When we are about to derealize a controller on v1 cgroup, we first
attempt to delete the controller cgroup and migrate afterwards. This
doesn't work in practice because populated cgroup cannot be deleted.
Furthermore, we leave out slices from migration completely, so
(un)setting a control value on them won't realize their controller
cgroup.
Rework actual realization, unit_create_cgroup() becomes
unit_update_cgroup() and make sure that controller hierarchies are
reduced when given controller cgroup ceased to be needed.
Note that with this we introduce slight deviation between v1 and v2 code
-- when a descendant unit turns off a delegated controller, we attempt
to disable it in ancestor slices. On v2 this may fail (kernel enforced,
because of child cgroups using the controller), on v1 we'll migrate
whole subtree and trim the subhierachy. (Previously, we wouldn't take
away delegated controller, however, derealization was broken anyway.)
Fixes: #14149
The description didn't really explain how the distribution mechanism
works exactly and the relationship of leaf and slice units.
Update the documentation and also explicitly explain the expected
behaviour as it is created by the memory_recursiveprot cgroup2 mount
option.
When available, enable memory_recursiveprot. Realistically it always
makes sense to delegate MemoryLow= and MemoryMin= to all children of a
slice/unit.
The kernel option is not enabled by default as it might cause
regressions in some setups. However, it is the better default in
general, and it results in a more flexible and obvious behaviour.
The alternative to using this option would be for user's to also set
DefaultMemoryLow= on slices when assigning MemoryLow=. However, this
makes the effect of MemoryLow= on some children less obvious, as it
could result in a lower protection rather than increasing it.
From the kernel documentation:
memory_recursiveprot
Recursively apply memory.min and memory.low protection to
entire subtrees, without requiring explicit downward
propagation into leaf cgroups. This allows protecting entire
subtrees from one another, while retaining free competition
within those subtrees. This should have been the default
behavior but is a mount-option to avoid regressing setups
relying on the original semantics (e.g. specifying bogusly
high 'bypass' protection values at higher tree levels).
This was added in kernel commit 8a931f801340c (mm: memcontrol:
recursive memory.low protection), which became available in 5.7 and was
subsequently fixed in kernel 5.7.7 (mm: memcontrol: handle div0 crash
race condition in memory.low).
It is possible that we will be running with an upgraded libseccomp, in which
case libseccomp might know the syscall name, even if the number is not known at
the time when systemd is being compiled. The guard only serves to break such
upgrades, by requiring that we also recompile systemd.
For s390-specific syscalls, use a define to exclude them, so that that we don't
try to filter them on other arches.
(cherry picked from commit 6cf852e79eb0eced2f77653941f9c75c3bd79386)
The kernel finally has a proper API to determine the mnt_id of a file.
Let's use it.
This adds support for the STATX_MNT_ID field of statx(), added in
kernel 5.8.
Also, let's move the glue for this to src/shared/ so that we later can
reuse this in sysemd-firstboot.
Given that libpwquality is a more a leaf dependency, let's make it
runtime optional, so that downstream distros can downgrade their package
deps from Required to Recommended.
The cryptsetup context pins the loop device even after deactivation.
Let's explicitly release the context to make sure the subsequent
loopback device detaching works cleanly.
This how this works on Linux: when atomically creating a file we need to
fully populate it under a temporary name and then when we are fully
done, sync it and the directory it is contained in, before renaming it
to the final name.
This was needed before commit 16e4fd87c5 added a
mode that opens the log fds for every single log message. This mode is used in
execute.c since then making the explicit call to log_open unnecessary.
This basically reverts ea89a119cd.
glibc 2.26 lifted restrictions on search domains count or length to
unlimited. This has also been backported to 2.17 in some distributions (RHEL 7
and derivatives). Other softwares may have their own limits for search domains,
but we should not restrict what is written out any more.
https://sourceware.org/legacy-ml/libc-announce/2017/msg00001.html
This comes up occasionally with new users. The phrase "Logs begin ..." is
ambiguous because it can be taken to mean the logs being displayed or all logs
(the intended meaning). Let's rephrase this as "Journal begins ..." to make
this clearer.
This information was already available in the debug output, but I think it
is good to include it in the message in the table. This makes it easier to wrap
one's head around the allowlist/denylist filtering.
I am pretty sure /etc/hosts (i.e. an explicitly configured, local,
trusted database) should be useful for overriding the automatic
myhostname logic.
resolved's internal logic handles it that way and hence we should
suggest it in the NSS fallback line, too.
Let's also bring the factory file back into sync with what the docs say.
And update the prose a bit too, to actually match what we recommend.
Unprivileged test-fs-util fails on my system since /sys/dev/block is
inaccessible for unprivileged users, so let's skip encrypted path test if we
get EACCES or similar.
If we don't succeed on the first try it's because another process is
opening the same device. Do a microsleep for 2ms to increase the
chances it has completed the next time around the loop.
The symlink /dev/mapper/dm_name is created by udev after a mapper
device is set up. So libdevmapper/libcrypsetup might tell us that
a verity device exists, but the symlink we use as the source for
the mount operation might not be there yet.
Instead of falling back to a new unique device set up, wait for
the udev event matching on the expected devlink for at least 100ms
(after which the benefits of sharing a device in terms of setup
time start to disappear - on my production machines, opening a new
verity device seems to take between 150ms and 300ms)
Michal Koutný
Benjamin Berg
Zbigniew Jędrzejewski-Szmek
Lennart Poettering
Anita Zhang
Alyssa Ross
Nicolas Fella
Yu Watanabe
Franck Bui
Dimitri John Ledkov
Grant Mathews
Haochen Tong
Topi Miettinen
Michael Scherer
Luca Boccassi