Systemd/src/core/unit.c

5388 lines
174 KiB
C
Raw Normal View History

/* SPDX-License-Identifier: LGPL-2.1+ */
/***
This file is part of systemd.
Copyright 2010 Lennart Poettering
systemd is free software; you can redistribute it and/or modify it
under the terms of the GNU Lesser General Public License as published by
the Free Software Foundation; either version 2.1 of the License, or
(at your option) any later version.
systemd is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public License
along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/
2010-01-26 21:39:06 +01:00
#include <errno.h>
2010-01-27 00:15:56 +01:00
#include <stdlib.h>
#include <string.h>
#include <sys/prctl.h>
#include <sys/stat.h>
#include <unistd.h>
2010-01-26 21:39:06 +01:00
#include "sd-id128.h"
#include "sd-messages.h"
#include "alloc-util.h"
#include "bus-common-errors.h"
#include "bus-util.h"
#include "cgroup-util.h"
#include "dbus-unit.h"
#include "dbus.h"
#include "dropin.h"
#include "escape.h"
#include "execute.h"
#include "fd-util.h"
#include "fileio-label.h"
#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"
core: add "invocation ID" concept to service manager This adds a new invocation ID concept to the service manager. The invocation ID identifies each runtime cycle of a unit uniquely. A new randomized 128bit ID is generated each time a unit moves from and inactive to an activating or active state. The primary usecase for this concept is to connect the runtime data PID 1 maintains about a service with the offline data the journal stores about it. Previously we'd use the unit name plus start/stop times, which however is highly racy since the journal will generally process log data after the service already ended. The "invocation ID" kinda matches the "boot ID" concept of the Linux kernel, except that it applies to an individual unit instead of the whole system. The invocation ID is passed to the activated processes as environment variable. It is additionally stored as extended attribute on the cgroup of the unit. The latter is used by journald to automatically retrieve it for each log logged message and attach it to the log entry. The environment variable is very easily accessible, even for unprivileged services. OTOH the extended attribute is only accessible to privileged processes (this is because cgroupfs only supports the "trusted." xattr namespace, not "user."). The environment variable may be altered by services, the extended attribute may not be, hence is the better choice for the journal. Note that reading the invocation ID off the extended attribute from journald is racy, similar to the way reading the unit name for a logging process is. This patch adds APIs to read the invocation ID to sd-id128: sd_id128_get_invocation() may be used in a similar fashion to sd_id128_get_boot(). PID1's own logging is updated to always include the invocation ID when it logs information about a unit. A new bus call GetUnitByInvocationID() is added that allows retrieving a bus path to a unit by its invocation ID. The bus path is built using the invocation ID, thus providing a path for referring to a unit that is valid only for the current runtime cycleof it. Outlook for the future: should the kernel eventually allow passing of cgroup information along AF_UNIX/SOCK_DGRAM messages via a unique cgroup id, then we can alter the invocation ID to be generated as hash from that rather than entirely randomly. This way we can derive the invocation race-freely from the messages.
2016-08-30 23:18:46 +02:00
#include "id128-util.h"
#include "io-util.h"
#include "load-dropin.h"
#include "load-fragment.h"
#include "log.h"
#include "macro.h"
#include "missing.h"
#include "mkdir.h"
#include "parse-util.h"
#include "path-util.h"
2015-04-10 19:10:00 +02:00
#include "process-util.h"
#include "set.h"
#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"
#include "special.h"
#include "specifier.h"
#include "stat-util.h"
#include "stdio-util.h"
#include "string-table.h"
#include "string-util.h"
#include "strv.h"
#include "umask-util.h"
#include "unit-name.h"
#include "unit.h"
#include "user-util.h"
#include "virt.h"
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,
[UNIT_SWAP] = &swap_vtable,
[UNIT_TIMER] = &timer_vtable,
[UNIT_PATH] = &path_vtable,
[UNIT_SLICE] = &slice_vtable,
[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);
Unit *unit_new(Manager *m, size_t size) {
2010-01-26 21:39:06 +01:00
Unit *u;
assert(m);
assert(size >= sizeof(Unit));
2010-01-26 21:39:06 +01:00
u = malloc0(size);
if (!u)
2010-01-26 21:39:06 +01:00
return NULL;
u->names = set_new(&string_hash_ops);
2016-10-17 00:28:30 +02:00
if (!u->names)
return mfree(u);
2010-01-26 21:39:06 +01:00
u->manager = m;
u->type = _UNIT_TYPE_INVALID;
u->default_dependencies = true;
u->unit_file_state = _UNIT_FILE_STATE_INVALID;
u->unit_file_preset = -1;
u->on_failure_job_mode = JOB_REPLACE;
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->cgroup_inotify_wd = -1;
u->job_timeout = USEC_INFINITY;
u->job_running_timeout = USEC_INFINITY;
u->ref_uid = UID_INVALID;
u->ref_gid = GID_INVALID;
u->cpu_usage_last = NSEC_INFINITY;
u->cgroup_bpf_state = UNIT_CGROUP_BPF_INVALIDATED;
2010-01-26 21:39:06 +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;
u->last_section_private = -1;
RATELIMIT_INIT(u->start_limit, m->default_start_limit_interval, m->default_start_limit_burst);
RATELIMIT_INIT(u->auto_stop_ratelimit, 10 * USEC_PER_SEC, 16);
2010-01-26 21:39:06 +01:00
return u;
}
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;
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)
return r;
*ret = u;
2018-03-09 21:34:28 +01:00
u = NULL;
return r;
}
bool unit_has_name(Unit *u, const char *name) {
assert(u);
assert(name);
return set_contains(u->names, (char*) name);
}
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;
core: add io controller support on the unified hierarchy On the unified hierarchy, blkio controller is renamed to io and the interface is changed significantly. * blkio.weight and blkio.weight_device are consolidated into io.weight which uses the standardized weight range [1, 10000] with 100 as the default value. * blkio.throttle.{read|write}_{bps|iops}_device are consolidated into io.max. Expansion of throttling features is being worked on to support work-conserving absolute limits (io.low and io.high). * All stats are consolidated into io.stats. This patchset adds support for the new interface. As the interface has been revamped and new features are expected to be added, it seems best to treat it as a separate controller rather than trying to expand the blkio settings although we might add automatic translation if only blkio settings are specified. * io.weight handling is mostly identical to blkio.weight[_device] handling except that the weight range is different. * Both read and write bandwidth settings are consolidated into CGroupIODeviceLimit which describes all limits applicable to the device. This makes it less painful to add new limits. * "max" can be used to specify the maximum limit which is equivalent to no config for max limits and treated as such. If a given CGroupIODeviceLimit doesn't contain any non-default configs, the config struct is discarded once the no limit config is applied to cgroup. * lookup_blkio_device() is renamed to lookup_block_device(). Signed-off-by: Tejun Heo <htejun@fb.com>
2016-05-05 22:42:55 +02:00
cc->io_accounting = u->manager->default_io_accounting;
cc->ip_accounting = u->manager->default_ip_accounting;
cc->blockio_accounting = u->manager->default_blockio_accounting;
cc->memory_accounting = u->manager->default_memory_accounting;
cc->tasks_accounting = u->manager->default_tasks_accounting;
cc->ip_accounting = u->manager->default_ip_accounting;
if (u->type != UNIT_SLICE)
cc->tasks_max = u->manager->default_tasks_max;
}
ec = unit_get_exec_context(u);
if (ec) {
exec_context_init(ec);
ec->keyring_mode = MANAGER_IS_SYSTEM(u->manager) ?
EXEC_KEYRING_SHARED : EXEC_KEYRING_INHERIT;
}
kc = unit_get_kill_context(u);
if (kc)
kill_context_init(kc);
if (UNIT_VTABLE(u)->init)
UNIT_VTABLE(u)->init(u);
}
2010-01-26 21:39:06 +01:00
int unit_add_name(Unit *u, const char *text) {
_cleanup_free_ char *s = NULL, *i = NULL;
2010-01-26 21:39:06 +01:00
UnitType t;
int r;
assert(u);
assert(text);
if (unit_name_is_valid(text, UNIT_NAME_TEMPLATE)) {
if (!u->instance)
2010-04-15 03:11:11 +02:00
return -EINVAL;
2010-01-26 21:39:06 +01:00
r = unit_name_replace_instance(text, u->instance, &s);
if (r < 0)
return r;
} else {
2010-04-15 03:11:11 +02:00
s = strdup(text);
if (!s)
return -ENOMEM;
}
2010-01-26 21:39:06 +01:00
if (set_contains(u->names, s))
return 0;
if (hashmap_contains(u->manager->units, s))
return -EEXIST;
if (!unit_name_is_valid(s, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE))
return -EINVAL;
t = unit_name_to_type(s);
if (t < 0)
return -EINVAL;
2010-01-26 21:39:06 +01:00
if (u->type != _UNIT_TYPE_INVALID && t != u->type)
return -EINVAL;
2010-01-26 21:39:06 +01:00
2013-10-22 01:54:10 +02:00
r = unit_name_to_instance(s, &i);
if (r < 0)
return r;
2010-01-26 21:39:06 +01:00
if (i && !unit_type_may_template(t))
return -EINVAL;
2010-04-15 03:11:11 +02:00
/* Ensure that this unit is either instanced or not instanced,
* but not both. Note that we do allow names with different
* instance names however! */
if (u->type != _UNIT_TYPE_INVALID && !u->instance != !i)
return -EINVAL;
2010-04-15 03:11:11 +02:00
if (!unit_type_may_alias(t) && !set_isempty(u->names))
return -EEXIST;
2010-04-15 03:11:11 +02:00
if (hashmap_size(u->manager->units) >= MANAGER_MAX_NAMES)
return -E2BIG;
2013-10-22 01:54:10 +02:00
r = set_put(u->names, s);
if (r < 0)
return r;
assert(r > 0);
2010-01-26 21:39:06 +01:00
2013-10-22 01:54:10 +02:00
r = hashmap_put(u->manager->units, s, u);
if (r < 0) {
(void) set_remove(u->names, s);
return r;
2010-01-26 21:39:06 +01:00
}
if (u->type == _UNIT_TYPE_INVALID) {
u->type = t;
u->id = s;
u->instance = i;
2010-04-15 03:11:11 +02:00
LIST_PREPEND(units_by_type, u->manager->units_by_type[t], u);
unit_init(u);
2010-01-26 21:39:06 +01:00
i = NULL;
}
2010-04-15 03:11:11 +02:00
s = NULL;
2010-04-15 03:11:11 +02:00
unit_add_to_dbus_queue(u);
return 0;
2010-01-26 21:39:06 +01:00
}
int unit_choose_id(Unit *u, const char *name) {
_cleanup_free_ char *t = NULL;
char *s, *i;
int r;
assert(u);
assert(name);
if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
2010-04-15 03:11:11 +02:00
if (!u->instance)
2010-04-15 03:11:11 +02:00
return -EINVAL;
r = unit_name_replace_instance(name, u->instance, &t);
if (r < 0)
return r;
2010-04-15 03:11:11 +02:00
name = t;
}
/* Selects one of the names of this unit as the id */
s = set_get(u->names, (char*) name);
2010-04-15 03:11:11 +02:00
if (!s)
return -ENOENT;
/* Determine the new instance from the new id */
2013-10-22 01:54:10 +02:00
r = unit_name_to_instance(s, &i);
if (r < 0)
return r;
u->id = s;
free(u->instance);
u->instance = i;
unit_add_to_dbus_queue(u);
2010-04-15 03:11:11 +02:00
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);
2010-01-29 03:18:09 +01:00
return 0;
}
bool unit_may_gc(Unit *u) {
UnitActiveState state;
int r;
assert(u);
/* 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
* 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.
*/
if (u->job)
return false;
if (u->nop_job)
return false;
state = unit_active_state(u);
/* If the unit is inactive and failed and no job is queued for it, then release its runtime resources */
if (UNIT_IS_INACTIVE_OR_FAILED(state) &&
UNIT_VTABLE(u)->release_resources)
UNIT_VTABLE(u)->release_resources(u);
if (u->perpetual)
return false;
if (sd_bus_track_count(u->bus_track) > 0)
return false;
/* 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)
return false;
break;
case COLLECT_INACTIVE_OR_FAILED:
if (!IN_SET(state, UNIT_INACTIVE, UNIT_FAILED))
return false;
break;
default:
assert_not_reached("Unknown garbage collection mode");
}
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)
return false;
}
if (UNIT_VTABLE(u)->may_gc && !UNIT_VTABLE(u)->may_gc(u))
return false;
return true;
}
2010-01-26 21:39:06 +01:00
void unit_add_to_load_queue(Unit *u) {
assert(u);
assert(u->type != _UNIT_TYPE_INVALID);
2010-01-26 21:39:06 +01:00
if (u->load_state != UNIT_STUB || u->in_load_queue)
2010-01-26 21:39:06 +01:00
return;
LIST_PREPEND(load_queue, u->manager->load_queue, u);
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);
if (u->in_cleanup_queue)
2010-04-06 02:43:58 +02:00
return;
LIST_PREPEND(cleanup_queue, u->manager->cleanup_queue, u);
u->in_cleanup_queue = true;
2010-04-06 02:43:58 +02:00
}
void unit_add_to_gc_queue(Unit *u) {
assert(u);
if (u->in_gc_queue || u->in_cleanup_queue)
return;
if (!unit_may_gc(u))
return;
LIST_PREPEND(gc_queue, u->manager->gc_unit_queue, u);
u->in_gc_queue = true;
}
void unit_add_to_dbus_queue(Unit *u) {
assert(u);
assert(u->type != _UNIT_TYPE_INVALID);
if (u->load_state == UNIT_STUB || u->in_dbus_queue)
return;
/* Shortcut things if nobody cares */
if (sd_bus_track_count(u->manager->subscribed) <= 0 &&
sd_bus_track_count(u->bus_track) <= 0 &&
set_isempty(u->manager->private_buses)) {
u->sent_dbus_new_signal = true;
return;
}
LIST_PREPEND(dbus_queue, u->manager->dbus_unit_queue, u);
u->in_dbus_queue = true;
}
static void bidi_set_free(Unit *u, Hashmap *h) {
2010-01-26 21:39:06 +01:00
Unit *other;
Iterator i;
void *v;
2010-01-26 21:39:06 +01:00
assert(u);
/* Frees the hashmap and makes sure we are dropped from the inverse pointers */
2010-01-26 21:39:06 +01:00
HASHMAP_FOREACH_KEY(v, other, h, i) {
2010-01-26 21:39:06 +01:00
UnitDependency d;
for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++)
hashmap_remove(other->dependencies[d], u);
unit_add_to_gc_queue(other);
2010-01-26 21:39:06 +01:00
}
hashmap_free(h);
2010-01-26 21:39:06 +01:00
}
static void unit_remove_transient(Unit *u) {
char **i;
assert(u);
if (!u->transient)
return;
if (u->fragment_path)
(void) unlink(u->fragment_path);
STRV_FOREACH(i, u->dropin_paths) {
_cleanup_free_ char *p = NULL, *pp = NULL;
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;
(void) unlink(*i);
(void) rmdir(p);
}
}
static void unit_free_requires_mounts_for(Unit *u) {
assert(u);
for (;;) {
_cleanup_free_ char *path;
path = hashmap_steal_first_key(u->requires_mounts_for);
if (!path)
break;
else {
char s[strlen(path) + 1];
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;
(void) set_remove(x, u);
if (set_isempty(x)) {
(void) hashmap_remove(u->manager->units_requiring_mounts_for, y);
free(y);
set_free(x);
}
}
}
}
u->requires_mounts_for = hashmap_free(u->requires_mounts_for);
}
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) {
UnitDependency d;
Iterator i;
char *t;
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);
if (!MANAGER_IS_RELOADING(u->manager))
unit_remove_transient(u);
bus_unit_send_removed_signal(u);
unit_done(u);
2010-06-04 22:31:33 +02:00
sd_bus_slot_unref(u->match_bus_slot);
sd_bus_track_unref(u->bus_track);
u->deserialized_refs = strv_free(u->deserialized_refs);
unit_free_requires_mounts_for(u);
SET_FOREACH(t, u->names, i)
hashmap_remove_value(u->manager->units, t, u);
2010-01-26 21:39:06 +01:00
core: add "invocation ID" concept to service manager This adds a new invocation ID concept to the service manager. The invocation ID identifies each runtime cycle of a unit uniquely. A new randomized 128bit ID is generated each time a unit moves from and inactive to an activating or active state. The primary usecase for this concept is to connect the runtime data PID 1 maintains about a service with the offline data the journal stores about it. Previously we'd use the unit name plus start/stop times, which however is highly racy since the journal will generally process log data after the service already ended. The "invocation ID" kinda matches the "boot ID" concept of the Linux kernel, except that it applies to an individual unit instead of the whole system. The invocation ID is passed to the activated processes as environment variable. It is additionally stored as extended attribute on the cgroup of the unit. The latter is used by journald to automatically retrieve it for each log logged message and attach it to the log entry. The environment variable is very easily accessible, even for unprivileged services. OTOH the extended attribute is only accessible to privileged processes (this is because cgroupfs only supports the "trusted." xattr namespace, not "user."). The environment variable may be altered by services, the extended attribute may not be, hence is the better choice for the journal. Note that reading the invocation ID off the extended attribute from journald is racy, similar to the way reading the unit name for a logging process is. This patch adds APIs to read the invocation ID to sd-id128: sd_id128_get_invocation() may be used in a similar fashion to sd_id128_get_boot(). PID1's own logging is updated to always include the invocation ID when it logs information about a unit. A new bus call GetUnitByInvocationID() is added that allows retrieving a bus path to a unit by its invocation ID. The bus path is built using the invocation ID, thus providing a path for referring to a unit that is valid only for the current runtime cycleof it. Outlook for the future: should the kernel eventually allow passing of cgroup information along AF_UNIX/SOCK_DGRAM messages via a unique cgroup id, then we can alter the invocation ID to be generated as hash from that rather than entirely randomly. This way we can derive the invocation race-freely from the messages.
