 +02:00
|
|
|
}
|
|
|
|
|
2019-10-01 15:53:42 +02:00
|
|
|
static int operation_to_signal(const KillContext *c, KillOperation k, bool *noteworthy) {
|
2016-07-20 11:16:05 +02:00
|
|
|
assert(c);
|
|
|
|
|
|
|
|
switch (k) {
|
|
|
|
|
|
|
|
case KILL_TERMINATE:
|
|
|
|
case KILL_TERMINATE_AND_LOG:
|
2019-10-01 15:53:42 +02:00
|
|
|
*noteworthy = false;
|
2016-07-20 11:16:05 +02:00
|
|
|
return c->kill_signal;
|
|
|
|
|
2019-10-01 15:15:06 +02:00
|
|
|
case KILL_RESTART:
|
2019-10-01 15:53:42 +02:00
|
|
|
*noteworthy = false;
|
2019-10-01 15:15:06 +02:00
|
|
|
return restart_kill_signal(c);
|
|
|
|
|
2016-07-20 11:16:05 +02:00
|
|
|
case KILL_KILL:
|
2019-10-01 15:53:42 +02:00
|
|
|
*noteworthy = true;
|
2018-07-20 17:22:43 +02:00
|
|
|
return c->final_kill_signal;
|
2016-07-20 11:16:05 +02:00
|
|
|
|
2018-09-19 21:03:01 +02:00
|
|
|
case KILL_WATCHDOG:
|
2019-10-01 15:53:42 +02:00
|
|
|
*noteworthy = true;
|
2018-09-19 21:03:01 +02:00
|
|
|
return c->watchdog_signal;
|
2016-07-20 11:16:05 +02:00
|
|
|
|
|
|
|
default:
|
|
|
|
assert_not_reached("KillOperation unknown");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-01-26 05:53:30 +01:00
|
|
|
int unit_kill_context(
|
|
|
|
Unit *u,
|
|
|
|
KillContext *c,
|
2014-10-28 16:35:40 +01:00
|
|
|
KillOperation k,
|
2013-01-26 05:53:30 +01:00
|
|
|
pid_t main_pid,
|
|
|
|
pid_t control_pid,
|
|
|
|
bool main_pid_alien) {
|
|
|
|
|
2016-07-20 11:16:05 +02:00
|
|
|
bool wait_for_exit = false, send_sighup;
|
2016-11-15 21:01:40 +01:00
|
|
|
cg_kill_log_func_t log_func = NULL;
|
2015-09-01 18:51:44 +02:00
|
|
|
int sig, r;
|
2013-01-26 05:53:30 +01:00
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(c);
|
|
|
|
|
2016-11-15 21:01:40 +01:00
|
|
|
/* Kill the processes belonging to this unit, in preparation for shutting the unit down.
|
|
|
|
* Returns > 0 if we killed something worth waiting for, 0 otherwise. */
|
2016-07-20 11:16:05 +02:00
|
|
|
|
2013-01-26 05:53:30 +01:00
|
|
|
if (c->kill_mode == KILL_NONE)
|
|
|
|
return 0;
|
|
|
|
|
2019-10-01 15:53:42 +02:00
|
|
|
bool noteworthy;
|
|
|
|
sig = operation_to_signal(c, k, ¬eworthy);
|
|
|
|
if (noteworthy)
|
|
|
|
log_func = log_kill;
|
2016-07-20 11:16:05 +02:00
|
|
|
|
|
|
|
send_sighup =
|
|
|
|
c->send_sighup &&
|
|
|
|
IN_SET(k, KILL_TERMINATE, KILL_TERMINATE_AND_LOG) &&
|
|
|
|
sig != SIGHUP;
|
|
|
|
|
2013-01-26 05:53:30 +01:00
|
|
|
if (main_pid > 0) {
|
2016-07-20 11:16:05 +02:00
|
|
|
if (log_func)
|
|
|
|
log_func(main_pid, sig, u);
|
2013-01-26 05:53:30 +01:00
|
|
|
|
2016-07-20 11:16:05 +02:00
|
|
|
r = kill_and_sigcont(main_pid, sig);
|
2013-01-26 05:53:30 +01:00
|
|
|
if (r < 0 && r != -ESRCH) {
|
|
|
|
_cleanup_free_ char *comm = NULL;
|
2016-07-20 11:16:05 +02:00
|
|
|
(void) get_process_comm(main_pid, &comm);
|
2013-01-26 05:53:30 +01:00
|
|
|
|
2015-09-01 18:51:44 +02:00
|
|
|
log_unit_warning_errno(u, r, "Failed to kill main process " PID_FMT " (%s), ignoring: %m", main_pid, strna(comm));
|
2013-07-30 01:54:59 +02:00
|
|
|
} else {
|
2014-01-29 20:12:18 +01:00
|
|
|
if (!main_pid_alien)
|
|
|
|
wait_for_exit = true;
|
2013-07-30 01:54:59 +02:00
|
|
|
|
2016-07-20 11:16:05 +02:00
|
|
|
if (r != -ESRCH && send_sighup)
|
2015-09-01 18:54:08 +02:00
|
|
|
(void) kill(main_pid, SIGHUP);
|
2013-07-30 01:54:59 +02:00
|
|
|
}
|
2013-01-26 05:53:30 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
if (control_pid > 0) {
|
2016-07-20 11:16:05 +02:00
|
|
|
if (log_func)
|
|
|
|
log_func(control_pid, sig, u);
|
2013-01-26 05:53:30 +01:00
|
|
|
|
2016-07-20 11:16:05 +02:00
|
|
|
r = kill_and_sigcont(control_pid, sig);
|
2013-01-26 05:53:30 +01:00
|
|
|
if (r < 0 && r != -ESRCH) {
|
|
|
|
_cleanup_free_ char *comm = NULL;
|
2016-07-20 11:16:05 +02:00
|
|
|
(void) get_process_comm(control_pid, &comm);
|
2013-01-26 05:53:30 +01:00
|
|
|
|
2015-09-01 18:51:44 +02:00
|
|
|
log_unit_warning_errno(u, r, "Failed to kill control process " PID_FMT " (%s), ignoring: %m", control_pid, strna(comm));
|
2013-07-30 01:54:59 +02:00
|
|
|
} else {
|
2013-01-26 05:53:30 +01:00
|
|
|
wait_for_exit = true;
|
2013-07-30 01:54:59 +02:00
|
|
|
|
2016-07-20 11:16:05 +02:00
|
|
|
if (r != -ESRCH && send_sighup)
|
2015-09-01 18:54:08 +02:00
|
|
|
(void) kill(control_pid, SIGHUP);
|
2013-07-30 01:54:59 +02:00
|
|
|
}
|
2013-01-26 05:53:30 +01:00
|
|
|
}
|
|
|
|
|
2015-09-01 18:51:44 +02:00
|
|
|
if (u->cgroup_path &&
|
|
|
|
(c->kill_mode == KILL_CONTROL_GROUP || (c->kill_mode == KILL_MIXED && k == KILL_KILL))) {
|
2013-01-26 05:53:30 +01:00
|
|
|
_cleanup_set_free_ Set *pid_set = NULL;
|
|
|
|
|
2013-07-30 01:54:59 +02:00
|
|
|
/* Exclude the main/control pids from being killed via the cgroup */
|
|
|
|
pid_set = unit_pid_set(main_pid, control_pid);
|
2013-01-26 05:53:30 +01:00
|
|
|
if (!pid_set)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2016-07-20 11:16:05 +02:00
|
|
|
r = cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
|
|
|
|
sig,
|
|
|
|
CGROUP_SIGCONT|CGROUP_IGNORE_SELF,
|
|
|
|
pid_set,
|
|
|
|
log_func, u);
|
2013-01-26 05:53:30 +01:00
|
|
|
if (r < 0) {
|
2017-10-04 16:01:32 +02:00
|
|
|
if (!IN_SET(r, -EAGAIN, -ESRCH, -ENOENT))
|
2015-09-01 18:51:44 +02:00
|
|
|
log_unit_warning_errno(u, r, "Failed to kill control group %s, ignoring: %m", u->cgroup_path);
|
|
|
|
|
2013-07-30 01:54:59 +02:00
|
|
|
} else if (r > 0) {
|
2014-01-29 20:12:18 +01:00
|
|
|
|
2018-02-06 11:57:35 +01:00
|
|
|
/* FIXME: For now, on the legacy hierarchy, we will not wait for the cgroup members to die if
|
|
|
|
* we are running in a container or if this is a delegation unit, simply because cgroup
|
|
|
|
* notification is unreliable in these cases. It doesn't work at all in containers, and outside
|
|
|
|
* of containers it can be confused easily by left-over directories in the cgroup — which
|
|
|
|
* however should not exist in non-delegated units. On the unified hierarchy that's different,
|
|
|
|
* there we get proper events. Hence rely on them. */
|
core: unified cgroup hierarchy support
This patch set adds full support the new unified cgroup hierarchy logic
of modern kernels.
A new kernel command line option "systemd.unified_cgroup_hierarchy=1" is
added. If specified the unified hierarchy is mounted to /sys/fs/cgroup
instead of a tmpfs. No further hierarchies are mounted. The kernel
command line option defaults to off. We can turn it on by default as
soon as the kernel's APIs regarding this are stabilized (but even then
downstream distros might want to turn this off, as this will break any
tools that access cgroupfs directly).
It is possibly to choose for each boot individually whether the unified
or the legacy hierarchy is used. nspawn will by default provide the
legacy hierarchy to containers if the host is using it, and the unified
otherwise. However it is possible to run containers with the unified
hierarchy on a legacy host and vice versa, by setting the
$UNIFIED_CGROUP_HIERARCHY environment variable for nspawn to 1 or 0,
respectively.
