]> git.ipfire.org Git - thirdparty/systemd.git/log
thirdparty/systemd.git
2 years agorepart: port to new "struct iovec" JSON + primitive helpers 29644/head
Lennart Poettering [Fri, 20 Oct 2023 12:16:01 +0000 (14:16 +0200)] 
repart: port to new "struct iovec" JSON + primitive helpers

2 years agojson: add helpers for dispatching/building JSON with base64 struct iovecs
Lennart Poettering [Thu, 19 Oct 2023 15:41:01 +0000 (17:41 +0200)] 
json: add helpers for dispatching/building JSON with base64 struct iovecs

2 years agoiovec-util: add some useful helpers for dealing with iovecs that refer to dynamic...
Lennart Poettering [Thu, 19 Oct 2023 15:04:04 +0000 (17:04 +0200)] 
iovec-util: add some useful helpers for dealing with iovecs that refer to dynamic memory

2 years agotpm2-util: add line break where appropriate
Lennart Poettering [Thu, 19 Oct 2023 15:55:09 +0000 (17:55 +0200)] 
tpm2-util: add line break where appropriate

2 years agotpm2-util: add a bunch of line breaks for an overly long bitmask
Lennart Poettering [Thu, 19 Oct 2023 15:45:48 +0000 (17:45 +0200)] 
tpm2-util: add a bunch of line breaks for an overly long bitmask

2 years agoMerge pull request #29627 from poettering/io-split
Luca Boccassi [Fri, 20 Oct 2023 11:19:22 +0000 (12:19 +0100)] 
Merge pull request #29627 from poettering/io-split

split up io-util.[ch], and some clean-ups

2 years agorm-rf: Make sure we rewinddir() before readdir()
Daan De Meyer [Thu, 19 Oct 2023 17:42:07 +0000 (19:42 +0200)] 
rm-rf: Make sure we rewinddir() before readdir()

We don't know what state the passed in file descriptor is in. Let's
make sure we rewind it before iterating over it.

Fixes #29606

2 years agocredentials: document that their path is stable for system services
Joerg Behrmann [Fri, 20 Oct 2023 09:35:04 +0000 (11:35 +0200)] 
credentials: document that their path is stable for system services

2 years agonetif-naming-scheme: disable NAMING_BRIDGE_MULTIFUNCTION_SLOT
Yu Watanabe [Wed, 23 Aug 2023 15:01:19 +0000 (00:01 +0900)] 
netif-naming-scheme: disable NAMING_BRIDGE_MULTIFUNCTION_SLOT

This effectively reverts 66425daf2c68793adf24a48a26d58add8662e83f.

The commit assumes that if the network interface has multifunctions,
then the function fields of the pci devices under the same PCI bridge
device are unique.
But it seems not, at least on some setups. See issue #28929.
Let's revert the change, and always refuse to set slot base naming if
a PCI bridge is detected.

Fixes #28929.

2 years agoMerge pull request #29618 from bonktree/fchmodat2
Luca Boccassi [Fri, 20 Oct 2023 09:49:46 +0000 (10:49 +0100)] 
Merge pull request #29618 from bonktree/fchmodat2

seccomp: get `fchmodat2(2)` covered by `RestrictSUIDSGID=`

2 years agoman: document the order in which we talk to DNS servers
Lennart Poettering [Wed, 18 Oct 2023 16:14:00 +0000 (18:14 +0200)] 
man: document the order in which we talk to DNS servers

2 years agoio-util: split out iovw_xyz into iovec-wrapper.h 29627/head
Lennart Poettering [Thu, 19 Oct 2023 14:24:20 +0000 (16:24 +0200)] 
io-util: split out iovw_xyz into iovec-wrapper.h

Let's split these APIs out, they are kinda their own thing.

2 years agoiovec-util: make IOVEC_INCREMENT a regular function too
Lennart Poettering [Thu, 19 Oct 2023 14:36:43 +0000 (16:36 +0200)] 
iovec-util: make IOVEC_INCREMENT a regular function too

Even more than with the previous commit, this is not a trivial function
and there's no reason to believe this will actually be inlined nor that
it would be beneficial.

2 years agoiovec-util: make IOVEC_MAKE_STRING() safer
Lennart Poettering [Thu, 19 Oct 2023 14:30:23 +0000 (16:30 +0200)] 
iovec-util: make IOVEC_MAKE_STRING() safer

Let's not accept arbitrary types, but only char* and const char*.

2 years agoiovec-util: make IOVEC_NULL const
Lennart Poettering [Thu, 19 Oct 2023 14:27:16 +0000 (16:27 +0200)] 
iovec-util: make IOVEC_NULL const

compound initialized structs can be placed im immutable memory if const,
hence there's benefit in making this one const too

2 years agoiovec-util: make IOVEC_TOTAL_SIZE() a regular function
Lennart Poettering [Thu, 19 Oct 2023 14:33:51 +0000 (16:33 +0200)] 
iovec-util: make IOVEC_TOTAL_SIZE() a regular function

The function isn't necessarily fast (it's O(n)), and there's no reason
to have it defined as inline function, since it's neither fast, nor
entirely trivial.