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);
if (u->job) {
Job *j = u->job;
job_uninstall(j);
job_free(j);
}
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);
}
for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++)
bidi_set_free(u, u->dependencies[d]);
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);
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);
unit_unref_uid_gid(u, false);
(void) manager_update_failed_units(u->manager, u, false);
set_remove(u->manager->startup_units, u);
unit_unwatch_all_pids(u);
2013-06-18 02:07:35 +02:00
unit_ref_unset(&u->slice);
while (u->refs_by_target)
unit_ref_unset(u->refs_by_target);
if (u->type != _UNIT_TYPE_INVALID)
LIST_REMOVE(units_by_type, u->manager->units_by_type[u->type], u);
2010-01-29 06:04:08 +01:00
if (u->in_load_queue)
LIST_REMOVE(load_queue, u->manager->load_queue, u);
2010-01-26 21:39:06 +01:00
if (u->in_dbus_queue)
LIST_REMOVE(dbus_queue, u->manager->dbus_unit_queue, u);
if (u->in_gc_queue)
LIST_REMOVE(gc_queue, u->manager->gc_unit_queue, u);
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
if (u->in_cgroup_empty_queue)
LIST_REMOVE(cgroup_empty_queue, u->manager->cgroup_empty_queue, u);
if (u->in_cleanup_queue)
LIST_REMOVE(cleanup_queue, u->manager->cleanup_queue, u);
safe_close(u->ip_accounting_ingress_map_fd);
safe_close(u->ip_accounting_egress_map_fd);
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
bpf_program_unref(u->ip_bpf_ingress);
bpf: rework how we keep track and attach cgroup bpf programs So, the kernel's management of cgroup/BPF programs is a bit misdesigned: if you attach a BPF program to a cgroup and close the fd for it it will stay pinned to the cgroup with no chance of ever removing it again (or otherwise getting ahold of it again), because the fd is used for selecting which BPF program to detach. The only way to get rid of the program again is to destroy the cgroup itself. This is particularly bad for root the cgroup (and in fact any other cgroup that we cannot realistically remove during runtime, such as /system.slice, /init.scope or /system.slice/dbus.service) as getting rid of the program only works by rebooting the system. To counter this let's closely keep track to which cgroup a BPF program is attached and let's implicitly detach the BPF program when we are about to close the BPF fd. This hence changes the bpf_program_cgroup_attach() function to track where we attached the program and changes bpf_program_cgroup_detach() to use this information. Moreover bpf_program_unref() will now implicitly call bpf_program_cgroup_detach(). In order to simplify things, bpf_program_cgroup_attach() will now implicitly invoke bpf_program_load_kernel() when necessary, simplifying the caller's side. Finally, this adds proper reference counting to BPF programs. This is useful for working with two BPF programs in parallel: the BPF program we are preparing for installation and the BPF program we so far installed, shortening the window when we detach the old one and reattach the new one.
2018-02-20 19:28:24 +01:00
bpf_program_unref(u->ip_bpf_ingress_installed);
bpf_program_unref(u->ip_bpf_egress);
bpf: rework how we keep track and attach cgroup bpf programs So, the kernel's management of cgroup/BPF programs is a bit misdesigned: if you attach a BPF program to a cgroup and close the fd for it it will stay pinned to the cgroup with no chance of ever removing it again (or otherwise getting ahold of it again), because the fd is used for selecting which BPF program to detach. The only way to get rid of the program again is to destroy the cgroup itself. This is particularly bad for root the cgroup (and in fact any other cgroup that we cannot realistically remove during runtime, such as /system.slice, /init.scope or /system.slice/dbus.service) as getting rid of the program only works by rebooting the system. To counter this let's closely keep track to which cgroup a BPF program is attached and let's implicitly detach the BPF program when we are about to close the BPF fd. This hence changes the bpf_program_cgroup_attach() function to track where we attached the program and changes bpf_program_cgroup_detach() to use this information. Moreover bpf_program_unref() will now implicitly call bpf_program_cgroup_detach(). In order to simplify things, bpf_program_cgroup_attach() will now implicitly invoke bpf_program_load_kernel() when necessary, simplifying the caller's side. Finally, this adds proper reference counting to BPF programs. This is useful for working with two BPF programs in parallel: the BPF program we are preparing for installation and the BPF program we so far installed, shortening the window when we detach the old one and reattach the new one.
2018-02-20 19:28:24 +01:00
bpf_program_unref(u->ip_bpf_egress_installed);
condition_free_list(u->conditions);
condition_free_list(u->asserts);
free(u->description);
strv_free(u->documentation);
free(u->fragment_path);
free(u->source_path);
strv_free(u->dropin_paths);
free(u->instance);
2010-01-26 21:39:06 +01:00
free(u->job_timeout_reboot_arg);
set_free_free(u->names);
2010-01-26 21:39:06 +01:00
free(u->reboot_arg);
2010-01-26 21:39:06 +01:00
free(u);
}
UnitActiveState unit_active_state(Unit *u) {
assert(u);
if (u->load_state == UNIT_MERGED)
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
* shortcut failed loading to UNIT_INACTIVE_FAILED. */
2010-01-26 21:39:06 +01:00
return UNIT_VTABLE(u)->active_state(u);
}
const char* unit_sub_state_to_string(Unit *u) {
assert(u);
return UNIT_VTABLE(u)->sub_state_to_string(u);
}
static int set_complete_move(Set **s, Set **other) {
assert(s);
assert(other);
if (!other)
return 0;
if (*s)
return set_move(*s, *other);
else
*s = TAKE_PTR(*other);
return 0;
}
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)
return 0;
2010-01-26 21:39:06 +01:00
if (*s)
return hashmap_move(*s, *other);
else
*s = TAKE_PTR(*other);
return 0;
2010-04-06 02:43:58 +02:00
}
2010-01-26 21:39:06 +01:00
static int merge_names(Unit *u, Unit *other) {
2010-04-06 02:43:58 +02:00
char *t;
Iterator i;
int r;
2010-01-26 21:39:06 +01:00
2010-04-06 02:43:58 +02:00
assert(u);
assert(other);
r = set_complete_move(&u->names, &other->names);
if (r < 0)
return r;
2010-04-06 02:43:58 +02:00
set_free_free(other->names);
other->names = NULL;
other->id = NULL;
2010-04-06 02:43:58 +02:00
SET_FOREACH(t, u->names, i)
assert_se(hashmap_replace(u->manager->units, t, u) == 0);
return 0;
2010-01-26 21:39:06 +01:00
}
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
* to reserve anything. In that case other's set will be transferred
* as a whole to u by complete_move().
*/
if (!u->dependencies[d])
return 0;
/* merge_dependencies() will skip a u-on-u dependency */
n_reserve = hashmap_size(other->dependencies[d]) - !!hashmap_get(other->dependencies[d], u);
return hashmap_reserve(u->dependencies[d], n_reserve);
}
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;
void *v;
2010-01-26 21:39:06 +01:00
int r;
2010-04-06 02:43:58 +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);
/* Fix backwards pointers. Let's iterate through all dependendent units of the other unit. */
HASHMAP_FOREACH_KEY(v, back, other->dependencies[d], i) {
2010-04-06 02:43:58 +02:00
UnitDependency k;
/* 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'. */
2013-10-22 01:54:10 +02:00
for (k = 0; k < _UNIT_DEPENDENCY_MAX; k++) {
if (back == u) {
/* 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);
} else {
UnitDependencyInfo di_u, di_other, di_merged;
/* 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);
di_merged = (UnitDependencyInfo) {
.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); */
}
2013-10-22 01:54:10 +02:00
}
2010-04-06 02:43:58 +02:00
}
/* Also do not move dependencies on u to itself */
back = hashmap_remove(other->dependencies[d], u);
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);
/* The move cannot fail. The caller must have performed a reservation. */
assert_se(hashmap_complete_move(&u->dependencies[d], &other->dependencies[d]) == 0);
2010-04-06 02:43:58 +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) {
2010-01-26 21:39:06 +01:00
UnitDependency d;
const char *other_id = NULL;
int r;
2010-01-26 21:39:06 +01:00
assert(u);
assert(other);
assert(u->manager == other->manager);
assert(u->type != _UNIT_TYPE_INVALID);
2010-01-26 21:39:06 +01:00
other = unit_follow_merge(other);
2010-04-06 02:43:58 +02:00
if (other == u)
return 0;
if (u->type != other->type)
2010-04-15 03:11:11 +02:00
return -EINVAL;
if (!u->instance != !other->instance)
2010-01-26 21:39:06 +01:00
return -EINVAL;
if (!unit_type_may_alias(u->type)) /* Merging only applies to unit names that support aliases */
return -EEXIST;
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
if (other->job)
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;
if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
return -EEXIST;
if (other->id)
other_id = strdupa(other->id);
/* Make reservations to ensure merge_dependencies() won't fail */
for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++) {
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 */
r = merge_names(u, other);
if (r < 0)
return r;
2010-01-26 21:39:06 +01:00
/* Redirect all references */
while (other->refs_by_target)
unit_ref_set(other->refs_by_target, other->refs_by_target->source, u);
2010-01-26 21:39:06 +01:00
/* Merge dependencies */
for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++)
merge_dependencies(u, other, other_id, d);
2010-01-26 21:39:06 +01:00
other->load_state = UNIT_MERGED;
other->merged_into = u;
2010-04-06 02:43:58 +02:00
/* If there is still some data attached to the other node, we
* don't need it anymore, and can free it. */
if (other->load_state != UNIT_STUB)
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) {
_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
assert(u);
assert(name);
if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE)) {
if (!u->instance)
2010-04-15 03:11:11 +02:00
return -EINVAL;
r = unit_name_replace_instance(name, u->instance, &s);
if (r < 0)
return r;
2010-04-15 03:11:11 +02:00
name = s;
}
other = manager_get_unit(u->manager, name);
if (other)
return unit_merge(u, other);
2010-04-06 02:43:58 +02:00
return unit_add_name(u, name);
2010-04-06 02:43:58 +02:00
}
Unit* unit_follow_merge(Unit *u) {
assert(u);
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) {
ExecDirectoryType dt;
char **dp;
2010-04-06 02:43:58 +02:00
int r;
assert(u);
assert(c);
if (c->working_directory) {
r = unit_require_mounts_for(u, c->working_directory, UNIT_DEPENDENCY_FILE);
if (r < 0)
return r;
}
if (c->root_directory) {
r = unit_require_mounts_for(u, c->root_directory, UNIT_DEPENDENCY_FILE);
if (r < 0)
return r;
}
if (c->root_image) {
r = unit_require_mounts_for(u, c->root_image, UNIT_DEPENDENCY_FILE);
if (r < 0)
return r;
}
for (dt = 0; dt < _EXEC_DIRECTORY_TYPE_MAX; dt++) {
if (!u->manager->prefix[dt])
continue;
STRV_FOREACH(dp, c->directories[dt].paths) {
_cleanup_free_ char *p;
p = strjoin(u->manager->prefix[dt], "/", *dp);
if (!p)
return -ENOMEM;
r = unit_require_mounts_for(u, p, UNIT_DEPENDENCY_FILE);
if (r < 0)
return r;
}
}
if (!MANAGER_IS_SYSTEM(u->manager))
return 0;
if (c->private_tmp) {
const char *p;
FOREACH_STRING(p, "/tmp", "/var/tmp") {
r = unit_require_mounts_for(u, p, UNIT_DEPENDENCY_FILE);
if (r < 0)
return r;
}
r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_TMPFILES_SETUP_SERVICE, NULL, true, UNIT_DEPENDENCY_FILE);
if (r < 0)
return r;
}
if (!IN_SET(c->std_output,
EXEC_OUTPUT_JOURNAL, EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
EXEC_OUTPUT_KMSG, EXEC_OUTPUT_KMSG_AND_CONSOLE,
EXEC_OUTPUT_SYSLOG, EXEC_OUTPUT_SYSLOG_AND_CONSOLE) &&
!IN_SET(c->std_error,
EXEC_OUTPUT_JOURNAL, EXEC_OUTPUT_JOURNAL_AND_CONSOLE,
EXEC_OUTPUT_KMSG, EXEC_OUTPUT_KMSG_AND_CONSOLE,
EXEC_OUTPUT_SYSLOG, EXEC_OUTPUT_SYSLOG_AND_CONSOLE))
2010-04-06 02:43:58 +02:00
return 0;
/* If syslog or kernel logging is requested, make sure our own
* logging daemon is run first. */
r = unit_add_dependency_by_name(u, UNIT_AFTER, SPECIAL_JOURNALD_SOCKET, NULL, true, UNIT_DEPENDENCY_FILE);
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);
if (u->description)
return u->description;
2010-01-26 21:39:06 +01:00
return strna(u->id);
2010-01-26 21:39:06 +01:00
}
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;
if ((mask & table[i].mask) == table[i].mask) {
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) {
char *t, **j;
2010-01-26 21:39:06 +01:00
UnitDependency d;
Iterator i;
2010-02-03 14:21:48 +01:00
const char *prefix2;
char
timestamp0[FORMAT_TIMESTAMP_MAX],
timestamp1[FORMAT_TIMESTAMP_MAX],
timestamp2[FORMAT_TIMESTAMP_MAX],
timestamp3[FORMAT_TIMESTAMP_MAX],
timestamp4[FORMAT_TIMESTAMP_MAX],
timespan[FORMAT_TIMESPAN_MAX];
Unit *following;
_cleanup_set_free_ Set *following_set = NULL;
const char *n;
CGroupMask m;
int r;
2010-01-26 21:39:06 +01:00
assert(u);
assert(u->type >= 0);
2010-01-26 21:39:06 +01:00
prefix = strempty(prefix);
prefix2 = strjoina(prefix, "\t");
2010-01-26 21:39:06 +01:00
fprintf(f,
"%s-> Unit %s:\n"
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"
"%s\tState Change Timestamp: %s\n"
"%s\tInactive Exit Timestamp: %s\n"
2010-04-10 04:43:57 +02:00
"%s\tActive Enter Timestamp: %s\n"
"%s\tActive Exit Timestamp: %s\n"
"%s\tInactive Enter Timestamp: %s\n"
"%s\tMay GC: %s\n"
"%s\tNeed Daemon Reload: %s\n"
"%s\tTransient: %s\n"
"%s\tPerpetual: %s\n"
"%s\tGarbage Collection Mode: %s\n"
"%s\tSlice: %s\n"
"%s\tCGroup: %s\n"
"%s\tCGroup realized: %s\n",
prefix, u->id,
2010-01-26 21:39:06 +01:00
prefix, unit_description(u),
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)),
prefix, strna(format_timestamp(timestamp0, sizeof(timestamp0), u->state_change_timestamp.realtime)),
prefix, strna(format_timestamp(timestamp1, sizeof(timestamp1), u->inactive_exit_timestamp.realtime)),
prefix, strna(format_timestamp(timestamp2, sizeof(timestamp2), u->active_enter_timestamp.realtime)),
prefix, strna(format_timestamp(timestamp3, sizeof(timestamp3), u->active_exit_timestamp.realtime)),
prefix, strna(format_timestamp(timestamp4, sizeof(timestamp4), u->inactive_enter_timestamp.realtime)),
prefix, yes_no(unit_may_gc(u)),
prefix, yes_no(unit_need_daemon_reload(u)),
prefix, yes_no(u->transient),
prefix, yes_no(u->perpetual),
prefix, collect_mode_to_string(u->collect_mode),
prefix, strna(unit_slice_name(u)),
prefix, strna(u->cgroup_path),
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);
fprintf(f, "%s\tCGroup realized mask: %s\n", prefix, strnull(s));
}
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));
}
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));
}
m = unit_get_members_mask(u);
if (m != 0) {
_cleanup_free_ char *s = NULL;
(void) cg_mask_to_string(m, &s);
fprintf(f, "%s\tCGroup members mask: %s\n", prefix, strnull(s));
}
2010-01-27 00:15:56 +01:00
SET_FOREACH(t, u->names, i)
2010-01-26 21:39:06 +01:00
fprintf(f, "%s\tName: %s\n", prefix, t);
core: add "invocation ID" concept to service manager This adds a new invocation ID concept to the service manager. The invocation ID identifies each runtime cycle of a unit uniquely. A new randomized 128bit ID is generated each time a unit moves from and inactive to an activating or active state. The primary usecase for this concept is to connect the runtime data PID 1 maintains about a service with the offline data the journal stores about it. Previously we'd use the unit name plus start/stop times, which however is highly racy since the journal will generally process log data after the service already ended. The "invocation ID" kinda matches the "boot ID" concept of the Linux kernel, except that it applies to an individual unit instead of the whole system. The invocation ID is passed to the activated processes as environment variable. It is additionally stored as extended attribute on the cgroup of the unit. The latter is used by journald to automatically retrieve it for each log logged message and attach it to the log entry. The environment variable is very easily accessible, even for unprivileged services. OTOH the extended attribute is only accessible to privileged processes (this is because cgroupfs only supports the "trusted." xattr namespace, not "user."). The environment variable may be altered by services, the extended attribute may not be, hence is the better choice for the journal. Note that reading the invocation ID off the extended attribute from journald is racy, similar to the way reading the unit name for a logging process is. This patch adds APIs to read the invocation ID to sd-id128: sd_id128_get_invocation() may be used in a similar fashion to sd_id128_get_boot(). PID1's own logging is updated to always include the invocation ID when it logs information about a unit. A new bus call GetUnitByInvocationID() is added that allows retrieving a bus path to a unit by its invocation ID. The bus path is built using the invocation ID, thus providing a path for referring to a unit that is valid only for the current runtime cycleof it. Outlook for the future: should the kernel eventually allow passing of cgroup information along AF_UNIX/SOCK_DGRAM messages via a unique cgroup id, then we can alter the invocation ID to be generated as hash from that rather than entirely randomly. This way we can derive the invocation race-freely from the messages.
