Re: [PATCH V2 2/2] fs/super.c: don't fool lockdep in freeze_super() and thaw_super() paths

From: Oleg Nesterov
Date: Wed Oct 05 2016 - 12:33:40 EST


On 10/05, Dave Chinner wrote:
>
> On Tue, Oct 04, 2016 at 06:58:27PM +0200, Oleg Nesterov wrote:
> > I removed this test and then the next run (after reboot) hangs at xfs/073 with
> > a lot of errors in dmesg like
> >
> > XFS (loop2): Failing async write on buffer block 0x9600790. Retrying async write.
> > blk_update_request: I/O error, dev loop2, sector 8389920
> > loop: Write error at byte offset 4295647232, length 4096.
>
> tests will dump lots of errors into dmesg. That doesn't mean there's
> a problem - many tests are designed to exercise error paths. That,
> however, looks like a loop device problem.
>
> xfs/073 is using loop devices internally itself, so this ends up
> with XFS on loop2 on XFS on loop1. That loop2 device is 100GB in
> size, and the test copies the xfstests source tree to a loopback
> filesystem image in $SCRATCH_DEV mounted on $TEST_DIR/$$.
>
> The issue is, most likely, that your TEST_DIR and SCRATCH_MNT are
> rooted in the xfstests source tree. Hence copying the xfstests
> source tree will also try to copy the 8GB image files into the
> filesystems that the image files contain. Which, clearly, will
> eventually result in ENOSPC errors when writing to the underlying
> loop device...
>
> Put your TEST_DIR and SCRATCHMNT mount points outside the xfstests
> directory, and this should go away.

Yes, thanks, xfs/073 no longer hangs.

Oleg.