So here's something we should always keep in mind:
systemd-udevd actually does *two* things with BSD file locks on block
devices:
1. While it probes a device it takes a LOCK_SH lock. Thus everyone else
taking a LOCK_EX lock will temporarily block udev from probing
devices, which is good when making changes to it.
2. Whenever a device is closed after write (detected via inotify), udevd
will issue BLKRRPART (requesting the kernel to reread the partition
table). It does this while holding a LOCK_EX lock on the block
device. Thus anyone else taking LOCK_SH or LOCK_EX will temporarily
block udevd from issuing that ioctl. And that's quite relevant, since
the kernel will temporarily flush out all partitions while re-reading
the partition table and then create them anew. Thus it is smart to
take LOCK_SH when dissecting a block device to ensure that no
BLKRRPART is issued in the background, until we mounted the devices.
This revisits the mess around waiting for partition block devices in
the image dissection code. It implements a nice little trick:
Instead of waiting for the kernel to probe the partition table for us
and generate the block devices from it, we'll just do that ourselves.
How can we do it? Via the BLKPG_ADD_PARTITION ioctl, that the kernel has
supported for a while. This ioctl allows creating partition block
devices off "whole" block devices from userspace, without the partitions
necessarily being present in the partition table at all.
So, whenever we want a partition to be there, we'll just issue
BLKPG_ADD_PARTITION. This can either work, in which case we know the
partition is there, and can use it. Yay. Or it can fail with EBUSY,
which the kernel returns if a partition by the selected partition index
already exists (or if an existing partition overlaps with the new one).
But if that's the case, then that's also OK, because the partition will
already exist.
So, regardless if we win or the kernel wins, for us the outcome is the
same: the partition block device will exist after invoking the ioctl.
Yay.
Net effect: we are not dependent on asynchronous uevent messages to wait
for the devices. Instead we synchronously get what we need. This makes
us independent of the (apparently less than reliable) netlink transport,
and should almost always be quicker.
Hopefully addresses #17469 even on older kernels.
Fixes: #17469
In Semaphore CI, for some reason, /run/systemd/resolve is busy so the umount
fails at the end of the test run:
Verify link states with Unmanaged= settings, cold-plug. ... umount: /run/systemd/resolve: target is busy.14:57
ok14:57
ERROR14:57
======================================================================14:57
ERROR: tearDownModule (__main__)14:57
----------------------------------------------------------------------14:57
Traceback (most recent call last):14:57
File /tmp/autopkgtest-lxc.6islza9t/downtmp/build.A9b/src/test/networkd-test.py, line 94, in tearDownModule14:57
subprocess.check_call([umount, d])14:57
File /usr/lib/python3.9/subprocess.py, line 373, in check_call14:57
raise CalledProcessError(retcode, cmd)14:57
subprocess.CalledProcessError: Command '['umount', '/run/systemd/resolve']' returned non-zero exit status 32.14:57
----------------------------------------------------------------------14:58
Ran 35 tests in 138.868s14:58
FAILED (errors=1, skipped=2)
Use lazy umount to avoid erroring out.
When looking at debug logs, it's helpful to know what type of server
address has been added.
For that, introduce a string lookup table for the ServerType type.
This new server type can only be set at runtime through a D-Bus method
and is exposed for reading through a D-Bus property.
`CAP_NET_ADMIN` and a PolKit acknowledge is required for setting
runtime servers.
Entries submitted that way are used before system and link servers
are being looked at.
* machine: update to use new-style sd-bus macros
Replace old SD_BUS_METHOD_WITH_NAMES and SD_BUS_SIGNAL_WITH_NAMES macros to
the new SD_BUS_METHOD_WITH_ARGS and SD_BUS_SIGNAL_WITH_ARGS macros.
Meson test, mkosi test image and running machinectl after build returned
no error. But since I don't have any virtual machines or containers, I'm not
sure how to test the changes thoroughly.
%R is already used in service manager specifier expansion (cgroup root),
hence use a different char, that was so far not used.
Follow-up for: 6ceb0a4094
To make sure we don't miss any _exit() calls let's move the
coverage-related tweaks into a separate header file and include it
explicitly on the compiler command line using -include when a coverage
build is requested.
Follow-up to c6552ad381.
attach_empty() file takes a BSD file lock on the device, and we really
should release that before going to sleep. hence explicitly close the
block device before the sleep instead of relying on _cleanup_ to close
it after the sleep.
This doesn't really make any real difference, given the file should only
contain a single line. But it's conceptually more correct to just remove
the trailing newline/whitespace then the whole lines coming after that.
i.e. if the file actually contains more lines than one, this should
probably be considered an error.
This makes sure that building ia32 EFI binaries on x86_64 works.
We force gnu-efi support to ensure it's not skipped by accident
and provide the lib32 dir manually, because clang does not support
'--print-multi-os-directory', which is used to auto-detect it.
This allows building 32bit versions of efi binaries on x86_64 machines
and vice-versa by passing "-Defi-cflags=-m32" to meson, provided the
32bit gnu-efi and gcc-multilib are available.
It is expected that distros that want to provide both ia32 and x64
versions to use a second build dir to build the non-native version
by adding -m32 to efi-cflags and then running the sd-boot/sd-stub
ninja target directly.
The implementation of MountImageUnit()/systemctl mount-image was
changed to use a /proc/self/fd path as the source, but that causes
the dm-verity files autodiscovery to fail, as it looks for files
in the same directory as the image.
Use the original file path when setting up dm-verity.
That file only exported one function, and it fits nicely within the scope of
"parse helpers". Let's move it there to reduce the file count a bit.
No functional change.
Sonali Srivastava
Yu Watanabe
Lennart Poettering
Ben Efros
Luca Boccassi
Daniel Mack
Amarjargal
Frantisek Sumsal
Michal Sekletar
Jan Janssen
Zbigniew Jędrzejewski-Szmek