Description of problem:

Paths to all devices on the system is created at the start of a command (saved in "dev-cache"), but while the command is running, some of those device paths may go away.  lvm was not detecting and dropping invalid paths in many cases, which could lead to segfaults or other undefined behavior.  Also, many places in the lvm code assumed that "dev_name()" always returned an actual device path name, and would attempt to use that path.  In fact this function has always been meant for printing messages, and may return "unknown" instead of an actual device path.

An upstream user reported repeatable crashes due to these issues, and reports that the problems are resolved with these fixes:

https://listman.redhat.com/archives/linux-lvm/2022-February/026122.html
https://listman.redhat.com/archives/linux-lvm/2022-April/026151.html

lvm commits in main branch:
cc73d99886df devices: only close PVs on LVs when scan_lvs is enabled
7b1a857d5ac4 devices: use dev-cache aliases handling from label scan functions
4eb04c8c05e5 devices: fix dev_name assumptions
00c3069872ab devices: initial use of existing option
7e70041e324e devices: drop incorrect paths from aliases list
1126be8f8dbc devices: simplify dev_cache_get_by_devt

(cc73d99886df is rhel9-only)

I have not been able to reproduce any of these issues, but based on the code that crashed, and the known problems with that code, we can say that it would be triggered by path names on the system being removed (e.g. devices being removed) while lvm commands were running.


Version-Release number of selected component (if applicable):


How reproducible:


Steps to Reproduce:
1.
2.
3.

Actual results:


Expected results:


Additional info: