On Wed, Aug 31, 2022 at 1:04 PM Zorro Lang <zlang@xxxxxxxxxx> wrote:
>
> On Wed, Aug 31, 2022 at 09:53:55AM +0800, Murphy Zhou wrote:
> > Oops.. Darrick left a workaround in the xfsprogs code for fstests. My
> > test setup missed TEST_DEV export somehow and the workaround was not
> > working.
> >
> > Nevermind for this patchset.. My bloody hours...
>
> Thanks for reminding me, I just checked that patch, and yes:
>
> + /*
> + * fstests has a large number of tests that create tiny filesystems to
> + * perform specific regression and resource depletion tests in a
> + * controlled environment. Avoid breaking fstests by allowing
> + * unsupported configurations if TEST_DIR, TEST_DEV, and QA_CHECK_FS
> + * are all set.
> + */
> + if (getenv("TEST_DIR") && getenv("TEST_DEV") && getenv("QA_CHECK_FS"))
> + return;
>
> So we need to set QA_CHECK_FS to use this workaround... that's a little tricky
> for xfsprogs, I never thought it would like to do this.
QA_CHECK_FS is already set in `check`. My setup was missing TEST_DEV somehow,
after adding export in local.config like the example, everything works fine.
>
> Your patchset is still helpful, I think it's still worth dealing with the minimal
> fs size situation, better to make it configurable, or can be detected automatically.
Unless it's a hard limit, I think it's not worth the effort. Like
Darrick said, backward
compatibility is also important. Random magic numbers may bring more test
coverage.
Thanks,
Murphy
> For example:
>
> # A workaround in xfsprogs can break the limitation of xfs minimal size
> if [ -n "$QA_CHECK_FS" ];then
> export XFS_MIN_SIZE=$((300 * 1024 * 1024))
> else
> export XFS_MIN_SIZE=$((16 * 1024 * 1024)) # or "unlimited"??
> fi
> ...
> init_min_fs_size()
> {
> if [ -n "$FS_MIN_SIZE" ];then
> return
> fi
>
> case $FSTYP in
> xfs)
> FS_MIN_SIZE=$XFS_MIN_SIZE
> ;;
> *)
> FS_MIN_SIZE="unlimited" # or a big enough size??
> ;;
> esac
> }
>
> But a configurable FS_MIN_SIZE might break some golden image. Hmm... need think
> about it more, any suggestions are welcome :)
>
> Thanks,
> Zorro
>
> >
> > On Wed, Aug 31, 2022 at 8:18 AM Murphy Zhou <jencce.kernel@xxxxxxxxx> wrote:
> > >
> > > On Wed, Aug 31, 2022 at 3:07 AM Zorro Lang <zlang@xxxxxxxxxx> wrote:
> > > >
> > > > On Tue, Aug 30, 2022 at 07:46:40AM -0700, Darrick J. Wong wrote:
> > > > > On Tue, Aug 30, 2022 at 03:36:34PM +0800, Zorro Lang wrote:
> > > > > > On Tue, Aug 30, 2022 at 12:44:30PM +0800, Murphy Zhou wrote:
> > > > > > > Since this xfsprogs commit:
> > > > > > > 6e0ed3d19c54 mkfs: stop allowing tiny filesystems
> > > > > > > XFS requires filesystem size bigger then 300m.
> > > > > >
> > > > > > I'm wondering if we can just use 300M, or 512M is better. CC linux-xfs to
> > > > > > get more discussion about how to deal with this change on mkfs.xfs.
> > > > > >
> > > > > > >
> > > > > > > Increase thoese numbers to 512M at least. There is no special
> > > > > > > reason for the magic number 512, just double it from original
> > > > > > > 256M and being reasonable small.
> > > > > >
> > > > > > Hmm... do we need a global parameter to define the minimal XFS size,
> > > > > > or even minimal local fs size? e.g. MIN_XFS_SIZE, or MIN_FS_SIZE ...
> > > > >
> > > > > I think it would be a convenient time to create a helper to capture
> > > > > that, seeing as the LTP developers recently let slip that they have such
> > > > > a thing somewhere, and min fs size logic is scattered around fstests.
> > > >
> > > > It's a little hard to find out all cases which use the minimal fs size.
> > > > But for xfs, I think we can do that with this chance. We can have:
> > > >
> > > > export XFS_MIN_SIZE=$((300 * 1024 * 1024))
> > > > export XFS_MIN_LOG_SIZE=$((64 * 1024 * 1024))
> > > >
> > > > at first, then init minimal $FSTYP size likes:
> > > >
> > > > init_min_fs_size()
> > > > {
> > > > case $FSTYP in
> > > > xfs)
> > > > FS_MIN_SIZE=$XFS_MIN_SIZE
> > > > ;;
> > > > *)
> > > > FS_MIN_SIZE="unlimited" # or a big enough size??
> > > > ;;
> > > > esac
> > > > }
> > > >
> > > > Then other fs can follow this to add their size limitation.
> > > > Any better ideas?
> > >
> > > In generic/042 f2fs has a similar kind of limitation.
> > >
> > > Let me check how LTP guys handle this.
> > >
> > > Thanks,
> > > Murphy
> > >
> > > >
> > > > Thanks,
> > > > Zorro
> > > >
> > > > >
> > > >
> > > snipped
> > > >
> >
>
