Systemd/src/core/device.c

949 lines
30 KiB
C
Raw Normal View History

/***
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/>.
***/
#include <errno.h>
2010-01-29 06:45:59 +01:00
#include <sys/epoll.h>
#include "libudev.h"
#include "alloc-util.h"
2010-04-18 03:08:16 +02:00
#include "dbus-device.h"
#include "device.h"
#include "log.h"
#include "parse-util.h"
2012-05-07 21:36:12 +02:00
#include "path-util.h"
#include "stat-util.h"
#include "string-util.h"
#include "swap.h"
#include "udev-util.h"
#include "unit-name.h"
#include "unit.h"
2010-01-29 03:18:09 +01:00
static const UnitActiveState state_translation_table[_DEVICE_STATE_MAX] = {
[DEVICE_DEAD] = UNIT_INACTIVE,
[DEVICE_TENTATIVE] = UNIT_ACTIVATING,
[DEVICE_PLUGGED] = UNIT_ACTIVE,
2010-01-29 03:18:09 +01:00
};
static int device_dispatch_io(sd_event_source *source, int fd, uint32_t revents, void *userdata);
static void device_unset_sysfs(Device *d) {
2013-11-25 15:25:01 +01:00
Hashmap *devices;
Device *first;
assert(d);
if (!d->sysfs)
return;
/* Remove this unit from the chain of devices which share the
* same sysfs path. */
2013-11-25 15:25:01 +01:00
devices = UNIT(d)->manager->devices_by_sysfs;
first = hashmap_get(devices, d->sysfs);
LIST_REMOVE(same_sysfs, first, d);
if (first)
2013-11-25 15:25:01 +01:00
hashmap_remove_and_replace(devices, d->sysfs, first->sysfs, first);
else
2013-11-25 15:25:01 +01:00
hashmap_remove(devices, d->sysfs);
d->sysfs = mfree(d->sysfs);
}
static int device_set_sysfs(Device *d, const char *sysfs) {
Device *first;
char *copy;
int r;
assert(d);
if (streq_ptr(d->sysfs, sysfs))
return 0;
r = hashmap_ensure_allocated(&UNIT(d)->manager->devices_by_sysfs, &string_hash_ops);
if (r < 0)
return r;
copy = strdup(sysfs);
if (!copy)
return -ENOMEM;
device_unset_sysfs(d);
first = hashmap_get(UNIT(d)->manager->devices_by_sysfs, sysfs);
LIST_PREPEND(same_sysfs, first, d);
r = hashmap_replace(UNIT(d)->manager->devices_by_sysfs, copy, first);
if (r < 0) {
LIST_REMOVE(same_sysfs, first, d);
free(copy);
return r;
}
d->sysfs = copy;
return 0;
}
static void device_init(Unit *u) {
Device *d = DEVICE(u);
assert(d);
assert(UNIT(d)->load_state == UNIT_STUB);
/* In contrast to all other unit types we timeout jobs waiting
* for devices by default. This is because they otherwise wait
* indefinitely for plugged in devices, something which cannot
* happen for the other units since their operations time out
* anyway. */
u->job_running_timeout = u->manager->default_timeout_start_usec;
2013-11-25 15:25:01 +01:00
u->ignore_on_isolate = true;
}
2010-01-26 21:39:06 +01:00
static void device_done(Unit *u) {
Device *d = DEVICE(u);
assert(d);
device_unset_sysfs(d);
}
2010-01-29 03:18:09 +01:00
static void device_set_state(Device *d, DeviceState state) {
DeviceState old_state;
assert(d);
2010-01-29 03:18:09 +01:00
old_state = d->state;
d->state = state;
if (state != old_state)
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(UNIT(d), "Changed %s -> %s", device_state_to_string(old_state), device_state_to_string(state));
2010-01-29 03:18:09 +01:00
unit_notify(UNIT(d), state_translation_table[old_state], state_translation_table[state], true);
2010-01-29 03:18:09 +01:00
}
static int device_coldplug(Unit *u) {
2010-01-29 03:18:09 +01:00
Device *d = DEVICE(u);
assert(d);
assert(d->state == DEVICE_DEAD);
if (d->found & DEVICE_FOUND_UDEV)
/* If udev says the device is around, it's around */
device_set_state(d, DEVICE_PLUGGED);
else if (d->found != DEVICE_NOT_FOUND && d->deserialized_state != DEVICE_PLUGGED)