2 years agoio-util: split out "struct iovec" related calls into their own .c/.h files
Lennart Poettering [Thu, 19 Oct 2023 14:01:38 +0000 (16:01 +0200)] 
io-util: split out "struct iovec" related calls into their own .c/.h files

This is preparation for #28891, which adds a bunch more helpers around
"struct iovec", at which point this really deserves its own .c/.h file.

The idea is that we sooner or later can consider "struct iovec" as an
entirely generic mechanism to reference some binary blob, and is the
go-to type for this purpose whenever we need one.

2 years agoexecutor: return instead of assert on invalid command line arguments
Luca Boccassi [Thu, 19 Oct 2023 19:53:10 +0000 (20:53 +0100)] 
executor: return instead of assert on invalid command line arguments

Before the split, it made sense to assert, as checks were on setup.
But now these come from deserialization, and the fuzzer hits the
asserts, so simply return an error instead.

2 years agotest-recurse-dir: work around nftw() ignoring symlinks()
Zbigniew Jędrzejewski-Szmek [Thu, 19 Oct 2023 14:46:56 +0000 (16:46 +0200)] 
test-recurse-dir: work around nftw() ignoring symlinks()

We have a test where we compare the results from nftw() and our own
resurce_dit_at(). nftw() skips a dangling symlink when running under mkosi and
the test fails. I don't understand why nftw() does that, but in our code we
don't need to test and care about the details of nftw(), which we don't use,
outside of the one test, so let's just skip symlinks in the test.

Closes #29603.

2 years agoMerge pull request #29628 from mrc0mmand/systemd-executor-test
Luca Boccassi [Thu, 19 Oct 2023 18:39:39 +0000 (19:39 +0100)] 
Merge pull request #29628 from mrc0mmand/systemd-executor-test

test: slightly improve sd-executor's coverage

2 years agopidref: make PIDREF_NULL const
Lennart Poettering [Thu, 19 Oct 2023 15:09:41 +0000 (17:09 +0200)] 
pidref: make PIDREF_NULL const

That way compiler can put it in an immutable section

2 years agoMerge pull request #29629 from bluca/mount_tunnel_pidref
Luca Boccassi [Thu, 19 Oct 2023 17:59:22 +0000 (18:59 +0100)] 
Merge pull request #29629 from bluca/mount_tunnel_pidref

mount tunnel: use PidRef

2 years agoMerge pull request #29623 from YHNdnzj/core-followup
Luca Boccassi [Thu, 19 Oct 2023 17:04:40 +0000 (18:04 +0100)] 
Merge pull request #29623 from YHNdnzj/core-followup

Follow-ups for recent changes to core/

2 years agotest: skip Protect{Home,System}= tests with coverage builds
Frantisek Sumsal [Thu, 19 Oct 2023 13:13:07 +0000 (15:13 +0200)] 
test: skip Protect{Home,System}= tests with coverage builds

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.

2 years agopid1,vconsole-setup: lock /dev/console instead of the tty device
Zbigniew Jędrzejewski-Szmek [Tue, 17 Oct 2023 17:43:31 +0000 (19:43 +0200)] 
pid1,vconsole-setup: lock /dev/console instead of the tty device

As requested in https://github.com/systemd/systemd/pull/27867#pullrequestreview-1567161854.
/dev/console, /dev/tty0, and /dev/ttyN are "different" device nodes
that may point to a single underlying device. We want to use a single
lock so that we don't get a race if different writers are using a different
device path, so let's just always lock around /dev/console.
This effectively makes the locking less granular.

Fixup for a0043bfa51281c2374878e2a98cf2a3ee10fd92c.
Fixes https://github.com/systemd/systemd/issues/28721.
Maybe fixes https://github.com/systemd/systemd/issues/28778 and
https://github.com/systemd/systemd/issues/28634.

2 years agotest: unify checking for user xattrs support in cgroupfs
Frantisek Sumsal [Thu, 19 Oct 2023 10:28:37 +0000 (12:28 +0200)] 
test: unify checking for user xattrs support in cgroupfs

Also, run the coredump forwarding test only if user xattrs are
supported.

2 years agomeson: generate proper version tag when git fails on permission errors
Zbigniew Jędrzejewski-Szmek [Thu, 19 Oct 2023 10:55:59 +0000 (12:55 +0200)] 
meson: generate proper version tag when git fails on permission errors

When building with mkosi I would get the following:

    [1/477] Generating version.h with a custom command
    fatal: detected dubious ownership in repository at '/work/src'
    To add an exception for this directory, call:

            git config --global --add safe.directory /work/src

and then the tag would be generated as 'v254-'. This is obviously some problem
with the setup, but we should handle this gracefully. Let's fall back to 'v254'
instead.