The unified hierarchy provides reliable cgroup empty notifications for
the first time, via inotify. To make use of this we maintain one
manager-wide inotify fd, and each cgroup to it.
This patch also removes cg_delete() which is unused now.
On kernel 4.2 only the "memory" controller is compatible with the
unified hierarchy, hence that's the only controller systemd exposes when
booted in unified heirarchy mode.
This introduces a new enum for enumerating supported controllers, plus a
related enum for the mask bits mapping to it. The core is changed to
make use of this everywhere.
This moves PID 1 into a new "init.scope" implicit scope unit in the root
slice. This is necessary since on the unified hierarchy cgroups may
either contain subgroups or processes but not both. PID 1 hence has to
move out of the root cgroup (strictly speaking the root cgroup is the
only one where processes and subgroups are still allowed, but in order
to support containers nicey, we move PID 1 into the new scope in all
cases.) This new unit is also used on legacy hierarchy setups. It's
actually pretty useful on all systems, as it can then be used to filter
journal messages coming from PID 1, and so on.
The root slice ("-.slice") is now implicitly created and started (and
does not require a unit file on disk anymore), since
that's where "init.scope" is located and the slice needs to be started
before the scope can.
To check whether we are in unified or legacy hierarchy mode we use
statfs() on /sys/fs/cgroup. If the .f_type field reports tmpfs we are in
legacy mode, if it reports cgroupfs we are in unified mode.
This patch set carefuly makes sure that cgls and cgtop continue to work
as desired.
When invoking nspawn as a service it will implicitly create two
subcgroups in the cgroup it is using, one to move the nspawn process
into, the other to move the actual container processes into. This is
done because of the requirement that cgroups may either contain
processes or other subgroups.
2015-09-01 19:22:36 +02:00
|
|
|
|
2017-02-24 18:00:04 +01:00
|
|
|
if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0 ||
|
2018-02-06 11:57:35 +01:00
|
|
|
(detect_container() == 0 && !unit_cgroup_delegate(u)))
|
2015-09-01 17:25:59 +02:00
|
|
|
wait_for_exit = true;
|
2014-01-29 13:38:55 +01:00
|
|
|
|
2016-07-20 11:16:05 +02:00
|
|
|
if (send_sighup) {
|
2013-07-30 01:54:59 +02:00
|
|
|
set_free(pid_set);
|
|
|
|
|
|
|
|
pid_set = unit_pid_set(main_pid, control_pid);
|
|
|
|
if (!pid_set)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2020-05-26 14:32:03 +02:00
|
|
|
(void) cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
|
|
|
|
SIGHUP,
|
|
|
|
CGROUP_IGNORE_SELF,
|
|
|
|
pid_set,
|
|
|
|
NULL, NULL);
|
2013-07-30 01:54:59 +02:00
|
|
|
}
|
|
|
|
}
|
2013-01-26 05:53:30 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
return wait_for_exit;
|
|
|
|
}
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
int unit_require_mounts_for(Unit *u, const char *path, UnitDependencyMask mask) {
|
2018-03-18 12:51:31 +01:00
|
|
|
_cleanup_free_ char *p = NULL;
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
UnitDependencyInfo di;
|
2013-09-26 20:14:24 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(path);
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
/* Registers a unit for requiring a certain path and all its prefixes. We keep a hashtable of these paths in
|
|
|
|
* the unit (from the path to the UnitDependencyInfo structure indicating how to the dependency came to
|
|
|
|
* be). However, we build a prefix table for all possible prefixes so that new appearing mount units can easily
|
|
|
|
* determine which units to make themselves a dependency of. */
|
2013-09-26 20:14:24 +02:00
|
|
|
|
2014-01-27 07:23:16 +01:00
|
|
|
if (!path_is_absolute(path))
|
|
|
|
return -EINVAL;
|
|
|
|
|
2018-02-08 18:58:35 +01:00
|
|
|
r = hashmap_ensure_allocated(&u->requires_mounts_for, &path_hash_ops);
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2013-09-26 20:14:24 +02:00
|
|
|
p = strdup(path);
|
|
|
|
if (!p)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2019-03-06 23:16:55 +01:00
|
|
|
path = path_simplify(p, true);
|
2013-09-26 20:14:24 +02:00
|
|
|
|
2018-03-18 12:51:31 +01:00
|
|
|
if (!path_is_normalized(path))
|
2013-09-26 20:14:24 +02:00
|
|
|
return -EPERM;
|
|
|
|
|
2018-03-18 12:51:31 +01:00
|
|
|
if (hashmap_contains(u->requires_mounts_for, path))
|
2013-09-26 20:14:24 +02:00
|
|
|
return 0;
|
|
|
|
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
di = (UnitDependencyInfo) {
|
|
|
|
.origin_mask = mask
|
|
|
|
};
|
|
|
|
|
2018-03-18 12:51:31 +01:00
|
|
|
r = hashmap_put(u->requires_mounts_for, path, di.data);
|
|
|
|
if (r < 0)
|
2013-09-26 20:14:24 +02:00
|
|
|
return r;
|
2018-03-18 12:51:31 +01:00
|
|
|
p = NULL;
|
2013-09-26 20:14:24 +02:00
|
|
|
|
2018-12-07 16:38:03 +01:00
|
|
|
char prefix[strlen(path) + 1];
|
2018-03-18 12:51:31 +01:00
|
|
|
PATH_FOREACH_PREFIX_MORE(prefix, path) {
|
2013-09-26 20:14:24 +02:00
|
|
|
Set *x;
|
|
|
|
|
|
|
|
x = hashmap_get(u->manager->units_requiring_mounts_for, prefix);
|
|
|
|
if (!x) {
|
2018-03-18 12:51:31 +01:00
|
|
|
_cleanup_free_ char *q = NULL;
|
2013-09-26 20:14:24 +02:00
|
|
|
|
2018-02-08 18:58:35 +01:00
|
|
|
r = hashmap_ensure_allocated(&u->manager->units_requiring_mounts_for, &path_hash_ops);
|
2015-04-24 19:54:29 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2013-09-26 20:14:24 +02:00
|
|
|
|
|
|
|
q = strdup(prefix);
|
|
|
|
if (!q)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2014-08-13 01:00:18 +02:00
|
|
|
x = set_new(NULL);
|
2018-03-18 12:51:31 +01:00
|
|
|
if (!x)
|
2013-09-26 20:14:24 +02:00
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
r = hashmap_put(u->manager->units_requiring_mounts_for, q, x);
|
|
|
|
if (r < 0) {
|
|
|
|
set_free(x);
|
|
|
|
return r;
|
|
|
|
}
|
2018-03-18 12:51:31 +01:00
|
|
|
q = NULL;
|
2013-09-26 20:14:24 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
r = set_put(x, u);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2013-11-27 20:23:18 +01:00
|
|
|
int unit_setup_exec_runtime(Unit *u) {
|
|
|
|
ExecRuntime **rt;
|
|
|
|
size_t offset;
|
|
|
|
Unit *other;
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
Iterator i;
|
|
|
|
void *v;
|
2018-02-06 08:00:34 +01:00
|
|
|
int r;
|
2013-11-27 20:23:18 +01:00
|
|
|
|
|
|
|
offset = UNIT_VTABLE(u)->exec_runtime_offset;
|
|
|
|
assert(offset > 0);
|
|
|
|
|
2014-08-30 17:13:16 +02:00
|
|
|
/* Check if there already is an ExecRuntime for this unit? */
|
2013-11-27 20:23:18 +01:00
|
|
|
rt = (ExecRuntime**) ((uint8_t*) u + offset);
|
|
|
|
if (*rt)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* Try to get it from somebody else */
|
core: track why unit dependencies came to be
This replaces the dependencies Set* objects by Hashmap* objects, where
the key is the depending Unit, and the value is a bitmask encoding why
the specific dependency was created.
The bitmask contains a number of different, defined bits, that indicate
why dependencies exist, for example whether they are created due to
explicitly configured deps in files, by udev rules or implicitly.
Note that memory usage is not increased by this change, even though we
store more information, as we manage to encode the bit mask inside the
value pointer each Hashmap entry contains.
Why this all? When we know how a dependency came to be, we can update
dependencies correctly when a configuration source changes but others
are left unaltered. Specifically:
1. We can fix UDEV_WANTS dependency generation: so far we kept adding
dependencies configured that way, but if a device lost such a
dependency we couldn't them again as there was no scheme for removing
of dependencies in place.
2. We can implement "pin-pointed" reload of unit files. If we know what
dependencies were created as result of configuration in a unit file,
then we know what to flush out when we want to reload it.
3. It's useful for debugging: "systemd-analyze dump" now shows
this information, helping substantially with understanding how
systemd's dependency tree came to be the way it came to be.