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));
STRV_FOREACH(j, u->documentation)
fprintf(f, "%s\tDocumentation: %s\n", prefix, *j);
following = unit_following(u);
if (following)
fprintf(f, "%s\tFollowing: %s\n", prefix, following->id);
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);
}
if (u->fragment_path)
fprintf(f, "%s\tFragment Path: %s\n", prefix, u->fragment_path);
2010-04-06 02:43:58 +02:00
if (u->source_path)
fprintf(f, "%s\tSource Path: %s\n", prefix, u->source_path);
STRV_FOREACH(j, u->dropin_paths)
2013-04-02 16:24:45 +02:00
fprintf(f, "%s\tDropIn Path: %s\n", prefix, *j);
if (u->failure_action != EMERGENCY_ACTION_NONE)
fprintf(f, "%s\tFailure Action: %s\n", prefix, emergency_action_to_string(u->failure_action));
if (u->success_action != EMERGENCY_ACTION_NONE)
fprintf(f, "%s\tSuccess Action: %s\n", prefix, emergency_action_to_string(u->success_action));
if (u->job_timeout != USEC_INFINITY)
fprintf(f, "%s\tJob Timeout: %s\n", prefix, format_timespan(timespan, sizeof(timespan), u->job_timeout, 0));
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));
if (u->job_timeout_reboot_arg)
fprintf(f, "%s\tJob Timeout Reboot Argument: %s\n", prefix, u->job_timeout_reboot_arg);
condition_dump_list(u->conditions, f, prefix, condition_type_to_string);
condition_dump_list(u->asserts, f, prefix, assert_type_to_string);
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",
prefix, strna(format_timestamp(timestamp1, sizeof(timestamp1), u->condition_timestamp.realtime)),
prefix, yes_no(u->condition_result));
2011-03-17 04:36:19 +01:00
if (dual_timestamp_is_set(&u->assert_timestamp))
fprintf(f,
"%s\tAssert Timestamp: %s\n"
"%s\tAssert Result: %s\n",
prefix, strna(format_timestamp(timestamp1, sizeof(timestamp1), u->assert_timestamp.realtime)),
prefix, yes_no(u->assert_result));
2010-01-26 21:39:06 +01:00
for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++) {
UnitDependencyInfo di;
2010-01-26 21:39:06 +01:00
Unit *other;
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
}
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;
fprintf(f, "%s\tRequiresMountsFor: %s (", prefix, path);
print_unit_dependency_mask(f, "origin", di.origin_mask, &space);
print_unit_dependency_mask(f, "destination", di.destination_mask, &space);
fputs(")\n", f);
}
}
if (u->load_state == UNIT_LOADED) {
fprintf(f,
"%s\tStopWhenUnneeded: %s\n"
"%s\tRefuseManualStart: %s\n"
"%s\tRefuseManualStop: %s\n"
2011-04-07 04:11:31 +02:00
"%s\tDefaultDependencies: %s\n"
"%s\tOnFailureJobMode: %s\n"
"%s\tIgnoreOnIsolate: %s\n",
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),
prefix, job_mode_to_string(u->on_failure_job_mode),
prefix, yes_no(u->ignore_on_isolate));
2010-04-06 02:43:58 +02:00
if (UNIT_VTABLE(u)->dump)
UNIT_VTABLE(u)->dump(u, f, prefix2);
} else if (u->load_state == UNIT_MERGED)
fprintf(f,
"%s\tMerged into: %s\n",
prefix, u->merged_into->id);
else if (u->load_state == UNIT_ERROR)
fprintf(f, "%s\tLoad Error Code: %s\n", prefix, strerror(-u->load_error));
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
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 */
int unit_load_fragment_and_dropin(Unit *u) {
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 */
r = unit_load_fragment(u);
if (r < 0)
2010-04-06 02:43:58 +02:00
return r;
if (u->load_state == UNIT_STUB)
2010-04-06 02:43:58 +02:00
return -ENOENT;
/* 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. */
return unit_load_dropin(unit_follow_merge(u));
2010-04-06 02:43:58 +02:00
}
/* Common implementation for multiple backends */
int unit_load_fragment_and_dropin_optional(Unit *u) {
2010-04-06 02:43:58 +02:00
int r;
2010-01-26 21:39:06 +01:00
assert(u);
2010-04-06 02:43:58 +02:00
/* Same as unit_load_fragment_and_dropin(), but whether
* something can be loaded or not doesn't matter. */
/* Load a .service file */
r = unit_load_fragment(u);
if (r < 0)
2010-01-26 21:39:06 +01:00
return r;
if (u->load_state == UNIT_STUB)
u->load_state = UNIT_LOADED;
2010-01-28 02:44:47 +01:00
/* Load drop-in directory data */
return unit_load_dropin(unit_follow_merge(u));
2010-01-26 21:39:06 +01:00
}
int unit_add_default_target_dependency(Unit *u, Unit *target) {
assert(u);
assert(target);
if (target->type != UNIT_TARGET)
return 0;
/* Only add the dependency if both units are loaded, so that
* that loop check below is reliable */
if (u->load_state != UNIT_LOADED ||
target->load_state != UNIT_LOADED)
return 0;
/* If either side wants no automatic dependencies, then let's
* skip this */
if (!u->default_dependencies ||
!target->default_dependencies)
return 0;
/* Don't create loops */
if (hashmap_get(target->dependencies[UNIT_BEFORE], u))
return 0;
return unit_add_dependency(target, UNIT_AFTER, u, true, UNIT_DEPENDENCY_DEFAULT);
}
static int unit_add_target_dependencies(Unit *u) {
static const UnitDependency deps[] = {
UNIT_REQUIRED_BY,
UNIT_REQUISITE_OF,
UNIT_WANTED_BY,
UNIT_BOUND_BY
};
unsigned k;
int r = 0;
assert(u);
for (k = 0; k < ELEMENTSOF(deps); k++) {
Unit *target;
Iterator i;
void *v;
HASHMAP_FOREACH_KEY(v, target, u->dependencies[deps[k]], i) {
r = unit_add_default_target_dependency(u, target);
if (r < 0)
return r;
}
}
return r;
}
static int unit_add_slice_dependencies(Unit *u) {
UnitDependencyMask mask;
assert(u);
if (!UNIT_HAS_CGROUP_CONTEXT(u))
return 0;
/* 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). */
mask = u->type == UNIT_SLICE ? UNIT_DEPENDENCY_IMPLICIT : UNIT_DEPENDENCY_FILE;
if (UNIT_ISSET(u->slice))
return unit_add_two_dependencies(u, UNIT_AFTER, UNIT_REQUIRES, UNIT_DEREF(u->slice), true, mask);
if (unit_has_name(u, SPECIAL_ROOT_SLICE))
return 0;
return unit_add_two_dependencies_by_name(u, UNIT_AFTER, UNIT_REQUIRES, SPECIAL_ROOT_SLICE, NULL, true, mask);
}
static int unit_add_mount_dependencies(Unit *u) {
UnitDependencyInfo di;
const char *path;
Iterator i;
int r;
assert(u);
HASHMAP_FOREACH_KEY(di.data, path, u->requires_mounts_for, i) {
char prefix[strlen(path) + 1];
PATH_FOREACH_PREFIX_MORE(prefix, path) {
_cleanup_free_ char *p = NULL;
Unit *m;
r = unit_name_from_path(prefix, ".mount", &p);
if (r < 0)
return r;
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);
continue;
}
if (m == u)
continue;
if (m->load_state != UNIT_LOADED)
continue;
r = unit_add_dependency(u, UNIT_AFTER, m, true, di.origin_mask);
if (r < 0)
return r;
if (m->fragment_path) {
r = unit_add_dependency(u, UNIT_REQUIRES, m, true, di.origin_mask);
if (r < 0)
return r;
}
}
}
return 0;
}
static int unit_add_startup_units(Unit *u) {
CGroupContext *c;
int r;
c = unit_get_cgroup_context(u);
if (!c)
return 0;
if (c->startup_cpu_shares == CGROUP_CPU_SHARES_INVALID &&
core: add io controller support on the unified hierarchy On the unified hierarchy, blkio controller is renamed to io and the interface is changed significantly. * blkio.weight and blkio.weight_device are consolidated into io.weight which uses the standardized weight range [1, 10000] with 100 as the default value. * blkio.throttle.{read|write}_{bps|iops}_device are consolidated into io.max. Expansion of throttling features is being worked on to support work-conserving absolute limits (io.low and io.high). * All stats are consolidated into io.stats. This patchset adds support for the new interface. As the interface has been revamped and new features are expected to be added, it seems best to treat it as a separate controller rather than trying to expand the blkio settings although we might add automatic translation if only blkio settings are specified. * io.weight handling is mostly identical to blkio.weight[_device] handling except that the weight range is different. * Both read and write bandwidth settings are consolidated into CGroupIODeviceLimit which describes all limits applicable to the device. This makes it less painful to add new limits. * "max" can be used to specify the maximum limit which is equivalent to no config for max limits and treated as such. If a given CGroupIODeviceLimit doesn't contain any non-default configs, the config struct is discarded once the no limit config is applied to cgroup. * lookup_blkio_device() is renamed to lookup_block_device(). Signed-off-by: Tejun Heo <htejun@fb.com>
2016-05-05 22:42:55 +02:00
c->startup_io_weight == CGROUP_WEIGHT_INVALID &&
c->startup_blockio_weight == CGROUP_BLKIO_WEIGHT_INVALID)
return 0;
r = set_ensure_allocated(&u->manager->startup_units, NULL);
if (r < 0)
return r;
return set_put(u->manager->startup_units, u);
}
2010-01-26 21:39:06 +01:00
int unit_load(Unit *u) {
int r;
assert(u);
if (u->in_load_queue) {
LIST_REMOVE(load_queue, u->manager->load_queue, u);
u->in_load_queue = false;
2010-01-26 21:39:06 +01:00
}
if (u->type == _UNIT_TYPE_INVALID)
return -EINVAL;
if (u->load_state != UNIT_STUB)
2010-01-26 21:39:06 +01:00
return 0;
if (u->transient_file) {
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);
u->fragment_mtime = now(CLOCK_REALTIME);
}
if (UNIT_VTABLE(u)->load) {
r = UNIT_VTABLE(u)->load(u);
if (r < 0)
2010-01-26 21:39:06 +01:00
goto fail;
}
2010-04-06 02:43:58 +02:00
if (u->load_state == UNIT_STUB) {
2010-04-06 02:43:58 +02:00
r = -ENOENT;
goto fail;
}
if (u->load_state == UNIT_LOADED) {
r = unit_add_target_dependencies(u);
if (r < 0)
goto fail;
r = unit_add_slice_dependencies(u);
if (r < 0)
goto fail;
r = unit_add_mount_dependencies(u);
if (r < 0)
goto fail;
r = unit_add_startup_units(u);
if (r < 0)
goto fail;
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.");
r = -EINVAL;
goto fail;
}
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.");
unit_update_cgroup_members_masks(u);
}
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));
unit_add_to_gc_queue(u);
2010-01-26 21:39:06 +01:00
return 0;
fail:
u->load_state = u->load_state == UNIT_STUB ? UNIT_NOT_FOUND : UNIT_ERROR;
u->load_error = r;
unit_add_to_dbus_queue(u);
unit_add_to_gc_queue(u);
2010-04-06 02:43:58 +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
log_unit_debug_errno(u, r, "Failed to load configuration: %m");
2010-04-06 02:43:58 +02:00
2010-01-26 21:39:06 +01:00
return r;
}
static bool unit_condition_test_list(Unit *u, Condition *first, const char *(*to_string)(ConditionType t)) {
Condition *c;
int triggered = -1;
assert(u);
assert(to_string);
/* If the condition list is empty, then it is true */
if (!first)
return true;
/* Otherwise, if all of the non-trigger conditions apply and
* if any of the trigger conditions apply (unless there are
* none) we return true */
LIST_FOREACH(conditions, c, first) {
int r;
r = condition_test(c);
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,
"Couldn't determine result for %s=%s%s%s, assuming failed: %m",