In the case where we have a repo but no tags, use --dirty=^ too, as in the case
with tags.

I tested four cases:
- normal checkout
- checkout with .git removed
- checkout with .git chowned to root
- checkout wiht all tags removed

2 years agoUpdate TODO 29629/head
Luca Boccassi [Thu, 19 Oct 2023 15:00:21 +0000 (16:00 +0100)] 
Update TODO

2 years agomount tunnel: use PidRef
Luca Boccassi [Thu, 19 Oct 2023 15:00:00 +0000 (16:00 +0100)] 
mount tunnel: use PidRef

2 years agocore: properly cleanup ExecParameter's prefix array 29628/head
Frantisek Sumsal [Thu, 19 Oct 2023 14:47:46 +0000 (16:47 +0200)] 
core: properly cleanup ExecParameter's prefix array

2 years agotest: add coverage for #29610
Frantisek Sumsal [Thu, 19 Oct 2023 14:23:28 +0000 (16:23 +0200)] 
test: add coverage for #29610

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

2 years agotest: add a real-world state to the corpus
Frantisek Sumsal [Thu, 19 Oct 2023 14:12:45 +0000 (16:12 +0200)] 
test: add a real-world state to the corpus

To give the fuzzer a nice head start.

2 years agocore/service: check error first and log about errno 29623/head
Mike Yuan [Wed, 18 Oct 2023 15:29:00 +0000 (23:29 +0800)] 
core/service: check error first and log about errno

Follow-up for becdfcb9f1cb555c50dcfe51894cb0b155f7f01e

2 years agocgroup-util: use RET_GATHER more, return first error
Mike Yuan [Thu, 19 Oct 2023 12:23:54 +0000 (20:23 +0800)] 
cgroup-util: use RET_GATHER more, return first error

2 years agocore/exec-invoke: use correct exit status
Mike Yuan [Wed, 18 Oct 2023 13:22:19 +0000 (21:22 +0800)] 
core/exec-invoke: use correct exit status

These calls can fail not only due to OOM.

2 years agocore/namespace: merge if blocks
Mike Yuan [Wed, 18 Oct 2023 11:23:42 +0000 (19:23 +0800)] 
core/namespace: merge if blocks

2 years agoexecutor: fix double free of MountOptions
Luca Boccassi [Thu, 19 Oct 2023 11:24:19 +0000 (12:24 +0100)] 
executor: fix double free of MountOptions

This list is owned by ExecContext, which is cleaned up when sd-executor
fails, but it is also cleaned up when namespace setup exits, so we get
a double free.

Fixes https://github.com/systemd/systemd/issues/29610

Follow-up for bb5232b6a3

2 years agoMerge pull request #29617 from keszybz/efi-no-xmalloc0
Luca Boccassi [Thu, 19 Oct 2023 13:23:54 +0000 (14:23 +0100)] 
Merge pull request #29617 from keszybz/efi-no-xmalloc0

efi: drop duplicate initialization to 0

2 years agoseccomp: also check the mode parameter of `fchmodat2(2)` 29618/head
Arseny Maslennikov [Sun, 15 Oct 2023 08:00:00 +0000 (11:00 +0300)] 
seccomp: also check the mode parameter of `fchmodat2(2)`

If there is no libseccomp support, just ban the entire syscall instead
so wrappers will fall back to older, supported syscalls.
Also reflect all of this in `test-seccomp.c`.

2 years agoseccomp: include `fchmodat2` in `@file-system`
Arseny Maslennikov [Sun, 15 Oct 2023 08:00:00 +0000 (11:00 +0300)] 
seccomp: include `fchmodat2` in `@file-system`

2 years agocgroup-util: drop dead code block
Yu Watanabe [Thu, 19 Oct 2023 09:40:10 +0000 (18:40 +0900)] 
cgroup-util: drop dead code block

Follow-up for 4d1b2df199227ed4b934bbcb054364e92e93a1a6.

Fixes CID#1522888.

2 years agobasic/missing_syscall: generate defs for `fchmodat2(2)`
Arseny Maslennikov [Sun, 15 Oct 2023 08:00:00 +0000 (11:00 +0300)] 
basic/missing_syscall: generate defs for `fchmodat2(2)`

We will need this to set seccomp filters on this system call regardless
of libseccomp or kernel support.

2 years agoUpdate system call tables for Linux 6.6
Arseny Maslennikov [Sun, 15 Oct 2023 08:00:00 +0000 (11:00 +0300)] 
Update system call tables for Linux 6.6

We are doing this to obtain the definition of fchmodat2.