2017-10-25 20:46:01 +02:00
|
|
|
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_JOINS_NAMESPACE_OF], i) {
|
2018-02-06 08:00:34 +01:00
|
|
|
r = exec_runtime_acquire(u->manager, NULL, other->id, false, rt);
|
|
|
|
if (r == 1)
|
|
|
|
return 1;
|
2013-11-27 20:23:18 +01:00
|
|
|
}
|
|
|
|
|
2018-02-06 08:00:34 +01:00
|
|
|
return exec_runtime_acquire(u->manager, unit_get_exec_context(u), u->id, true, rt);
|
2013-11-27 20:23:18 +01:00
|
|
|
}
|
|
|
|
|
2016-07-14 12:37:28 +02:00
|
|
|
int unit_setup_dynamic_creds(Unit *u) {
|
|
|
|
ExecContext *ec;
|
|
|
|
DynamicCreds *dcreds;
|
|
|
|
size_t offset;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
offset = UNIT_VTABLE(u)->dynamic_creds_offset;
|
|
|
|
assert(offset > 0);
|
|
|
|
dcreds = (DynamicCreds*) ((uint8_t*) u + offset);
|
|
|
|
|
|
|
|
ec = unit_get_exec_context(u);
|
|
|
|
assert(ec);
|
|
|
|
|
|
|
|
if (!ec->dynamic_user)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
return dynamic_creds_acquire(dcreds, u->manager, ec->user, ec->group);
|
|
|
|
}
|
|
|
|
|
2015-04-30 01:29:00 +02:00
|
|
|
bool unit_type_supported(UnitType t) {
|
|
|
|
if (_unlikely_(t < 0))
|
|
|
|
return false;
|
|
|
|
if (_unlikely_(t >= _UNIT_TYPE_MAX))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
if (!unit_vtable[t]->supported)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
return unit_vtable[t]->supported();
|
|
|
|
}
|
|
|
|
|
2015-05-11 22:28:52 +02:00
|
|
|
void unit_warn_if_dir_nonempty(Unit *u, const char* where) {
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(where);
|
|
|
|
|
|
|
|
r = dir_is_empty(where);
|
2018-02-06 16:35:52 +01:00
|
|
|
if (r > 0 || r == -ENOTDIR)
|
2015-05-11 22:28:52 +02:00
|
|
|
return;
|
|
|
|
if (r < 0) {
|
|
|
|
log_unit_warning_errno(u, r, "Failed to check directory %s: %m", where);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
log_struct(LOG_NOTICE,
|
tree-wide: add SD_ID128_MAKE_STR, remove LOG_MESSAGE_ID
Embedding sd_id128_t's in constant strings was rather cumbersome. We had
SD_ID128_CONST_STR which returned a const char[], but it had two problems:
- it wasn't possible to statically concatanate this array with a normal string
- gcc wasn't really able to optimize this, and generated code to perform the
"conversion" at runtime.
Because of this, even our own code in coredumpctl wasn't using
SD_ID128_CONST_STR.
Add a new macro to generate a constant string: SD_ID128_MAKE_STR.
It is not as elegant as SD_ID128_CONST_STR, because it requires a repetition
of the numbers, but in practice it is more convenient to use, and allows gcc
to generate smarter code:
$ size .libs/systemd{,-logind,-journald}{.old,}
text data bss dec hex filename
1265204 149564 4808 1419576 15a938 .libs/systemd.old
1260268 149564 4808 1414640 1595f0 .libs/systemd
246805 13852 209 260866 3fb02 .libs/systemd-logind.old
240973 13852 209 255034 3e43a .libs/systemd-logind
146839 4984 34 151857 25131 .libs/systemd-journald.old
146391 4984 34 151409 24f71 .libs/systemd-journald
It is also much easier to check if a certain binary uses a certain MESSAGE_ID:
$ strings .libs/systemd.old|grep MESSAGE_ID
MESSAGE_ID=%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x
MESSAGE_ID=%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x
MESSAGE_ID=%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x
MESSAGE_ID=%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x
$ strings .libs/systemd|grep MESSAGE_ID
MESSAGE_ID=c7a787079b354eaaa9e77b371893cd27
MESSAGE_ID=b07a249cd024414a82dd00cd181378ff
MESSAGE_ID=641257651c1b4ec9a8624d7a40a9e1e7
MESSAGE_ID=de5b426a63be47a7b6ac3eaac82e2f6f
MESSAGE_ID=d34d037fff1847e6ae669a370e694725
MESSAGE_ID=7d4958e842da4a758f6c1cdc7b36dcc5
MESSAGE_ID=1dee0369c7fc4736b7099b38ecb46ee7
MESSAGE_ID=39f53479d3a045ac8e11786248231fbf
MESSAGE_ID=be02cf6855d2428ba40df7e9d022f03d
MESSAGE_ID=7b05ebc668384222baa8881179cfda54
MESSAGE_ID=9d1aaa27d60140bd96365438aad20286
2016-11-06 18:48:23 +01:00
|
|
|
"MESSAGE_ID=" SD_MESSAGE_OVERMOUNTING_STR,
|
2015-05-11 22:28:52 +02:00
|
|
|
LOG_UNIT_ID(u),
|
2017-09-20 18:27:53 +02:00
|
|
|
LOG_UNIT_INVOCATION_ID(u),
|
2015-05-11 22:28:52 +02:00
|
|
|
LOG_UNIT_MESSAGE(u, "Directory %s to mount over is not empty, mounting anyway.", where),
|
2018-06-04 12:59:22 +02:00
|
|
|
"WHERE=%s", where);
|
2015-05-11 22:28:52 +02:00
|
|
|
}
|
|
|
|
|
2018-01-19 18:28:38 +01:00
|
|
|
int unit_fail_if_noncanonical(Unit *u, const char* where) {
|
2018-12-14 08:16:31 +01:00
|
|
|
_cleanup_free_ char *canonical_where = NULL;
|
2015-05-11 22:28:52 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(where);
|
|
|
|
|
2019-10-24 10:33:20 +02:00
|
|
|
r = chase_symlinks(where, NULL, CHASE_NONEXISTENT, &canonical_where, NULL);
|
2015-05-11 22:28:52 +02:00
|
|
|
if (r < 0) {
|
2018-01-19 18:28:38 +01:00
|
|
|
log_unit_debug_errno(u, r, "Failed to check %s for symlinks, ignoring: %m", where);
|
2015-05-11 22:28:52 +02:00
|
|
|
return 0;
|
|
|
|
}
|
2018-01-19 18:28:38 +01:00
|
|
|
|
|
|
|
/* We will happily ignore a trailing slash (or any redundant slashes) */
|
|
|
|
if (path_equal(where, canonical_where))
|
2015-05-11 22:28:52 +02:00
|
|
|
return 0;
|
|
|
|
|
2018-01-19 18:28:38 +01:00
|
|
|
/* No need to mention "." or "..", they would already have been rejected by unit_name_from_path() */
|
2015-05-11 22:28:52 +02:00
|
|
|
log_struct(LOG_ERR,
|
tree-wide: add SD_ID128_MAKE_STR, remove LOG_MESSAGE_ID
Embedding sd_id128_t's in constant strings was rather cumbersome. We had
SD_ID128_CONST_STR which returned a const char[], but it had two problems:
- it wasn't possible to statically concatanate this array with a normal string
- gcc wasn't really able to optimize this, and generated code to perform the
"conversion" at runtime.
Because of this, even our own code in coredumpctl wasn't using
SD_ID128_CONST_STR.
Add a new macro to generate a constant string: SD_ID128_MAKE_STR.
It is not as elegant as SD_ID128_CONST_STR, because it requires a repetition
of the numbers, but in practice it is more convenient to use, and allows gcc
to generate smarter code:
$ size .libs/systemd{,-logind,-journald}{.old,}
text data bss dec hex filename
1265204 149564 4808 1419576 15a938 .libs/systemd.old
1260268 149564 4808 1414640 1595f0 .libs/systemd
246805 13852 209 260866 3fb02 .libs/systemd-logind.old
240973 13852 209 255034 3e43a .libs/systemd-logind
146839 4984 34 151857 25131 .libs/systemd-journald.old
146391 4984 34 151409 24f71 .libs/systemd-journald
It is also much easier to check if a certain binary uses a certain MESSAGE_ID:
$ strings .libs/systemd.old|grep MESSAGE_ID
MESSAGE_ID=%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x
MESSAGE_ID=%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x
MESSAGE_ID=%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x
MESSAGE_ID=%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x
$ strings .libs/systemd|grep MESSAGE_ID
MESSAGE_ID=c7a787079b354eaaa9e77b371893cd27
MESSAGE_ID=b07a249cd024414a82dd00cd181378ff
MESSAGE_ID=641257651c1b4ec9a8624d7a40a9e1e7
MESSAGE_ID=de5b426a63be47a7b6ac3eaac82e2f6f
MESSAGE_ID=d34d037fff1847e6ae669a370e694725
MESSAGE_ID=7d4958e842da4a758f6c1cdc7b36dcc5
MESSAGE_ID=1dee0369c7fc4736b7099b38ecb46ee7
MESSAGE_ID=39f53479d3a045ac8e11786248231fbf
MESSAGE_ID=be02cf6855d2428ba40df7e9d022f03d
MESSAGE_ID=7b05ebc668384222baa8881179cfda54
MESSAGE_ID=9d1aaa27d60140bd96365438aad20286
2016-11-06 18:48:23 +01:00
|
|
|
"MESSAGE_ID=" SD_MESSAGE_OVERMOUNTING_STR,
|
2015-05-11 22:28:52 +02:00
|
|
|
LOG_UNIT_ID(u),
|
2017-09-20 18:27:53 +02:00
|
|
|
LOG_UNIT_INVOCATION_ID(u),
|
2018-01-19 18:28:38 +01:00
|
|
|
LOG_UNIT_MESSAGE(u, "Mount path %s is not canonical (contains a symlink).", where),
|
2018-06-04 12:59:22 +02:00
|
|
|
"WHERE=%s", where);
|
2015-05-11 22:28:52 +02:00
|
|
|
|
|
|
|
return -ELOOP;
|
|
|
|
}
|
2015-11-17 14:04:40 +01:00
|
|
|
|
|
|
|
bool unit_is_pristine(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
2015-11-17 14:09:16 +01:00
|
|
|
/* Check if the unit already exists or is already around,
|
2015-11-17 14:04:40 +01:00
|
|
|
* in a number of different ways. Note that to cater for unit
|
|
|
|
* types such as slice, we are generally fine with units that
|
2018-04-04 14:36:56 +02:00
|
|
|
* are marked UNIT_LOADED even though nothing was actually
|
|
|
|
* loaded, as those unit types don't require a file on disk. */
|
2015-11-17 14:04:40 +01:00
|
|
|
|
|
|
|
return !(!IN_SET(u->load_state, UNIT_NOT_FOUND, UNIT_LOADED) ||
|
|
|
|
u->fragment_path ||
|
|
|
|
u->source_path ||
|
|
|
|
!strv_isempty(u->dropin_paths) ||
|
|
|
|
u->job ||
|
|
|
|
u->merged_into);
|
|
|
|
}
|
2016-04-20 15:28:28 +02:00
|
|
|
|
|
|
|
pid_t unit_control_pid(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (UNIT_VTABLE(u)->control_pid)
|
|
|
|
return UNIT_VTABLE(u)->control_pid(u);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
pid_t unit_main_pid(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (UNIT_VTABLE(u)->main_pid)
|
|
|
|
return UNIT_VTABLE(u)->main_pid(u);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
2016-08-01 19:24:40 +02:00
|
|
|
|
|
|
|
static void unit_unref_uid_internal(
|
|
|
|
Unit *u,
|
|
|
|
uid_t *ref_uid,
|
|
|
|
bool destroy_now,
|
|
|
|
void (*_manager_unref_uid)(Manager *m, uid_t uid, bool destroy_now)) {
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(ref_uid);
|
|
|
|
assert(_manager_unref_uid);
|
|
|
|
|
|
|
|
/* Generic implementation of both unit_unref_uid() and unit_unref_gid(), under the assumption that uid_t and
|
|
|
|
* gid_t are actually the same time, with the same validity rules.
