We have three states:
- ENABLE_COREDUMP and systemd-coredump is installed,
- ENABLE_COREDUMP but systemd-coredump is not installed,
- !ENABLE_COREDUMP.
In the last case we would not do any coredumping-related setup in pid1, which
means that coredumps would go to to the working directory of the process, but
actually limits are set to 0. This is inherited by children of pid1.
As discussed extensively in https://github.com/systemd/systemd/pull/26607, this
default is bad: dumps are written to arbitrary directories and not cleaned up.
Nevertheless, the kernel cannot really fix it. It doesn't know where to write,
and it doesn't know when that place would become available. It is only the
userspace that can tell this to the kernel. So the only sensible change in the
kernel would be to default to '|/bin/false', i.e. do what we do now.
In the middle case, we disabled writing of coredumps via a pattern, but raise
the RLIMIT_CORE. We need to raise the limit because we can't raise it later
after processes have been forked off. This means we behave correctly, but allow
coredumping to be enabled at a later point without a reboot.
This patch makes the last case behave like the middle case. This means that
even if systemd is compiled with systemd-coredump, it still does the usual
setup. If users want to restore the kernel default, they need to provide two
drop-in files:
for sysctl.d, with 'kernel.core_pattern=core'
for systemd.conf, with 'DefaultLimitCORE=0'.
The general idea is that pid1 does the safe thing. A distro may want to use
something different than the systemd-coredump machinery, and then that would
could packaged together with the drop-ins to change the configuration.
Alternative-for: #26607
- allow to run without $PROJECT_BUILD_ROOT,
- drop unnecessary export for bootctl,
- enable -x option to show commands,
- use 'test ! -e' to check the nonexistence of files,
- show more debugging logs.
It's a bit nicer if we only write the sysctl core_pattern once the
coredump socket is established, since it's the backend for the handler.
Given the systemd-coredump.socket basically has no dependencies that run
before it this should not really make things slower or so, it just
removes the tiny window where core pattern is in effect that wants to
connect to the backend socket but cannot.
The status quo isn't terrible, and not too different in effect: either
way, until the socket unit is up we won't process coredumps. It's mostly
what kind of behaviour you get then: an error due to /bin/false being
invoked, or an error because systemd-coredump can't connect to its
socket. After this patch we'll exclusively see the former.
Let's use more _cleanup_ expressions. Various other modernizations. No
actual code changes, except for maybe a conversion to use heap memory
when generating an array of fds, instead of stack as before. Given that
fdstores are typically user controlled, that should be a wise idea.
When a service deactivates and is then automatically restarted via
Restart= we currently quickly transition through
SERVICE_DEAD/SERVICE_FAILED. Which is weird given it's not the
normal ("permanent") dead/failed state, but a transitory one we
immediately leave from again. We do this so that software that looks for
failures/successes can take notice, even if we restart as a consequence
of the deactivation.
Let's clean this up a bit: let's introduce two new states:
SERVICE_DEAD_BEFORE_AUTO_RESTART and SERVICE_FAILED_BEFORE_AUTO_RESTART
that are used for the transitory states. Both the SERVICE_DEAD and
SERVICE_DEAD_BEFORE_AUTO_RESTART will map to the high-level
UNIT_INACTIVE state though. (and similar for the respective failed
states). This means the high-level state machine won't change by this,
only the low-level one.
This clearly seperates the substates, which makes the state engine
cleaner, and allows clients to follow precisely whether we are in a
transitory dead/failed state, or a permanent one, by looking at the
service substate. Moreover it allows us to remove the 'n_keep_fd_store'
which so far we used to ensure the fdstore was not released during this
transitory dead/failed state but only during the permanent one. Since we
can now distinguish these states properly we can just use that.
This has been bugging me for a while. Let's clean this up.
Note that the unit restart logic is already nicely covered in the
testsiute, hence this adds no new tests for that.
And yes, this could be considered a compat break, but sofar we took the
liberty to make changes to the low-level state machine (i.e. SERVICE_xyz
states, sometimes called "substates") without considering this a bad
breakage – the high-level state machine (i.e. UNIT_xyz states) should
be considered API that cannot be changed.
Zbigniew Jędrzejewski-Szmek
Lennart Poettering
Xiaotian Wu
Daan De Meyer
Yu Watanabe
Mike Yuan