/* If a device is found in /proc/self/mountinfo or
* /proc/swaps, and was not yet announced via udev,
* it's "tentatively" around. */
device_set_state(d, DEVICE_TENTATIVE);
2010-01-29 03:18:09 +01:00
return 0;
}
static int device_serialize(Unit *u, FILE *f, FDSet *fds) {
Device *d = DEVICE(u);
assert(u);
assert(f);
assert(fds);
unit_serialize_item(u, f, "state", device_state_to_string(d->state));
return 0;
}
static int device_deserialize_item(Unit *u, const char *key, const char *value, FDSet *fds) {
Device *d = DEVICE(u);
assert(u);
assert(key);
assert(value);
assert(fds);
if (streq(key, "state")) {
DeviceState state;
state = device_state_from_string(value);
if (state < 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 state value: %s", value);
else
d->deserialized_state = state;
} 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, "Unknown serialization key: %s", key);
return 0;
}
2010-01-29 03:18:09 +01:00
static void device_dump(Unit *u, FILE *f, const char *prefix) {
Device *d = DEVICE(u);
assert(d);
fprintf(f,
"%sDevice State: %s\n"
"%sSysfs Path: %s\n",
2010-04-21 03:27:44 +02:00
prefix, device_state_to_string(d->state),
2010-01-29 03:18:09 +01:00
prefix, strna(d->sysfs));
}
_pure_ static UnitActiveState device_active_state(Unit *u) {
2010-01-29 03:18:09 +01:00
assert(u);
return state_translation_table[DEVICE(u)->state];
}
_pure_ static const char *device_sub_state_to_string(Unit *u) {
assert(u);
2010-04-21 03:27:44 +02:00
return device_state_to_string(DEVICE(u)->state);
}
static int device_update_description(Unit *u, struct udev_device *dev, const char *path) {
const char *model;
int r;
assert(u);
assert(dev);
assert(path);
model = udev_device_get_property_value(dev, "ID_MODEL_FROM_DATABASE");
if (!model)
model = udev_device_get_property_value(dev, "ID_MODEL");
if (model) {
const char *label;
/* Try to concatenate the device model string with a label, if there is one */
label = udev_device_get_property_value(dev, "ID_FS_LABEL");
if (!label)
label = udev_device_get_property_value(dev, "ID_PART_ENTRY_NAME");
if (!label)
label = udev_device_get_property_value(dev, "ID_PART_ENTRY_NUMBER");
if (label) {
_cleanup_free_ char *j;
j = strjoin(model, " ", label);
if (j)
r = unit_set_description(u, j);
2015-02-28 23:39:55 +01:00
else
r = -ENOMEM;
} else
r = unit_set_description(u, model);
} else
r = unit_set_description(u, path);
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 set device description: %m");
return r;
}
static int device_add_udev_wants(Unit *u, struct udev_device *dev) {
const char *wants, *property;
int r;
assert(u);
assert(dev);
property = MANAGER_IS_USER(u->manager) ? "SYSTEMD_USER_WANTS" : "SYSTEMD_WANTS";
wants = udev_device_get_property_value(dev, property);
if (!wants)
return 0;
for (;;) {
_cleanup_free_ char *word = NULL, *k = NULL;
r = extract_first_word(&wants, &word, NULL, EXTRACT_QUOTES);
if (r == 0)
return 0;
if (r == -ENOMEM)
return log_oom();
if (r < 0)
return log_unit_error_errno(u, r, "Failed to add parse %s: %m", property);
r = unit_name_mangle(word, UNIT_NAME_NOGLOB, &k);
if (r < 0)
return log_unit_error_errno(u, r, "Failed to mangle unit name \"%s\": %m", word);
r = unit_add_dependency_by_name(u, UNIT_WANTS, k, NULL, true, UNIT_DEPENDENCY_UDEV);
if (r < 0)
return log_unit_error_errno(u, r, "Failed to add Wants= dependency: %m");
}
}
static bool device_is_bound_by_mounts(Device *d, struct udev_device *dev) {
const char *bound_by;