2 years agoseccomp: fix debug logging typo
Arseny Maslennikov [Sun, 15 Oct 2023 08:00:00 +0000 (11:00 +0300)] 
seccomp: fix debug logging typo

Fixes: da4dc9a67487 ("seccomp: rework how the S[UG]ID filter is installed")
2 years agoMerge pull request #29595 from YHNdnzj/systemctl-failed-system
Luca Boccassi [Thu, 19 Oct 2023 11:07:37 +0000 (12:07 +0100)] 
Merge pull request #29595 from YHNdnzj/systemctl-failed-system

systemctl: is-failed: check if system is degraded when no unit given

2 years agoefi: drop unused xmalloc0() 29617/head
Zbigniew Jędrzejewski-Szmek [Tue, 17 Oct 2023 08:56:45 +0000 (10:56 +0200)] 
efi: drop unused xmalloc0()

Keeping this separate to make it easy to revert if it's ever needed again.

2 years agoefi: do not memzero fields before initializing them
Zbigniew Jędrzejewski-Szmek [Tue, 17 Oct 2023 08:55:57 +0000 (10:55 +0200)] 
efi: do not memzero fields before initializing them

In all three cases we immediately overwrite the whole field anyway,
so the call to memzero is not needed.

2 years agovarlink: Add new varlink_collect method
Arthur Shau [Wed, 20 Sep 2023 01:52:39 +0000 (18:52 -0700)] 
varlink: Add new varlink_collect method

varlink_collect is meant to be used when the client is willing to wait for the reply from the varlink method, much like varlink_call.
However, unlike varlink_call, it allows the client to collect all "more" replies that may be sent by method before the "final" reply is enqueued.
It aggregates all of these replies into a json variant array that it returns to the client.

2 years agofuzz: unify logging setup
Frantisek Sumsal [Wed, 18 Oct 2023 21:11:13 +0000 (23:11 +0200)] 
fuzz: unify logging setup

Make sure we don't log anything when running in "fuzzing" mode. Also,
when at it, unify the setup logic into a helper, pretty similar to
the test_setup_logging() one.

Addresses:
  - https://github.com/systemd/systemd/pull/29558#pullrequestreview-1676060607
  - https://github.com/systemd/systemd/pull/29558#discussion_r1358940663

2 years agoMerge pull request #29611 from mrc0mmand/execute-serialize-fuzz
Luca Boccassi [Wed, 18 Oct 2023 22:33:00 +0000 (23:33 +0100)] 
Merge pull request #29611 from mrc0mmand/execute-serialize-fuzz

test: add a fuzzer for exec_{serialize,deserialize}_invocation()

2 years agocore: don't assert when serializing malformed state 29611/head
Frantisek Sumsal [Wed, 18 Oct 2023 17:57:06 +0000 (19:57 +0200)] 
core: don't assert when serializing malformed state

2 years agotest: add a fuzzer for exec_{serialize,deserialize}_invocation()
Frantisek Sumsal [Fri, 28 Jul 2023 08:33:08 +0000 (10:33 +0200)] 
test: add a fuzzer for exec_{serialize,deserialize}_invocation()

2 years agosystemctl: is-failed: check if system is degraded when no unit given 29595/head
Mike Yuan [Tue, 17 Oct 2023 12:47:54 +0000 (20:47 +0800)] 
systemctl: is-failed: check if system is degraded when no unit given

Closes #3335

2 years agosystemctl: minor modernization
Mike Yuan [Tue, 17 Oct 2023 12:35:31 +0000 (20:35 +0800)] 
systemctl: minor modernization

2 years agoMerge pull request #29594 from poettering/cgroup-rename-ret-params
Lennart Poettering [Wed, 18 Oct 2023 15:25:44 +0000 (17:25 +0200)] 
Merge pull request #29594 from poettering/cgroup-rename-ret-params

more pidfdification

2 years agotest: use Type=exec for the test unit
Frantisek Sumsal [Wed, 18 Oct 2023 12:11:33 +0000 (14:11 +0200)] 
test: use Type=exec for the test unit

Make sure everything we need is set up before continuing further with
the test. This should, hopefully, help with a race where we check
a dynamic user before it's created by NFTSet= stuff.

