man: describe setting of the clock by systemd and systemd-timesyncd
The setting of systemd clock is important and deserves an accurate description,
see for example:
https://discussion.fedoraproject.org/t/f38-to-f39-40-dnf-system-upgrade-can-fail-on-raspberry-pi/92403
https://bugzilla.redhat.com/show_bug.cgi?id=2242759
The meat of the description was in systemd-timesyncd.service(8), but
actually it's systemd that sets the clock. In particular, systemd-timesyncd
doesn't know anything about /usr/lib/clock-epoch, and since systemd sets
the clock to the epoch when initializing, systemd-timesyncd would only
get to advance the clock to the epoch under special circumstances.
Also, systemd-timesyncd is an optional component, so we can't even rely
on its man page being installed in all circumstances. The description needs
to be moved to systemd(1).
The description is updated to describe the changes that were made in
previous commits.
manager: add structured log message about clock bump
Requested in https://github.com/systemd/systemd/pull/33214#discussion_r1630251308.
Also, reword error messages a bit. When /usr/lib/clock-epoch was introduced,
"build time" stopped being acurate. Just say "epoch" instead.
The same message ID is used in the manager and timesyncd. The event is
essentially equivalent for the user, and it seems reasonable that to search for
both at the same time.
The catalog entry is dropped. It provided almost no additional information above
the message. When the same message ID is now applied to messages from PID1 and
timesyncd, and the clock can be both advanced and rewound, it becomes very hard
to make the catalog entry provide something useful, because catalog entries don't
allow conditionalization.
We would attempt to take the built-in epoch twice. Since
advance_tstamp() is only called from one place, we don't need to do that.
Also, just pass usec_t instead of a pointer to stat buf.
Don't say we set the clock to "recorded timestamp" if we just set it
to the built-in epoch. Also, consistently say "advance" to make it clear
that we'll not attempt to rewind the clock here.
If we're updating on a system with an invalid clock, and we're installing
a newer system version with a higher update, adjust the clock. This
way the invariant that the clock is always later than
max(compile time, timestamp file, other timestamp file) is maintained.
Also, adjust the wording of messages. When /usr/lib/clock-epoch was
introduced, "build time" stopped being acurate. Just say "epoch" instead.
manager: use max of: compile epoch, epoch file, timesyncd file
Previously systemd would not use /var/lib/systemd/timesync/clock. This means
that even if /var/ is mounted when systemd is started and the file is
available, we would potentially make one time jump and than another time jump.
From a user's POV, this doesn't seem useful at all.
Also, we would always let /usr/lib/clock-epoch take priority over the built-in
epoch. But there is no guarantee that this file is actually fresh. In
particular, a user may touch /usr/lib/clock-epoch to work around a broken clock
during installation (as recommended in [1]), and then this file will grow stale
over time.
So just load the three timestamps and use the highest one as the epoch.
Move two functions only used in pid1 from libshared to the binary
Anything that is part of src/shared becomes part of the shared library, which
is wasteful, because the library is linked into almost all binaries. The
library is also forms a separate link unit, so this prevents the function from
being inlined or such.
Also, move logging into the function. Now that it's not a library function,
there is no reason not to do that.
Section "Description" didn't actually say what systemd does. And we had a giant
"Concepts" section that actually described units types and other details about
them. So let's move the basic description of functionality to "Description" and
rename the following section to "Units".
The link to the Original Design Document is moved to "See Also", it is of
historical interest mostly at this point.
The only actual change is that when talking about API filesystems, /dev is also
mentioned. (I think /sys+/proc+/dev are the canonical set and should be always
listed on one breath.)
Luca Boccassi [Fri, 31 May 2024 15:40:12 +0000 (16:40 +0100)]
journald: enable persistent FD Store to fix logging during soft-reboot
A unit with StandardOutput=journal (the default) will get its stdout/stderr sockets
disconnected when journald stops, as the file descriptors on journald's side are
not preserved (it works on restart, as the FD Store keeps them open during restarts).
Set FileDescriptorStorePreserve=yes so that the journal FD's stay open during a soft
reboot, and applications don't get broken stdout/stderr.
Daan De Meyer [Sat, 1 Jun 2024 12:45:22 +0000 (14:45 +0200)]
mkosi: Allow clearing meson cache with WIPE=1
meson does not support changing compilation flags on the fly, when
doing so, the entire build directory has to be cleared explicitly, so
let's add a way to do that by setting WIPE=1.
Let's also allow developers to specify their own meson options via
$MESON_OPTIONS.
Daan De Meyer [Mon, 3 Jun 2024 09:17:48 +0000 (11:17 +0200)]
mkosi: Make system dependencies conditional on format
If the none format is specified, we don't use the extra trees at all
so let's make the dependencies on the subimages conditional on Format=
not being set to "none".
Trying to install with dnf5 correctly filters out grubby and sdubby
as they are conditional dependencies and shouldn't be installed. However,
dnf doesn't do the right thing and tries to install both grubby and sdubby,
and since they conflict this causes the build to fail.
Let's filter out sdubby and grubby explicitly to work around the bug in dnf
as it's unlikely to get fixed since all development effort is now focused on
dnf5.
Daan De Meyer [Thu, 16 May 2024 15:18:38 +0000 (17:18 +0200)]
mkosi: Sanitizer improvements
- Let's set the environment on the kernel command line so it applies
to initrd and main system.
- Let's add the necessary wrappers that are also added in test-functions.
Unlike test-functions we don't use gcc/clang to get the library path as
that requires installing gcc/clang in the initrd.
- Let's drop the hack to get journald writing to the console and have
it write to kmsg instead. We'll get the output either way.
- Stop removing libstdc++ and sanitizer libraries from Arch Linux
initrds and other images as it's required by the sanitizer libraries.
- Add a workaround for specifying extra meson options for opensuse
- Add a leak sanitizer suppression file as a workaround for a false
positive leak in verify_selinuxmnt() in libselinux. We do a soname match
because the stacktrace can't be properly symbolized on Debian.
Mike Yuan [Fri, 31 May 2024 04:41:31 +0000 (12:41 +0800)]
core/unit: refuse to spawn units under frozen cgroup
Currently, Unit.freezer_state is always initialized to
FREEZER_RUNNING. While realizing cgroups for frozen units
was disabled in 7923e9493c48694b32d1a6de7b9a996c0194bf17,
the commit only checked for freezer_state of the unit inself,
meaning that newly-loaded units might be started and the kernel
would hang pid1 when trying to spawn sd-executor into sub-cgroup.
This can be easily reproduced by the following:
projects / thirdparty / systemd.git / log