int r;
assert(d);
assert(dev);
bound_by = udev_device_get_property_value(dev, "SYSTEMD_MOUNT_DEVICE_BOUND");
if (bound_by) {
r = parse_boolean(bound_by);
if (r < 0)
log_warning_errno(r, "Failed to parse SYSTEMD_MOUNT_DEVICE_BOUND='%s' udev property of %s, ignoring: %m", bound_by, strna(d->sysfs));
d->bind_mounts = r > 0;
} else
d->bind_mounts = false;
return d->bind_mounts;
}
static int device_upgrade_mount_deps(Unit *u) {
Unit *other;
Iterator i;
void *v;
int r;
/* Let's upgrade Requires= to BindsTo= on us. (Used when SYSTEMD_MOUNT_DEVICE_BOUND is set) */
HASHMAP_FOREACH_KEY(v, other, u->dependencies[UNIT_REQUIRED_BY], i) {
if (other->type != UNIT_MOUNT)
continue;
r = unit_add_dependency(other, UNIT_BINDS_TO, u, true, UNIT_DEPENDENCY_UDEV);
if (r < 0)
return r;
}
return 0;
}
static int device_setup_unit(Manager *m, struct udev_device *dev, const char *path, bool main) {
_cleanup_free_ char *e = NULL;
const char *sysfs = NULL;
Unit *u = NULL;
bool delete;
int r;
assert(m);
assert(path);
if (dev) {
sysfs = udev_device_get_syspath(dev);
if (!sysfs)
return 0;
}
r = unit_name_from_path(path, ".device", &e);
if (r < 0)
2015-05-06 19:56:29 +02:00
return log_error_errno(r, "Failed to generate unit name from device path: %m");
u = manager_get_unit(m, e);
/* The device unit can still be present even if the device was
* unplugged: a mount unit can reference it hence preventing
* the GC to have garbaged it. That's desired since the device
* unit may have a dependency on the mount unit which was
* added during the loading of the later. */
if (dev && u && DEVICE(u)->state == DEVICE_PLUGGED) {
/* This unit is in plugged state: we're sure it's
* attached to a device. */
if (!path_equal(DEVICE(u)->sysfs, sysfs)) {
log_unit_debug(u, "Dev %s appeared twice with different sysfs paths %s and %s",
e, DEVICE(u)->sysfs, sysfs);
return -EEXIST;
}
}
if (!u) {
delete = true;
r = unit_new_for_name(m, sizeof(Device), e, &u);
if (r < 0)
goto fail;
unit_add_to_load_queue(u);
} else
delete = false;
/* If this was created via some dependency and has not
* actually been seen yet ->sysfs will not be
* initialized. Hence initialize it if necessary. */
if (sysfs) {
r = device_set_sysfs(DEVICE(u), sysfs);
if (r < 0)
goto fail;
(void) device_update_description(u, dev, path);
/* The additional systemd udev properties we only interpret
* for the main object */
if (main)
(void) device_add_udev_wants(u, dev);
}
/* So the user wants the mount units to be bound to the device but a mount unit might has been seen by systemd
* before the device appears on its radar. In this case the device unit is partially initialized and includes
* the deps on the mount unit but at that time the "bind mounts" flag wasn't not present. Fix this up now. */
if (dev && device_is_bound_by_mounts(DEVICE(u), dev))
device_upgrade_mount_deps(u);
2010-01-29 06:45:59 +01:00
2013-11-25 15:25:01 +01:00
/* Note that this won't dispatch the load queue, the caller
* has to do that if needed and appropriate */
unit_add_to_dbus_queue(u);
return 0;
fail:
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_errno(u, r, "Failed to set up device unit: %m");
2010-07-13 02:17:26 +02:00
if (delete)
unit_free(u);
2010-07-13 02:17:26 +02:00
return r;
}
static int device_process_new(Manager *m, struct udev_device *dev) {
2013-11-25 15:25:01 +01:00
const char *sysfs, *dn, *alias;
struct udev_list_entry *item = NULL, *first = NULL;
int r;
assert(m);
sysfs = udev_device_get_syspath(dev);
if (!sysfs)