Before:
$ journalctl -o short-monotonic --grep '(test-nft|NFT|testsuite-75)' --no-hostname
...
[ 3657.929223] testsuite-75.sh[48]: + systemd-run -u test-nft.service -p DynamicUser=yes -p 'NFTSet=cgroup:inet:sd_test:c user:inet:sd_test:u group:inet:sd_test:g' sleep 10000
...
[ 3657.977372] systemd[1]: test-nft.service: Changed dead -> running
[ 3657.977388] systemd[1]: test-nft.service: Job 376 test-nft.service/start finished, result=done
[ 3657.977407] testsuite-75.sh[853]: Running as unit: test-nft.service; invocation ID: 8776af2ec7864a60a058cb5d403d1ca6
[ 3657.982437] testsuite-75.sh[856]:                 elements = { "system.slice/test-nft.service" }
[ 3657.984570] testsuite-75.sh[48]: + grep -qF test-nft.service /tmp/tmp.uqXKfyzcpJ
[ 3657.985400] testsuite-75.sh[859]: ++ getent passwd test-nft
[ 3657.986434] systemd[1]: varlink-12: New incoming message: {"method":"io.systemd.UserDatabase.GetUserRecord","parameters":{"userName":"test-nft","service":"io.systemd.DynamicUser"}}
[ 3657.986503] systemd[1]: varlink-12: Sending message: {"error":"io.systemd.UserDatabase.NoRecordFound","parameters":{}}
[ 3657.986643] systemd[1]: testsuite-75.service: Child 48 belongs to testsuite-75.service.
[ 3657.986660] systemd[1]: testsuite-75.service: Main process exited, code=exited, status=2/INVALIDARGUMENT
[ 3657.986681] systemd[1]: testsuite-75.service: Failed with result 'exit-code'.
...
[ 3657.989885] systemd[1]: test-nft.service: User lookup succeeded: uid=64568 gid=64568
[ 3657.989936] systemd[1]: Added NFT set: family inet, table sd_test, set u, ID 64568
[ 3657.989965] systemd[1]: Added NFT set: family inet, table sd_test, set g, ID 64568

After:
$ journalctl -o short-monotonic --grep '(test-nft|NFT|testsuite-75)' --no-hostname
[ 1523.874408] testsuite-75.sh[49]: + systemd-run -u test-nft.service --service-type=exec -p DynamicUser=yes -p 'NFTSet=cgroup:inet:sd_test:c user:inet:sd_test:u group:inet:sd_test:g' sleep 10000
...
[ 1523.924091] systemd[1]: Starting test-nft.service...
[ 1523.928891] systemd[1]: test-nft.service: User lookup succeeded: uid=64568 gid=64568
[ 1523.929102] systemd[1]: Added NFT set: family inet, table sd_test, set u, ID 64568
[ 1523.929220] systemd[1]: Added NFT set: family inet, table sd_test, set g, ID 64568
...
[ 1523.933263] systemd[1]: Started test-nft.service.
[ 1523.933626] testsuite-75.sh[849]: Running as unit: test-nft.service; invocation ID: d1a5c3eacbc647a7a5fcbd46c7b2f863
[ 1523.940810] testsuite-75.sh[852]:                 elements = { "system.slice/test-nft.service" }
[ 1523.942880] testsuite-75.sh[49]: + grep -qF test-nft.service /tmp/tmp.gWka2x3mQq
[ 1523.944031] testsuite-75.sh[855]: ++ getent passwd test-nft
[ 1523.945496] systemd[1]: varlink-18: New incoming message: {"method":"io.systemd.UserDatabase.GetUserRecord","parameters":{"userName":"test-nft","service":"io.systemd.DynamicUser"}}
[ 1523.951593] systemd[1]: varlink-18: Sending message: {"parameters":{"record":{"userName":"test-nft","uid":64568,"gid":64568,"realName":"Dynamic User",...

2 years agomanager: move idle_pipe allocation to manager.c and make it atomic
Lennart Poettering [Wed, 18 Oct 2023 11:37:45 +0000 (13:37 +0200)] 
manager: move idle_pipe allocation to manager.c and make it atomic

Let's make sure it either fails or suceeds, but never fails half-way
leaving a half-initialized array around.

2 years agoMerge pull request #29583 from poettering/deserialize-fd-many
Luca Boccassi [Wed, 18 Oct 2023 15:06:22 +0000 (16:06 +0100)] 
Merge pull request #29583 from poettering/deserialize-fd-many

add serialize_fd_many() + deserialize_fd_many() helpers

2 years agoupdate TODO 29594/head
Lennart Poettering [Tue, 17 Oct 2023 08:57:26 +0000 (10:57 +0200)] 
update TODO

2 years agotest-cgroup: make test case a bit more robust towards previous aborted runs
Lennart Poettering [Tue, 17 Oct 2023 12:31:11 +0000 (14:31 +0200)] 
test-cgroup: make test case a bit more robust towards previous aborted runs

2 years agokillall: port killing spree loop over to PidRef
Lennart Poettering [Tue, 17 Oct 2023 11:43:29 +0000 (13:43 +0200)] 
killall: port killing spree loop over to PidRef

2 years agotest: port tests over to new /proc/ enumeration API
Lennart Poettering [Tue, 17 Oct 2023 11:43:59 +0000 (13:43 +0200)] 
test: port tests over to new /proc/ enumeration API

2 years agoprocess-util: add API for enumerating processes in /proc/ and pinning them via PidRef
Lennart Poettering [Tue, 17 Oct 2023 11:41:08 +0000 (13:41 +0200)] 
process-util: add API for enumerating processes in /proc/ and pinning them via PidRef

2 years agoprocess-util: add pidref_is_unwaited() and make pid_is_unwaited() return errors
Lennart Poettering [Tue, 17 Oct 2023 10:32:00 +0000 (12:32 +0200)] 
process-util: add pidref_is_unwaited() and make pid_is_unwaited() return errors

2 years agoprocess-util: add pidref_is_my_child()
Lennart Poettering [Tue, 17 Oct 2023 10:20:16 +0000 (12:20 +0200)] 
process-util: add pidref_is_my_child()

2 years agoprocess-util: change pid_is_alive() to not eat up errors, and add pidref_is_alive()
Lennart Poettering [Tue, 17 Oct 2023 10:12:05 +0000 (12:12 +0200)] 
process-util: change pid_is_alive() to not eat up errors, and add pidref_is_alive()

Let's no eat up errors, but propagate unexpected ones.