|
|
|
|
*
|
|
|
|
* Drops a reference to UID/GID from a unit. */
|
|
|
|
|
|
|
|
assert_cc(sizeof(uid_t) == sizeof(gid_t));
|
|
|
|
assert_cc(UID_INVALID == (uid_t) GID_INVALID);
|
|
|
|
|
|
|
|
if (!uid_is_valid(*ref_uid))
|
|
|
|
return;
|
|
|
|
|
|
|
|
_manager_unref_uid(u->manager, *ref_uid, destroy_now);
|
|
|
|
*ref_uid = UID_INVALID;
|
|
|
|
}
|
|
|
|
|
2020-01-21 11:51:45 +01:00
|
|
|
static void unit_unref_uid(Unit *u, bool destroy_now) {
|
2016-08-01 19:24:40 +02:00
|
|
|
unit_unref_uid_internal(u, &u->ref_uid, destroy_now, manager_unref_uid);
|
|
|
|
}
|
|
|
|
|
2020-01-21 11:51:45 +01:00
|
|
|
static void unit_unref_gid(Unit *u, bool destroy_now) {
|
2016-08-01 19:24:40 +02:00
|
|
|
unit_unref_uid_internal(u, (uid_t*) &u->ref_gid, destroy_now, manager_unref_gid);
|
|
|
|
}
|
|
|
|
|
2020-01-21 11:51:45 +01:00
|
|
|
void unit_unref_uid_gid(Unit *u, bool destroy_now) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
unit_unref_uid(u, destroy_now);
|
|
|
|
unit_unref_gid(u, destroy_now);
|
|
|
|
}
|
|
|
|
|
2016-08-01 19:24:40 +02:00
|
|
|
static int unit_ref_uid_internal(
|
|
|
|
Unit *u,
|
|
|
|
uid_t *ref_uid,
|
|
|
|
uid_t uid,
|
|
|
|
bool clean_ipc,
|
|
|
|
int (*_manager_ref_uid)(Manager *m, uid_t uid, bool clean_ipc)) {
|
|
|
|
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(ref_uid);
|
|
|
|
assert(uid_is_valid(uid));
|
|
|
|
assert(_manager_ref_uid);
|
|
|
|
|
|
|
|
/* Generic implementation of both unit_ref_uid() and unit_ref_guid(), under the assumption that uid_t and gid_t
|
|
|
|
* are actually the same type, and have the same validity rules.
|
|
|
|
*
|
|
|
|
* Adds a reference on a specific UID/GID to this unit. Each unit referencing the same UID/GID maintains a
|
|
|
|
* reference so that we can destroy the UID/GID's IPC resources as soon as this is requested and the counter
|
|
|
|
* drops to zero. */
|
|
|
|
|
|
|
|
assert_cc(sizeof(uid_t) == sizeof(gid_t));
|
|
|
|
assert_cc(UID_INVALID == (uid_t) GID_INVALID);
|
|
|
|
|
|
|
|
if (*ref_uid == uid)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (uid_is_valid(*ref_uid)) /* Already set? */
|
|
|
|
return -EBUSY;
|
|
|
|
|
|
|
|
r = _manager_ref_uid(u->manager, uid, clean_ipc);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
*ref_uid = uid;
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2020-01-21 11:51:45 +01:00
|
|
|
static int unit_ref_uid(Unit *u, uid_t uid, bool clean_ipc) {
|
2016-08-01 19:24:40 +02:00
|
|
|
return unit_ref_uid_internal(u, &u->ref_uid, uid, clean_ipc, manager_ref_uid);
|
|
|
|
}
|
|
|
|
|
2020-01-21 11:51:45 +01:00
|
|
|
static int unit_ref_gid(Unit *u, gid_t gid, bool clean_ipc) {
|
2016-08-01 19:24:40 +02:00
|
|
|
return unit_ref_uid_internal(u, (uid_t*) &u->ref_gid, (uid_t) gid, clean_ipc, manager_ref_gid);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int unit_ref_uid_gid_internal(Unit *u, uid_t uid, gid_t gid, bool clean_ipc) {
|
|
|
|
int r = 0, q = 0;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Reference both a UID and a GID in one go. Either references both, or neither. */
|
|
|
|
|
|
|
|
if (uid_is_valid(uid)) {
|
|
|
|
r = unit_ref_uid(u, uid, clean_ipc);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (gid_is_valid(gid)) {
|
|
|
|
q = unit_ref_gid(u, gid, clean_ipc);
|
|
|
|
if (q < 0) {
|
|
|
|
if (r > 0)
|
|
|
|
unit_unref_uid(u, false);
|
|
|
|
|
|
|
|
return q;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return r > 0 || q > 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_ref_uid_gid(Unit *u, uid_t uid, gid_t gid) {
|
|
|
|
ExecContext *c;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
c = unit_get_exec_context(u);
|
|
|
|
|
|
|
|
r = unit_ref_uid_gid_internal(u, uid, gid, c ? c->remove_ipc : false);
|
|
|
|
if (r < 0)
|
|
|
|
return log_unit_warning_errno(u, r, "Couldn't add UID/GID reference to unit, proceeding without: %m");
|
|
|
|
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
void unit_notify_user_lookup(Unit *u, uid_t uid, gid_t gid) {
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* This is invoked whenever one of the forked off processes let's us know the UID/GID its user name/group names
|
|
|
|
* resolved to. We keep track of which UID/GID is currently assigned in order to be able to destroy its IPC
|
|
|
|
* objects when no service references the UID/GID anymore. */
|
|
|
|
|
|
|
|
r = unit_ref_uid_gid(u, uid, gid);
|
|
|
|
if (r > 0)
|
2018-11-27 20:09:10 +01:00
|
|
|
unit_add_to_dbus_queue(u);
|
2016-08-01 19:24:40 +02:00
|
|
|
}
|
2016-08-30 23:18:46 +02:00
|
|
|
|
|
|
|
int unit_acquire_invocation_id(Unit *u) {
|
|
|
|
sd_id128_t id;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
r = sd_id128_randomize(&id);
|
|
|
|
if (r < 0)
|
|
|
|
return log_unit_error_errno(u, r, "Failed to generate invocation ID for unit: %m");
|
|
|
|
|
|
|
|
r = unit_set_invocation_id(u, id);
|
|
|
|
if (r < 0)
|
|
|
|
return log_unit_error_errno(u, r, "Failed to set invocation ID for unit: %m");
|
|
|
|
|
2018-11-29 16:39:18 +01:00
|
|
|
unit_add_to_dbus_queue(u);
|
2016-08-30 23:18:46 +02:00
|
|
|
return 0;
|
|
|
|
}
|
2017-08-01 11:02:30 +02:00
|
|
|
|
2018-10-31 15:49:19 +01:00
|
|
|
int unit_set_exec_params(Unit *u, ExecParameters *p) {
|
|
|
|
int r;
|
|
|
|
|
2017-09-22 20:02:23 +02:00
|
|
|
assert(u);
|
|
|
|
assert(p);
|
2017-08-01 11:02:30 +02:00
|
|
|
|
2018-02-06 13:09:52 +01:00
|
|
|
/* Copy parameters from manager */
|
2018-10-31 15:49:19 +01:00
|
|
|
r = manager_get_effective_environment(u->manager, &p->environment);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2018-02-06 13:09:52 +01:00
|
|
|
p->confirm_spawn = manager_get_confirm_spawn(u->manager);
|
|
|
|
p->cgroup_supported = u->manager->cgroup_supported;
|
|
|
|
p->prefix = u->manager->prefix;
|
|
|
|
SET_FLAG(p->flags, EXEC_PASS_LOG_UNIT|EXEC_CHOWN_DIRECTORIES, MANAGER_IS_SYSTEM(u->manager));
|
|
|
|
|
2019-04-27 02:22:40 +02:00
|
|
|
/* Copy parameters from unit */
|
2017-09-22 20:02:23 +02:00
|
|
|
p->cgroup_path = u->cgroup_path;
|
2018-02-06 11:57:35 +01:00
|
|
|
SET_FLAG(p->flags, EXEC_CGROUP_DELEGATE, unit_cgroup_delegate(u));
|
2018-10-31 15:49:19 +01:00
|
|
|
|
2020-07-23 08:49:52 +02:00
|
|
|
p->received_credentials = u->manager->received_credentials;
|
|
|
|
|
2018-10-31 15:49:19 +01:00
|
|
|
return 0;
|
2017-08-01 11:02:30 +02:00
|
|
|
}
|
2017-09-07 11:17:43 +02:00
|
|
|
|
2017-12-22 13:08:14 +01:00
|
|
|
int unit_fork_helper_process(Unit *u, const char *name, pid_t *ret) {
|
2017-09-07 11:17:43 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(ret);
|
|
|
|
|
|
|
|
/* Forks off a helper process and makes sure it is a member of the unit's cgroup. Returns == 0 in the child,
|
|
|
|
* and > 0 in the parent. The pid parameter is always filled in with the child's PID. */
|
|
|
|
|
|
|
|
(void) unit_realize_cgroup(u);
|
|
|
|
|
2017-12-22 13:08:14 +01:00
|
|
|
r = safe_fork(name, FORK_REOPEN_LOG, ret);
|
|
|
|
if (r != 0)
|
|
|
|
return r;
|
2017-09-07 11:17:43 +02:00
|
|
|
|
2017-12-22 13:08:14 +01:00
|
|
|
(void) default_signals(SIGNALS_CRASH_HANDLER, SIGNALS_IGNORE, -1);
|
|
|
|
(void) ignore_signals(SIGPIPE, -1);
|
2017-09-07 11:17:43 +02:00
|
|
|
|
2017-12-22 13:08:14 +01:00
|
|
|
(void) prctl(PR_SET_PDEATHSIG, SIGTERM);
|
2017-09-07 11:17:43 +02:00
|
|
|
|
2017-12-22 13:08:14 +01:00
|
|
|
if (u->cgroup_path) {
|
|
|
|
r = cg_attach_everywhere(u->manager->cgroup_supported, u->cgroup_path, 0, NULL, NULL);
|
|
|
|
if (r < 0) {
|
|
|
|
log_unit_error_errno(u, r, "Failed to join unit cgroup %s: %m", u->cgroup_path);
|
|
|
|
_exit(EXIT_CGROUP);
|
2017-09-07 11:17:43 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-12-22 13:08:14 +01:00