2013-11-25 15:25:01 +01:00
return 0;
/* Add the main unit named after the sysfs path */
r = device_setup_unit(m, dev, sysfs, true);
if (r < 0)
return r;
/* Add an additional unit for the device node */
2013-11-25 15:25:01 +01:00
dn = udev_device_get_devnode(dev);
if (dn)
(void) device_setup_unit(m, dev, dn, false);
/* Add additional units for all symlinks */
first = udev_device_get_devlinks_list_entry(dev);
udev_list_entry_foreach(item, first) {
const char *p;
struct stat st;
/* Don't bother with the /dev/block links */
p = udev_list_entry_get_name(item);
if (path_startswith(p, "/dev/block/") ||
path_startswith(p, "/dev/char/"))
continue;
/* Verify that the symlink in the FS actually belongs
* to this device. This is useful to deal with
* conflicting devices, e.g. when two disks want the
* same /dev/disk/by-label/xxx link because they have
* the same label. We want to make sure that the same
* device that won the symlink wins in systemd, so we
* check the device node major/minor */
if (stat(p, &st) >= 0)
if ((!S_ISBLK(st.st_mode) && !S_ISCHR(st.st_mode)) ||
st.st_rdev != udev_device_get_devnum(dev))
continue;
(void) device_setup_unit(m, dev, p, false);
}
2013-11-25 15:25:01 +01:00
/* Add additional units for all explicitly configured
* aliases */
alias = udev_device_get_property_value(dev, "SYSTEMD_ALIAS");
for (;;) {
2016-12-06 00:07:34 +01:00
_cleanup_free_ char *word = NULL;
r = extract_first_word(&alias, &word, NULL, EXTRACT_QUOTES);
if (r == 0)
return 0;
if (r == -ENOMEM)
return log_oom();
if (r < 0)
return log_warning_errno(r, "Failed to add parse SYSTEMD_ALIAS for %s: %m", sysfs);
2013-11-25 15:25:01 +01:00
if (path_is_absolute(word))
(void) device_setup_unit(m, dev, word, false);
else
log_warning("SYSTEMD_ALIAS for %s is not an absolute path, ignoring: %s", sysfs, word);
}
}
static void device_update_found_one(Device *d, bool add, DeviceFound found, bool now) {
DeviceFound n, previous;
assert(d);
n = add ? (d->found | found) : (d->found & ~found);
if (n == d->found)
return;
previous = d->found;
d->found = n;
if (!now)
return;
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
/* Didn't exist before, but does now? if so, generate a new invocation ID for it */
if (previous == DEVICE_NOT_FOUND && d->found != DEVICE_NOT_FOUND)
(void) unit_acquire_invocation_id(UNIT(d));
if (d->found & DEVICE_FOUND_UDEV)
/* When the device is known to udev we consider it
* plugged. */
device_set_state(d, DEVICE_PLUGGED);
else if (d->found != DEVICE_NOT_FOUND && (previous & DEVICE_FOUND_UDEV) == 0)
/* If the device has not been seen by udev yet, but is
* now referenced by the kernel, then we assume the
* kernel knows it now, and udev might soon too. */
device_set_state(d, DEVICE_TENTATIVE);
else {
/* If nobody sees the device, or if the device was
* previously seen by udev and now is only referenced
* from the kernel, then we consider the device is
* gone, the kernel just hasn't noticed it yet. */
device_set_state(d, DEVICE_DEAD);
device_unset_sysfs(d);
}
}
static int device_update_found_by_sysfs(Manager *m, const char *sysfs, bool add, DeviceFound found, bool now) {
Device *d, *l, *n;
assert(m);
assert(sysfs);
if (found == DEVICE_NOT_FOUND)
return 0;
2013-11-25 15:25:01 +01:00
l = hashmap_get(m->devices_by_sysfs, sysfs);
LIST_FOREACH_SAFE(same_sysfs, d, n, l)
device_update_found_one(d, add, found, now);
return 0;
}
static int device_update_found_by_name(Manager *m, const char *path, bool add, DeviceFound found, bool now) {