2 years agoprocess-util: add pidref_get_uid() and rename get_process_uid() → pidref_get_uid()
Lennart Poettering [Tue, 17 Oct 2023 09:27:06 +0000 (11:27 +0200)] 
process-util: add pidref_get_uid() and rename get_process_uid() → pidref_get_uid()

2 years agocore: port unit/process kill logic to pidref
Lennart Poettering [Tue, 17 Oct 2023 08:34:00 +0000 (10:34 +0200)] 
core: port unit/process kill logic to pidref

2 years agopidref: add new pidref_is_self() helper
Lennart Poettering [Tue, 17 Oct 2023 08:17:49 +0000 (10:17 +0200)] 
pidref: add new pidref_is_self() helper

This simply checks if the specified PidRef refers to the process we are
running in.

(In case you wonder why this is not a static inline: to avoid cyclic
header inclusion problems between pidref.h + process-util.h)

2 years agoprocess-util: add pidref_get_comm() and rename get_process_comm() to pid_get_comm()
Lennart Poettering [Tue, 17 Oct 2023 08:16:31 +0000 (10:16 +0200)] 
process-util: add pidref_get_comm() and rename get_process_comm() to pid_get_comm()

2 years agodbus-unit: make sure GetProcesses() D-Bus call internally uses pidfd
Lennart Poettering [Mon, 16 Oct 2023 21:18:37 +0000 (23:18 +0200)] 
dbus-unit: make sure GetProcesses() D-Bus call internally uses pidfd

This way we can be sure to provide an atomic view of the process tree of
a unit.

2 years agoprocess-util: add pidref_is_kernel_thread()
Lennart Poettering [Mon, 16 Oct 2023 21:18:05 +0000 (23:18 +0200)] 
process-util: add pidref_is_kernel_thread()

2 years agoprocess-util: add pidref_get_cmdline()
Lennart Poettering [Mon, 16 Oct 2023 21:17:29 +0000 (23:17 +0200)] 
process-util: add pidref_get_cmdline()

2 years agocgroup-show: use size_t for array sizes
Lennart Poettering [Mon, 16 Oct 2023 21:13:29 +0000 (23:13 +0200)] 
cgroup-show: use size_t for array sizes

2 years agopidref: we never have to verify PID 1
Lennart Poettering [Mon, 16 Oct 2023 21:11:58 +0000 (23:11 +0200)] 
pidref: we never have to verify PID 1

The process exists as long as the kernel/userns exists at all, hence we
don't have to verify a pidfd to it.

2 years agocgroup-util: make cg_pidref_get_path() PidRef parameter const
Lennart Poettering [Wed, 18 Oct 2023 08:24:28 +0000 (10:24 +0200)] 
cgroup-util: make cg_pidref_get_path() PidRef parameter const

2 years agopidref: make signal sending calls take const PidRef
Lennart Poettering [Wed, 18 Oct 2023 12:38:07 +0000 (14:38 +0200)] 
pidref: make signal sending calls take const PidRef

2 years agofd-util: use close_many() where appropriate 29583/head
Lennart Poettering [Wed, 18 Oct 2023 12:03:47 +0000 (14:03 +0200)] 
fd-util: use close_many() where appropriate

2 years agofd-util: add new helper close_many_unset()
Lennart Poettering [Wed, 18 Oct 2023 12:03:36 +0000 (14:03 +0200)] 
fd-util: add new helper close_many_unset()

2 years agoexecute: use close_many_and_free() more
Lennart Poettering [Wed, 18 Oct 2023 10:12:30 +0000 (12:12 +0200)] 
execute: use close_many_and_free() more

2 years agoserialize: add serialize_fd_many() helper
Lennart Poettering [Mon, 16 Oct 2023 13:56:58 +0000 (15:56 +0200)] 
serialize: add serialize_fd_many() helper

2 years agoserialize: change order of deserialize_strv() parameters 29481/head
Lennart Poettering [Fri, 6 Oct 2023 15:56:30 +0000 (17:56 +0200)] 
serialize: change order of deserialize_strv() parameters

The other deserializers put value first, and return parameter second,
let's do so here too.