to_string(c->type),
c->trigger ? "|" : "",
c->negate ? "!" : "",
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
c->parameter);
else
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,
"%s=%s%s%s %s.",
to_string(c->type),
c->trigger ? "|" : "",
c->negate ? "!" : "",
c->parameter,
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
condition_result_to_string(c->result));
if (!c->trigger && r <= 0)
return false;
if (c->trigger && triggered <= 0)
triggered = r > 0;
}
return triggered != 0;
}
static bool unit_condition_test(Unit *u) {
assert(u);
dual_timestamp_get(&u->condition_timestamp);
u->condition_result = unit_condition_test_list(u, u->conditions, condition_type_to_string);
return u->condition_result;
}
static bool unit_assert_test(Unit *u) {
assert(u);
dual_timestamp_get(&u->assert_timestamp);
u->assert_result = unit_condition_test_list(u, u->asserts, assert_type_to_string);
return u->assert_result;
}
void unit_status_printf(Unit *u, const char *status, const char *unit_status_msg_format) {
DISABLE_WARNING_FORMAT_NONLITERAL;
manager_status_printf(u->manager, STATUS_TYPE_NORMAL, status, unit_status_msg_format, unit_description(u));
REENABLE_WARNING;
}
_pure_ static const char* unit_get_status_message_format(Unit *u, JobType t) {
const char *format;
const UnitStatusMessageFormats *format_table;
assert(u);
assert(IN_SET(t, JOB_START, JOB_STOP, JOB_RELOAD));
if (t != JOB_RELOAD) {
format_table = &UNIT_VTABLE(u)->status_message_formats;
if (format_table) {
format = format_table->starting_stopping[t == JOB_STOP];
if (format)
return format;
}
}
/* Return generic strings */
if (t == JOB_START)
return "Starting %s.";
else if (t == JOB_STOP)
return "Stopping %s.";
else
return "Reloading %s.";
}
static void unit_status_print_starting_stopping(Unit *u, JobType t) {
const char *format;
assert(u);
/* Reload status messages have traditionally not been printed to console. */
if (!IN_SET(t, JOB_START, JOB_STOP))
return;
format = unit_get_status_message_format(u, t);
DISABLE_WARNING_FORMAT_NONLITERAL;
unit_status_printf(u, "", format);
REENABLE_WARNING;
}
static void unit_status_log_starting_stopping_reloading(Unit *u, JobType t) {
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
const char *format, *mid;
char buf[LINE_MAX];
assert(u);
if (!IN_SET(t, JOB_START, JOB_STOP, JOB_RELOAD))
return;
if (log_on_console())
return;
/* We log status messages for all units and all operations. */
format = unit_get_status_message_format(u, t);
DISABLE_WARNING_FORMAT_NONLITERAL;
(void) snprintf(buf, sizeof buf, format, unit_description(u));
REENABLE_WARNING;
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
mid = t == JOB_START ? "MESSAGE_ID=" SD_MESSAGE_UNIT_STARTING_STR :
t == JOB_STOP ? "MESSAGE_ID=" SD_MESSAGE_UNIT_STOPPING_STR :
"MESSAGE_ID=" SD_MESSAGE_UNIT_RELOADING_STR;
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
/* Note that we deliberately use LOG_MESSAGE() instead of
* LOG_UNIT_MESSAGE() here, since this is supposed to mimic
* closely what is written to screen using the status output,
* which is supposed the highest level, friendliest output
* possible, which means we should avoid the low-level unit
* name. */
log_struct(LOG_INFO,
LOG_MESSAGE("%s", buf),
LOG_UNIT_ID(u),
LOG_UNIT_INVOCATION_ID(u),
mid,
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
NULL);
}
void unit_status_emit_starting_stopping_reloading(Unit *u, JobType t) {
assert(u);
assert(t >= 0);
assert(t < _JOB_TYPE_MAX);
unit_status_log_starting_stopping_reloading(u, t);
unit_status_print_starting_stopping(u, t);
}
int unit_start_limit_test(Unit *u) {
assert(u);
if (ratelimit_test(&u->start_limit)) {
u->start_limit_hit = false;
return 0;
}
log_unit_warning(u, "Start request repeated too quickly.");
u->start_limit_hit = true;
return emergency_action(u->manager, u->start_limit_action, u->reboot_arg, "unit failed");
}
bool unit_shall_confirm_spawn(Unit *u) {
assert(u);
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;
}
static bool unit_verify_deps(Unit *u) {
Unit *other;
Iterator j;
void *v;
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. */
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BINDS_TO], j) {
if (!hashmap_contains(u->dependencies[UNIT_AFTER], other))
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;
}
2010-01-26 21:39:06 +01:00
/* Errors:
* -EBADR: This unit type does not support starting.
* -EALREADY: Unit is already started.
* -EAGAIN: An operation is already in progress. Retry later.
* -ECANCELED: Too many requests for now.
* -EPROTO: Assert failed
* -EINVAL: Unit not loaded
* -EOPNOTSUPP: Unit type not supported
* -ENOLINK: The necessary dependencies are not fulfilled.
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);
/* 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;
/* Units that aren't loaded cannot be started */
if (u->load_state != UNIT_LOADED)
return -EINVAL;
/* 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. */
if (state != UNIT_ACTIVATING &&
!unit_condition_test(u)) {
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, "Starting requested but condition failed. Not starting unit.");
return -EALREADY;
}
/* If the asserts failed, fail the entire job */
if (state != UNIT_ACTIVATING &&
!unit_assert_test(u)) {
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_notice(u, "Starting requested but asserts failed.");
return -EPROTO;
}
/* 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).
*/
if (!unit_supported(u))
return -EOPNOTSUPP;
/* 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. */
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. */
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;
2010-01-26 21:39:06 +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. */
unit_add_to_dbus_queue(u);
return UNIT_VTABLE(u)->start(u);
2010-01-26 21:39:06 +01:00
}
bool unit_can_start(Unit *u) {
assert(u);
if (u->load_state != UNIT_LOADED)
return false;
if (!unit_supported(u))
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) &&
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);
if (UNIT_IS_INACTIVE_OR_FAILED(state))
2010-01-26 21:39:06 +01:00
return -EALREADY;
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);
}
if (!UNIT_VTABLE(u)->stop)
return -EBADR;
unit_add_to_dbus_queue(u);
return UNIT_VTABLE(u)->stop(u);
2010-01-26 21:39:06 +01:00
}
bool unit_can_stop(Unit *u) {
assert(u);
if (!unit_supported(u))
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);
if (u->load_state != UNIT_LOADED)
return -EINVAL;
2010-01-26 21:39:06 +01:00
if (!unit_can_reload(u))
return -EBADR;
state = unit_active_state(u);
if (state == UNIT_RELOADING)
2010-01-26 21:39:06 +01:00
return -EALREADY;
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;
}
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);
}
unit_add_to_dbus_queue(u);
if (!UNIT_VTABLE(u)->reload) {
/* Unit doesn't have a reload function, but we need to propagate the reload anyway */
unit_notify(u, unit_active_state(u), unit_active_state(u), true);
return 0;
}
return UNIT_VTABLE(u)->reload(u);
2010-01-26 21:39:06 +01:00
}
bool unit_can_reload(Unit *u) {
assert(u);
if (UNIT_VTABLE(u)->can_reload)
return UNIT_VTABLE(u)->can_reload(u);
2010-01-26 21:39:06 +01:00
if (!hashmap_isempty(u->dependencies[UNIT_PROPAGATES_RELOAD_TO]))
2010-01-26 21:39:06 +01:00
return true;
return UNIT_VTABLE(u)->reload;
2010-01-26 21:39:06 +01:00
}
static void unit_check_unneeded(Unit *u) {
_cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
static const UnitDependency needed_dependencies[] = {
UNIT_REQUIRED_BY,
UNIT_REQUISITE_OF,
UNIT_WANTED_BY,
UNIT_BOUND_BY,
};
unsigned j;
int r;
assert(u);
/* If this service shall be shut down when unneeded then do
* so. */
if (!u->stop_when_unneeded)
return;
if (!UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)))
return;
for (j = 0; j < ELEMENTSOF(needed_dependencies); j++) {
Unit *other;
Iterator i;
void *v;
HASHMAP_FOREACH_KEY(v, other, u->dependencies[needed_dependencies[j]], i)
if (unit_active_or_pending(other) || unit_will_restart(other))
return;
}
/* If stopping a unit fails continuously we might enter a stop
* loop here, hence stop acting on the service being
* unnecessary after a while. */
if (!ratelimit_test(&u->auto_stop_ratelimit)) {
log_unit_warning(u, "Unit not needed anymore, but not stopping since we tried this too often recently.");
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, "Unit not needed anymore. Stopping.");
/* Ok, nobody needs us anymore. Sniff. Then let's commit suicide */
r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, &error, NULL);
if (r < 0)
log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
}
static void unit_check_binds_to(Unit *u) {
_cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
bool stop = false;
Unit *other;
Iterator i;
void *v;
int r;
assert(u);
if (u->job)
return;
if (unit_active_state(u) != UNIT_ACTIVE)
return;
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BINDS_TO], i) {
if (other->job)
continue;
systemd: do not stop units bound to inactive units while coldplugging (#6316) When running systemd-analyze verify I would get a random subset of warnings (sometimes none, sometimes one or two): dev-mapper-luks\x2d8db85dcf\x2d6230\x2d4e88\x2d940d\x2dba176d062b31.swap: Unit is bound to inactive unit dev-mapper-luks\x2d8db85dcf\x2d6230\x2d4e88\x2d940d\x2dba176d062b31.device. Stopping, too. home.mount: Unit is bound to inactive unit dev-disk-by\x2duuid-75751556\x2d6e31\x2d438b\x2d99c9\x2dd626330d9a1b.device. Stopping, too. boot.mount: Unit is bound to inactive unit dev-disk-by\x2duuid-56c56bfd\x2d93f0\x2d48fb\x2dbc4b\x2d90aa67144ea5.device. Stopping, too. When running with debug on, it's pretty obvious what is happening: home.mount: Changed dead -> mounted home.mount: Unit is bound to inactive unit dev-disk-by\x2duuid-75751556\x2d6e31\x2d438b\x2d99c9\x2dd626330d9a1b.device. Stopping, too. home.mount: Trying to enqueue job home.mount/stop/fail home.mount: Installed new job home.mount/stop as 27 home.mount: Enqueued job home.mount/stop as 27 ... dev-disk-by\x2duuid-75751556\x2d6e31\x2d438b\x2d99c9\x2dd626330d9a1b.device: Installed new job dev-disk-by\x2duuid-75751556\x2d6e31\x2d438b\x2d99c9\x2dd626330d9a1b.device/start as 47 dev-disk-by\x2duuid-75751556\x2d6e31\x2d438b\x2d99c9\x2dd626330d9a1b.device: Changed dead -> plugged dev-disk-by\x2duuid-75751556\x2d6e31\x2d438b\x2d99c9\x2dd626330d9a1b.device: Job dev-disk-by\x2duuid-75751556\x2d6e31\x2d438b\x2d99c9\x2dd626330d9a1b.device/start finished, result=done Fixes #2206, https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=808151.
