So far we didn't enable the cpu controller because of overhead of the
accounting. If I'm reading things correctly, delegation was enabled for a while
for the units with user and pam context set, i.e. for user@.service too.
a931ad47a8 added the explicit Delegate=yes|no
switch, but it was initially set to 'yes'.
acc8059129 disabled delegation for user@.service
with the justication that CPU accounting is expensive, but half a year later
a88c5b8ac4 changed DefaultCPUAccounting=yes for
kernels >=4.15 with the justification that CPU accounting is inexpensive there.
In my (very noncomprehensive) testing, I don't see a measurable overhead if the
cpu controller is enabled for user slices. I tried some repeated compilations,
and there is was no statistical difference, but the noise level was fairly
high. Maybe better benchmarking would reveal a difference.
The goal of this change is very simple: currently all of the user session,
including services like the display server and pipewire are under user@.service.
This means that when e.g. a compilation job is started in the session's
app.slice, the processes in session.slice compete for CPU and can be starved.
In particular, audio starts to stutter, etc. With CPU controller enabled,
I can start start 'ninja -C build -j40' in a tab and this doesn't have any
noticable effect on audio.
I don't think the particular values matter too much: the CPU controller is
work-convserving, and presumably the session slice would never need more than
e.g. one 1 full CPU, i.e. half or a quarter of available CPU resources on even
the smallest of today's machines. app.slice and session.slice are assigned
equal weights, background.slice is assigned a smaller fraction. CPUWeight=100
is the default, but I wrote it explicitly to make it easier for users to see
how the split is done. So effectively this should result in session.slice
getting as much power as it needs.
If if turns out that this does have a noticable overhead, we could make it
opt-in. But I think that the benefit to usability is important enough to enable
it by default. W/o something like this the session is not really usable with
background tasks.
The "Networking" section has a lonely single document listed right now,
even though the "Concepts" section has two more network related docs.
Move them over, let's end this loneliness.
On Ubuntu, cryptsetup does not link against libgcc_s.so.1 which leads to
the following test failure in TEST-70-TPM2:
systemd[1]: Starting testsuite-70.service...
systemd[329]: testsuite-70.service: Executing: /usr/lib/systemd/tests/testdata/units/testsuite-70.sh
testsuite-70.sh[329]: + export SYSTEMD_LOG_LEVEL=debug
testsuite-70.sh[329]: + SYSTEMD_LOG_LEVEL=debug
testsuite-70.sh[329]: + img=/var/tmp/test.img
testsuite-70.sh[329]: + dd if=/dev/zero of=/var/tmp/test.img bs=1024k count=20 status=none
testsuite-70.sh[329]: + echo -n passphrase
testsuite-70.sh[329]: + cryptsetup luksFormat -q --use-urandom /var/tmp/test.img /tmp/passphrase
testsuite-70.sh[333]: libgcc_s.so.1 must be installed for pthread_exit to work
testsuite-70.sh[329]: /usr/lib/systemd/tests/testdata/units/testsuite-70.sh:
line 12: 333 Aborted (core dumped) cryptsetup luksFormat -q --use-urandom $img /tmp/passphrase
To fix this, manually install libgcc_s.so.1 to the test image if running
on Debian-like systems.
gcc was warning that found_fs_uuid was used unitialized. The issue stemmed from
the call to open(), where gcc seemingly didn't know that errno must be negative.
When that is set, we can drop some unnecessary initializations without warnings.
Fixes a bug introduced by 846f1da465.
The commit 846f1da465 made systemd.unit=
filtered out from the command line. That causes debug-generator does not
work as expected on daemon-reexecute, and we cannot call `systemctl
daemon-reexecute` in our test suite running on nspawn.
Fixes issue reported in https://github.com/systemd/systemd/pull/23851#issuecomment-1170992052.
In function 'sd_id128_equal',
inlined from 'journal_file_verify' at ../src/libsystemd/sd-journal/journal-verify.c:1047:29:
../src/systemd/sd-id128.h:119:43: error: 'entry_boot_id.qwords[0]' may be used uninitialized [-Werror=maybe-uninitialized]
119 | return a.qwords[0] == b.qwords[0] && a.qwords[1] == b.qwords[1];
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../src/libsystemd/sd-journal/journal-verify.c: In function 'journal_file_verify':
../src/libsystemd/sd-journal/journal-verify.c:823:20: note: 'entry_boot_id.qwords[0]' was declared here
823 | sd_id128_t entry_boot_id;
| ^~~~~~~~~~~~~
cc1: all warnings being treated as errors
entry_boot_id is only used when entry_monotonic_set has been set, and that's
only done in one place where entry_boot_id is also initalized.