|
|
|
return 0;
|
2017-09-07 11:17:43 +02:00
|
|
|
}
|
2017-10-26 16:39:35 +02:00
|
|
|
|
2019-08-25 10:57:08 +02:00
|
|
|
int unit_fork_and_watch_rm_rf(Unit *u, char **paths, pid_t *ret_pid) {
|
|
|
|
pid_t pid;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(ret_pid);
|
|
|
|
|
|
|
|
r = unit_fork_helper_process(u, "(sd-rmrf)", &pid);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
if (r == 0) {
|
|
|
|
int ret = EXIT_SUCCESS;
|
|
|
|
char **i;
|
|
|
|
|
|
|
|
STRV_FOREACH(i, paths) {
|
|
|
|
r = rm_rf(*i, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_MISSING_OK);
|
|
|
|
if (r < 0) {
|
|
|
|
log_error_errno(r, "Failed to remove '%s': %m", *i);
|
|
|
|
ret = EXIT_FAILURE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
_exit(ret);
|
|
|
|
}
|
|
|
|
|
|
|
|
r = unit_watch_pid(u, pid, true);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
*ret_pid = pid;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2017-10-26 16:39:35 +02:00
|
|
|
static void unit_update_dependency_mask(Unit *u, UnitDependency d, Unit *other, UnitDependencyInfo di) {
|
|
|
|
assert(u);
|
|
|
|
assert(d >= 0);
|
|
|
|
assert(d < _UNIT_DEPENDENCY_MAX);
|
|
|
|
assert(other);
|
|
|
|
|
|
|
|
if (di.origin_mask == 0 && di.destination_mask == 0) {
|
|
|
|
/* No bit set anymore, let's drop the whole entry */
|
|
|
|
assert_se(hashmap_remove(u->dependencies[d], other));
|
2019-11-26 11:33:08 +01:00
|
|
|
log_unit_debug(u, "lost dependency %s=%s", unit_dependency_to_string(d), other->id);
|
2017-10-26 16:39:35 +02:00
|
|
|
} else
|
|
|
|
/* Mask was reduced, let's update the entry */
|
|
|
|
assert_se(hashmap_update(u->dependencies[d], other, di.data) == 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
void unit_remove_dependencies(Unit *u, UnitDependencyMask mask) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Removes all dependencies u has on other units marked for ownership by 'mask'. */
|
|
|
|
|
|
|
|
if (mask == 0)
|
|
|
|
return;
|
|
|
|
|
2020-05-28 15:25:22 +02:00
|
|
|
for (UnitDependency d = 0; d < _UNIT_DEPENDENCY_MAX; d++) {
|
2017-10-26 16:39:35 +02:00
|
|
|
bool done;
|
|
|
|
|
|
|
|
do {
|
|
|
|
UnitDependencyInfo di;
|
|
|
|
Unit *other;
|
|
|
|
Iterator i;
|
|
|
|
|
|
|
|
done = true;
|
|
|
|
|
|
|
|
HASHMAP_FOREACH_KEY(di.data, other, u->dependencies[d], i) {
|
|
|
|
if ((di.origin_mask & ~mask) == di.origin_mask)
|
|
|
|
continue;
|
|
|
|
di.origin_mask &= ~mask;
|
|
|
|
unit_update_dependency_mask(u, d, other, di);
|
|
|
|
|
|
|
|
/* We updated the dependency from our unit to the other unit now. But most dependencies
|
|
|
|
* imply a reverse dependency. Hence, let's delete that one too. For that we go through
|
|
|
|
* all dependency types on the other unit and delete all those which point to us and
|
|
|
|
* have the right mask set. */
|
|
|
|
|
2020-05-28 15:25:22 +02:00
|
|
|
for (UnitDependency q = 0; q < _UNIT_DEPENDENCY_MAX; q++) {
|
2017-10-26 16:39:35 +02:00
|
|
|
UnitDependencyInfo dj;
|
|
|
|
|
|
|
|
dj.data = hashmap_get(other->dependencies[q], u);
|
|
|
|
if ((dj.destination_mask & ~mask) == dj.destination_mask)
|
|
|
|
continue;
|
|
|
|
dj.destination_mask &= ~mask;
|
|
|
|
|
|
|
|
unit_update_dependency_mask(other, q, u, dj);
|
|
|
|
}
|
|
|
|
|
|
|
|
unit_add_to_gc_queue(other);
|
|
|
|
|
|
|
|
done = false;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
} while (!done);
|
|
|
|
}
|
|
|
|
}
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
|
2019-12-12 06:15:42 +01:00
|
|
|
static int unit_get_invocation_path(Unit *u, char **ret) {
|
|
|
|
char *p;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(ret);
|
|
|
|
|
|
|
|
if (MANAGER_IS_SYSTEM(u->manager))
|
|
|
|
p = strjoin("/run/systemd/units/invocation:", u->id);
|
|
|
|
else {
|
|
|
|
_cleanup_free_ char *user_path = NULL;
|
|
|
|
r = xdg_user_runtime_dir(&user_path, "/systemd/units/invocation:");
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
p = strjoin(user_path, u->id);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!p)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
*ret = p;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
static int unit_export_invocation_id(Unit *u) {
|
2019-12-12 06:15:42 +01:00
|
|
|
_cleanup_free_ char *p = NULL;
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (u->exported_invocation_id)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (sd_id128_is_null(u->invocation_id))
|
|
|
|
return 0;
|
|
|
|
|
2019-12-12 06:15:42 +01:00
|
|
|
r = unit_get_invocation_path(u, &p);
|
|
|
|
if (r < 0)
|
|
|
|
return log_unit_debug_errno(u, r, "Failed to get invocation path: %m");
|
|
|
|
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
r = symlink_atomic(u->invocation_id_string, p);
|
|
|
|
if (r < 0)
|
|
|
|
return log_unit_debug_errno(u, r, "Failed to create invocation ID symlink %s: %m", p);
|
|
|
|
|
|
|
|
u->exported_invocation_id = true;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int unit_export_log_level_max(Unit *u, const ExecContext *c) {
|
|
|
|
const char *p;
|
|
|
|
char buf[2];
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(c);
|
|
|
|
|
|
|
|
if (u->exported_log_level_max)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (c->log_level_max < 0)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
assert(c->log_level_max <= 7);
|
|
|
|
|
|
|
|
buf[0] = '0' + c->log_level_max;
|
|
|
|
buf[1] = 0;
|
|
|
|
|
|
|
|
p = strjoina("/run/systemd/units/log-level-max:", u->id);
|
|
|
|
r = symlink_atomic(buf, p);
|
|
|
|
if (r < 0)
|
|
|
|
return log_unit_debug_errno(u, r, "Failed to create maximum log level symlink %s: %m", p);
|
|
|
|
|
|
|
|
u->exported_log_level_max = true;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int unit_export_log_extra_fields(Unit *u, const ExecContext *c) {
|
|
|
|
_cleanup_close_ int fd = -1;
|
|
|
|
struct iovec *iovec;
|
|
|
|
const char *p;
|
|
|
|
char *pattern;
|
|
|
|
le64_t *sizes;
|
|
|
|
ssize_t n;
|
|
|
|
size_t i;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
if (u->exported_log_extra_fields)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (c->n_log_extra_fields <= 0)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
sizes = newa(le64_t, c->n_log_extra_fields);
|
|
|
|
iovec = newa(struct iovec, c->n_log_extra_fields * 2);
|
|
|
|
|
|
|
|
for (i = 0; i < c->n_log_extra_fields; i++) {
|
|
|
|
sizes[i] = htole64(c->log_extra_fields[i].iov_len);
|
|
|
|
|
|
|
|
iovec[i*2] = IOVEC_MAKE(sizes + i, sizeof(le64_t));
|
|
|
|
iovec[i*2+1] = c->log_extra_fields[i];
|
|
|
|
}
|
|
|
|
|
|
|
|
p = strjoina("/run/systemd/units/log-extra-fields:", u->id);
|
|
|
|
pattern = strjoina(p, ".XXXXXX");
|
|
|
|
|
|
|
|
fd = mkostemp_safe(pattern);
|
|
|
|
if (fd < 0)
|
|
|
|
return log_unit_debug_errno(u, fd, "Failed to create extra fields file %s: %m", p);
|
|
|
|
|
|
|
|
n = writev(fd, iovec, c->n_log_extra_fields*2);
|
|
|
|
if (n < 0) {
|
|
|
|
r = log_unit_debug_errno(u, errno, "Failed to write extra fields: %m");
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
(void) fchmod(fd, 0644);
|
|
|
|
|
|
|
|
if (rename(pattern, p) < 0) {
|
|
|
|
r = log_unit_debug_errno(u, errno, "Failed to rename extra fields file: %m");
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
u->exported_log_extra_fields = true;
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
fail:
|
|
|
|
(void) unlink(pattern);
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
static int unit_export_log_ratelimit_interval(Unit *u, const ExecContext *c) {
|
2018-10-08 05:28:36 +02:00
|
|
|
_cleanup_free_ char *buf = NULL;
|
|
|
|
