Filesystem: try umount first during stop-action, and avoid potential "Argument list too long" for force_unmount=safe

[lge]

See individual commit messages.

• fix stop: try umount first, and only if that fails try to find users of that file system
• in the "safe" mode of get_pids(), avoid a potential "Argument list too long" problem

[SatomiOSAWA]

Hi, @lge @oalbrigt

Impressive work! Thank you for your consideration. I understand that the issue with RHBZ:2207567 is serious. However, as Mr. Lars noted, if the dependencies are properly configured, it seems unlikely that unmounting first would cause any problems. I have come to recognize the importance of adding proper comments once again. I believe many users will appreciate this improvement.

[oalbrigt]

Thanks.

[lge]

if scanning /proc twice should become a problem, we could either use bash and its more advance redirections, or use fifos

procs=$(
    exec 2>/dev/null
    D=$(mktemp -d) # (preferably somwhere on a tmpfs, so set TMPDIR or add -p or an explicit template)
    mkfifo "$D/maps";
    (
    < "$D/maps" xargs -r grep -l " $dir/"  | cut -d/ -f3 | uniq &
    exec 3> "$D/maps";
    rm -rf "$D";
    find /proc -path '/proc/[0-9]*' \
        \( -name maps -fprint "/dev/fd/3" \
        -o -type l \( -lname "$dir/*" -o -lname "$dir" \) -print \
        \) | cut -d/ -f 3 | uniq
    ) | sort -u
)

:stuck_out_tongue_winking_eye:

[oalbrigt]

Feel free to make a patch when you got the time.

We should keep it non-bash specific, as the agent can also be used on *BSD.

[lge]

Feel free to make a patch when you got the time.

We should keep it non-bash specific, as the agent can also be used on *BSD.

https://github.com/ClusterLabs/resource-agents/pull/1978 (hope that is sufficiently *BSD compatible ...)