We already had it on the socket units, so it's possible that
systemd-journald.service would be stopped and then restarted when trafic hits
the sockets when something logs. Let's not try to stop it. It is supposed to
run until the end and be eventually killed in the final killing spree.
This might (or not) help with #23287.
Currently kde installs a fake utmp session to listen for this. This provides an
alternative mechanism as discussed in #23574.
Example with 'shutdown 6 -r' and shutdown -c':
PRIORITY=6
SYSLOG_FACILITY=4
SYSLOG_IDENTIFIER=systemd-logind
...
CODE_FILE=src/login/logind-utmp.c
CODE_LINE=90
CODE_FUNC=warn_wall
MESSAGE_ID=9e7066279dc8403da79ce4b1a69064b2
OPERATOR=root
MESSAGE=The system will reboot at Thu 2022-06-30 12:16:43 CEST!
ACTION=reboot
PRIORITY=5
SYSLOG_FACILITY=4
SYSLOG_IDENTIFIER=systemd-logind
...
OPERATOR=root
CODE_FILE=src/login/logind-dbus.c
CODE_LINE=2407
CODE_FUNC=method_cancel_scheduled_shutdown
MESSAGE=System shutdown has been cancelled
MESSAGE_ID=249f6fb9e6e2428c96f3f0875681ffa3
ACTION=reboot
We have RxBufferSize= and TxBufferSize= in .link files. Let's use the same
abbreviation here. OTOH, "inc" could be short for "increment" or "increase",
let's avoid that.
Same story as before: disabling a non-existent event source shouldn't
need to be guarded by an if. I retained the wrapper so that that we don't
have to say SD_EVENT_OFF in the many places where this is called.
This is a natural use case, and instead of defining a wrapper to do this
for us, let's just make this part of the API. Calling with NULL was not
allowed, so this is not a breaking change to the interface.
(After sd_event_source_is_enabled was originally added, we introduced
sd_event_source_disable_unref() and other similar functions which accept
NULL. So not accepting NULL here is likely to confuse people. Let's just
make the API usable with minimal fuss.)
Before, sd_netlink_message_read() expected to fill a buffer completely,
and would return -EIO if the attribute being read was shorter than the
buffer. This means that the function can be used to "peek" into attributes
(by specifying a short buffer to just read part of the attribute), but
cannot be used to read something into a union without knowing beforehand
which specific field in the union is being filled. That latter operation
seems more useful (messages are short, so we don't really need to do partial
reads), so let's allow reads that don't fill the output buffer completely.
In places the text was overly formal, e.g. "an 128-bit ID" was repeated, even
though it is clear from the context that we're talking about this type of ID.
OTOH, in other places the text was informal, e.g. "You can use …".
Also, "you may use f() to frob" → "f() frobs". The text without all the
flourishes is easier to read.
sd_id128_in_set_sentinel() was described only in passing when taking about
sd_id128_in_set(), now it gets is own brief paragraph.
The synopsis was missing.
We find this function useful in our code, so no reason not to export it.
I changed the order of last two words in the name to match the arguments.
(With "equal_string" I expected sd_id128_t first, string second, but in
actual use, the second argument is usually a long constant so it's nice
to keep this order of arguments.)
The usual: if we find that function useful, other users of the library
will too. In particular, the v-variants are necessary to build pass-thru
wrappers.
It had two symbols which were not actually exported because they were not
listed in libsystemd.sym. They were also entirely unused in our codebase.
I don't think it makes much sense to export just those two functions, and
it doesn't make to build a string processing library in systemd either.
History of the file shows that it was created in
faaa5728d9 'utf8: export utf8 validation functions as part of sd-bus'
and hasn't gone even one non-trivial change since then ;)
It was added originally in 65f568bbeb. The API is
has stabilized pretty much, and generally follows the usual style for
libsystemd. We've held it as a public-but-private library for almost 10 years,
let's export it.
sd_netlink_sendv() and sd_nfnl_nft_*() are excluded.
libsystemd.so seems to grow by 12k.
