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.
It is useful for clients to be able to read the last CPU usage counter value of
a unit even if the unit is already terminated. Hence, before destroying a
cgroup's cgroup cache the last CPU usage counter and return it if the cgroup is
gone.
This adds two (privileged) bus calls Ref() and Unref() to the Unit interface.
The two calls may be used by clients to pin a unit into memory, so that various
runtime properties aren't flushed out by the automatic GC. This is necessary
to permit clients to race-freely acquire runtime results (such as process exit
status/code or accumulated CPU time) on successful service termination.
Ref() and Unref() are fully recursive, hence act like the usual reference
counting concept in C. Taking a reference is a privileged operation, as this
allows pinning units into memory which consumes resources.
Transient units may also gain a reference at the time of creation, via the new
AddRef property (that is only defined for transient units at the time of
creation).
This adds the boolean RemoveIPC= setting to service, socket, mount and swap
units (i.e. all unit types that may invoke processes). if turned on, and the
unit's user/group is not root, all IPC objects of the user/group are removed
when the service is shut down. The life-cycle of the IPC objects is hence bound
to the unit life-cycle.
This is particularly relevant for units with dynamic users, as it is essential
that no objects owned by the dynamic users survive the service exiting. In
fact, this patch adds code to imply RemoveIPC= if DynamicUser= is set.
In order to communicate the UID/GID of an executed process back to PID 1 this
adds a new "user lookup" socket pair, that is inherited into the forked
processes, and closed before the exec(). This is needed since we cannot do NSS
from PID 1 due to deadlock risks, However need to know the used UID/GID in
order to clean up IPC owned by it if the unit shuts down.
Unfortunately, due to the disagreements in the kernel development community,
CPU controller cgroup v2 support has not been merged and enabling it requires
applying two small out-of-tree kernel patches. The situation is explained in
the following documentation.
https://git.kernel.org/cgit/linux/kernel/git/tj/cgroup.git/tree/Documentation/cgroup-v2-cpu.txt?h=cgroup-v2-cpu
While it isn't clear what will happen with CPU controller cgroup v2 support,
there are critical features which are possible only on cgroup v2 such as
buffered write control making cgroup v2 essential for a lot of workloads. This
commit implements systemd CPU controller support on the unified hierarchy so
that users who choose to deploy CPU controller cgroup v2 support can easily
take advantage of it.
On the unified hierarchy, "cpu.weight" knob replaces "cpu.shares" and "cpu.max"
replaces "cpu.cfs_period_us" and "cpu.cfs_quota_us". [Startup]CPUWeight config
options are added with the usual compat translation. CPU quota settings remain
unchanged and apply to both legacy and unified hierarchies.
v2: - Error in man page corrected.
- CPU config application in cgroup_context_apply() refactored.
- CPU accounting now works on unified hierarchy.
This adds a new boolean setting DynamicUser= to service files. If set, a new
user will be allocated dynamically when the unit is started, and released when
it is stopped. The user ID is allocated from the range 61184..65519. The user
will not be added to /etc/passwd (but an NSS module to be added later should
make it show up in getent passwd).
For now, care should be taken that the service writes no files to disk, since
this might result in files owned by UIDs that might get assigned dynamically to
a different service later on. Later patches will tighten sandboxing in order to
ensure that this cannot happen, except for a few selected directories.
A simple way to test this is:
systemd-run -p DynamicUser=1 /bin/sleep 99999
Let's lot at LOG_NOTICE about any processes that we are going to
SIGKILL/SIGABRT because clean termination of them didn't work.
This turns the various boolean flag parameters to cg_kill(), cg_migrate() and
related calls into a single binary flags parameter, simply because the function
now gained even more parameters and the parameter listed shouldn't get too
long.
Logging for killing processes is done either when the kill signal is SIGABRT or
SIGKILL, or on explicit request if KILL_TERMINATE_AND_LOG instead of LOG_TERMINATE
is passed. This isn't used yet in this patch, but is made use of in a later
patch.
By default, each iteration of manager_dispatch_sigchld() results in a unit level
sigchld event being invoked. For scope units, this results in a scope_sigchld_event()
which can seemingly stall for workloads that have a large number of PIDs within the
scope. The stall exhibits itself as a SIG_0 being initiated for each u->pids entry
as a result of pid_is_unwaited().
v2:
This patch resolves this condition by only paying to cost of a sigchld in the underlying
scope unit once per sigchld iteration. A new "sigchldgen" member resides within the
Unit struct. The Manager is incremented via the sd event loop, accessed via
sd_event_get_iteration, and the Unit member is set to the same value as the manager each
time that a sigchld event is invoked. If the Manager iteration value and Unit member
match, the sigchld event is not invoked for that iteration.
Let's move the enforcement of the per-unit start limit from unit.c into the
type-specific files again. For unit types that know a concept of "result" codes
this allows us to hook up the start limit condition to it with an explicit
result code. Also, this makes sure that the state checks in clal like
service_start() may be done before the start limit is checked, as the start
limit really should be checked last, right before everything has been verified
to be in order.
The generic start limit logic is left in unit.c, but the invocation of it is
moved into the per-type files, in the various xyz_start() functions, so that
they may place the check at the right location.
Note that this change drops the enforcement entirely from device, slice, target
and scope units, since these unit types generally may not fail activation, or
may only be activated a single time. This is also documented now.
Note that restores the "start-limit-hit" result code that existed before
6bf0f408e4 already in the service code. However,
it's not introduced for all units that have a result code concept.
Fixes#3166.
unit_has_mask_realized() determines whether the specified unit has its cgroups
set up properly given the desired target_mask; however, on the unified
hierarchy, controllers need to be enabled explicitly for children and the mask
of enabled controllers can deviate from target_mask. Only considering
target_mask in unit_has_mask_realized() can lead to false positives and
skipping enabling the requested controllers.
This patch adds unit->cgroup_enabled_mask to track which controllers are
enabled and updates unit_has_mask_realized() to also consider enable_mask.
Signed-off-by: Tejun Heo <htejun@fb.com>
This adds a new GetProcesses() bus call to the Unit object which returns an
array consisting of all PIDs, their process names, as well as their full cgroup
paths. This is then used by "systemctl status" to show the per-unit process
tree.
This has the benefit that the client-side no longer needs to access the
cgroupfs directly to show the process tree of a unit. Instead, it now uses this
new API, which means it also works if -H or -M are used correctly, as the
information from the specific host is used, and not the one from the local
system.
Fixes: #2945
With this change the logic for placing transient unit files and drop-ins
generated via "systemctl set-property" is reworked.
The latter are now placed in the newly introduced "control" unit file
directory. The fomer are now placed in the "transient" unit file directory.
Note that the properties originally set when a transient unit was created will
be written to and stay in the transient unit file directory, while later
changes are done via drop-ins.
This is preparation for a later "systemctl revert" addition, where existing
drop-ins are flushed out, but the original transient definition is restored.
This moves the StartLimitBurst=, StartLimitInterval=, StartLimitAction=, RebootArgument= from the [Service] section
into the [Unit] section of unit files, and thus support it in all unit types, not just in services.
This way we can enforce the start limit much earlier, in particular before testing the unit conditions, so that
repeated start-up failure due to failed conditions is also considered for the start limit logic.
For compatibility the four options may also be configured in the [Service] section still, but we only document them in
their new section [Unit].
This also renamed the socket unit failure code "service-failed-permanent" into "service-start-limit-hit" to express
more clearly what it is about, after all it's only triggered through the start limit being hit.
Finally, the code in busname_trigger_notify() and socket_trigger_notify() is altered to become more alike.
Fixes: #2467
This adds a new timestamp field to the Unit struct, storing when the last low-level state change took place, and make
sure this is restored after a daemon reload. This new field is useful to allow restarting of per-state timers exactly
where they originally started.
If a mount unit is bound to a device, systemd tries to umount the
mount point, if it thinks the device has gone away.
Due to the uevent queue and inotify of /proc/self/mountinfo being two
different sources, systemd can never get the ordering reliably correct.
It can happen, that in the uevent queue ADD,REMOVE,ADD is queued
and an inotify of mountinfo (or libmount event) happend with the
device in question.
systemd cannot know, at which point of time the mount happend in the
ADD,REMOVE,ADD sequence.
The real ordering might have been ADD,REMOVE,ADD,mount
and systemd might think ADD,mount,REMOVE,ADD and would umount the
mountpoint.
A test script which triggered this behaviour is:
rm -f test-efi-disk.img
dd if=/dev/null of=test-efi-disk.img bs=1M seek=512 count=1
parted --script test-efi-disk.img \
"mklabel gpt" \
"mkpart ESP fat32 1MiB 511MiB" \
"set 1 boot on"
LOOP=$(losetup --show -f -P test-efi-disk.img)
udevadm settle
mkfs.vfat -F32 ${LOOP}p1
mkdir -p mnt
mount ${LOOP}p1 mnt
... <dostuffwith mnt>
Without the "udevadm settle" systemd unmounted mnt while the script was
operating on mnt.
Of course the question is, why there was a REMOVE in the first place,
but this is not part of this patch.
Lets introduce unit_is_pristine() that verifies whether a unit is
suitable to become a transient unit, by checking that it is no
referenced yet and has no data on disk assigned.
Snapshots were never useful or used for anything. Many systemd
developers that I spoke to at systemd.conf2015, didn't even know they
existed, so it is fairly safe to assume that this type can be deleted
without harm.
The fundamental problem with snapshots is that the state of the system
is dynamic, devices come and go, users log in and out, timers fire...
and restoring all units to some state from the past would "undo"
those changes, which isn't really possible.
Tested by creating a snapshot, running the new binary, and checking
that the transition did not cause errors, and the snapshot is gone,
and snapshots cannot be created anymore.
New systemctl says:
Unknown operation snapshot.
Old systemctl says:
Failed to create snapshot: Support for snapshots has been removed.
IgnoreOnSnaphost settings are warned about and ignored:
Support for option IgnoreOnSnapshot= has been removed and it is ignored
http://lists.freedesktop.org/archives/systemd-devel/2015-November/034872.html
We can't handle errors of thisc all sanely anyway, and we never actually
return any errors from the unit type that implements the call. Hence,
let's make this void, in order to simplify things.
We cannot handle enumeration failures in a sensible way, hence let's try
hard to continue without making such failures fatal, and log about it
with precise error messages.
Let's use strjoina() rather than strjoin() for construct dbus match
strings.
Also, while we are at it, fix parameter ordering, so that our functions
always put the object first, like it is customary for OO-like
programming.
When starting a transient service, allow setting stdin/stdout/stderr fds
for it, by passing them in via the bus.
This also simplifies some of the serialization code for units.
Add a new config directive called NetClass= to CGroup enabled units.
Allowed values are positive numbers for fix assignments and "auto" for
picking a free value automatically, for which we need to keep track of
dynamically assigned net class IDs of units. Introduce a hash table for
this, and also record the last ID that was given out, so the allocator
can start its search for the next 'hole' from there. This could
eventually be optimized with something like an irb.
The class IDs up to 65536 are considered reserved and won't be
assigned automatically by systemd. This barrier can be made a config
directive in the future.
Values set in unit files are stored in the CGroupContext of the
unit and considered read-only. The actually assigned number (which
may have been chosen dynamically) is stored in the unit itself and
is guaranteed to remain stable as long as the unit is active.
In the CGroup controller, set the configured CGroup net class to
net_cls.classid. Multiple unit may share the same net class ID,
and those which do are linked together.
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.
Let's add a way to get the type-specific D-Bus interface of a unit from
either its type or name to src/basic/unit-name.[ch]. That way we can
share it with the client side, where it is useful in tools like cgls or
machinectl.
Also ports over machinectl to make use of this.
Currently, PID1 installs an unfiltered NameOwnerChanged signal match, and
dispatches the signals itself. This does not scale, as right now, PID1
wakes up every time a bus client connects.
To fix this, install individual matches once they are requested by
unit_watch_bus_name(), and remove the watches again through their slot in
unit_unwatch_bus_name().
If the bus is not available during unit_watch_bus_name(), just store
name in the 'watch_bus' hashmap, and let bus_setup_api() do the installing
later.
For instantaneous jobs (e.g. starting of targets, sockets, slices, or
Type=simple services) the log shows the job completion
before starting:
systemd[1]: Created slice -.slice.
systemd[1]: Starting -.slice.
systemd[1]: Created slice System Slice.
systemd[1]: Starting System Slice.
systemd[1]: Listening on Journal Audit Socket.
systemd[1]: Starting Journal Audit Socket.
systemd[1]: Reached target Timers.
systemd[1]: Starting Timers.
...
The reason is that the job completes before the ->start() method returns
and only then does unit_start() print the "Starting ..." message.
The same thing happens when stopping units.
Rather than fixing the order of the messages, let's just not emit the
Starting/Stopping message at all when the job completes instantaneously.
The job completion message is sufficient in this case.
No distro ships that old systemd versions anyway, hence let's drop
support for live-upgrades for them. Offline updates are still supported.
And live-upgrades will only lose the job queue, hence basically still
work...
The call is only used by the mount and automount unit types, but that's
already enough to consider it generic unit functionality, hence move it
out of mount.c and into unit.c.
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
