src/slot: fix mount point detection for symlinks #1722
Conversation
It may be desirable to specify the device of a slot not as the real device path, but as a symlink to it (e.g. by using a udev-provided symlink such as /dev/disk/by-id/... instead of /dev/mmcblk...). The slot mount point detection failed in these cases because the device passed to r_slot_find_by_device() is always the real path of the device. Fix this by checking the realpath() of each slot device in addition to the specified path. Signed-off-by: Andreas Mützel <andreas.muetzel@emlix.com>
|
Hmm. I would have thought that we already have something like this... We could resolve the path during config loading and store it in a new field of Resolving during config loading would rely on the assumption that the symlinks don't change at runtime, but I think that's reasonable. If |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #1722 +/- ##
==========================================
- Coverage 84.50% 84.49% -0.01%
==========================================
Files 76 76
Lines 22381 22387 +6
==========================================
+ Hits 18913 18917 +4
- Misses 3468 3470 +2
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
|
This is somewhat similar to #406, which was supposed to fix #388 but never got finished. The major difference seems to be that #406 also normalizes the path of the mounted device (not just of the mount point specified in the config). That might indeed lead to somewhat strange behaviour if a symlink called like a special filesystem exists in the current RAUC working directory. For example (not tested, just deduced from the related discussions):
This whole situation is avoided in this PR because only the mount point specified in the config is resolved/normalized, not the mount point returned by Resolving this at load time and storing it as an additional field ( Pro:
Con:
I think the last point is the most important point. Are such use-cases relevant? IMHO, this is mostly an architectural decision - if you prefer, I could adapt this MR to the alternative approach. |
In multiple use cases, it is desirable to use a symlink to a device in the
device=/dev/<device>attribute of a slot insystem.confinstead of the actual path of the device:One might want to identify the slots using attributes such as the disk id or path by using udev-provided symlinks such as
/dev/disk/by-id/...or/dev/disk/by-path.Or, on devices that support booting from multiple media, dynamically create a symlink such as
/dev/disk/rootfs-0that points to the appropriate partition on the currently-active boot device.This currently makes the mount point detection fail because RAUC cannot identify that such a device is actually mounted to a slot. (
rauc statusdoes not identify such a slot asmounted: /<mountpoint>.)The reason seems to be the way that
update_external_mount_points()matches mount points with slots:To get the currently-mounted devices,
g_unix_mounts_get()is used and returns the real paths of all mounted devices.Then,
find_config_slot_by_device()is used and callsr_slot_find_by_device()to match active mounts with config entries.This iterates over all slots and checks whether the
slot->deviceis equal to the givendevicename. In the symlink case, this will fail because the path of theslot->deviceis never resolved. This MR adds handling for the symlink case tor_slot_find_by_device()by usingrealpathto resolve the slot's device path.I could imagine two alternative approaches:
realpath()instead of only when the direct comparison fails. This would introduce a small additional cost in the regular case.Also, it would make the function not identify a slot as mounted if the
slot->devicedoes not point to an existing file that is listed as a mount point nonetheless - which is relevant at least in one of thebootchooserunit tests (but most likely not in real-world use-cases because non-existing devices should not be returned byg_unix_mounts_get()).