2017-07-11 10:45:03 +02:00
if (!other->coldplugged)
/* We might yet create a job for the other unit… */
continue;
if (!UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(other)))
continue;
stop = true;
break;
}
if (!stop)
return;
/* If stopping a unit fails continuously we might enter a stop
* loop here, hence stop acting on the service being
* unnecessary after a while. */
if (!ratelimit_test(&u->auto_stop_ratelimit)) {
log_unit_warning(u, "Unit is bound to inactive unit %s, but not stopping since we tried this too often recently.", other->id);
return;
}
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);
/* A unit we need to run is gone. Sniff. Let's stop this. */
r = manager_add_job(u->manager, JOB_STOP, u, JOB_FAIL, &error, NULL);
if (r < 0)
log_unit_warning_errno(u, r, "Failed to enqueue stop job, ignoring: %s", bus_error_message(&error, r));
}
2010-01-26 21:39:06 +01:00
static void retroactively_start_dependencies(Unit *u) {
Iterator i;
Unit *other;
void *v;
2010-01-26 21:39:06 +01:00
assert(u);
assert(UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)));
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_REQUIRES], i)
if (!hashmap_get(u->dependencies[UNIT_AFTER], other) &&
!UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
manager_add_job(u->manager, JOB_START, other, JOB_REPLACE, NULL, NULL);
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BINDS_TO], i)
if (!hashmap_get(u->dependencies[UNIT_AFTER], other) &&
!UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
manager_add_job(u->manager, JOB_START, other, JOB_REPLACE, NULL, NULL);
2010-01-26 21:39:06 +01:00
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_WANTS], i)
if (!hashmap_get(u->dependencies[UNIT_AFTER], other) &&
!UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(other)))
manager_add_job(u->manager, JOB_START, other, JOB_FAIL, NULL, NULL);
2010-01-26 21:39:06 +01:00
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_CONFLICTS], i)
if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL);
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_CONFLICTED_BY], i)
if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL);
2010-01-26 21:39:06 +01:00
}
static void retroactively_stop_dependencies(Unit *u) {
Unit *other;
Iterator i;
void *v;
2010-01-26 21:39:06 +01:00
assert(u);
assert(UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)));
/* Pull down units which are bound to us recursively if enabled */
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BOUND_BY], i)
if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
manager_add_job(u->manager, JOB_STOP, other, JOB_REPLACE, NULL, NULL);
}
static void check_unneeded_dependencies(Unit *u) {
Unit *other;
Iterator i;
void *v;
assert(u);
assert(UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)));
/* Garbage collect services that might not be needed anymore, if enabled */
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_REQUIRES], i)
2010-01-26 21:39:06 +01:00
if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
unit_check_unneeded(other);
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_WANTS], i)
if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
unit_check_unneeded(other);
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_REQUISITE], i)
if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
unit_check_unneeded(other);
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_BINDS_TO], i)
if (!UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(other)))
unit_check_unneeded(other);
2010-01-26 21:39:06 +01:00
}
void unit_start_on_failure(Unit *u) {
Unit *other;
Iterator i;
void *v;
assert(u);
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
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_ON_FAILURE], i) {
2011-04-07 04:11:31 +02:00
int r;
r = manager_add_job(u->manager, JOB_START, other, u->on_failure_job_mode, NULL, NULL);
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_error_errno(u, r, "Failed to enqueue OnFailure= job: %m");
2011-04-07 04:11:31 +02:00
}
}
void unit_trigger_notify(Unit *u) {
Unit *other;
Iterator i;
void *v;
assert(u);
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_TRIGGERED_BY], i)
if (UNIT_VTABLE(other)->trigger_notify)
UNIT_VTABLE(other)->trigger_notify(other, u);
}
static int unit_log_resources(Unit *u) {
struct iovec iovec[1 + _CGROUP_IP_ACCOUNTING_METRIC_MAX + 4];
size_t n_message_parts = 0, n_iovec = 0;
char* message_parts[3 + 1], *t;
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",
};
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);
t = strjoin(n_message_parts > 0 ? "consumed " : "Consumed ", buf, " CPU time");
if (!t) {
r = log_oom();
goto finish;
}
message_parts[n_message_parts++] = t;
}
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;
/* 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) */
if (m == CGROUP_IP_INGRESS_BYTES)
t = strjoin(n_message_parts > 0 ? "received " : "Received ",
format_bytes(buf, sizeof(buf), value),
" IP traffic");
else if (m == CGROUP_IP_EGRESS_BYTES)
t = strjoin(n_message_parts > 0 ? "sent " : "Sent ",
format_bytes(buf, sizeof(buf), value),
" IP traffic");
else
continue;
if (!t) {
r = log_oom();
goto finish;
}
message_parts[n_message_parts++] = t;
}
/* Is there any accounting data available at all? */
if (n_iovec == 0) {
r = 0;
goto finish;
}
if (n_message_parts == 0)
t = strjoina("MESSAGE=", u->id, ": Completed");
else {
_cleanup_free_ char *joined;
message_parts[n_message_parts] = NULL;
joined = strv_join(message_parts, ", ");
if (!joined) {
r = log_oom();
goto finish;
}
t = strjoina("MESSAGE=", u->id, ": ", joined);
}
/* 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);
log_struct_iovec(LOG_INFO, iovec, n_iovec + 4);
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;
}
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);
}
void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, bool reload_success) {
bool unexpected;
Manager *m;
2010-01-26 21:39:06 +01:00
assert(u);
assert(os < _UNIT_ACTIVE_STATE_MAX);
assert(ns < _UNIT_ACTIVE_STATE_MAX);
/* 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
m = u->manager;
/* Update timestamps for state changes */
if (!MANAGER_IS_RELOADING(m)) {
dual_timestamp_get(&u->state_change_timestamp);
if (UNIT_IS_INACTIVE_OR_FAILED(os) && !UNIT_IS_INACTIVE_OR_FAILED(ns))
u->inactive_exit_timestamp = u->state_change_timestamp;
else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && UNIT_IS_INACTIVE_OR_FAILED(ns))
u->inactive_enter_timestamp = u->state_change_timestamp;
if (!UNIT_IS_ACTIVE_OR_RELOADING(os) && UNIT_IS_ACTIVE_OR_RELOADING(ns))
u->active_enter_timestamp = u->state_change_timestamp;
else if (UNIT_IS_ACTIVE_OR_RELOADING(os) && !UNIT_IS_ACTIVE_OR_RELOADING(ns))
u->active_exit_timestamp = u->state_change_timestamp;
}
2010-01-26 21:39:06 +01:00
2014-05-23 20:56:42 +02:00
/* Keep track of failed units */
(void) manager_update_failed_units(u->manager, u, ns == UNIT_FAILED);
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
unit_update_on_console(u);
if (u->job) {
unexpected = false;
2010-01-26 21:39:06 +01:00
if (u->job->state == JOB_WAITING)
2010-01-26 21:39:06 +01:00
/* 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(u->job);
2010-01-26 21:39:06 +01:00
/* Let's check whether this state change constitutes a
* finished job, or maybe contradicts a running job and
* hence needs to invalidate jobs. */
2010-01-26 21:39:06 +01:00
switch (u->job->type) {
2010-01-26 21:39:06 +01:00
case JOB_START:
case JOB_VERIFY_ACTIVE:
2010-01-26 21:39:06 +01:00
if (UNIT_IS_ACTIVE_OR_RELOADING(ns))
job_finish_and_invalidate(u->job, JOB_DONE, true, false);
else if (u->job->state == JOB_RUNNING && ns != UNIT_ACTIVATING) {
unexpected = true;
if (UNIT_IS_INACTIVE_OR_FAILED(ns))
job_finish_and_invalidate(u->job, ns == UNIT_FAILED ? JOB_FAILED : JOB_DONE, true, false);
}
2010-01-26 21:39:06 +01:00
break;
2010-01-26 21:39:06 +01:00
case JOB_RELOAD:
case JOB_RELOAD_OR_START:
case JOB_TRY_RELOAD:
2010-01-26 21:39:06 +01:00
if (u->job->state == JOB_RUNNING) {
if (ns == UNIT_ACTIVE)
job_finish_and_invalidate(u->job, reload_success ? JOB_DONE : JOB_FAILED, true, false);
else if (!IN_SET(ns, UNIT_ACTIVATING, UNIT_RELOADING)) {
unexpected = true;
if (UNIT_IS_INACTIVE_OR_FAILED(ns))
job_finish_and_invalidate(u->job, ns == UNIT_FAILED ? JOB_FAILED : JOB_DONE, true, false);
}
}
2010-01-26 21:39:06 +01:00
break;
2010-01-26 21:39:06 +01:00
case JOB_STOP:
case JOB_RESTART:
case JOB_TRY_RESTART:
2010-01-26 21:39:06 +01:00
if (UNIT_IS_INACTIVE_OR_FAILED(ns))
job_finish_and_invalidate(u->job, JOB_DONE, true, false);
else if (u->job->state == JOB_RUNNING && ns != UNIT_DEACTIVATING) {
unexpected = true;
job_finish_and_invalidate(u->job, JOB_FAILED, true, false);
}
2010-01-26 21:39:06 +01:00
break;
2010-01-26 21:39:06 +01:00
default:
assert_not_reached("Job type unknown");
2010-01-26 21:39:06 +01:00
}
} else
unexpected = true;
if (!MANAGER_IS_RELOADING(m)) {
/* 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. */
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);
}
/* stop unneeded units regardless if going down was expected or not */
if (UNIT_IS_INACTIVE_OR_DEACTIVATING(ns))
check_unneeded_dependencies(u);
if (ns != os && ns == UNIT_FAILED) {
log_unit_debug(u, "Unit entered failed state.");
unit_start_on_failure(u);
}
}
if (UNIT_IS_ACTIVE_OR_RELOADING(ns)) {
if (u->type == UNIT_SERVICE &&
!UNIT_IS_ACTIVE_OR_RELOADING(os) &&
!MANAGER_IS_RELOADING(m)) {
/* Write audit record if we have just finished starting up */
manager_send_unit_audit(m, u, AUDIT_SERVICE_START, true);
u->in_audit = true;
}
if (!UNIT_IS_ACTIVE_OR_RELOADING(os))
manager_send_unit_plymouth(m, u);
} else {
if (UNIT_IS_INACTIVE_OR_FAILED(ns) &&
!UNIT_IS_INACTIVE_OR_FAILED(os)
&& !MANAGER_IS_RELOADING(m)) {
/* This unit just stopped/failed. */
if (u->type == UNIT_SERVICE) {
/* Hmm, if there was no start record written
* write it now, so that we always have a nice
* pair */
if (!u->in_audit) {
manager_send_unit_audit(m, u, AUDIT_SERVICE_START, ns == UNIT_INACTIVE);
if (ns == UNIT_INACTIVE)
manager_send_unit_audit(m, u, AUDIT_SERVICE_STOP, true);
} else
/* Write audit record if we have just finished shutting down */
manager_send_unit_audit(m, u, AUDIT_SERVICE_STOP, ns == UNIT_INACTIVE);
u->in_audit = false;
}
/* Write a log message about consumed resources */
unit_log_resources(u);
}
}
manager_recheck_journal(m);
manager_recheck_dbus(m);
unit_trigger_notify(u);
if (!MANAGER_IS_RELOADING(u->manager)) {
/* Maybe we finished startup and are now ready for being stopped because unneeded? */
unit_check_unneeded(u);
/* 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,
* without ever entering started.) */
unit_check_binds_to(u);
if (os != UNIT_FAILED && ns == UNIT_FAILED)
(void) emergency_action(u->manager, u->failure_action, u->reboot_arg, "unit failed");
else if (!UNIT_IS_INACTIVE_OR_FAILED(os) && ns == UNIT_INACTIVE)
(void) emergency_action(u->manager, u->success_action, u->reboot_arg, "unit succeeded");
}
unit_add_to_dbus_queue(u);
unit_add_to_gc_queue(u);
2010-01-26 21:39:06 +01:00
}
int unit_watch_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
int r;
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 */
r = set_ensure_allocated(&u->pids, NULL);
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);
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;
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. */
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;
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 */
assert(hashmap_remove(u->manager->watch_pids, PID_TO_PTR(-pid)) == array);
free(array);
}
}
2010-01-26 21:39:06 +01:00
(void) set_remove(u->pids, PID_TO_PTR(pid));
}
void unit_unwatch_all_pids(Unit *u) {
assert(u);
while (!set_isempty(u->pids))
unit_unwatch_pid(u, PTR_TO_PID(set_first(u->pids)));
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);
}
void unit_tidy_watch_pids(Unit *u, pid_t except1, pid_t except2) {
Iterator i;
void *e;
assert(u);
/* Cleans dead PIDs from our list */
SET_FOREACH(e, u->pids, i) {
pid_t pid = PTR_TO_PID(e);
if (pid == except1 || pid == except2)
continue;
if (!pid_is_unwaited(pid))
unit_unwatch_pid(u, pid);
}
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:
/* Note that we don't check unit_can_start() here. That's because .device units and suchlike are not
* startable by us but may appear due to external events, and it thus makes sense to permit enqueing
* jobs for it. */
2010-01-26 21:39:06 +01:00
return true;
case JOB_STOP:
/* Similar as above. However, perpetual units can never be stopped (neither explicitly nor due to
* external events), hence it makes no sense to permit enqueing such a request either. */
return !u->perpetual;
2010-01-26 21:39:06 +01:00
case JOB_RESTART:
case JOB_TRY_RESTART:
return unit_can_stop(u) && unit_can_start(u);
2010-01-26 21:39:06 +01:00
case JOB_RELOAD:
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);
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);
}
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. */
if ((info.origin_mask & origin_mask) == info.origin_mask &&
(info.destination_mask & destination_mask) == info.destination_mask)
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,
[UNIT_REQUISITE] = UNIT_REQUISITE_OF,
[UNIT_BINDS_TO] = UNIT_BOUND_BY,
[UNIT_PART_OF] = UNIT_CONSISTS_OF,
[UNIT_REQUIRED_BY] = UNIT_REQUIRES,
[UNIT_REQUISITE_OF] = UNIT_REQUISITE,
[UNIT_WANTED_BY] = UNIT_WANTS,
[UNIT_BOUND_BY] = UNIT_BINDS_TO,
[UNIT_CONSISTS_OF] = UNIT_PART_OF,
[UNIT_CONFLICTS] = UNIT_CONFLICTED_BY,
[UNIT_CONFLICTED_BY] = UNIT_CONFLICTS,
2010-01-26 21:39:06 +01:00
[UNIT_BEFORE] = UNIT_AFTER,
[UNIT_AFTER] = UNIT_BEFORE,
[UNIT_ON_FAILURE] = _UNIT_DEPENDENCY_INVALID,
[UNIT_REFERENCES] = UNIT_REFERENCED_BY,
[UNIT_REFERENCED_BY] = UNIT_REFERENCES,
[UNIT_TRIGGERS] = UNIT_TRIGGERED_BY,
[UNIT_TRIGGERED_BY] = UNIT_TRIGGERS,
[UNIT_PROPAGATES_RELOAD_TO] = UNIT_RELOAD_PROPAGATED_FROM,
[UNIT_RELOAD_PROPAGATED_FROM] = UNIT_PROPAGATES_RELOAD_TO,
[UNIT_JOINS_NAMESPACE_OF] = UNIT_JOINS_NAMESPACE_OF,
2010-01-26 21:39:06 +01:00
};
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);
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. */
if (u == other) {
maybe_warn_about_dependency(original_u, original_other->id, d);
2010-01-26 21:39:06 +01:00
return 0;
}
2010-01-26 21:39:06 +01:00
if ((d == UNIT_BEFORE && other->type == UNIT_DEVICE) ||
(d == UNIT_AFTER && u->type == UNIT_DEVICE)) {
log_unit_warning(u, "Dependency Before=%s ignored (.device units cannot be delayed)", other->id);
return 0;
}
r = unit_add_dependency_hashmap(u->dependencies + d, other, mask, 0);
if (r < 0)
2010-01-26 21:39:06 +01:00
return r;
if (inverse_table[d] != _UNIT_DEPENDENCY_INVALID && inverse_table[d] != d) {
r = unit_add_dependency_hashmap(other->dependencies + inverse_table[d], u, 0, mask);
if (r < 0)
return r;
}
if (add_reference) {
r = unit_add_dependency_hashmap(u->dependencies + UNIT_REFERENCES, other, mask, 0);
if (r < 0)
return r;
r = unit_add_dependency_hashmap(other->dependencies + UNIT_REFERENCED_BY, u, 0, mask);
if (r < 0)
return r;
}
2010-01-26 21:39:06 +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
int unit_add_two_dependencies(Unit *u, UnitDependency d, UnitDependency e, Unit *other, bool add_reference, UnitDependencyMask mask) {
int r;
assert(u);
r = unit_add_dependency(u, d, other, add_reference, mask);
2014-08-18 22:25:24 +02:00
if (r < 0)
return r;
return unit_add_dependency(u, e, other, add_reference, mask);
}
static int resolve_template(Unit *u, const char *name, const char*path, char **buf, const char **ret) {
int r;
2010-04-15 03:11:11 +02:00
assert(u);
assert(name || path);
assert(buf);
assert(ret);
2010-04-15 03:11:11 +02:00
if (!name)
name = basename(path);
2010-04-15 03:11:11 +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
}
if (u->instance)
r = unit_name_replace_instance(name, u->instance, buf);
2010-04-15 03:11:11 +02:00
else {
_cleanup_free_ char *i = NULL;
2010-04-15 03:11:11 +02:00
r = unit_name_to_prefix(u->id, &i);
if (r < 0)
return r;
2010-04-15 03:11:11 +02:00
r = unit_name_replace_instance(name, i, buf);
2010-04-15 03:11:11 +02:00
}
if (r < 0)
return r;
2010-04-15 03:11:11 +02:00
*ret = *buf;
return 0;
2010-04-15 03:11:11 +02:00
}
int unit_add_dependency_by_name(Unit *u, UnitDependency d, const char *name, const char *path, bool add_reference, UnitDependencyMask mask) {
_cleanup_free_ char *buf = NULL;
Unit *other;
int r;
2010-04-15 03:11:11 +02:00
assert(u);
assert(name || path);
r = resolve_template(u, name, path, &buf, &name);
if (r < 0)
return r;
r = manager_load_unit(u->manager, name, path, NULL, &other);
if (r < 0)
return r;
2010-04-15 03:11:11 +02:00
return unit_add_dependency(u, d, other, add_reference, mask);
}
int unit_add_two_dependencies_by_name(Unit *u, UnitDependency d, UnitDependency e, const char *name, const char *path, bool add_reference, UnitDependencyMask mask) {
_cleanup_free_ char *buf = NULL;
Unit *other;
int r;
assert(u);
assert(name || path);
r = resolve_template(u, name, path, &buf, &name);
if (r < 0)
return r;
2014-08-18 22:25:24 +02:00
r = manager_load_unit(u->manager, name, path, NULL, &other);
if (r < 0)
return r;
return unit_add_two_dependencies(u, d, e, other, add_reference, mask);
}
2010-01-27 00:15:56 +01:00
int set_unit_path(const char *p) {
/* This is mostly for debug purposes */
if (setenv("SYSTEMD_UNIT_PATH", p, 1) < 0)
return -errno;
2010-01-27 00:15:56 +01:00
return 0;
}
2010-01-28 06:44:30 +01:00
char *unit_dbus_path(Unit *u) {
assert(u);
if (!u->id)
return NULL;
return unit_dbus_path_from_name(u->id);
}
core: add "invocation ID" concept to service manager This adds a new invocation ID concept to the service manager. The invocation ID identifies each runtime cycle of a unit uniquely. A new randomized 128bit ID is generated each time a unit moves from and inactive to an activating or active state. The primary usecase for this concept is to connect the runtime data PID 1 maintains about a service with the offline data the journal stores about it. Previously we'd use the unit name plus start/stop times, which however is highly racy since the journal will generally process log data after the service already ended. The "invocation ID" kinda matches the "boot ID" concept of the Linux kernel, except that it applies to an individual unit instead of the whole system. The invocation ID is passed to the activated processes as environment variable. It is additionally stored as extended attribute on the cgroup of the unit. The latter is used by journald to automatically retrieve it for each log logged message and attach it to the log entry. The environment variable is very easily accessible, even for unprivileged services. OTOH the extended attribute is only accessible to privileged processes (this is because cgroupfs only supports the "trusted." xattr namespace, not "user."). The environment variable may be altered by services, the extended attribute may not be, hence is the better choice for the journal. Note that reading the invocation ID off the extended attribute from journald is racy, similar to the way reading the unit name for a logging process is. This patch adds APIs to read the invocation ID to sd-id128: sd_id128_get_invocation() may be used in a similar fashion to sd_id128_get_boot(). PID1's own logging is updated to always include the invocation ID when it logs information about a unit. A new bus call GetUnitByInvocationID() is added that allows retrieving a bus path to a unit by its invocation ID. The bus path is built using the invocation ID, thus providing a path for referring to a unit that is valid only for the current runtime cycleof it. Outlook for the future: should the kernel eventually allow passing of cgroup information along AF_UNIX/SOCK_DGRAM messages via a unique cgroup id, then we can alter the invocation ID to be generated as hash from that rather than entirely randomly. This way we can derive the invocation race-freely from the messages.
