Re: [PATCH] generic/558: limit the number of spawned subprocesses

["Theodore Ts'o" <tytso@mit.edu>]

On Thu, Jul 13, 2023 at 02:40:51AM +0800, Zorro Lang wrote:
> The generic/558 was shared/006, it was written for specific fs (e.g. xfs), then
> shared with other similar localfs.
> 
> After we changed it to a generic test case, the `_scratch_mkfs_sized` helps to
> avoid running this case on nfs/cifs and any other fs which can't be mkfs sized.
> Originally we thought a fs has a small specific size generally has limited
> free inodes. It works for long time, but now bcachefs looks like an exception :)
> 
> I think we must limit the number of processes, then let each process create more
> files if it need more inodes, that helps to avoid the forkbomb problem, and helps
> this case to work with bcachefs and other fs have lots of free inodes in 1G space.
> 
> But we'd better to limit the number of free inodes too, we don't want to run this
> case too long time. If a fs shows too many free inodes, _notrun "The 1G $FSTYP has
> too many free inodes!".

Alternatively we could have some fs-specific code which uses a
different sized scratch file system depending on the file system.  So
for bcachefs, maybe it should only be, say, 1 MiB (or whatever the
smallest file system bcachefs can support).

This test is designed to test what happens when the file system is
filled 100% with inodes, to make sure the file system passes fsck at
100%, and then after deleting all of the inodes, the file system
should be self-consistent afterwards as well.  That's a pretty
straightforward test, and for a file system where inodes can be
anywhere and are variably sized, it's certainly a completely valid
thing to do.  And if it turns out that "too many free inodes", then
maybe the test should have some special case sizes for different file
systems.

	`				- Ted