2 years agoserialize: modernize deserialize_{usec|dual_timestamp}() a bit
Lennart Poettering [Fri, 6 Oct 2023 15:53:13 +0000 (17:53 +0200)] 
serialize: modernize deserialize_{usec|dual_timestamp}() a bit

Rename the return parameters "ret", and use compound initialization. Add
an assert() on input.

2 years agoserialize: add new helper deserialize_fd()
Lennart Poettering [Fri, 6 Oct 2023 15:50:48 +0000 (17:50 +0200)] 
serialize: add new helper deserialize_fd()

Currently, when we deserialize an fd we do a lot of manual work. Add a
common helper that makes this more robust and uniform.

Note that this sometimes changes behaviour slightly, but in ways that
shouldn't really matter: if we fail to deserialize an fd correctly we'll
unset (i.e. set to -EBADF) the fd in the deserialized data structure.
Previously, we'd leave the old value in place.

This should not change effective result (as in either case we'll be in a
bad state afterwards, just once we mix old/invalidated state with new
state, while now we'll reset the state explicitly to invalidated state
on failure). In particular as deserialization starts from an empty
structure generally, hence the old value should be unset anyway.

Another slight change is that if we fail to deserialize some object half
way, and we already have taken out one fd from the serialized fdset
we'll now just close it instead of returning it to/leaving it in the
fdset. Given that such "orphaned" fds are blanket closed after
deserialization finishes this also shouldn't change behaviour IRL.

Also, the idle_pipe was previously incorrectly serialized: we'd
serialize invalidated fds, which would fail, but because parsing errors
on this were ignored on the deserializatin noone noticed. This is fixed.

2 years agopidref: make pidref_verify() parameter const
Lennart Poettering [Wed, 18 Oct 2023 08:26:11 +0000 (10:26 +0200)] 
pidref: make pidref_verify() parameter const

2 years agocgroup-util: rename all return parameters in cgroup-util to ret_xyz
Lennart Poettering [Mon, 16 Oct 2023 20:45:56 +0000 (22:45 +0200)] 
cgroup-util: rename all return parameters in cgroup-util to ret_xyz

2 years agotree-wide: fix typo
Yu Watanabe [Wed, 18 Oct 2023 03:40:25 +0000 (12:40 +0900)] 
tree-wide: fix typo

2 years agonetwork: include SSID in ipv6 stable prefix address generation
Ronan Pigott [Sat, 14 Oct 2023 03:22:49 +0000 (20:22 -0700)] 
network: include SSID in ipv6 stable prefix address generation

The SSID fills the role of the optional Network_ID input parameter
suggested by RFC7217. Including the SSID allows networkd to generate a
different pseudorandom address for different wireless networks, which
should help to obscure the host's identity when roaming between multiple
networks.

2 years agorepart: avoid use of uninitialized TPM2B_PUBLIC data
Daniel P. Berrangé [Tue, 17 Oct 2023 15:45:52 +0000 (16:45 +0100)] 
repart: avoid use of uninitialized TPM2B_PUBLIC data

The 'TPM2B public' struct is only initialized if the public key
is non-NULL, however, it is unconditionally passed to
tpm2_calculate_sealing_policy, resulting in use of uninitialized
data. If the uninitialized data is lucky enough to be all zeroes,
this results eventually results in an error message from
tpm2_calculate_name about an unsupported nameAlg field value.

Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2 years agoMerge pull request #29441 from evelikov/no-input-delay
Luca Boccassi [Tue, 17 Oct 2023 22:45:44 +0000 (23:45 +0100)] 
Merge pull request #29441 from evelikov/no-input-delay

sd-boot: add "menu-disabled" option to "timeout", disabling the 100ms input polling

2 years agocgroup: turn device cgroup controller "rwm" strings into proper flags
Lennart Poettering [Tue, 10 Oct 2023 14:01:16 +0000 (16:01 +0200)] 
cgroup: turn device cgroup controller "rwm" strings into proper flags

We generally prefer dealing with parsed data instead of original
strings, do so for the "rwm" strings too. We have to convert this to
flags for the primary backend implementation (BPF) anyway, hence we
can do this early to have simpler, shorter and more normalized code.

2 years agoMerge pull request #29586 from poettering/read-stripped-line
Luca Boccassi [Tue, 17 Oct 2023 18:54:14 +0000 (19:54 +0100)] 
Merge pull request #29586 from poettering/read-stripped-line

add read_stripped_line() as trivial read_line() + strstrip() combo

2 years agomeson: add build option for install path of main config files
Franck Bui [Mon, 21 Aug 2023 10:37:00 +0000 (12:37 +0200)] 
meson: add build option for install path of main config files

This allows distros to install configuration file templates in /usr/lib/systemd
for example.

Currently we install "empty" config files in /etc/systemd/. They serve two
purposes:

- The file contains commented-out values that show the default settings.
- It is easier to edit the right file if it is already there, the user doesn't
  have to type in the path correctly, and the basic file structure is already in
  place so it's easier to edit.