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);
}
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;
if (unit_active_state(u) != UNIT_INACTIVE)
return -EBUSY;
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;
if (UNIT_DEREF(u->slice) == slice)
return 0;
/* Disallow slice changes if @u is already bound to cgroups */
if (UNIT_ISSET(u->slice) && u->cgroup_realized)
return -EBUSY;
unit_ref_set(&u->slice, u, slice);
return 1;
}
int unit_set_default_slice(Unit *u) {
_cleanup_free_ char *b = NULL;
const char *slice_name;
Unit *slice;
int r;
assert(u);
if (UNIT_ISSET(u->slice))
return 0;
if (u->instance) {
_cleanup_free_ char *prefix = NULL, *escaped = NULL;
/* Implicitly place all instantiated units in their
* own per-template slice */
r = unit_name_to_prefix(u->id, &prefix);
if (r < 0)
return r;
/* The prefix is already escaped, but it might include
* "-" which has a special meaning for slice units,
* hence escape it here extra. */
escaped = unit_name_escape(prefix);
if (!escaped)
return -ENOMEM;
if (MANAGER_IS_SYSTEM(u->manager))
b = strjoin("system-", escaped, ".slice");
else
b = strappend(escaped, ".slice");
if (!b)
return -ENOMEM;
slice_name = b;
} else
slice_name =
MANAGER_IS_SYSTEM(u->manager) && !unit_has_name(u, SPECIAL_INIT_SCOPE)
? SPECIAL_SYSTEM_SLICE
: SPECIAL_ROOT_SLICE;
r = manager_load_unit(u->manager, slice_name, NULL, NULL, &slice);
if (r < 0)
return r;
return unit_set_slice(u, slice);
}
const char *unit_slice_name(Unit *u) {
assert(u);
if (!UNIT_ISSET(u->slice))
return NULL;
return UNIT_DEREF(u->slice)->id;
}
int unit_load_related_unit(Unit *u, const char *type, Unit **_found) {
_cleanup_free_ char *t = NULL;
int r;
assert(u);
assert(type);
assert(_found);
r = unit_name_change_suffix(u->id, type, &t);
if (r < 0)
return r;
if (unit_has_name(u, t))
return -EINVAL;
r = manager_load_unit(u->manager, t, NULL, NULL, _found);
2010-04-15 03:11:11 +02:00
assert(r < 0 || *_found != u);
return r;
}
static int signal_name_owner_changed(sd_bus_message *message, void *userdata, sd_bus_error *error) {
const char *name, *old_owner, *new_owner;
Unit *u = userdata;
int r;
assert(message);
assert(u);
r = sd_bus_message_read(message, "sss", &name, &old_owner, &new_owner);
if (r < 0) {
bus_log_parse_error(r);
return 0;
}
2017-12-07 12:13:00 +01:00
old_owner = empty_to_null(old_owner);
new_owner = empty_to_null(new_owner);
if (UNIT_VTABLE(u)->bus_name_owner_change)
UNIT_VTABLE(u)->bus_name_owner_change(u, name, old_owner, new_owner);
return 0;
}
int unit_install_bus_match(Unit *u, sd_bus *bus, const char *name) {
const char *match;
assert(u);
assert(bus);
assert(name);
if (u->match_bus_slot)
return -EBUSY;
match = strjoina("type='signal',"
"sender='org.freedesktop.DBus',"
"path='/org/freedesktop/DBus',"
"interface='org.freedesktop.DBus',"
"member='NameOwnerChanged',"
"arg0='", name, "'");
return sd_bus_add_match_async(bus, &u->match_bus_slot, match, signal_name_owner_changed, NULL, u);
}
int unit_watch_bus_name(Unit *u, const char *name) {
int r;
assert(u);
assert(name);
/* Watch a specific name on the bus. We only support one unit
* watching each name for now. */
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. */
r = unit_install_bus_match(u, u->manager->api_bus, name);
if (r < 0)
return log_warning_errno(r, "Failed to subscribe to NameOwnerChanged signal for '%s': %m", name);
}
r = hashmap_put(u->manager->watch_bus, name, u);
if (r < 0) {
u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
return log_warning_errno(r, "Failed to put bus name to hashmap: %m");
}
return 0;
}
void unit_unwatch_bus_name(Unit *u, const char *name) {
assert(u);
assert(name);
(void) hashmap_remove_value(u->manager->watch_bus, name, u);
u->match_bus_slot = sd_bus_slot_unref(u->match_bus_slot);
}
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;
}
static int unit_serialize_cgroup_mask(FILE *f, const char *key, CGroupMask mask) {
_cleanup_free_ char *s = NULL;
int r = 0;
assert(f);
assert(key);
if (mask != 0) {
r = cg_mask_to_string(mask, &s);
if (r >= 0) {
fputs(key, f);
fputc('=', f);
fputs(s, f);
fputc('\n', f);
}
}
return r;
}
static const char *ip_accounting_metric_field[_CGROUP_IP_ACCOUNTING_METRIC_MAX] = {
[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",
};
int unit_serialize(Unit *u, FILE *f, FDSet *fds, bool serialize_jobs) {
CGroupIPAccountingMetric m;
2010-04-21 03:27:44 +02:00
int r;
assert(u);
assert(f);
assert(fds);
if (unit_can_serialize(u)) {
r = UNIT_VTABLE(u)->serialize(u, f, fds);
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
}
dual_timestamp_serialize(f, "state-change-timestamp", &u->state_change_timestamp);
dual_timestamp_serialize(f, "inactive-exit-timestamp", &u->inactive_exit_timestamp);
dual_timestamp_serialize(f, "active-enter-timestamp", &u->active_enter_timestamp);
dual_timestamp_serialize(f, "active-exit-timestamp", &u->active_exit_timestamp);
dual_timestamp_serialize(f, "inactive-enter-timestamp", &u->inactive_enter_timestamp);
dual_timestamp_serialize(f, "condition-timestamp", &u->condition_timestamp);
dual_timestamp_serialize(f, "assert-timestamp", &u->assert_timestamp);
2011-03-17 04:36:19 +01:00
if (dual_timestamp_is_set(&u->condition_timestamp))
unit_serialize_item(u, f, "condition-result", yes_no(u->condition_result));
2010-10-27 00:01:12 +02:00
if (dual_timestamp_is_set(&u->assert_timestamp))
unit_serialize_item(u, f, "assert-result", yes_no(u->assert_result));
unit_serialize_item(u, f, "transient", yes_no(u->transient));
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_serialize_item(u, f, "exported-invocation-id", yes_no(u->exported_invocation_id));
unit_serialize_item(u, f, "exported-log-level-max", yes_no(u->exported_log_level_max));
unit_serialize_item(u, f, "exported-log-extra-fields", yes_no(u->exported_log_extra_fields));
unit_serialize_item_format(u, f, "cpu-usage-base", "%" PRIu64, u->cpu_usage_base);
if (u->cpu_usage_last != NSEC_INFINITY)
unit_serialize_item_format(u, f, "cpu-usage-last", "%" PRIu64, u->cpu_usage_last);
if (u->cgroup_path)
unit_serialize_item(u, f, "cgroup", u->cgroup_path);
unit_serialize_item(u, f, "cgroup-realized", yes_no(u->cgroup_realized));
(void) unit_serialize_cgroup_mask(f, "cgroup-realized-mask", u->cgroup_realized_mask);
(void) unit_serialize_cgroup_mask(f, "cgroup-enabled-mask", u->cgroup_enabled_mask);
unit_serialize_item_format(u, f, "cgroup-bpf-realized", "%i", u->cgroup_bpf_state);
if (uid_is_valid(u->ref_uid))
unit_serialize_item_format(u, f, "ref-uid", UID_FMT, u->ref_uid);
if (gid_is_valid(u->ref_gid))
unit_serialize_item_format(u, f, "ref-gid", GID_FMT, u->ref_gid);
core: add "invocation ID" concept to service manager This adds a new invocation ID concept to the service manager. The invocation ID identifies each runtime cycle of a unit uniquely. A new randomized 128bit ID is generated each time a unit moves from and inactive to an activating or active state. The primary usecase for this concept is to connect the runtime data PID 1 maintains about a service with the offline data the journal stores about it. Previously we'd use the unit name plus start/stop times, which however is highly racy since the journal will generally process log data after the service already ended. The "invocation ID" kinda matches the "boot ID" concept of the Linux kernel, except that it applies to an individual unit instead of the whole system. The invocation ID is passed to the activated processes as environment variable. It is additionally stored as extended attribute on the cgroup of the unit. The latter is used by journald to automatically retrieve it for each log logged message and attach it to the log entry. The environment variable is very easily accessible, even for unprivileged services. OTOH the extended attribute is only accessible to privileged processes (this is because cgroupfs only supports the "trusted." xattr namespace, not "user."). The environment variable may be altered by services, the extended attribute may not be, hence is the better choice for the journal. Note that reading the invocation ID off the extended attribute from journald is racy, similar to the way reading the unit name for a logging process is. This patch adds APIs to read the invocation ID to sd-id128: sd_id128_get_invocation() may be used in a similar fashion to sd_id128_get_boot(). PID1's own logging is updated to always include the invocation ID when it logs information about a unit. A new bus call GetUnitByInvocationID() is added that allows retrieving a bus path to a unit by its invocation ID. The bus path is built using the invocation ID, thus providing a path for referring to a unit that is valid only for the current runtime cycleof it. Outlook for the future: should the kernel eventually allow passing of cgroup information along AF_UNIX/SOCK_DGRAM messages via a unique cgroup id, then we can alter the invocation ID to be generated as hash from that rather than entirely randomly. This way we can derive the invocation race-freely from the messages.
2016-08-30 23:18:46 +02:00
if (!sd_id128_is_null(u->invocation_id))
unit_serialize_item_format(u, f, "invocation-id", SD_ID128_FORMAT_STR, SD_ID128_FORMAT_VAL(u->invocation_id));
bus_track_serialize(u->bus_track, f, "ref");
for (m = 0; m < _CGROUP_IP_ACCOUNTING_METRIC_MAX; m++) {
uint64_t v;
r = unit_get_ip_accounting(u, m, &v);
if (r >= 0)
unit_serialize_item_format(u, f, ip_accounting_metric_field[m], "%" PRIu64, v);
}
if (serialize_jobs) {
if (u->job) {
fprintf(f, "job\n");
job_serialize(u->job, f);
}
if (u->nop_job) {
fprintf(f, "job\n");
job_serialize(u->nop_job, f);
}
}
2010-04-21 03:27:44 +02:00
/* End marker */
fputc('\n', f);
return 0;
}
int unit_serialize_item(Unit *u, FILE *f, const char *key, const char *value) {
assert(u);
assert(f);
assert(key);
if (!value)
return 0;
fputs(key, f);
fputc('=', f);
fputs(value, f);
fputc('\n', f);
return 1;
}
int unit_serialize_item_escaped(Unit *u, FILE *f, const char *key, const char *value) {
_cleanup_free_ char *c = NULL;
assert(u);
assert(f);
assert(key);
if (!value)
return 0;
c = cescape(value);
if (!c)
return -ENOMEM;
fputs(key, f);
fputc('=', f);
fputs(c, f);
fputc('\n', f);
return 1;
}
int unit_serialize_item_fd(Unit *u, FILE *f, FDSet *fds, const char *key, int fd) {
int copy;
assert(u);
assert(f);
assert(key);
if (fd < 0)
return 0;
copy = fdset_put_dup(fds, fd);
if (copy < 0)
return copy;
fprintf(f, "%s=%i\n", key, copy);
return 1;
}
2010-04-21 03:27:44 +02:00
void unit_serialize_item_format(Unit *u, FILE *f, const char *key, const char *format, ...) {
va_list ap;
assert(u);
assert(f);
assert(key);
assert(format);
fputs(key, f);
fputc('=', f);
va_start(ap, format);
vfprintf(f, format, ap);
va_end(ap);
fputc('\n', f);
}
int unit_deserialize(Unit *u, FILE *f, FDSet *fds) {
int r;
assert(u);
assert(f);
assert(fds);
for (;;) {
char line[LINE_MAX], *l, *v;
CGroupIPAccountingMetric m;
2010-04-21 03:27:44 +02:00
size_t k;
if (!fgets(line, sizeof(line), f)) {
if (feof(f))
return 0;
return -errno;
}
char_array_0(line);
2010-04-21 03:27:44 +02:00
l = strstrip(line);
/* End marker */
if (isempty(l))
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;
if (streq(l, "job")) {
if (v[0] == '\0') {
/* new-style serialized job */
Job *j;
j = job_new_raw(u);
if (!j)
return log_oom();
r = job_deserialize(j, f);
if (r < 0) {
job_free(j);
return r;
}
r = hashmap_put(u->manager->jobs, UINT32_TO_PTR(j->id), j);
if (r < 0) {
job_free(j);
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
r = job_install_deserialized(j);
if (r < 0) {
hashmap_remove(u->manager->jobs, UINT32_TO_PTR(j->id));
job_free(j);
return r;
}
} else /* legacy for pre-44 */
log_unit_warning(u, "Update from too old systemd versions are unsupported, cannot deserialize job: %s", v);
continue;
} else if (streq(l, "state-change-timestamp")) {
dual_timestamp_deserialize(v, &u->state_change_timestamp);
continue;
} else if (streq(l, "inactive-exit-timestamp")) {
dual_timestamp_deserialize(v, &u->inactive_exit_timestamp);
continue;
} else if (streq(l, "active-enter-timestamp")) {
dual_timestamp_deserialize(v, &u->active_enter_timestamp);
continue;
} else if (streq(l, "active-exit-timestamp")) {
dual_timestamp_deserialize(v, &u->active_exit_timestamp);
continue;
} else if (streq(l, "inactive-enter-timestamp")) {
dual_timestamp_deserialize(v, &u->inactive_enter_timestamp);
continue;
2011-03-17 04:36:19 +01:00
} else if (streq(l, "condition-timestamp")) {
dual_timestamp_deserialize(v, &u->condition_timestamp);
2011-03-17 04:36:19 +01:00
continue;
} else if (streq(l, "assert-timestamp")) {
dual_timestamp_deserialize(v, &u->assert_timestamp);
continue;
2011-03-17 04:36:19 +01:00
} else if (streq(l, "condition-result")) {
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
u->condition_result = r;
2011-03-29 23:32:31 +02:00
continue;
} else if (streq(l, "assert-result")) {
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);
else
u->assert_result = r;
continue;
} else if (streq(l, "transient")) {
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);
else
u->transient = 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;
} else if (STR_IN_SET(l, "cpu-usage-base", "cpuacct-usage-base")) {
r = safe_atou64(v, &u->cpu_usage_base);
if (r < 0)
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);
continue;
} else if (streq(l, "cgroup")) {
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);
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);
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;
continue;
} 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;
} else if (streq(l, "cgroup-bpf-realized")) {
int i;
r = safe_atoi(v, &i);
if (r < 0)
log_unit_debug(u, "Failed to parse cgroup BPF state %s, ignoring.", v);
else
u->cgroup_bpf_state =
i < 0 ? UNIT_CGROUP_BPF_INVALIDATED :
i > 0 ? UNIT_CGROUP_BPF_ON :
UNIT_CGROUP_BPF_OFF;
continue;
} 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);
} else if (streq(l, "ref")) {
r = strv_extend(&u->deserialized_refs, v);
if (r < 0)
log_oom();
core: add "invocation ID" concept to service manager This adds a new invocation ID concept to the service manager. The invocation ID identifies each runtime cycle of a unit uniquely. A new randomized 128bit ID is generated each time a unit moves from and inactive to an activating or active state. The primary usecase for this concept is to connect the runtime data PID 1 maintains about a service with the offline data the journal stores about it. Previously we'd use the unit name plus start/stop times, which however is highly racy since the journal will generally process log data after the service already ended. The "invocation ID" kinda matches the "boot ID" concept of the Linux kernel, except that it applies to an individual unit instead of the whole system. The invocation ID is passed to the activated processes as environment variable. It is additionally stored as extended attribute on the cgroup of the unit. The latter is used by journald to automatically retrieve it for each log logged message and attach it to the log entry. The environment variable is very easily accessible, even for unprivileged services. OTOH the extended attribute is only accessible to privileged processes (this is because cgroupfs only supports the "trusted." xattr namespace, not "user."). The environment variable may be altered by services, the extended attribute may not be, hence is the better choice for the journal. Note that reading the invocation ID off the extended attribute from journald is racy, similar to the way reading the unit name for a logging process is. This patch adds APIs to read the invocation ID to sd-id128: sd_id128_get_invocation() may be used in a similar fashion to sd_id128_get_boot(). PID1's own logging is updated to always include the invocation ID when it logs information about a unit. A new bus call GetUnitByInvocationID() is added that allows retrieving a bus path to a unit by its invocation ID. The bus path is built using the invocation ID, thus providing a path for referring to a unit that is valid only for the current runtime cycleof it. Outlook for the future: should the kernel eventually allow passing of cgroup information along AF_UNIX/SOCK_DGRAM messages via a unique cgroup id, then we can alter the invocation ID to be generated as hash from that rather than entirely randomly. This way we can derive the invocation race-freely from the messages.
