I've run into an odd bug when testing the clone / reflink functionality,
and managed to isolate a testcase not relying on clones. The issue is
that after a lot of direct I/O hitting ENOSPC on a NFS mount I run
into a silly rename that never gets cleaned up after cleaning up the
directory.
Note that so far I've only managed to reproduce it against a Linux
server running XFS (ext4 and btrfs seem fine) despite looking like
a client bug, and although once in a while a single run of the testcase
cause the bug I usually need a few iterations.
All over it seems like some odd dentry refcount race where we see
a spurious reference for one reason or another.
Below is a patch to xfstests to add this reproducer:
diff --git a/tests/generic/700 b/tests/generic/700
new file mode 100755
index 0000000..33505b0
--- /dev/null
+++ b/tests/generic/700
@@ -0,0 +1,130 @@
+#! /bin/bash
+# FS QA Test No. 700
+#
+# Dumbed down Copy of the original version of generic/187 that reproduced
+# NFS silly rename issues.
+#-----------------------------------------------------------------------
+# Copyright (c) 2015, Oracle and/or its affiliates. All Rights Reserved.
+#
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License as
+# published by the Free Software Foundation.
+#
+# This program is distributed in the hope that it would be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write the Free Software Foundation,
+# Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+#-----------------------------------------------------------------------
+
+seq=`basename $0`
+seqres=$RESULT_DIR/$seq
+echo "QA output created by $seq"
+
+here=`pwd`
+tmp=/tmp/$$
+status=1 # failure is the default!
+trap "_cleanup; exit \$status" 0 1 2 3 15
+
+_cleanup()
+{
+ cd /
+ rm -rf $tmp.* $testdir
+}
+
+# get standard environment, filters and checks
+. ./common/rc
+. ./common/filter
+
+# real QA test starts here
+_supported_os Linux
+_require_xfs_io_command "falloc"
+_require_xfs_io_command "fpunch"
+test $FSTYP = "btrfs" && _notrun "Can't fragment free space on btrfs."
+_require_odirect
+
+rm -f $seqres.full
+
+_fragment_freesp()
+{
+ file=$1
+
+ # consume nearly all available space (leave ~1MB)
+ avail=`_get_available_space $SCRATCH_MNT`
+ echo "$avail bytes left"
+ filesize=$((avail - 1048576))
+ $XFS_IO_PROG -fc "truncate $filesize" $file
+
+ chunks=20
+ chunksizemb=$((filesize / chunks / 1048576))
+ seq 1 $chunks | while read f; do
+ echo "$((f * chunksizemb)) file size $f / 20"
+ $XFS_IO_PROG -fc "falloc -k $(( (f - 1) * chunksizemb))m ${chunksizemb}m" $file
+ done
+
+ chunks=100
+ chunksizemb=$((filesize / chunks / 1048576))
+ seq 80 $chunks | while read f; do
+ echo "$((f * chunksizemb)) file size $f / $chunks"
+ $XFS_IO_PROG -fc "falloc -k $(( (f - 1) * chunksizemb))m ${chunksizemb}m" $file
+ done
+
+ filesizemb=$((filesize / 1048576))
+ $XFS_IO_PROG -fc "falloc -k 0 ${filesizemb}m" $file
+
+ # Try again anyway
+ avail=`_get_available_space $SCRATCH_MNT`
+ $XFS_IO_PROG -fc "pwrite -S 0x65 0 $avail" ${file}.${i}
+
+ # Punch out whatever we need
+ seq 1 $((nr * 4)) | while read f; do
+ $XFS_IO_PROG -f -c "fpunch $((f * 2 * blksz)) $blksz" $file
+ done
+}
+
+echo "Format and mount"
+_scratch_mkfs > $seqres.full 2>&1
+_scratch_mount >> $seqres.full 2>&1
+
+testdir=$SCRATCH_MNT/test-$seq
+mkdir $testdir
+
+echo "Create the original files"
+blksz=65536
+nr=1024
+filesize=$((blksz * nr))
+_pwrite_byte 0x61 0 $filesize $testdir/file1 >> $seqres.full
+_pwrite_byte 0x62 0 $filesize $testdir/file2 >> $seqres.full
+seq 0 2 $((nr-1)) | while read f; do
+# _reflink_range $testdir/file1 $((blksz * f)) $testdir/file3 $((blksz * f)) $blksz >> $seqres.full
+ touch $testdir/file3
+ _pwrite_byte 0x61 $((blksz * f)) $blksz $testdir/file3.chk >> $seqres.full
+done
+seq 1 2 $((nr-1)) | while read f; do
+ touch $testdir/file3
+# _reflink_range $testdir/file2 $((blksz * f)) $testdir/file3 $((blksz * f)) $blksz >> $seqres.full
+ _pwrite_byte 0x62 $((blksz * f)) $blksz $testdir/file3.chk >> $seqres.full
+done
+_scratch_cycle_mount
+_fragment_freesp $testdir/bigfile >> $seqres.full 2>&1
+filesize=$((blksz * nr))
+_scratch_cycle_mount
+
+echo "Overwrite"
+cowoff=$((filesize / 4))
+cowsz=$((filesize / 2))
+$XFS_IO_PROG -d -f -c "pwrite -S 0x63 -b $cowsz $cowoff $cowsz" $testdir/file3 >> $seqres.full
+_pwrite_byte 0x63 $cowoff $cowsz $testdir/file3.chk >> $seqres.full
+
+# mount / unmount
+_scratch_cycle_mount
+
+echo "Deleting"
+rm -r $testdir
+
+# success, all done
+status=0
+exit
diff --git a/tests/generic/700.out b/tests/generic/700.out
new file mode 100644
index 0000000..9f7ce0f
--- /dev/null
+++ b/tests/generic/700.out
@@ -0,0 +1,5 @@
+QA output created by 700
+Format and mount
+Create the original files
+Overwrite
+Deleting
diff --git a/tests/generic/group b/tests/generic/group
index 2bd40a1..46a796a 100644
--- a/tests/generic/group
+++ b/tests/generic/group
@@ -354,6 +354,7 @@
349 blockdev quick rw
350 blockdev quick rw
351 blockdev quick rw
+700 auto nfs
923 auto quick clone
924 auto quick clone
925 auto quick clone
On Thu, Jun 16, 2016 at 08:01:47AM -0700, Christoph Hellwig wrote:
> I've run into an odd bug when testing the clone / reflink functionality,
> and managed to isolate a testcase not relying on clones. The issue is
> that after a lot of direct I/O hitting ENOSPC on a NFS mount I run
> into a silly rename that never gets cleaned up after cleaning up the
> directory.