const char *p;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(c);
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
if (u->exported_log_ratelimit_interval)
|
2018-10-08 05:28:36 +02:00
|
|
|
return 0;
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
if (c->log_ratelimit_interval_usec == 0)
|
2018-10-08 05:28:36 +02:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
p = strjoina("/run/systemd/units/log-rate-limit-interval:", u->id);
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
if (asprintf(&buf, "%" PRIu64, c->log_ratelimit_interval_usec) < 0)
|
2018-10-08 05:28:36 +02:00
|
|
|
return log_oom();
|
|
|
|
|
|
|
|
r = symlink_atomic(buf, p);
|
|
|
|
if (r < 0)
|
|
|
|
return log_unit_debug_errno(u, r, "Failed to create log rate limit interval symlink %s: %m", p);
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
u->exported_log_ratelimit_interval = true;
|
2018-10-08 05:28:36 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
static int unit_export_log_ratelimit_burst(Unit *u, const ExecContext *c) {
|
2018-10-08 05:28:36 +02:00
|
|
|
_cleanup_free_ char *buf = NULL;
|
|
|
|
const char *p;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(c);
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
if (u->exported_log_ratelimit_burst)
|
2018-10-08 05:28:36 +02:00
|
|
|
return 0;
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
if (c->log_ratelimit_burst == 0)
|
2018-10-08 05:28:36 +02:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
p = strjoina("/run/systemd/units/log-rate-limit-burst:", u->id);
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
if (asprintf(&buf, "%u", c->log_ratelimit_burst) < 0)
|
2018-10-08 05:28:36 +02:00
|
|
|
return log_oom();
|
|
|
|
|
|
|
|
r = symlink_atomic(buf, p);
|
|
|
|
if (r < 0)
|
|
|
|
return log_unit_debug_errno(u, r, "Failed to create log rate limit burst symlink %s: %m", p);
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
u->exported_log_ratelimit_burst = true;
|
2018-10-08 05:28:36 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
void unit_export_state_files(Unit *u) {
|
|
|
|
const ExecContext *c;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (!u->id)
|
|
|
|
return;
|
|
|
|
|
2018-10-09 16:15:54 +02:00
|
|
|
if (MANAGER_IS_TEST_RUN(u->manager))
|
2018-04-19 10:34:58 +02:00
|
|
|
return;
|
|
|
|
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
/* Exports a couple of unit properties to /run/systemd/units/, so that journald can quickly query this data
|
|
|
|
* from there. Ideally, journald would use IPC to query this, like everybody else, but that's hard, as long as
|
|
|
|
* the IPC system itself and PID 1 also log to the journal.
|
|
|
|
*
|
|
|
|
* Note that these files really shouldn't be considered API for anyone else, as use a runtime file system as
|
|
|
|
* IPC replacement is not compatible with today's world of file system namespaces. However, this doesn't really
|
|
|
|
* apply to communication between the journal and systemd, as we assume that these two daemons live in the same
|
|
|
|
* namespace at least.
|
|
|
|
*
|
|
|
|
* Note that some of the "files" exported here are actually symlinks and not regular files. Symlinks work
|
|
|
|
* better for storing small bits of data, in particular as we can write them with two system calls, and read
|
|
|
|
* them with one. */
|
|
|
|
|
|
|
|
(void) unit_export_invocation_id(u);
|
|
|
|
|
2019-12-12 06:15:42 +01:00
|
|
|
if (!MANAGER_IS_SYSTEM(u->manager))
|
|
|
|
return;
|
|
|
|
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
c = unit_get_exec_context(u);
|
|
|
|
if (c) {
|
|
|
|
(void) unit_export_log_level_max(u, c);
|
|
|
|
(void) unit_export_log_extra_fields(u, c);
|
2019-09-19 17:49:14 +02:00
|
|
|
(void) unit_export_log_ratelimit_interval(u, c);
|
|
|
|
(void) unit_export_log_ratelimit_burst(u, c);
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void unit_unlink_state_files(Unit *u) {
|
|
|
|
const char *p;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (!u->id)
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* Undoes the effect of unit_export_state() */
|
|
|
|
|
|
|
|
if (u->exported_invocation_id) {
|
2019-12-12 06:15:42 +01:00
|
|
|
_cleanup_free_ char *invocation_path = NULL;
|
|
|
|
int r = unit_get_invocation_path(u, &invocation_path);
|
|
|
|
if (r >= 0) {
|
|
|
|
(void) unlink(invocation_path);
|
|
|
|
u->exported_invocation_id = false;
|
|
|
|
}
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
}
|
|
|
|
|
2019-12-12 06:15:42 +01:00
|
|
|
if (!MANAGER_IS_SYSTEM(u->manager))
|
|
|
|
return;
|
|
|
|
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
if (u->exported_log_level_max) {
|
|
|
|
p = strjoina("/run/systemd/units/log-level-max:", u->id);
|
|
|
|
(void) unlink(p);
|
|
|
|
|
|
|
|
u->exported_log_level_max = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (u->exported_log_extra_fields) {
|
|
|
|
p = strjoina("/run/systemd/units/extra-fields:", u->id);
|
|
|
|
(void) unlink(p);
|
|
|
|
|
|
|
|
u->exported_log_extra_fields = false;
|
|
|
|
}
|
2018-10-08 05:28:36 +02:00
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
if (u->exported_log_ratelimit_interval) {
|
2018-10-08 05:28:36 +02:00
|
|
|
p = strjoina("/run/systemd/units/log-rate-limit-interval:", u->id);
|
|
|
|
(void) unlink(p);
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
u->exported_log_ratelimit_interval = false;
|
2018-10-08 05:28:36 +02:00
|
|
|
}
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
if (u->exported_log_ratelimit_burst) {
|
2018-10-08 05:28:36 +02:00
|
|
|
p = strjoina("/run/systemd/units/log-rate-limit-burst:", u->id);
|
|
|
|
(void) unlink(p);
|
|
|
|
|
2019-09-19 17:49:14 +02:00
|
|
|
u->exported_log_ratelimit_burst = false;
|
2018-10-08 05:28:36 +02:00
|
|
|
}
|
core: implement /run/systemd/units/-based path for passing unit info from PID 1 to journald
And let's make use of it to implement two new unit settings with it:
1. LogLevelMax= is a new per-unit setting that may be used to configure
log priority filtering: set it to LogLevelMax=notice and only
messages of level "notice" and lower (i.e. more important) will be
processed, all others are dropped.
2. LogExtraFields= is a new per-unit setting for configuring per-unit
journal fields, that are implicitly included in every log record
generated by the unit's processes. It takes field/value pairs in the
form of FOO=BAR.
Also, related to this, one exisiting unit setting is ported to this new
facility:
3. The invocation ID is now pulled from /run/systemd/units/ instead of
cgroupfs xattrs. This substantially relaxes requirements of systemd
on the kernel version and the privileges it runs with (specifically,
cgroupfs xattrs are not available in containers, since they are
stored in kernel memory, and hence are unsafe to permit to lesser
privileged code).
/run/systemd/units/ is a new directory, which contains a number of files
and symlinks encoding the above information. PID 1 creates and manages
these files, and journald reads them from there.
Note that this is supposed to be a direct path between PID 1 and the
journal only, due to the special runtime environment the journal runs
in. Normally, today we shouldn't introduce new interfaces that (mis-)use
a file system as IPC framework, and instead just an IPC system, but this
is very hard to do between the journal and PID 1, as long as the IPC
system is a subject PID 1 manages, and itself a client to the journal.
This patch cleans up a couple of types used in journal code:
specifically we switch to size_t for a couple of memory-sizing values,
as size_t is the right choice for everything that is memory.