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");
}
continue;
}
/* Check if this is an IP accounting metric serialization field */
for (m = 0; m < _CGROUP_IP_ACCOUNTING_METRIC_MAX; m++)
if (streq(l, ip_accounting_metric_field[m]))
break;
if (m < _CGROUP_IP_ACCOUNTING_METRIC_MAX) {
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;
}
if (unit_can_serialize(u)) {
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;
}
/* Returns positive if key was handled by the call */
if (r > 0)
continue;
r = UNIT_VTABLE(u)->deserialize_item(u, l, v, fds);
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);
}
2010-04-21 03:27:44 +02:00
}
/* 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. */
if (!dual_timestamp_is_set(&u->state_change_timestamp))
dual_timestamp_get(&u->state_change_timestamp);
/* 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);
return 0;
2010-04-21 03:27:44 +02:00
}
void unit_deserialize_skip(FILE *f) {
assert(f);
/* Skip serialized data for this unit. We don't know what it is. */
for (;;) {
char line[LINE_MAX], *l;
if (!fgets(line, sizeof line, f))
return;
char_array_0(line);
l = strstrip(line);
/* End marker */
if (isempty(l))
return;
}
}
int unit_add_node_dependency(Unit *u, const char *what, bool wants, UnitDependency dep, UnitDependencyMask mask) {
Unit *device;
_cleanup_free_ char *e = NULL;
int r;
assert(u);
/* Adds in links to the device node that this unit is based on */
if (isempty(what))
return 0;
if (!is_device_path(what))
return 0;
/* When device units aren't supported (such as in a
* container), don't create dependencies on them. */
if (!unit_type_supported(UNIT_DEVICE))
return 0;
r = unit_name_from_path(what, ".device", &e);
if (r < 0)
return r;
r = manager_load_unit(u->manager, e, NULL, NULL, &device);
if (r < 0)
return r;
if (dep == UNIT_REQUIRES && device_shall_be_bound_by(device, u))
dep = UNIT_BINDS_TO;
r = unit_add_two_dependencies(u, UNIT_AFTER,
MANAGER_IS_SYSTEM(u->manager) ? dep : UNIT_WANTS,
device, true, mask);
if (r < 0)
return r;
if (wants) {
r = unit_add_dependency(device, UNIT_WANTS, u, false, mask);
if (r < 0)
return r;
}
return 0;
}
2010-04-21 03:27:44 +02:00
int unit_coldplug(Unit *u) {
int r = 0, q;
char **i;
assert(u);
/* Make sure we don't enter a loop, when coldplugging
* recursively. */
if (u->coldplugged)
return 0;
u->coldplugged = true;
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);
if (UNIT_VTABLE(u)->coldplug) {
q = UNIT_VTABLE(u)->coldplug(u);
if (q < 0 && r >= 0)
r = q;
}
if (u->job) {
q = job_coldplug(u->job);
if (q < 0 && r >= 0)
r = q;
}
return r;
}
static bool fragment_mtime_newer(const char *path, usec_t mtime, bool path_masked) {
struct stat st;
if (!path)
return false;
/* 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;
if (stat(path, &st) < 0)
/* What, cannot access this anymore? */
return true;
if (path_masked)
/* For masked files check if they are still so */
return !null_or_empty(&st);
else
/* For non-empty files check the mtime */
return timespec_load(&st.st_mtim) > mtime;
return false;
}
bool unit_need_daemon_reload(Unit *u) {
_cleanup_strv_free_ char **t = NULL;
char **path;
assert(u);
/* For unit files, we allow masking… */
if (fragment_mtime_newer(u->fragment_path, u->fragment_mtime,
u->load_state == UNIT_MASKED))
return true;
/* Source paths should not be masked… */
if (fragment_mtime_newer(u->source_path, u->source_mtime, false))
return true;
if (u->load_state == UNIT_LOADED)
(void) unit_find_dropin_paths(u, &t);
if (!strv_equal(u->dropin_paths, t))
return true;
/* … any drop-ins that are masked are simply omitted from the list. */
STRV_FOREACH(path, u->dropin_paths)
if (fragment_mtime_newer(*path, u->dropin_mtime, false))
return true;
return false;
}
void unit_reset_failed(Unit *u) {
assert(u);
if (UNIT_VTABLE(u)->reset_failed)
UNIT_VTABLE(u)->reset_failed(u);
RATELIMIT_RESET(u->start_limit);
u->start_limit_hit = false;
}
Unit *unit_following(Unit *u) {
assert(u);
if (UNIT_VTABLE(u)->following)
return UNIT_VTABLE(u)->following(u);
return NULL;
}
bool unit_stop_pending(Unit *u) {
2010-09-01 03:35:04 +02:00
assert(u);
/* 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
return u->job && u->job->type == JOB_STOP;
}
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
if (UNIT_IS_INACTIVE_OR_DEACTIVATING(unit_active_state(u)))
return true;
if (unit_stop_pending(u))
2010-09-01 03:35:04 +02:00
return true;
return false;
}
bool unit_active_or_pending(Unit *u) {
assert(u);
/* Returns true if the unit is active or going up */
if (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)))
return true;
if (u->job &&
IN_SET(u->job->type, JOB_START, JOB_RELOAD_OR_START, JOB_RESTART))
return true;
return false;
}
bool unit_will_restart(Unit *u) {
assert(u);
if (!UNIT_VTABLE(u)->will_restart)
return false;
return UNIT_VTABLE(u)->will_restart(u);
}
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);
assert(SIGNAL_VALID(signo));
2010-10-22 16:11:50 +02:00
if (!UNIT_VTABLE(u)->kill)
return -EOPNOTSUPP;
2010-10-22 16:11:50 +02:00
return UNIT_VTABLE(u)->kill(u, w, signo, error);
2010-10-22 16:11:50 +02:00
}
static Set *unit_pid_set(pid_t main_pid, pid_t control_pid) {
Set *pid_set;
int r;
pid_set = set_new(NULL);
if (!pid_set)
return NULL;
/* Exclude the main/control pids from being killed via the cgroup */
if (main_pid > 0) {
r = set_put(pid_set, PID_TO_PTR(main_pid));
if (r < 0)
goto fail;
}
if (control_pid > 0) {
r = set_put(pid_set, PID_TO_PTR(control_pid));
if (r < 0)
goto fail;
}
return pid_set;
fail:
set_free(pid_set);
return NULL;
}
int unit_kill_common(
Unit *u,
KillWho who,
int signo,
pid_t main_pid,
pid_t control_pid,
sd_bus_error *error) {
int r = 0;
bool killed = false;
if (IN_SET(who, KILL_MAIN, KILL_MAIN_FAIL)) {
if (main_pid < 0)
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)
return sd_bus_error_set_const(error, BUS_ERROR_NO_SUCH_PROCESS, "No main process to kill");
}
if (IN_SET(who, KILL_CONTROL, KILL_CONTROL_FAIL)) {
if (control_pid < 0)
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)
return sd_bus_error_set_const(error, BUS_ERROR_NO_SUCH_PROCESS, "No control process to kill");
}
if (IN_SET(who, KILL_CONTROL, KILL_CONTROL_FAIL, KILL_ALL, KILL_ALL_FAIL))
if (control_pid > 0) {
if (kill(control_pid, signo) < 0)
r = -errno;
else
killed = true;
}
if (IN_SET(who, KILL_MAIN, KILL_MAIN_FAIL, KILL_ALL, KILL_ALL_FAIL))
if (main_pid > 0) {
if (kill(main_pid, signo) < 0)
r = -errno;
else
killed = true;
}
if (IN_SET(who, KILL_ALL, KILL_ALL_FAIL) && u->cgroup_path) {
_cleanup_set_free_ Set *pid_set = NULL;
int q;
/* Exclude the main/control pids from being killed via the cgroup */
pid_set = unit_pid_set(main_pid, control_pid);
if (!pid_set)
return -ENOMEM;
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))
r = q;
else
killed = true;
}
if (r == 0 && !killed && IN_SET(who, KILL_ALL_FAIL, KILL_CONTROL_FAIL))
return -ESRCH;
return r;
}
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(
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,
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
return u->unit_file_state;
2011-07-31 18:28:02 +02:00
}
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(
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,
basename(u->fragment_path));
return u->unit_file_preset;
}
Unit* unit_ref_set(UnitRef *ref, Unit *source, Unit *target) {
assert(ref);
assert(source);
assert(target);
if (ref->target)
unit_ref_unset(ref);
ref->source = source;
ref->target = target;
LIST_PREPEND(refs_by_target, target->refs_by_target, ref);
return target;
}
void unit_ref_unset(UnitRef *ref) {
assert(ref);
if (!ref->target)
return;
/* 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. */
unit_add_to_gc_queue(ref->target);
LIST_REMOVE(refs_by_target, ref->target->refs_by_target, ref);
ref->source = ref->target = NULL;
}
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;
if (valid_user_group_name(n)) {
*ret = TAKE_PTR(n);
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;
}
int unit_patch_contexts(Unit *u) {
CGroupContext *cc;
ExecContext *ec;
unsigned i;
int r;
assert(u);
/* Patch in the manager defaults into the exec and cgroup
* contexts, _after_ the rest of the settings have been
* initialized */
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;
}
if (MANAGER_IS_USER(u->manager) &&
!ec->working_directory) {
r = get_home_dir(&ec->working_directory);
if (r < 0)
return r;
/* Allow user services to run, even if the
* home directory is missing */
ec->working_directory_missing_ok = true;
}
if (ec->private_devices)
ec->capability_bounding_set &= ~((UINT64_C(1) << CAP_MKNOD) | (UINT64_C(1) << CAP_SYS_RAWIO));
if (ec->protect_kernel_modules)
ec->capability_bounding_set &= ~(UINT64_C(1) << CAP_SYS_MODULE);
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;
}
/* 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. */
ec->private_tmp = true;
ec->remove_ipc = true;
ec->protect_system = PROTECT_SYSTEM_STRICT;
if (ec->protect_home == PROTECT_HOME_NO)
ec->protect_home = PROTECT_HOME_READ_ONLY;
}
}
cc = unit_get_cgroup_context(u);
if (cc) {
if (ec &&
ec->private_devices &&
cc->device_policy == CGROUP_AUTO)
cc->device_policy = CGROUP_CLOSED;
}
return 0;
}
ExecContext *unit_get_exec_context(Unit *u) {
size_t offset;
assert(u);
if (u->type < 0)
return NULL;
offset = UNIT_VTABLE(u)->exec_context_offset;
if (offset <= 0)
return NULL;
return (ExecContext*) ((uint8_t*) u + offset);
}
KillContext *unit_get_kill_context(Unit *u) {
size_t offset;
assert(u);
if (u->type < 0)
return NULL;
offset = UNIT_VTABLE(u)->kill_context_offset;
if (offset <= 0)
return NULL;
return (KillContext*) ((uint8_t*) u + offset);
}
CGroupContext *unit_get_cgroup_context(Unit *u) {
size_t offset;
if (u->type < 0)
return NULL;
offset = UNIT_VTABLE(u)->cgroup_context_offset;
if (offset <= 0)
return NULL;
return (CGroupContext*) ((uint8_t*) u + offset);
}
ExecRuntime *unit_get_exec_runtime(Unit *u) {
size_t offset;
if (u->type < 0)
return NULL;
offset = UNIT_VTABLE(u)->exec_runtime_offset;
if (offset <= 0)
return NULL;
return *(ExecRuntime**) ((uint8_t*) u + offset);
}
static const char* unit_drop_in_dir(Unit *u, UnitWriteFlags flags) {
assert(u);
if (UNIT_WRITE_FLAGS_NOOP(flags))
return NULL;
if (u->transient) /* Redirect drop-ins for transient units always into the transient directory. */
return u->manager->lookup_paths.transient;
if (flags & UNIT_PERSISTENT)
return u->manager->lookup_paths.persistent_control;
if (flags & UNIT_RUNTIME)
return u->manager->lookup_paths.runtime_control;
return NULL;
}
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;
char **i;
/* 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;
return TAKE_PTR(result);
}
int unit_write_setting(Unit *u, UnitWriteFlags flags, const char *name, const char *data) {
_cleanup_free_ char *p = NULL, *q = NULL, *escaped = NULL;
const char *dir, *wrapped;
int r;
assert(u);
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);
}
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);
if (!endswith(data, "\n"))
fputc('\n', u->transient_file);
/* Remember which section we wrote this entry to */
u->last_section_private = !!(flags & UNIT_PRIVATE);
return 0;
}
dir = unit_drop_in_dir(u, flags);
if (!dir)
return -EINVAL;
wrapped = strjoina("# This is a drop-in unit file extension, created via \"systemctl set-property\"\n"
"# or an equivalent operation. Do not edit.\n",
data,
"\n");
r = drop_in_file(dir, u->id, 50, name, &p, &q);
if (r < 0)
return r;
(void) mkdir_p_label(p, 0755);
r = write_string_file_atomic_label(q, wrapped);
if (r < 0)
return r;
r = strv_push(&u->dropin_paths, q);
if (r < 0)
return r;
q = NULL;
strv_uniq(u->dropin_paths);
u->dropin_mtime = now(CLOCK_REALTIME);
return 0;
}
int unit_write_settingf(Unit *u, UnitWriteFlags flags, const char *name, const char *format, ...) {
_cleanup_free_ char *p = NULL;
va_list ap;
int r;
assert(u);
assert(name);
assert(format);
if (UNIT_WRITE_FLAGS_NOOP(flags))
return 0;
va_start(ap, format);
r = vasprintf(&p, format, ap);
va_end(ap);
if (r < 0)
return -ENOMEM;
return unit_write_setting(u, flags, name, p);
}
int unit_make_transient(Unit *u) {
2017-11-23 17:39:25 +01:00
_cleanup_free_ char *path = NULL;
FILE *f;
assert(u);
if (!UNIT_VTABLE(u)->can_transient)
return -EOPNOTSUPP;
(void) mkdir_p_label(u->manager->lookup_paths.transient, 0755);
path = strjoin(u->manager->lookup_paths.transient, "/", u->id);
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. */
RUN_WITH_UMASK(0022) {
f = fopen(path, "we");
2017-11-23 17:39:25 +01:00
if (!f)
return -errno;
}
2017-11-23 17:39:25 +01:00
safe_fclose(u->transient_file);
u->transient_file = f;
2017-11-23 17:39:25 +01:00
free_and_replace(u->fragment_path, path);
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;
u->load_state = UNIT_STUB;
u->load_error = 0;
u->transient = true;
unit_add_to_dbus_queue(u);
unit_add_to_gc_queue(u);
fputs("# This is a transient unit file, created programmatically via the systemd API. Do not edit.\n",
u->transient_file);
return 0;
}
static void log_kill(pid_t pid, int sig, void *userdata) {
_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] == '(')
return;
log_unit_notice(userdata,
"Killing process " PID_FMT " (%s) with signal SIG%s.",
pid,
strna(comm),
signal_to_string(sig));
}
static int operation_to_signal(KillContext *c, KillOperation k) {