_cleanup_free_ char *e = NULL;
Unit *u;
int r;
2010-01-29 06:45:59 +01:00
assert(m);
assert(path);
2010-01-29 06:45:59 +01:00
if (found == DEVICE_NOT_FOUND)
return 0;
2010-01-29 06:45:59 +01:00
r = unit_name_from_path(path, ".device", &e);
if (r < 0)
return log_error_errno(r, "Failed to generate unit name from device path: %m");
2010-01-29 06:45:59 +01:00
u = manager_get_unit(m, e);
if (!u)
return 0;
device_update_found_one(DEVICE(u), add, found, now);
2010-01-29 06:45:59 +01:00
return 0;
}
2013-11-25 15:25:01 +01:00
static bool device_is_ready(struct udev_device *dev) {
const char *ready;
assert(dev);
ready = udev_device_get_property_value(dev, "SYSTEMD_READY");
if (!ready)
return true;
return parse_boolean(ready) != 0;
}
static Unit *device_following(Unit *u) {
Device *d = DEVICE(u);
Device *other, *first = NULL;
assert(d);
if (startswith(u->id, "sys-"))
return NULL;
/* Make everybody follow the unit that's named after the sysfs path */
for (other = d->same_sysfs_next; other; other = other->same_sysfs_next)
if (startswith(UNIT(other)->id, "sys-"))
return UNIT(other);
for (other = d->same_sysfs_prev; other; other = other->same_sysfs_prev) {
if (startswith(UNIT(other)->id, "sys-"))
return UNIT(other);
first = other;
}
return UNIT(first);
}
2013-11-25 15:25:01 +01:00
static int device_following_set(Unit *u, Set **_set) {
Device *d = DEVICE(u), *other;
Set *set;
int r;
assert(d);
2013-11-25 15:25:01 +01:00
assert(_set);
2013-11-25 15:25:01 +01:00
if (LIST_JUST_US(same_sysfs, d)) {
*_set = NULL;
return 0;
}
set = set_new(NULL);
2013-11-25 15:25:01 +01:00
if (!set)
return -ENOMEM;
2013-11-25 15:25:01 +01:00
LIST_FOREACH_AFTER(same_sysfs, other, d) {
r = set_put(set, other);
if (r < 0)
goto fail;
2013-11-25 15:25:01 +01:00
}
2013-11-25 15:25:01 +01:00
LIST_FOREACH_BEFORE(same_sysfs, other, d) {
r = set_put(set, other);
if (r < 0)
goto fail;
2013-11-25 15:25:01 +01:00
}
2013-11-25 15:25:01 +01:00
*_set = set;
return 1;
fail:
2013-11-25 15:25:01 +01:00
set_free(set);
return r;
}
static void device_shutdown(Manager *m) {
assert(m);
m->udev_event_source = sd_event_source_unref(m->udev_event_source);
2010-04-21 03:27:44 +02:00
if (m->udev_monitor) {
2010-01-29 06:45:59 +01:00
udev_monitor_unref(m->udev_monitor);
2010-04-21 03:27:44 +02:00
m->udev_monitor = NULL;
}
2010-01-29 06:45:59 +01:00
m->devices_by_sysfs = hashmap_free(m->devices_by_sysfs);
}
static void device_enumerate(Manager *m) {
_cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
struct udev_list_entry *item = NULL, *first = NULL;
int r;
assert(m);
if (!m->udev_monitor) {
m->udev_monitor = udev_monitor_new_from_netlink(m->udev, "udev");
if (!m->udev_monitor) {
log_oom();
2010-04-21 03:27:44 +02:00
goto fail;
}
2010-01-29 06:45:59 +01:00
/* This will fail if we are unprivileged, but that
* should not matter much, as user instances won't run
* during boot. */
2015-03-14 03:21:41 +01:00
(void) udev_monitor_set_receive_buffer_size(m->udev_monitor, 128*1024*1024);
r = udev_monitor_filter_add_match_tag(m->udev_monitor, "systemd");
if (r < 0) {
log_error_errno(r, "Failed to add udev tag match: %m");
goto fail;
}
r = udev_monitor_enable_receiving(m->udev_monitor);
if (r < 0) {
log_error_errno(r, "Failed to enable udev event reception: %m");
2010-04-21 03:27:44 +02:00
goto fail;
}
2010-01-29 06:45:59 +01:00
r = sd_event_add_io(m->event, &m->udev_event_source, udev_monitor_get_fd(m->udev_monitor), EPOLLIN, device_dispatch_io, m);
if (r < 0) {
