Some drivers do not announce the diskseq change.
E.g. for md devices, the kernel increments the diskseq *after*
emitting a 'change' uevent when backing block devices are added to
a md device, and udevd does not receive no uevent which contains
the new diskseq.
No actual code changes, just splitting out of some dev_t handling
related calls from stat-util.[ch], they are quite a number already, and
deserve their own module now I think.
Also, try to settle on the name "devnum" as the name for the concept,
instead of "devno" or "dev" or "devid". "devnum" is the name exported in
udev APIs, hence probably best to stick to that. (this just renames a
few symbols to "devum", local variables are left untouched, to make the
patch not too invasive)
No actual code changes.
Currently if enumerating with debug logging you'll likely see something
like this:
sd-device: the syspath "/sys/class/devcoredump/disabled" is not a directory.
sd-device: the syspath "/sys/class/firmware/timeout" is not a directory.
sd-device: the syspath "/sys/class/zram-control/hot_remove" is not a directory.
sd-device: the syspath "/sys/class/zram-control/hot_add" is not a directory.
sd-device: the syspath "/sys/class/drm/version" is not a directory.
This is because these sysfs classes place regular files in these
directories, which we so far didn't expect.
Let's filter them early, and only bother with enumerated inodes if they
actually are dirs or symlinks, i.e. can be kobject dirs. Regular file
definitely never are kobject dirs...
Currently, for sysfs paths outside of /sys/devices/ we do better
checking if something is a suitable path: we check if it's actually a
directory, and if not return ENODEV.
Let's make the codepath for nodes *inside* of /sys/device/ similar:
let's also return ENODEV if the path supplied is not a directory.
Previously, we'd return ENOTDIR in that case, which is quite confusing I
think.
There are some file systems mounted below /sys/ that are not actually
sysfs, i.e. are not arranged in a sysfs/kobject style. Let's refuse
those early. (Example, /sys/fs/cgroup/ and similar.)
(Also, let's add an env var for this, so that it can be turned off for
test cases.)
chase_symlinks() can return its last O_PATH fd to us. Let's use that and
make the access() check a bit tighter by going via faccessat() on the
O_PATH fd.
This doesn't really change too much, but is nice in context of the next
commit, which uses the O_PATH fd in one other way.
Simplify generated sysfs paths, since we might get data passed that
includes extra // in the middle.
Also, let's not assume /sys/ prefix without verification.
GIT_VERSION is not available as a config.h variable, because it's rendered
into version.h during builds. Let's rework jinja2 rendering to also
parse version.h. No functional change, the new variable is so far unused.
I guess this will make partial rebuilds a bit slower, but it's useful
to be able to use the full version string.
This can be problematic especially when there's no more free disk
space. Consider the following:.
When disk space becomes sparse, writting to the system journal can lead to
error. In this case journald attempts to make room by rotating the journals,
which consists in archiving online journals and opening new ones.
However opening new files is likely to fail too and in this case
journal_file_open() leaves half initialized file around but in online
state. Then the error is propagated and journald switches into volatile mode.
Next time a new message is received by journald, it tries to open the
persistent system journal file to switch automatically back to persistent
mode.
When opening the system journal, journal_file_open(), called by
managed_journal_file_open_reliably(), finds the persistent system journal left
previously and assumes that it was uncleanly closed and considers it as
corrupted. The error is reported to managed_journal_file_open_reliably(), which
backs the file up and attempts to create a new system file, which fails and
leaves a corrupted system file again.
Since this is done for each message received by journald, /var/log/message can
be filled with backup files pretty quickly.
To prevent this, the patch makes sure to delete the newly created file in case
of error.
We usually open() device node obtained by sd_device_get_devname().
However, the device node corresponds to the sd-device object may be
already removed, and another device node with the same path may be
created, hence an unexpected device may be opened.
The sd_device_open() opens device node, and checks the devnum and
diskseq of opened devnum, to avoid the above possibility.
Prompted by https://github.com/systemd/systemd/issues/22906#issuecomment-1082736443.
Yu Watanabe
Lennart Poettering
Zbigniew Jędrzejewski-Szmek
Franck Bui