check_disk: hangs on stale mount points which should be ignored

[praffeck]

Since the update to Debian Bookworm, we're noticing hang-ups of check disk, notably when trying to stat a potentially stale file system mounted by autofs. We know, that these mount points might be stale, so they're ignored both by type (-X) and path (-i).

Unfortunately, the changes introduced by #1820 add calls to statvfs via get_fs_usage to np_set_best_match. When --all is given on the command line, np_set_best_match is called in process_arguments before the path_select_list is filtered with the paths given by -i (or the types given by -X, which happens much later).

Unfortunately, I don't have a solution also addressing the original issue #1819 besides explicitly checking the permissions of the best matches (and searching for a next best match if the permission check fails) after -i is handled. The current state of calling stat on explicitly ignored pathes, however, seems unwanted to me.

[RincewindsHat]

Hi @praffeck,
without looking at it, this sounds like a reasonable explanation. I can't promise anything, but I will try to have a look at it and try to think of something which would solve all of the involved concerns.

[gollux]

The current behavior is very impractical -- check_disk accesses all kinds of filesystems, even though -K or -i is used to exclude them. This leads to lock-ups, for example with inaccessible remote filesystems, and violates reasonable expectations of users.

I suspect that the fix for #1829 brought much worse problems than the original issue.