log_error_errno(r, "Failed to watch udev file descriptor: %m");
goto fail;
}
2015-04-29 16:05:32 +02:00
(void) sd_event_source_set_description(m->udev_event_source, "device");
2010-04-21 03:27:44 +02:00
}
2010-01-29 06:45:59 +01:00
e = udev_enumerate_new(m->udev);
if (!e) {
log_oom();
goto fail;
}
r = udev_enumerate_add_match_tag(e, "systemd");
if (r < 0) {
log_error_errno(r, "Failed to create udev tag enumeration: %m");
goto fail;
}
r = udev_enumerate_add_match_is_initialized(e);
if (r < 0) {
log_error_errno(r, "Failed to install initialization match into enumeration: %m");
goto fail;
}
r = udev_enumerate_scan_devices(e);
if (r < 0) {
log_error_errno(r, "Failed to enumerate devices: %m");
goto fail;
}
first = udev_enumerate_get_list_entry(e);
udev_list_entry_foreach(item, first) {
_cleanup_udev_device_unref_ struct udev_device *dev = NULL;
const char *sysfs;
sysfs = udev_list_entry_get_name(item);
dev = udev_device_new_from_syspath(m->udev, sysfs);
if (!dev) {
log_oom();
continue;
}
if (!device_is_ready(dev))
continue;
(void) device_process_new(m, dev);
device_update_found_by_sysfs(m, sysfs, true, DEVICE_FOUND_UDEV, false);
}
return;
fail:
device_shutdown(m);
}
static int device_dispatch_io(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
_cleanup_udev_device_unref_ struct udev_device *dev = NULL;
Manager *m = userdata;
const char *action, *sysfs;
int r;
2010-01-29 06:45:59 +01:00
assert(m);
if (revents != EPOLLIN) {
static RATELIMIT_DEFINE(limit, 10*USEC_PER_SEC, 5);
if (!ratelimit_test(&limit))
log_error_errno(errno, "Failed to get udev event: %m");
if (!(revents & EPOLLIN))
return 0;
}
2010-01-29 06:45:59 +01:00
/*
* libudev might filter-out devices which pass the bloom
* filter, so getting NULL here is not necessarily an error.
*/
dev = udev_monitor_receive_device(m->udev_monitor);
if (!dev)
return 0;
2010-01-29 06:45:59 +01:00
sysfs = udev_device_get_syspath(dev);
if (!sysfs) {
log_error("Failed to get udev sys path.");
return 0;
}
action = udev_device_get_action(dev);
if (!action) {
2010-01-29 06:45:59 +01:00
log_error("Failed to get udev action string.");
return 0;
2010-01-29 06:45:59 +01:00
}
if (streq(action, "change")) {
_cleanup_free_ char *e = NULL;
Unit *u;
r = unit_name_from_path(sysfs, ".device", &e);
if (r < 0)
log_error_errno(r, "Failed to generate unit name from device path: %m");
else {
u = manager_get_unit(m, e);
if (u && UNIT_VTABLE(u)->active_state(u) == UNIT_ACTIVE) {
r = manager_propagate_reload(m, u, JOB_REPLACE, NULL);
if (r < 0)
log_error_errno(r, "Failed to propagate reload: %m");
}
}
}
/* A change event can signal that a device is becoming ready, in particular if
* the device is using the SYSTEMD_READY logic in udev
* so we need to reach the else block of the follwing if, even for change events */
if (streq(action, "remove")) {
r = swap_process_device_remove(m, dev);
if (r < 0)
log_error_errno(r, "Failed to process swap device remove event: %m");
/* If we get notified that a device was removed by
* udev, then it's completely gone, hence unset all
* found bits */
device_update_found_by_sysfs(m, sysfs, false, DEVICE_FOUND_UDEV|DEVICE_FOUND_MOUNT|DEVICE_FOUND_SWAP, true);
2013-11-25 15:25:01 +01:00
} else if (device_is_ready(dev)) {
(void) device_process_new(m, dev);
r = swap_process_device_new(m, dev);
if (r < 0)
log_error_errno(r, "Failed to process swap device new event: %m");
2013-11-25 15:25:01 +01:00
manager_dispatch_load_queue(m);
/* The device is found now, set the udev found bit */
device_update_found_by_sysfs(m, sysfs, true, DEVICE_FOUND_UDEV, true);
} else {
/* The device is nominally around, but not ready for
* us. Hence unset the udev bit, but leave the rest
* around. */
device_update_found_by_sysfs(m, sysfs, false, DEVICE_FOUND_UDEV, true);
2010-01-29 06:45:59 +01:00
}
return 0;
2010-01-29 06:45:59 +01:00
}
static bool device_supported(void) {
static int read_only = -1;
/* If /sys is read-only we don't support device units, and any
* attempts to start one should fail immediately. */
if (read_only < 0)
read_only = path_is_read_only_fs("/sys");
return read_only <= 0;
}
int device_found_node(Manager *m, const char *node, bool add, DeviceFound found, bool now) {
_cleanup_udev_device_unref_ struct udev_device *dev = NULL;
struct stat st;
assert(m);
assert(node);
if (!device_supported())
return 0;
/* This is called whenever we find a device referenced in
* /proc/swaps or /proc/self/mounts. Such a device might be
* mounted/enabled at a time where udev has not finished
* probing it yet, and we thus haven't learned about it
* yet. In this case we will set the device unit to
* "tentative" state. */
if (add) {
if (!path_startswith(node, "/dev"))
return 0;
/* We make an extra check here, if the device node
* actually exists. If it's missing, then this is an
* indication that device was unplugged but is still
* referenced in /proc/swaps or
* /proc/self/mountinfo. Note that this check doesn't
* really cover all cases where a device might be gone
* away, since drives that can have a medium inserted
* will still have a device node even when the medium
* is not there... */
if (stat(node, &st) >= 0) {
if (!S_ISBLK(st.st_mode) && !S_ISCHR(st.st_mode))
return 0;
dev = udev_device_new_from_devnum(m->udev, S_ISBLK(st.st_mode) ? 'b' : 'c', st.st_rdev);
if (!dev && errno != ENOENT)
return log_error_errno(errno, "Failed to get udev device from devnum %u:%u: %m", major(st.st_rdev), minor(st.st_rdev));
} else if (errno != ENOENT)
return log_error_errno(errno, "Failed to stat device node file %s: %m", node);
/* If the device is known in the kernel and newly
* appeared, then we'll create a device unit for it,
* under the name referenced in /proc/swaps or
* /proc/self/mountinfo. */
(void) device_setup_unit(m, dev, node, false);
}
/* Update the device unit's state, should it exist */
return device_update_found_by_name(m, node, add, found, now);
}
bool device_shall_be_bound_by(Unit *device, Unit *u) {
if (u->type != UNIT_MOUNT)
return false;
return DEVICE(device)->bind_mounts;
}
2010-01-26 21:39:06 +01:00
const UnitVTable device_vtable = {
.object_size = sizeof(Device),
.sections =
"Unit\0"
"Device\0"
"Install\0",
.gc_jobs = true,
.init = device_init,
.done = device_done,
.load = unit_load_fragment_and_dropin_optional,
2010-01-29 03:18:09 +01:00
.coldplug = device_coldplug,
.serialize = device_serialize,
.deserialize_item = device_deserialize_item,
.dump = device_dump,
2010-01-29 03:18:09 +01:00
.active_state = device_active_state,
.sub_state_to_string = device_sub_state_to_string,
.bus_vtable = bus_device_vtable,
2010-04-18 03:08:16 +02:00
.following = device_following,
.following_set = device_following_set,
2010-01-29 03:18:09 +01:00
.enumerate = device_enumerate,
.shutdown = device_shutdown,
.supported = device_supported,
.status_message_formats = {
.starting_stopping = {
[0] = "Expecting device %s...",
},
.finished_start_job = {
[JOB_DONE] = "Found device %s.",
[JOB_TIMEOUT] = "Timed out waiting for device %s.",
},
},
};