closes#37602, see there for extra motivation and considered
alternatives.
On typical systems, only few services need to create SUID/SGID files.
This often is limited to the user explicitly setting suid/sgid, the
`systemd-tmpfiles*` services, and the package manager. Allowing a
default to globally restrict creation of suid/sgid files makes it easier
to apply this restriction precisely.
## testing done
- built on aarch64-linux and x86_64-linux
- ran a VM test on x86_64-linux, checking for:
- VM system boots successfully
- defaults apply (both `yes`, `no`, and undefined)
- systemd tmpfiles can set suid/sgid on journal log path
- Other services explicitly defining `RestrictSUIDSGID=no` can create
suid files
The recently added test case TEST-07-PID1.subgroup-kill.sh surfaced a
race: if we enumerate PIDs in a cgroup, and the cgroup is unlinked at
the very same time reading will result in ENODEV. We need to handle that
gracefully. Hence let's do so.
Noticed while looking at:
https://github.com/systemd/systemd/actions/runs/16143084441/job/45554929264?pr=38120
closes#37602
On typical systems, only few services need to create SUID/SGID files.
This often is limited to the user explicitly setting suid/sgid, the
`systemd-tmpfiles*` services, and the package manager. Allowing a default
to globally restrict creation of suid/sgid files makes it easier to apply
this restriction precisely.
This is very similar to 327cd2d3db:
If emergency.target is started while initrd-cleanup.service/start is queued,
the initrd-cleanup job did not get canceled. In parallel to the emergency
units, it eventually runs the service, which in turn isolates and starts
initrd-switch-root.target. This stops the emergency units and effectively
starts the initrd boot process again, which likely fails again like the
initial attempt. The system is thus stuck in a loop, never really reaching
emergency.target.
This can be triggered if a service in between initrd-parse-etc.service
and initrd.target fails.
With this conflict added, starting emergency.target automatically cancels
initrd-cleanup.service/start, avoiding the loop.
Add a new option `PrivateBPF=` to mount a private instance of bpffs.
Add also four configuration options
`BPFDelegate{Commands,Maps,Programs,Attachments}=` which set the
corresponding bpffs mount options in order to create BPF tokens:
https://lwn.net/Articles/947173/
Closes#35108.
Add a new option PrivateBPF= to mount a new instance of bpffs within a
namespace.
PrivateBPF= can be set to "no" to use the host bpffs in readonly mode
and "yes" to do a new mount.
The mount is done with the new fsopen()/fsmount() API because in future
we'll hook some commands between the two calls.
I added them in 41afb5eb72 without too
much explanation. Most likely the idea was to get rid of unused code
in libsystemd.so [1]. But now that I'm testing this, it doesn't seem
to have an effect. LTO is needed to get rid of unused functions, and
it's enough to have LTO without those options. Those options might have
some downsides [2], so let's disable them since there are doubts and no
particularly good reason to have them.
But keep the -Wl,--gc-sections option. Without this, libsystemd.so
grows a little:
-rwxr-xr-x 1 zbyszek zbyszek 5532424 07-08 13:24 build/libsystemd.so.0.40.0-orig
-rwxr-xr-x 1 zbyszek zbyszek 5614472 07-08 13:26 build/libsystemd.so.0.40.0-no-sections
-rwxr-xr-x 1 zbyszek zbyszek 5532392 07-08 13:27 build/libsystemd.so.0.40.0
Let's apply the --gc-sections option always to make the debug and final
builds more similar.
We need to verify that distro packages don't unexpectedly grow after this.
[1] https://unix.stackexchange.com/a/715901
[2] https://stackoverflow.com/a/36033811
We put the name of the variable in the message, but it is a local variable
and the name does not have global meaning. We end up with pointless copies
of the error string:
$ strings build/libsystemd.so.0.40.0 | grep 'big enough'
xsprintf: p[] must be big enough
xsprintf: error[] must be big enough
xsprintf: prefix[] must be big enough
xsprintf: pty[] must be big enough
xsprintf: mode[] must be big enough
xsprintf: t[] must be big enough
xsprintf: s[] must be big enough
xsprintf: spid[] must be big enough
xsprintf: header_priority[] must be big enough
xsprintf: header_pid[] must be big enough
xsprintf: path[] must be big enough
xsprintf: buf[] must be big enough
The error message already shows the file, line, and function name, which
is enough to identify the problem:
Assertion 'xsprintf: buffer too small' failed at src/test/test-string-util.c:20, function test_xsprintf(). Aborting.
Suggested in
https://github.com/systemd/systemd/pull/35892#discussion_r2180322856.
This is a tiny amount of code and does not warrant having a separate file
and spawning a separate instance of the compiler during the build.
Note: it took me a while to confirm that the contents of that table and
function don't end up in libsystemd.so. The issue is that they _are_ present in
it, unless LTO is used. We actually use link_whole[libbasic_static] for
libsystemd, so we end up with all that code there. LTO is needed to clean
that up.
As is usually the case, the bitfields don't create the expected space savings,
because the field that follows needs to be aligned. But we don't want to fully
drop the bitfields here, because then ConditionType and ConditionResult are
each 4 bytes, and the whole struct grows from 32 to 40 bytes (on amd64). We
potentially have lots of little Conditions and that'd waste some memory.
Make each of the four fields one byte. This still allows the compiler to
generate simpler code without changing the struct size:
E.g. in condition_test:
c->result = CONDITION_ERROR;
- 78fab: 48 8b 45 e8 mov -0x18(%rbp),%rax
- 78faf: 0f b6 50 01 movzbl 0x1(%rax),%edx
- 78fb3: 83 e2 03 and $0x3,%edx
- 78fb6: 83 ca 0c or $0xc,%edx
- 78fb9: 88 50 01 mov %dl,0x1(%rax)
+ 78f8b: 48 8b 45 e8 mov -0x18(%rbp),%rax
+ 78f8f: c6 40 03 03 movb $0x3,0x3(%rax)
Christian made this possible in Linux 6.15 with a new system call
open_tree_attr() that combines open_tree() and mount_setattr(). Because
idmapped mounts are (rightfully) not nested, we have to do some extra
shenanigans to make source we're putting the right source uid in the
userns for any idmapped mounts that we do in nspawn.
Of course we also add the necessary boilerplate to make open_tree_attr()
available in our code and wrap open_tree_attr() and the corresponding
fallback in a new function which we then use everywhere else.
The EXIT_STATUS is supposed to encapuslate an ANSI C process exit
status, which is 8bit unsigned. Hence parse it as such, do not accept
negative values, or values > 255.
All other status output lines use tabs, use that for the ID shift line
too. otherwise output will appear unaligned if log viewers have fixed
tab stop positions.
Lennart Poettering
Antonio Alvarez Feijoo
Yu Watanabe
Matteo Croce
Grimmauld
Fabian Vogt
Mike Yuan
Zbigniew Jędrzejewski-Szmek
