When a late error occurs in sd-executor, the cleanup-on-close of the
context structs happen, but at that time all FDs might have already
been closed via close_all_fds(), so a double-close happens. This
can be seen when DynamicUser is enabled, with a non-existing
WorkingDirectory.
Invalidate the FDs in the context structs if close_all_fds succeeds.
Prompted by #29705
Note that x-systemd.wanted-by= and x-systemd.required-by= are not
dropped, since we ignore them because they are unnecessary rather
than unapplicable.
When a mount is gracefully skipped (e.g.: BindReadOnlyPaths=-/nonexistent)
we still post-process it, like making it read-only. Except if nothing
has been mounted, the mount point will be made read-only for no reason.
Track when mounts are skipped and avoid post-processing.
One day we'll switch all of this to the new mount api and do these
operations atomically or not at all.
Fixes https://github.com/systemd/systemd/issues/29725
Don't hardcode the event number, so the test works correctly even if
someone wrote to the event log before us. Also, explicitly pick the
sha256 bank when checking digests, as the indexing may vary depending on
current TPM's capabilities.
When looking at configuration, often a user wants to suppress the comments and
just look at the parts that actually configure something, roughly equivalent to
systemd-analyze cat-config … | rg -v '^(#|;|$)
This switch implements this natively, skipping lines that start with a comment
character or only contain whitespace.
For formats that have section headers, section headers are skipped, if only
followed by stuff that would be skipped. (The last section header is printed
when we're about to print some actual output.)
Note that the caller doesn't know if the format has headers or not. We do format
type detection in pretty-print.c. So the caller only specifies tldr=true|false, and
conf_files_cat() figures out if the format has headers and whether those should
be handled specially.
The comments that show the file name are always printed, even if all of the file
is suppressed.
This is a partial answer to the discussions in
https://github.com/systemd/systemd/pull/28919,
https://github.com/systemd/systemd/pull/29248. If the default config is shown in
config files, the user can conveniently use '--tldr' to show the relevant parts.
Automatically softreboot if the nextroot has been set up with an OS
tree, or automatically kexec if a kernel has been loaded with kexec
--load.
Add SYSTEMCTL_SKIP_AUTO_KEXEC and SYSTEMCTL_SKIP_AUTO_SOFT_REBOOT to
skip the automated switchover.
With coverage builds we disable Protect{Home,System}= via a service.d
dropin in /etc, which has, unfortunately, higher priority than our
transient systemd-run stuff. Let's just skip the affected tests in that
case instead of making the test setup even more complicated.
As reproducing it is actually pretty easy, with the benefit of hindsight:
~# systemd-run -P -p MountImages="/this/should/definitely/not/exist.img:/run/img2\:3:nosuid" false
Running as unit: run-u42.service
free(): double free detected in tcache 2
In c08bec1587 the journal-upload unit gained Restart=on-fail, which goes
against this one particular test that expects the unit to fail, making
the test flaky. Let's disable the automatic restarts just for this test
to make it stable once again.
Let's support the same filtering options that we also support in
udevadm trigger in udevadm info to filter the devices produced by
--export-db.
One difference is that all properties specified by --propery-match=
have to be satisfied in udevadm info unlike udevadm trigger where just
one of them has to be satisfied.
Instead of mounting over, do an atomic swap using mount beneath, if
available. This way assets can be mounted again and again (e.g.:
updates) without leaking mounts.
Currently we spawn services by forking a child process, doing a bunch
of work, and then exec'ing the service executable.
There are some advantages to this approach:
- quick: we immediately have access to all the enourmous amount of
state simply by virtue of sharing the memory with the parent
- easy to refactor and add features
- part of the same binary, will never be out of sync
There are however significant drawbacks:
- doing work after fork and before exec is against glibc's supported
case for several APIs we call
- copy-on-write trap: anytime any memory is touched in either parent
or child, a copy of that page will be triggered
- memory footprint of the child process will be memory footprint of
PID1, but using the cgroup memory limits of the unit
The last issue is especially problematic on resource constrained
systems where hard memory caps are enforced and swap is not allowed.
As soon as PID1 is under load, with no page out due to no swap, and a
service with a low MemoryMax= tries to start, hilarity ensues.
Add a new systemd-executor binary, that is able to receive all the
required state via memfd, deserialize it, prepare the appropriate
data structures and call exec_child.
Use posix_spawn which uses CLONE_VM + CLONE_VFORK, to ensure there is
no copy-on-write (same address space will be used, and parent process
will be frozen, until exec).
The sd-executor binary is pinned by FD on startup, so that we can
guarantee there will be no incompatibilities during upgrades.
This reworks the image discovery logic, and conceptually allows DDIs
that are both confext and sysext to exist. Previously we'd only extract
one type of exension data from a DDI, with this we allow to extract both
if both exist.
This doesn't add support for true "multi-modal" DDIs, that qualify as
various things at once, it just lays some ground work that ensures we at
least can dissect such images.
This reworks 484d26dac1 quite a bit.
This changes systemd-dissect's JSON output, but given the
version with the fields it changes/dops has never been released (as the
above patch was merged post-v254) this shouldn't be an issue.
In TEST-70-TPM2, test systemd-cryptenroll --tpm2-seal-key-handle using the
default (0) as well as the SRK handle (0x81000001), and test using a non-SRK
handle index after creating and persisting a primary key.
In test/test-tpm2, test tpm2_seal() and tpm2_unseal() using default (0), the SRK
handle, and a transient handle.
Otherwise they might leave stuff behind if they don't respond fast
enough to the first SIGTERM and get SIGKILLEd, which then breaks reusing
the unit name further in the test:
[ 2993.620849] H testsuite-82.sh[43]: + systemd-run -p Type=exec -p DefaultDependencies=no -p IgnoreOnIsolate=yes --unit=testsuite-82-nosurvive.service sleep infinity
[ 2993.628686] H systemd[1]: testsuite-82-nosurvive.service: About to execute: /usr/bin/sleep infinity
[ 2993.628886] H systemd[1]: testsuite-82-nosurvive.service: Forked /usr/bin/sleep as 65
[ 2993.629328] H systemd[1]: testsuite-82-nosurvive.service: Changed dead -> start
...
[ 2993.699892] H testsuite-82.sh[43]: + systemctl --no-block --check-inhibitors=yes soft-reboot
[ 2993.704326] H systemd-logind[41]: The system will soft-reboot now!
...
[ 3001.249302] H systemd[1]: Sending SIGKILL to PID 65 (sleep).
...
[ 3001.303158] H testsuite-82.sh[136]: + systemd-notify '--status=Second Boot'
...
[ 3001.409504] H testsuite-82.sh[136]: + systemd-run -p Type=exec --unit=testsuite-82-nosurvive.service sleep infinity
[ 3001.414061] H testsuite-82.sh[165]: Failed to start transient service unit: Unit testsuite-82-nosurvive.service was already loaded or has a fragment file.
Spotted in Ubuntu CI.
Peter Hutterer
Frantisek Sumsal
Yu Watanabe
Luca Boccassi
Mike Yuan
Zbigniew Jędrzejewski-Szmek
Daan De Meyer
Lennart Poettering
Nick Rosbrook
Michael Biebl
Dan Streetman