assert(c);
switch (k) {
case KILL_TERMINATE:
case KILL_TERMINATE_AND_LOG:
return c->kill_signal;
case KILL_KILL:
return SIGKILL;
case KILL_ABORT:
return SIGABRT;
default:
assert_not_reached("KillOperation unknown");
}
}
int unit_kill_context(
Unit *u,
KillContext *c,
KillOperation k,
pid_t main_pid,
pid_t control_pid,
bool main_pid_alien) {
bool wait_for_exit = false, send_sighup;
cg_kill_log_func_t log_func = NULL;
int sig, r;
assert(u);
assert(c);
/* 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. */
if (c->kill_mode == KILL_NONE)
return 0;
sig = operation_to_signal(c, k);
send_sighup =
c->send_sighup &&
IN_SET(k, KILL_TERMINATE, KILL_TERMINATE_AND_LOG) &&
sig != SIGHUP;
if (k != KILL_TERMINATE || IN_SET(sig, SIGKILL, SIGABRT))
log_func = log_kill;
if (main_pid > 0) {
if (log_func)
log_func(main_pid, sig, u);
r = kill_and_sigcont(main_pid, sig);
if (r < 0 && r != -ESRCH) {
_cleanup_free_ char *comm = NULL;
(void) get_process_comm(main_pid, &comm);
log_unit_warning_errno(u, r, "Failed to kill main process " PID_FMT " (%s), ignoring: %m", main_pid, strna(comm));
} else {
if (!main_pid_alien)
wait_for_exit = true;
if (r != -ESRCH && send_sighup)
(void) kill(main_pid, SIGHUP);
}
}
if (control_pid > 0) {
if (log_func)
log_func(control_pid, sig, u);
r = kill_and_sigcont(control_pid, sig);
if (r < 0 && r != -ESRCH) {
_cleanup_free_ char *comm = NULL;
(void) get_process_comm(control_pid, &comm);
log_unit_warning_errno(u, r, "Failed to kill control process " PID_FMT " (%s), ignoring: %m", control_pid, strna(comm));
} else {
wait_for_exit = true;
if (r != -ESRCH && send_sighup)
(void) kill(control_pid, SIGHUP);
}
}
if (u->cgroup_path &&
(c->kill_mode == KILL_CONTROL_GROUP || (c->kill_mode == KILL_MIXED && k == KILL_KILL))) {
_cleanup_set_free_ Set *pid_set = NULL;
/* Exclude the main/control pids from being killed via the cgroup */
pid_set = unit_pid_set(main_pid, control_pid);
if (!pid_set)
return -ENOMEM;
r = cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
sig,
CGROUP_SIGCONT|CGROUP_IGNORE_SELF,
pid_set,
log_func, u);
if (r < 0) {
2017-10-04 16:01:32 +02:00
if (!IN_SET(r, -EAGAIN, -ESRCH, -ENOENT))
log_unit_warning_errno(u, r, "Failed to kill control group %s, ignoring: %m", u->cgroup_path);
} else if (r > 0) {
/* 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
if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0 ||
(detect_container() == 0 && !unit_cgroup_delegate(u)))
wait_for_exit = true;
if (send_sighup) {
set_free(pid_set);
pid_set = unit_pid_set(main_pid, control_pid);
if (!pid_set)
return -ENOMEM;
cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path,
SIGHUP,
CGROUP_IGNORE_SELF,
pid_set,
NULL, NULL);
}
}
}
return wait_for_exit;
}
int unit_require_mounts_for(Unit *u, const char *path, UnitDependencyMask mask) {
_cleanup_free_ char *p = NULL;
char *prefix;
UnitDependencyInfo di;
int r;
assert(u);
assert(path);
/* 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. */
if (!path_is_absolute(path))
return -EINVAL;
r = hashmap_ensure_allocated(&u->requires_mounts_for, &path_hash_ops);
if (r < 0)
return r;
p = strdup(path);
if (!p)
return -ENOMEM;
path = path_kill_slashes(p);
if (!path_is_normalized(path))
return -EPERM;
if (hashmap_contains(u->requires_mounts_for, path))
return 0;
di = (UnitDependencyInfo) {
.origin_mask = mask
};
r = hashmap_put(u->requires_mounts_for, path, di.data);
if (r < 0)
return r;
p = NULL;
prefix = alloca(strlen(path) + 1);
PATH_FOREACH_PREFIX_MORE(prefix, path) {
Set *x;
x = hashmap_get(u->manager->units_requiring_mounts_for, prefix);
if (!x) {
_cleanup_free_ char *q = NULL;
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;
q = strdup(prefix);
if (!q)
return -ENOMEM;
x = set_new(NULL);
if (!x)
return -ENOMEM;
r = hashmap_put(u->manager->units_requiring_mounts_for, q, x);
if (r < 0) {
set_free(x);
return r;
}
q = NULL;
}
r = set_put(x, u);
if (r < 0)
return r;
}
return 0;
}
int unit_setup_exec_runtime(Unit *u) {
ExecRuntime **rt;
size_t offset;
Unit *other;
Iterator i;
void *v;
int r;
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? */
rt = (ExecRuntime**) ((uint8_t*) u + offset);
if (*rt)
return 0;
/* Try to get it from somebody else */
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_JOINS_NAMESPACE_OF], i) {
r = exec_runtime_acquire(u->manager, NULL, other->id, false, rt);
if (r == 1)
return 1;
}
return exec_runtime_acquire(u->manager, unit_get_exec_context(u), u->id, true, rt);
}
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);
}
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();
}
void unit_warn_if_dir_nonempty(Unit *u, const char* where) {
int r;
assert(u);
assert(where);
r = dir_is_empty(where);
if (r > 0 || r == -ENOTDIR)
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,
LOG_UNIT_ID(u),
LOG_UNIT_INVOCATION_ID(u),
LOG_UNIT_MESSAGE(u, "Directory %s to mount over is not empty, mounting anyway.", where),
"WHERE=%s", where,
NULL);
}
int unit_fail_if_noncanonical(Unit *u, const char* where) {
_cleanup_free_ char *canonical_where;
int r;
assert(u);
assert(where);
r = chase_symlinks(where, NULL, CHASE_NONEXISTENT, &canonical_where);
if (r < 0) {
log_unit_debug_errno(u, r, "Failed to check %s for symlinks, ignoring: %m", where);
return 0;
}
/* We will happily ignore a trailing slash (or any redundant slashes) */
if (path_equal(where, canonical_where))
return 0;
/* No need to mention "." or "..", they would already have been rejected by unit_name_from_path() */
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,
LOG_UNIT_ID(u),
LOG_UNIT_INVOCATION_ID(u),
LOG_UNIT_MESSAGE(u, "Mount path %s is not canonical (contains a symlink).", where),
"WHERE=%s", where,
NULL);
return -ELOOP;
}
bool unit_is_pristine(Unit *u) {
assert(u);
/* Check if the unit already exists or is already around,
* in a number of different ways. Note that to cater for unit
* types such as slice, we are generally fine with units that
* are marked UNIT_LOADED even though nothing was
* actually loaded, as those unit types don't require a file
* on disk to validly load. */
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);
}
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;
}
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;
}
void unit_unref_uid(Unit *u, bool destroy_now) {
unit_unref_uid_internal(u, &u->ref_uid, destroy_now, manager_unref_uid);
}
void unit_unref_gid(Unit *u, bool destroy_now) {
unit_unref_uid_internal(u, (uid_t*) &u->ref_gid, destroy_now, manager_unref_gid);
}
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;
}
int unit_ref_uid(Unit *u, uid_t uid, bool clean_ipc) {
return unit_ref_uid_internal(u, &u->ref_uid, uid, clean_ipc, manager_ref_uid);
}
int unit_ref_gid(Unit *u, gid_t gid, bool clean_ipc) {
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_unref_uid_gid(Unit *u, bool destroy_now) {
assert(u);
unit_unref_uid(u, destroy_now);
unit_unref_gid(u, destroy_now);
}
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)
bus_unit_send_change_signal(u);
}
core: add "invocation ID" concept to service manager This adds a new invocation ID concept to the service manager. The invocation ID identifies each runtime cycle of a unit uniquely. A new randomized 128bit ID is generated each time a unit moves from and inactive to an activating or active state. The primary usecase for this concept is to connect the runtime data PID 1 maintains about a service with the offline data the journal stores about it. Previously we'd use the unit name plus start/stop times, which however is highly racy since the journal will generally process log data after the service already ended. The "invocation ID" kinda matches the "boot ID" concept of the Linux kernel, except that it applies to an individual unit instead of the whole system. The invocation ID is passed to the activated processes as environment variable. It is additionally stored as extended attribute on the cgroup of the unit. The latter is used by journald to automatically retrieve it for each log logged message and attach it to the log entry. The environment variable is very easily accessible, even for unprivileged services. OTOH the extended attribute is only accessible to privileged processes (this is because cgroupfs only supports the "trusted." xattr namespace, not "user."). The environment variable may be altered by services, the extended attribute may not be, hence is the better choice for the journal. Note that reading the invocation ID off the extended attribute from journald is racy, similar to the way reading the unit name for a logging process is. This patch adds APIs to read the invocation ID to sd-id128: sd_id128_get_invocation() may be used in a similar fashion to sd_id128_get_boot(). PID1's own logging is updated to always include the invocation ID when it logs information about a unit. A new bus call GetUnitByInvocationID() is added that allows retrieving a bus path to a unit by its invocation ID. The bus path is built using the invocation ID, thus providing a path for referring to a unit that is valid only for the current runtime cycleof it. Outlook for the future: should the kernel eventually allow passing of cgroup information along AF_UNIX/SOCK_DGRAM messages via a unique cgroup id, then we can alter the invocation ID to be generated as hash from that rather than entirely randomly. This way we can derive the invocation race-freely from the messages.
2016-08-30 23:18:46 +02:00
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;
}
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");
return 0;
}
void unit_set_exec_params(Unit *u, ExecParameters *p) {
assert(u);
assert(p);
/* Copy parameters from manager */
p->environment = u->manager->environment;
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));
/* Copy paramaters from unit */
p->cgroup_path = u->cgroup_path;
SET_FLAG(p->flags, EXEC_CGROUP_DELEGATE, unit_cgroup_delegate(u));
}
int unit_fork_helper_process(Unit *u, const char *name, pid_t *ret) {
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);
r = safe_fork(name, FORK_REOPEN_LOG, ret);
if (r != 0)
return r;
(void) default_signals(SIGNALS_CRASH_HANDLER, SIGNALS_IGNORE, -1);
(void) ignore_signals(SIGPIPE, -1);
(void) prctl(PR_SET_PDEATHSIG, SIGTERM);
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);
}
}
return 0;
}
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));
log_unit_debug(u, "%s lost dependency %s=%s", u->id, unit_dependency_to_string(d), other->id);
} 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) {
UnitDependency d;
assert(u);
/* Removes all dependencies u has on other units marked for ownership by 'mask'. */
if (mask == 0)
return;
for (d = 0; d < _UNIT_DEPENDENCY_MAX; d++) {
bool done;
do {
UnitDependencyInfo di;
Unit *other;
Iterator i;
done = true;
HASHMAP_FOREACH_KEY(di.data, other, u->dependencies[d], i) {
UnitDependency q;
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. */
for (q = 0; q < _UNIT_DEPENDENCY_MAX; q++) {
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
static int unit_export_invocation_id(Unit *u) {
const char *p;
int r;
assert(u);
if (u->exported_invocation_id)
return 0;
if (sd_id128_is_null(u->invocation_id))
return 0;
p = strjoina("/run/systemd/units/invocation:", u->id);
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;
}
void unit_export_state_files(Unit *u) {
const ExecContext *c;
assert(u);
if (!u->id)
return;
if (!MANAGER_IS_SYSTEM(u->manager))
return;
/* 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);
c = unit_get_exec_context(u);
if (c) {
(void) unit_export_log_level_max(u, c);
(void) unit_export_log_extra_fields(u, c);
}
}
void unit_unlink_state_files(Unit *u) {
const char *p;
assert(u);
if (!u->id)
return;
if (!MANAGER_IS_SYSTEM(u->manager))
return;
/* Undoes the effect of unit_export_state() */
if (u->exported_invocation_id) {
p = strjoina("/run/systemd/units/invocation:", u->id);
(void) unlink(p);
u->exported_invocation_id = false;
}
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;
}
}
int unit_prepare_exec(Unit *u) {
int r;
assert(u);
/* Prepares everything so that we can fork of a process for this unit */
(void) unit_realize_cgroup(u);
if (u->reset_accounting) {
(void) unit_reset_cpu_accounting(u);
(void) unit_reset_ip_accounting(u);
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;
}
static void log_leftover(pid_t pid, int sig, void *userdata) {
_cleanup_free_ char *comm = NULL;
(void) get_process_comm(pid, &comm);
if (comm && comm[0] == '(') /* Most likely our own helper process (PAM?), ignore */
return;
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));
}
void unit_warn_leftover_processes(Unit *u) {
assert(u);
(void) unit_pick_cgroup_path(u);
if (!u->cgroup_path)
return;
(void) cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, 0, 0, NULL, log_leftover, u);
}
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);
}
const char *unit_label_path(Unit *u) {
const char *p;
/* 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 */
if (path_equal(p, "/dev/null"))
return NULL;
return p;
}
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())
return sd_bus_error_setf(error, SD_BUS_ERROR_INVALID_ARGS, "Process " PID_FMT " is a manager processs, refusing.", pid);
/* 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;
}
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);