Re: [RFC] fstests: add fsstress + writeback + debugfs folio split test

[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]



On Thu, Apr 25, 2024 at 01:31:50PM +0800, Zorro Lang wrote:
> On Wed, Apr 24, 2024 at 03:46:48PM -0700, Luis Chamberlain wrote:
> > This is dangerous at the moment as its using
> > a debugfs API which requires two out of tree fixes [0] [1] which will
> > be posted shortly.
> I can't find "boot_params" in /sys/kernel/debug/. So we might need to
> fix this helper, and then call it in debugfs related test cases.
> 
> I'll send another patch to talk about that fix. For this case, it needs
> _require_debugfs and $DEBUGFS_MNT.

Sure I'll wait until you send patches to fix this.

> > +_begin_fstest long_rw stress soak smoketest dangerous_fuzzers
> 
> Just double check, is it a dangerous_fuzzers test, and not good to be in auto group?

Yes for the explanation quoted above from the commit log, today the
existing code requires fixes which I just posted. The sscanf bug is also
hard to reproduce but I manged to reproduce it so this simple fix is
still being discussed. So for now only those in the know of what they
are doing should run this test.

> > +blocksize=$(_get_file_block_size $SCRATCH_MNT)
> > +export XFS_DIO_MIN=$((blocksize * 2))
> 
> Oh, I even forgot we have this parameter in fsstress.c :)

Yes, but its not well documented for good reason too, its only for DIO.

> > +test -n "$SOAK_DURATION" && fsstress_args+=(--duration="$SOAK_DURATION")
> > +
> > +# Our general goal here is to race with ops which can stress folio addition,
> > +# removal, edits, and writeback.
> > +
> > +# zero frequencies for write ops to minimize writeback
> > +fsstress_args+=(-R)
> 
> But you set "fsstress_args=(-w -d $SCRATCH_MNT/test -n $nr_ops -p $nr_cpus)" above,
> so you zeros frequencies of non-write operations (-w) and then zeros frequencies of
> write operations (-R). Do you want to use "-z" directly, to zeros frequencies of
> all operations ?

You know, it took a whileto figure out the right values here but I've
done more experimentation and we can just go with the defaults, what we
also needed is to add compaction so to help reproduce the real crash I
was trying to aim for. Now its cooked up.

For now I'll just post a v2 RFC and leave the cleanup stuff you've
requested for a v3. Right now this is just useful for folks who want
to help debug truncation + writeback + compaction races we are
uncovering with min order.

  Luis




[Index of Archives]     [Linux Filesystems Development]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux