Units would fail to start, incl. systemd-journald.service and systemd-udevd.service.
Since unit->manager->restrict_fs will be set if and only if we can use it,
we can just check for that and remove the other checks.
Follow-up for 299d941723.
BPF LSM is cgroup unaware and it's set up is happening in core manager.
It occures that the current implementation is too restrictive and causes
pid 1 to freeze.
Instead:
* in bpf_lsm_setup set manager->restrict_fs pointer last,
so it is an indicator that the set up was successful
* check for manager->restrict_fs before applying unit options
The code assume that meson's cpu_family can be mapped directly to
'-D__<cpu_family>__'. This works in a surprising number of cases, but not for a
few architectures. PPC uses "powerpc", and RISC-V omits the trailing underscores.
ARM and RISC-V require a second define too.
Fixes#21900.
(I don't think this matters too much: we need *something* so that gnu/stubs.h
can be successfully included. But we don't actually call syscalls or depend too
much on the host environment, so things should be fine as long as we don't get
a compilation error.)
The code was written unidiomatically, using r as a boolean value, and
confusing errno and r in some places. AFAICS, there wasn't any actual
problem: even in the one place where errno was used instead of r, it would
almost certainly be initialized.
It seems that some libbpf functions set errno, while others return the
error, possibly encoded. Since there are almost no docs, the only way to
know is to read the code of the function. To make matters worse, there is
a global libbpf_mode which can be set to change the convention. With
LIBBPF_STRICT_DIRECT_ERRS in libbpf_mode, some functions set errno while others
return a negative error, and the only way to know is to read the code, except
that the split is now different. We currently don't set
LIBBPF_STRICT_DIRECT_ERRS, but even the possibility makes everything harder
to grok.
This is all very error-prone. Let's at least add some asserts to make sure that
the returned values are as expected.
This fixes a bug introduced by 6d93265955.
The commit makes several functions skipped if the manager is already
in finished state, as
> In manager_check_finished(), more steps are skipped if MANAGER_IS_FINISHED().
> Those steps are idempotent, but no need to waste cycles trying to do them
> more than once.
However, the idle pipe may be re-opened after manager is finished:
manager_dispatch_run_queue() -> manager_watch_idle_pipe().
So, the closing the pipe is not idempotent here.
Fixes#21889.
We have /usr/lib/systemd/libsystemd-{shared,core}-nnn.so. With this
path the 'nnn' part can be changed to something different. The idea
is that during a package build this will be set to the package version.
This way during in-place upgrades with the same major version both
the new and old libraries can cooexit. This should fix the issue
when systemd programs are called during package upgrades and fail
to exec because the expect different symbols in the library they
are linked to.
This should fix https://bugzilla.redhat.com/show_bug.cgi?id=1906010.
The scheme is very similar to libsystemd-shared.so: instead of building a
static library, we build a shared library from the same objects and link the
two users to it. Both systemd and systemd-analyze consist mostly of the fairly
big code in libcore, so we save a bit on the installation:
(-0g, no strip)
-rwxr-xr-x 5238864 Dec 14 12:52 /var/tmp/inst1/usr/lib/systemd/systemd
-rwxr-xr-x 5399600 Dec 14 12:52 /var/tmp/inst1/usr/bin/systemd-analyze
-rwxr-xr-x 244912 Dec 14 13:17 /var/tmp/inst2/usr/lib/systemd/systemd
-rwxr-xr-x 461224 Dec 14 13:17 /var/tmp/inst2/usr/bin/systemd-analyze
-rwxr-xr-x 5271568 Dec 14 13:17 /var/tmp/inst2/usr/lib/systemd/libsystemd-core-250.so
(-0g, strip)
-rwxr-xr-x 2522080 Dec 14 13:19 /var/tmp/inst1/usr/lib/systemd/systemd
-rwxr-xr-x 2604160 Dec 14 13:19 /var/tmp/inst1/usr/bin/systemd-analyze
-rwxr-xr-x 113304 Dec 14 13:19 /var/tmp/inst2/usr/lib/systemd/systemd
-rwxr-xr-x 207656 Dec 14 13:19 /var/tmp/inst2/usr/bin/systemd-analyze
-rwxr-xr-x 2648520 Dec 14 13:19 /var/tmp/inst2/usr/lib/systemd/libsystemd-core-250.so
So for systemd itself we grow a bit (2522080 → 2648520+113304=2761824), but
overall we save. The most is saved on all the test files that link to libcore,
if they are installed, because there's 15 of them:
$ du -s /var/tmp/inst?
220096 /var/tmp/inst1
122960 /var/tmp/inst2
I also considered making systemd-analyze a symlink to /usr/lib/systemd/systemd
and turning systemd into a multicall binary. We did something like this with
udevd and udevadm. But that solution doesn't fit well in this case.
systemd-analyze has a bunch of functionality that is not used in systemd,
so the systemd binary would need to grow quite a bit. And we're likely to
add new types of verification or introspection features in analyze, and this
baggage would only grow. In addition, there are the test binaries which also
benefit from this.
When conditions fail on a service unit, a path unit can cause
PID 1 to busy loop as it keeps trying to activate the service unit.
To avoid this from happening, add a trigger limit to the path unit,
identical to the trigger limit we have for socket units.
Initially, let's start with a high limit and not make it configurable.
If needed, we can add properties to configure the rate limit similar
to the ones we have for socket units.
The meson default for static_library() are:
build_by_default=true, install=false. We never interact with the
static libraries, and we only care about them as a stepping-stone towards
the installable executables or libraries. Thus let's only build them if
they are a dependency of something else we are building.
While at it, let's drop install:false, since this appears to be the default.
This change would have fixed the issue with lib_import_common failing
to build too: we wouldn't attempt to build it.
In practice this changes very little, because we generally only declare static
libraries where there's something in the default target that will make use of
them. But it seems to be a better pattern to set build_by_default to false.
The variable is not useful outside of the loop (it'll always be null
after the loop is finished), so we can declare it inline in the loop.
This saves one variable declaration and reduces the chances that somebody
tries to use the variable outside of the loop.
For consistency, 'de' is used everywhere for the var name.
When a service exits and triggers either an OnFailure= or OnSuccess=
dependency we now set a new environment variable for the ExecStart= and
ExecStartPre= process. This variable $MONITOR_METADATA exposes the
metadata relating to the service which triggered the dependency.
MONITOR_METADATA takes the following form:
MONITOR_METADATA="SERVICE_RESULT=<result-string0>,EXIT_CODE=<exit-code0>,EXIT_STATUS=<exit-status0>,INVOCATION_ID=<id>,UNIT=<triggering-unit0.service>;SERVICE_RESULT=<result-stringN>,EXIT_CODE=<exit-codeN>,=EXIT_STATUS=<exit-statusN>,INVOCATION_ID=<id>,UNIT=<triggering-unitN.service>"
MONITOR_METADATA is space separated set of metadata relating to the
service(s) which triggered the dependency. This is a list since if we
have 2 services which trigger the same dependency then the dependency
start job may be merged. In this case we need to pass both service
metadata to the triggered service. If there is no job merging then
MONITOR_METADATA will be a single entry.
For example, in the case we had a service "failer.service" which
triggers "failer-handler.service", the following variable is exported to
the ExecStart= and ExecStartPre= processes in failer-handler.service:
MONITOR_METADATA="SERVICE_RESULT=exit-code,EXIT_CODE=exited,EXIT_STATUS=1,INVOCATION_ID=67c657ed7b34466ea369abdf994c6393,UNIT=failer.service"
In another example where we have failer.service and failer2.service
which both also trigger failer-handler.service then the start job for
failer-handler.service may be merged and we might get the following:
MONITOR_METADATA="SERVICE_RESULT=exit-code,EXIT_CODE=exited,EXIT_STATUS=1,INVOCATION_ID=16a93ad196c94109990fb8b9aa5eef5f,UNIT=failer.service;SERVICE_RESULT=exit-code,EXIT_CODE=exited,EXIT_STATUS=1,INVOCATION_ID=ff70131e4cc145e994fb621de25a3e8f,UNIT=failer2.service"
Lower priority of RUN, so that TMPFS and especially the mount flags given with
`TemporaryFileSystem=` are used.
This allows making `/run` private with drop-ins such as:
```
[Service]
BindReadOnlyPaths=/run/systemd:/run/systemd:norbind
TemporaryFileSystem=/run:nodev,noexec,nosuid,rw,size=32k,nr_inodes=10,mode=0755
```
When an activation job is skipped because of a Condition*= setting failing,
currently the JobRemoved signal lists 'done' as the result, just as with
a successful job.
This is a problem when doing dbus activation: dbus-broker will receive a
signal that says the job was successful, so then it moves into a state where
it waits for the requested name to appear on the bus, but that never happens
because the job was actually skipped.
Add a new StartUnitWithFlags that changes the behaviour of the JobRemoved
signal to list 'done' or 'skipped'.
Fixes#21520
raise_level() takes the info condition as second argument and the notice
one as third. For the consumed CPU time these conditions are swapped.
Fixes: 37109b856a ("pid1: use LOG_DEBUG/INFO/NOTICE for unit resource consumption message")
Support for BPF might not have been checked, since it's not necessary
in test mode (eg: running offline analysis of units). This causes an
assert:
Assertion '(_error) != 0' failed at src/core/bpf-firewall.c:914, function emit_bpf_firewall_warning(). Aborting.
Export SYSTEMD_LOG_LEVEl=debug in TEST-65-ANALYZE is enough to trigger
this assert while doing an offline analysis of a unit that has some
firewall/network restrictions set.
Skip the warning if the manager is in test mode.
By default checks PSI on /proc/pressure, and causes a unit to be skipped
if the threshold is above the given configuration for the avg300
measurement.
Also allow to pass a custom timespan, and a particular slice unit to
check under.
Fixes#20139
Assumption in edc027b was that job we first skipped because of active
ratelimit is still in run_queue. Hence we trigger the queue and dispatch
it in the next iteration. Actually we remove jobs from run_queue in
job_run_and_invalidate() before we call unit_start(). Hence if we want
to attempt to run the job again in the future we need to add it back
to run_queue.
Fixes#21458
In case `/proc` is successfully mounted with pid tree subset only due to
`ProcSubset=pid`, the protective mounts for `ProtectKernelTunables=yes` and
`ProtectKernelLogs=yes` to non-pid `/proc` paths are failing because the paths
don't exist. But the pid only option may have failed gracefully (for example
because of ancient kernel), so let's try the mounts but it's not fatal if they
don't succeed.
This macro is like JSON_BUILD_STRING() but uses our json library's
ability to use literal strings directly as JsonVariant objects.
The changes all our codebase to use this new macro whenever we build
JSON objects from literal strings.
(I tried to make this automatic, i.e. to detect in JSON_BUILD_STRING()
whether something is a literal string nicely and thus do this stuff
automatically, but I couldn't find a way.)
This should reduce memory usage of our JSON code a bit. Constant strings
we use very often will now be shared and mapped directly from the ELF
image.
Zbigniew Jędrzejewski-Szmek
Albert Brox
Yu Watanabe
Julia Kartseva
Evgeny Vereshchagin
Luca Boccassi
Daan De Meyer
Peter Morrow
Topi Miettinen
James Hilliard
Jan Macku
Christian Göttsche
Frantisek Sumsal
yangmingtai
Michal Sekletar
Lennart Poettering