On Wed, 10 Apr 2024 08:28:06 -0700, Darrick J. Wong wrote:
> On Wed, Apr 10, 2024 at 05:07:20PM +1000, David Disseldorp wrote:
> > As of xfsprogs commit 6e0ed3d1 ("mkfs: stop allowing tiny filesystems")
> > attempts to create XFS filesystems sized under 300M fail.
> >
> > Lacking MIN_FSSIZE configuration, failure of _scratch_mkfs_sized() in
> > generic/027 doesn't halt the test, leading _scratch_mount() to mount any
> > previously created (full-size) filesystem. As a result create_file()
> > may loop near-endlessly awaiting ENOSPC.
> >
> > Use a 300M sized filesystem to support XFS' limit, and explicitly check
> > for _scratch_mkfs_sized() failure to avoid endless loops if the limit
> > changes again in future.
>
> Hmmm. Does your test environment export (as in, the bash keyword)
> TEST_DEV and TEST_DIR? mkfs.xfs is supposed to detect fstests and allow
> sub-300M filesystems for testing.
Ah, I missed that getenv() logic in xfs_mkfs.
> The detection breaks if TEST_{DEV,DIR} are set but not exported in the
> config file; perhaps fstests ought to be exporting them like it does for
> the <cough> other detection key.
Yep, in my case I have a configs/$hostname.config environment and
common.config doesn't explicitly export TEST_DEV. I'll rework this
fix.