Fixes: #4089
Fixes: #3041
Fixes: #4441
2017-11-02 19:43:32 +01:00
|
|
|
}
|
2017-11-13 17:14:07 +01:00
|
|
|
|
2017-11-17 16:43:08 +01:00
|
|
|
int unit_prepare_exec(Unit *u) {
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
2019-04-23 12:14:20 +02:00
|
|
|
/* Load any custom firewall BPF programs here once to test if they are existing and actually loadable.
|
|
|
|
* Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */
|
|
|
|
r = bpf_firewall_load_custom(u);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2017-11-17 16:43:08 +01:00
|
|
|
/* Prepares everything so that we can fork of a process for this unit */
|
|
|
|
|
|
|
|
(void) unit_realize_cgroup(u);
|
|
|
|
|
|
|
|
if (u->reset_accounting) {
|
2019-03-22 11:25:49 +01:00
|
|
|
(void) unit_reset_accounting(u);
|
2017-11-17 16:43:08 +01:00
|
|
|
u->reset_accounting = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
unit_export_state_files(u);
|
|
|
|
|
|
|
|
r = unit_setup_exec_runtime(u);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
r = unit_setup_dynamic_creds(u);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2020-05-26 14:29:46 +02:00
|
|
|
static bool ignore_leftover_process(const char *comm) {
|
|
|
|
return comm && comm[0] == '('; /* Most likely our own helper process (PAM?), ignore */
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_log_leftover_process_start(pid_t pid, int sig, void *userdata) {
|
2017-11-24 22:02:22 +01:00
|
|
|
_cleanup_free_ char *comm = NULL;
|
|
|
|
|
|
|
|
(void) get_process_comm(pid, &comm);
|
|
|
|
|
2020-05-26 14:29:46 +02:00
|
|
|
if (ignore_leftover_process(comm))
|
2019-01-16 11:20:18 +01:00
|
|
|
return 0;
|
2017-11-24 22:02:22 +01:00
|
|
|
|
2020-05-26 14:29:46 +02:00
|
|
|
/* During start we print a warning */
|
|
|
|
|
2017-11-24 22:02:22 +01:00
|
|
|
log_unit_warning(userdata,
|
|
|
|
"Found left-over process " PID_FMT " (%s) in control group while starting unit. Ignoring.\n"
|
|
|
|
"This usually indicates unclean termination of a previous run, or service implementation deficiencies.",
|
|
|
|
pid, strna(comm));
|
2019-01-16 11:20:18 +01:00
|
|
|
|
|
|
|
return 1;
|
2017-11-24 22:02:22 +01:00
|
|
|
}
|
|
|
|
|
2020-05-26 14:29:46 +02:00
|
|
|
int unit_log_leftover_process_stop(pid_t pid, int sig, void *userdata) {
|
|
|
|
_cleanup_free_ char *comm = NULL;
|
|
|
|
|
|
|
|
(void) get_process_comm(pid, &comm);
|
|
|
|
|
|
|
|
if (ignore_leftover_process(comm))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* During stop we only print an informational message */
|
|
|
|
|
|
|
|
log_unit_info(userdata,
|
|
|
|
"Unit process " PID_FMT " (%s) remains running after unit stopped.",
|
|
|
|
pid, strna(comm));
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_warn_leftover_processes(Unit *u, cg_kill_log_func_t log_func) {
|
2017-11-24 22:02:22 +01:00
|
|
|
assert(u);
|
|
|
|
|
|
|
|
(void) unit_pick_cgroup_path(u);
|
|
|
|
|
|
|
|
if (!u->cgroup_path)
|
2019-01-16 11:20:18 +01:00
|
|
|
return 0;
|
2017-11-24 22:02:22 +01:00
|
|
|
|
2020-05-26 14:29:46 +02:00
|
|
|
return cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, 0, 0, NULL, log_func, u);
|
2017-11-24 22:02:22 +01:00
|
|
|
}
|
|
|
|
|
2018-01-24 19:54:26 +01:00
|
|
|
bool unit_needs_console(Unit *u) {
|
|
|
|
ExecContext *ec;
|
|
|
|
UnitActiveState state;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
state = unit_active_state(u);
|
|
|
|
|
|
|
|
if (UNIT_IS_INACTIVE_OR_FAILED(state))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
if (UNIT_VTABLE(u)->needs_console)
|
|
|
|
return UNIT_VTABLE(u)->needs_console(u);
|
|
|
|
|
|
|
|
/* If this unit type doesn't implement this call, let's use a generic fallback implementation: */
|
|
|
|
ec = unit_get_exec_context(u);
|
|
|
|
if (!ec)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
return exec_context_may_touch_console(ec);
|
|
|
|
}
|
|
|
|
|
2019-12-23 23:06:38 +01:00
|
|
|
const char *unit_label_path(const Unit *u) {
|
2018-01-31 19:53:43 +01:00
|
|
|
const char *p;
|
|
|
|
|
2019-12-23 23:06:38 +01:00
|
|
|
assert(u);
|
|
|
|
|
2018-01-31 19:53:43 +01:00
|
|
|
/* Returns the file system path to use for MAC access decisions, i.e. the file to read the SELinux label off
|
|
|
|
* when validating access checks. */
|
|
|
|
|
|
|
|
p = u->source_path ?: u->fragment_path;
|
|
|
|
if (!p)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/* If a unit is masked, then don't read the SELinux label of /dev/null, as that really makes no sense */
|
2020-05-28 21:09:32 +02:00
|
|
|
if (null_or_empty_path(p) > 0)
|
2018-01-31 19:53:43 +01:00
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return p;
|
|
|
|
}
|
|
|
|
|
2018-02-07 22:52:52 +01:00
|
|
|
int unit_pid_attachable(Unit *u, pid_t pid, sd_bus_error *error) {
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Checks whether the specified PID is generally good for attaching, i.e. a valid PID, not our manager itself,
|
|
|
|
* and not a kernel thread either */
|
|
|
|
|
|
|
|
/* First, a simple range check */
|
|
|
|
if (!pid_is_valid(pid))
|
|
|
|
return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process identifier " PID_FMT " is not valid.", pid);
|
|
|
|
|
|
|
|
/* Some extra safety check */
|
|
|
|
if (pid == 1 || pid == getpid_cached())
|
2018-06-18 22:43:12 +02:00
|
|
|
return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a manager process, refusing.", pid);
|
2018-02-07 22:52:52 +01:00
|
|
|
|
|
|
|
/* Don't even begin to bother with kernel threads */
|
|
|
|
r = is_kernel_thread(pid);
|
|
|
|
if (r == -ESRCH)
|
|
|
|
return sd_bus_error_setf(error, SD_BUS_ERROR_UNIX_PROCESS_ID_UNKNOWN, "Process with ID " PID_FMT " does not exist.", pid);
|
|
|
|
if (r < 0)
|
|
|
|
return sd_bus_error_set_errnof(error, r, "Failed to determine whether process " PID_FMT " is a kernel thread: %m", pid);
|
|
|
|
if (r > 0)
|
|
|
|
return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a kernel thread, refusing.", pid);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2018-11-13 23:28:09 +01:00
|
|
|
void unit_log_success(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
log_struct(LOG_INFO,
|
|
|
|
"MESSAGE_ID=" SD_MESSAGE_UNIT_SUCCESS_STR,
|
|
|
|
LOG_UNIT_ID(u),
|
|
|
|
LOG_UNIT_INVOCATION_ID(u),
|
|
|
|
LOG_UNIT_MESSAGE(u, "Succeeded."));
|
|
|
|
}
|
|
|
|
|
2018-11-13 21:25:22 +01:00
|
|
|
void unit_log_failure(Unit *u, const char *result) {
|
|
|
|
assert(u);
|
|
|
|
assert(result);
|
|
|
|
|
|
|
|
log_struct(LOG_WARNING,
|
|
|
|
"MESSAGE_ID=" SD_MESSAGE_UNIT_FAILURE_RESULT_STR,
|
|
|
|
LOG_UNIT_ID(u),
|
|
|
|
LOG_UNIT_INVOCATION_ID(u),
|
|
|
|
LOG_UNIT_MESSAGE(u, "Failed with result '%s'.", result),
|
|
|
|
"UNIT_RESULT=%s", result);
|
|
|
|
}
|
|
|
|
|
2019-06-29 02:02:30 +02:00
|
|
|
void unit_log_skip(Unit *u, const char *result) {
|
|
|
|
assert(u);
|
|
|
|
assert(result);
|
|
|
|
|
|
|
|
log_struct(LOG_INFO,
|
|
|
|
"MESSAGE_ID=" SD_MESSAGE_UNIT_SKIPPED_STR,
|
|
|
|
LOG_UNIT_ID(u),
|
|
|
|
LOG_UNIT_INVOCATION_ID(u),
|
|
|
|
LOG_UNIT_MESSAGE(u, "Skipped due to '%s'.", result),
|
|
|
|
"UNIT_RESULT=%s", result);
|
|
|
|
}
|
|
|
|
|
2018-11-13 22:10:38 +01:00
|
|
|
void unit_log_process_exit(
|
|
|
|
Unit *u,
|
|
|
|
const char *kind,
|
|
|
|
const char *command,
|
2019-08-21 16:20:59 +02:00
|
|
|
bool success,
|
2018-11-13 22:10:38 +01:00
|
|
|
int code,
|
|
|
|
int status) {
|
|
|
|
|
2019-08-21 16:20:59 +02:00
|
|
|
int level;
|
|
|
|
|
2018-11-13 22:10:38 +01:00
|
|
|
assert(u);
|
|
|
|
assert(kind);
|
|
|
|
|
2019-08-21 16:20:59 +02:00
|
|
|
/* If this is a successful exit, let's log about the exit code on DEBUG level. If this is a failure
|
|
|
|
* and the process exited on its own via exit(), then let's make this a NOTICE, under the assumption
|
|
|
|
* that the service already logged the reason at a higher log level on its own. Otherwise, make it a
|
|
|
|
* WARNING. */
|
|
|
|
if (success)
|
|
|
|
level = LOG_DEBUG;
|
|
|
|
else if (code == CLD_EXITED)
|
|
|
|
level = LOG_NOTICE;
|
|
|
|
else
|
2018-11-13 22:10:38 +01:00
|
|
|
level = LOG_WARNING;
|
|
|
|
|
|
|
|
log_struct(level,
|
|
|
|
"MESSAGE_ID=" SD_MESSAGE_UNIT_PROCESS_EXIT_STR,
|
|
|
|
LOG_UNIT_MESSAGE(u, "%s exited, code=%s, status=%i/%s",
|
|
|
|
kind,
|
|
|
|
sigchld_code_to_string(code), status,
|
|
|
|
strna(code == CLD_EXITED
|
|
|
|
? exit_status_to_string(status, EXIT_STATUS_FULL)
|
|
|
|
: signal_to_string(status))),
|
|
|
|
"EXIT_CODE=%s", sigchld_code_to_string(code),
|
|
|
|
"EXIT_STATUS=%i", status,
|
|
|
|
"COMMAND=%s", strna(command),
|
|
|
|
LOG_UNIT_ID(u),
|
|
|
|
LOG_UNIT_INVOCATION_ID(u));
|
|
|
|
}
|
|
|
|
|
2018-11-16 11:41:18 +01:00
|
|
|
int unit_exit_status(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Returns the exit status to propagate for the most recent cycle of this unit. Returns a value in the range
|
|
|
|
* 0…255 if there's something to propagate. EOPNOTSUPP if the concept does not apply to this unit type, ENODATA
|
|
|
|
* if no data is currently known (for example because the unit hasn't deactivated yet) and EBADE if the main
|
|
|
|
* service process has exited abnormally (signal/coredump). */
|
|
|
|
|
|
|
|
if (!UNIT_VTABLE(u)->exit_status)
|
|
|
|
return -EOPNOTSUPP;
|
|
|
|
|
|
|
|
return UNIT_VTABLE(u)->exit_status(u);
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_failure_action_exit_status(Unit *u) {
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Returns the exit status to propagate on failure, or an error if there's nothing to propagate */
|
|
|
|
|
|
|
|
if (u->failure_action_exit_status >= 0)
|
|
|
|
return u->failure_action_exit_status;
|
|
|
|
|
|
|
|
r = unit_exit_status(u);
|
|
|
|
if (r == -EBADE) /* Exited, but not cleanly (i.e. by signal or such) */
|
|
|
|
return 255;
|
|
|
|
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_success_action_exit_status(Unit *u) {
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Returns the exit status to propagate on success, or an error if there's nothing to propagate */
|
|
|
|
|
|
|
|
if (u->success_action_exit_status >= 0)
|
|
|
|
return u->success_action_exit_status;
|
|
|
|
|
|
|
|
r = unit_exit_status(u);
|
|
|
|
if (r == -EBADE) /* Exited, but not cleanly (i.e. by signal or such) */
|
|
|
|
return 255;
|
|
|
|
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2019-03-18 12:29:08 +01:00
|
|
|
int unit_test_trigger_loaded(Unit *u) {
|
|
|
|
Unit *trigger;
|
|
|
|
|
|
|
|
/* Tests whether the unit to trigger is loaded */
|
|
|
|
|
|
|
|
trigger = UNIT_TRIGGER(u);
|
|
|
|
if (!trigger)
|
2019-07-28 13:00:01 +02:00
|
|
|
return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
|
|
|
|
"Refusing to start, no unit to trigger.");
|
2019-03-18 12:29:08 +01:00
|
|
|
if (trigger->load_state != UNIT_LOADED)
|
2019-07-28 13:00:01 +02:00
|
|
|
return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
|
|
|
|
"Refusing to start, unit %s to trigger not loaded.", trigger->id);
|
2019-03-18 12:29:08 +01:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2020-07-23 08:49:52 +02:00
|
|
|
void unit_destroy_runtime_data(Unit *u, const ExecContext *context) {
|
|
|
|
assert(u);
|
|
|
|
assert(context);
|
|
|
|
|
2019-08-22 16:49:49 +02:00
|
|
|
if (context->runtime_directory_preserve_mode == EXEC_PRESERVE_NO ||
|
|
|
|
(context->runtime_directory_preserve_mode == EXEC_PRESERVE_RESTART && !unit_will_restart(u)))
|
|
|
|
exec_context_destroy_runtime_directory(context, u->manager->prefix[EXEC_DIRECTORY_RUNTIME]);
|
2020-07-23 08:49:52 +02:00
|
|
|
|
|
|
|
exec_context_destroy_credentials(context, u->manager->prefix[EXEC_DIRECTORY_RUNTIME], u->id);
|
2019-08-22 16:49:49 +02:00
|
|
|
}
|
|
|
|
|
2019-06-25 11:31:28 +02:00
|
|
|
int unit_clean(Unit *u, ExecCleanMask mask) {
|
|
|
|
UnitActiveState state;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
/* Special return values:
|
|
|
|
*
|
|
|
|
* -EOPNOTSUPP → cleaning not supported for this unit type
|
|
|
|
* -EUNATCH → cleaning not defined for this resource type
|
|
|
|
* -EBUSY → unit currently can't be cleaned since it's running or not properly loaded, or has
|
|
|
|
* a job queued or similar
|
|
|
|
*/
|
|
|
|
|
|
|
|
if (!UNIT_VTABLE(u)->clean)
|
|
|
|
return -EOPNOTSUPP;
|
|
|
|
|
|
|
|
if (mask == 0)
|
|
|
|
return -EUNATCH;
|
|
|
|
|
|
|
|
if (u->load_state != UNIT_LOADED)
|
|
|
|
return -EBUSY;
|
|
|
|
|
|
|
|
if (u->job)
|
|
|
|
return -EBUSY;
|
|
|
|
|
|
|
|
state = unit_active_state(u);
|
|
|
|
if (!IN_SET(state, UNIT_INACTIVE))
|
|
|
|
return -EBUSY;
|
|
|
|
|
|
|
|
return UNIT_VTABLE(u)->clean(u, mask);
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_can_clean(Unit *u, ExecCleanMask *ret) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (!UNIT_VTABLE(u)->clean ||
|
|
|
|
u->load_state != UNIT_LOADED) {
|
|
|
|
*ret = 0;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* When the clean() method is set, can_clean() really should be set too */
|
|
|
|
assert(UNIT_VTABLE(u)->can_clean);
|
|
|
|
|
|
|
|
return UNIT_VTABLE(u)->can_clean(u, ret);
|
|
|
|
}
|
|
|
|
|
2020-04-29 17:53:43 +02:00
|
|
|
bool unit_can_freeze(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
if (UNIT_VTABLE(u)->can_freeze)
|
|
|
|
return UNIT_VTABLE(u)->can_freeze(u);
|
|
|
|
|
|
|
|
return UNIT_VTABLE(u)->freeze;
|
|
|
|
}
|
|
|
|
|
|
|
|
void unit_frozen(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
u->freezer_state = FREEZER_FROZEN;
|
|
|
|
|
|
|
|
bus_unit_send_pending_freezer_message(u);
|
|
|
|
}
|
|
|
|
|
|
|
|
void unit_thawed(Unit *u) {
|
|
|
|
assert(u);
|
|
|
|
|
|
|
|
u->freezer_state = FREEZER_RUNNING;
|
|
|
|
|
|
|
|
bus_unit_send_pending_freezer_message(u);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int unit_freezer_action(Unit *u, FreezerAction action) {
|
|
|
|
UnitActiveState s;
|
|
|
|
int (*method)(Unit*);
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(u);
|
|
|
|
assert(IN_SET(action, FREEZER_FREEZE, FREEZER_THAW));
|
|
|
|
|
|
|
|
method = action == FREEZER_FREEZE ? UNIT_VTABLE(u)->freeze : UNIT_VTABLE(u)->thaw;
|
|
|
|
if (!method || !cg_freezer_supported())
|
|
|
|
return -EOPNOTSUPP;
|
|
|
|
|
|
|
|
if (u->job)
|
|
|
|
return -EBUSY;
|
|
|
|
|
|
|
|
if (u->load_state != UNIT_LOADED)
|
|
|
|
return -EHOSTDOWN;
|
|
|
|
|
|
|
|
s = unit_active_state(u);
|
|
|
|
if (s != UNIT_ACTIVE)
|
|
|
|
return -EHOSTDOWN;
|
|
|
|
|
|
|
|
if (IN_SET(u->freezer_state, FREEZER_FREEZING, FREEZER_THAWING))
|
|
|
|
return -EALREADY;
|
|
|
|
|
|
|
|
r = method(u);
|
|
|
|
if (r <= 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_freeze(Unit *u) {
|
|
|
|
return unit_freezer_action(u, FREEZER_FREEZE);
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_thaw(Unit *u) {
|
|
|
|
return unit_freezer_action(u, FREEZER_THAW);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Wrappers around low-level cgroup freezer operations common for service and scope units */
|
|
|
|
int unit_freeze_vtable_common(Unit *u) {
|
|
|
|
return unit_cgroup_freezer_action(u, FREEZER_FREEZE);
|
|
|
|
}
|
|
|
|
|
|
|
|
int unit_thaw_vtable_common(Unit *u) {
|
|
|
|
return unit_cgroup_freezer_action(u, FREEZER_THAW);
|
|
|
|
}
|
|
|
|
|
2017-11-13 17:14:07 +01:00
|
|
|
static const char* const collect_mode_table[_COLLECT_MODE_MAX] = {
|
|
|
|
[COLLECT_INACTIVE] = "inactive",
|
|
|
|
[COLLECT_INACTIVE_OR_FAILED] = "inactive-or-failed",
|
|
|
|
};
|
|
|
|
|
|
|
|
DEFINE_STRING_TABLE_LOOKUP(collect_mode, CollectMode);
|