Things that have happened since this approach was put in place:

- We started supporting drop-ins for config files, and drop-ins are the
  recommended way to create local configuration overrides.
- We have systemd-analyze cat-config which takes care of iterating over
  all possible locations (/etc, /run, /usr, /usr/local) and figuring out
  the right file.
- Because of the first two points, systemd-analyze cat-config is much better,
  because it takes care of finding all the drop-ins and figuring out the
  precedence. Looking at files manually is still possible of course, but not
  very convenient.

The disadvantages of the current approach with "empty" files in /etc:

- We clutter up /etc so it's harder to see what the local configuration actually is.
- If a user edits the file, package updates will not override the file (e.g.
  systemd.rpm uses %config(noreplace). This means that the "documented defaults"
  will become stale over time, if the user ever edits the main config file.

Thus, I think that it's reasonable to:

- Install the main config file to /usr/lib so that it serves as reference for
  syntax and option names and default values and is properly updated on package
  upgrades.
- Recommend to users to always use drop-ins for configuration and
  systemd-analyze cat-config to view the documentation.

This setting makes this change opt-in.

Fixes #18420.

[zjs: add more text to the description]

2 years agoMerge pull request #29242 from fbuihuu/update-main-config-file-headers
Zbigniew Jędrzejewski-Szmek [Tue, 17 Oct 2023 16:56:09 +0000 (18:56 +0200)] 
Merge pull request #29242 from fbuihuu/update-main-config-file-headers

config files: update their header to reflect that they can be install…

2 years agosd-boot: add way to disable the 100ms delay when timeout=0 29441/head
Emil Velikov [Wed, 4 Oct 2023 11:55:52 +0000 (12:55 +0100)] 
sd-boot: add way to disable the 100ms delay when timeout=0

Currently we have a 100ms delay which allows for people to enter/show
the boot menu even when timeout is set to zero.

In a handful of cases, that may not be needed - both in terms of access
policy, as well as latency.

For example: the option to provide the boot menu may be hidden behind an
"expert only" UX in the OS, to avoid end users from accidentally
entering it.

In addition, the current 100ms input polling may cause unexpected
additional delays in the boot. Some example numbers from my SteamDeck:

 - boot counting/rename/flush doubles 300us -> 600us
 - seed/hash setup doubles 900us -> 1800us
 - kernel/image load gets ~40% slower 107ms -> 167ms

It's not entirely clear why the UEFI calls gets slower, nevertheless the
information in itself proves useful.

This commit introduces a new option "menu-disabled", which omits the
100ms delay. The option is documented throughout the manual pages as
well as the Boot Loader Specification.

v2:
 - use STR_IN_SET

v3:
 - drop erroneous whitespace

v4:
 - add a new LoaderFeature bit,
 - don't change ABI keep TIMEOUT_* tokens the same
 - move new token in the 64bit range, update API and storage for it
 - change inc/dec behaviour to TIMEOUT_MIN : TIMEOUT_MENU_FORCE
 - user cannot opt-in from sd-boot itself, add assert_not_reached()

v5:
 - s/Menu disablement control/Menu can be disabled/
 - rewrap comments to 109
 - use SYNTHETIC_ERRNO(EOPNOTSUPP)

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
2 years agobootctl, sd-boot: cross document the menu-hidden/force ABI
Emil Velikov [Thu, 12 Oct 2023 15:24:12 +0000 (16:24 +0100)] 
bootctl, sd-boot: cross document the menu-hidden/force ABI

Document the current state and how we got here.

v2:
 - rewrap command to 109 columns

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
2 years agodocs/BOOT_LOADER_INTERFACE: mention that menu-* options are strings
Emil Velikov [Wed, 4 Oct 2023 11:55:52 +0000 (12:55 +0100)] 
docs/BOOT_LOADER_INTERFACE: mention that menu-* options are strings

To be on the safe side, explicitly mention that apart from the numerical
entries we can allow string ones.

Implementation-wise, bootctl will use internal numerical values that
match sd-boot's ABI. The latter also accepts the string options.

Going forward we'd like to avoid adding more internal magic and be more
explicit.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
2 years agotest: don't restart journal-upload on an expected fail
Frantisek Sumsal [Tue, 17 Oct 2023 10:49:03 +0000 (12:49 +0200)] 
test: don't restart journal-upload on an expected fail

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.

2 years agotree-wide: port various parsers over to read_stripped_line() 29586/head
Lennart Poettering [Mon, 16 Oct 2023 16:01:00 +0000 (18:01 +0200)] 
tree-wide: port various parsers over to read_stripped_line()

2 years agofileio: add read_stripped_line() as trivial read_line() + strstrip() combo
Lennart Poettering [Mon, 16 Oct 2023 15:55:07 +0000 (17:55 +0200)] 
fileio: add read_stripped_line() as trivial read_line() + strstrip() combo