I have a redhat bug for what sounds like the same behavior. I spent
some time looking at it but I don't think I got anywere.
https://bugzilla.redhat.com/show_bug.cgi?id=1132610
(But you probably can't see that right now; I'll see if I can open it.)
--b.
>
> Note that so far I've only managed to reproduce it against a Linux
> server running XFS (ext4 and btrfs seem fine) despite looking like
> a client bug, and although once in a while a single run of the testcase
> cause the bug I usually need a few iterations.
>
> All over it seems like some odd dentry refcount race where we see
> a spurious reference for one reason or another.
>
>
> Below is a patch to xfstests to add this reproducer:
>
>
> diff --git a/tests/generic/700 b/tests/generic/700
> new file mode 100755
> index 0000000..33505b0
> --- /dev/null
> +++ b/tests/generic/700
> @@ -0,0 +1,130 @@
> +#! /bin/bash
> +# FS QA Test No. 700
> +#
> +# Dumbed down Copy of the original version of generic/187 that reproduced
> +# NFS silly rename issues.
> +#-----------------------------------------------------------------------
> +# Copyright (c) 2015, Oracle and/or its affiliates. All Rights Reserved.
> +#
> +# This program is free software; you can redistribute it and/or
> +# modify it under the terms of the GNU General Public License as
> +# published by the Free Software Foundation.
> +#
> +# This program is distributed in the hope that it would be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program; if not, write the Free Software Foundation,
> +# Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
> +#-----------------------------------------------------------------------
> +
> +seq=`basename $0`
> +seqres=$RESULT_DIR/$seq
> +echo "QA output created by $seq"
> +
> +here=`pwd`
> +tmp=/tmp/$$
> +status=1 # failure is the default!
> +trap "_cleanup; exit \$status" 0 1 2 3 15
> +
> +_cleanup()
> +{
> + cd /
> + rm -rf $tmp.* $testdir
> +}
> +
> +# get standard environment, filters and checks
> +. ./common/rc
> +. ./common/filter
> +
> +# real QA test starts here
> +_supported_os Linux
> +_require_xfs_io_command "falloc"
> +_require_xfs_io_command "fpunch"
> +test $FSTYP = "btrfs" && _notrun "Can't fragment free space on btrfs."
> +_require_odirect
> +
> +rm -f $seqres.full
> +
> +_fragment_freesp()
> +{
> + file=$1
> +
> + # consume nearly all available space (leave ~1MB)
> + avail=`_get_available_space $SCRATCH_MNT`
> + echo "$avail bytes left"
> + filesize=$((avail - 1048576))
> + $XFS_IO_PROG -fc "truncate $filesize" $file
> +
> + chunks=20
> + chunksizemb=$((filesize / chunks / 1048576))
> + seq 1 $chunks | while read f; do
> + echo "$((f * chunksizemb)) file size $f / 20"
> + $XFS_IO_PROG -fc "falloc -k $(( (f - 1) * chunksizemb))m ${chunksizemb}m" $file
> + done
> +
> + chunks=100
> + chunksizemb=$((filesize / chunks / 1048576))
> + seq 80 $chunks | while read f; do
> + echo "$((f * chunksizemb)) file size $f / $chunks"
> + $XFS_IO_PROG -fc "falloc -k $(( (f - 1) * chunksizemb))m ${chunksizemb}m" $file
> + done
> +
> + filesizemb=$((filesize / 1048576))
> + $XFS_IO_PROG -fc "falloc -k 0 ${filesizemb}m" $file
> +
> + # Try again anyway
> + avail=`_get_available_space $SCRATCH_MNT`
> + $XFS_IO_PROG -fc "pwrite -S 0x65 0 $avail" ${file}.${i}
> +
> + # Punch out whatever we need
> + seq 1 $((nr * 4)) | while read f; do
> + $XFS_IO_PROG -f -c "fpunch $((f * 2 * blksz)) $blksz" $file
> + done
> +}
> +
> +echo "Format and mount"
> +_scratch_mkfs > $seqres.full 2>&1
> +_scratch_mount >> $seqres.full 2>&1
> +
> +testdir=$SCRATCH_MNT/test-$seq
> +mkdir $testdir
> +
> +echo "Create the original files"
> +blksz=65536
> +nr=1024
> +filesize=$((blksz * nr))
> +_pwrite_byte 0x61 0 $filesize $testdir/file1 >> $seqres.full
> +_pwrite_byte 0x62 0 $filesize $testdir/file2 >> $seqres.full
> +seq 0 2 $((nr-1)) | while read f; do
> +# _reflink_range $testdir/file1 $((blksz * f)) $testdir/file3 $((blksz * f)) $blksz >> $seqres.full
> + touch $testdir/file3
> + _pwrite_byte 0x61 $((blksz * f)) $blksz $testdir/file3.chk >> $seqres.full
> +done
> +seq 1 2 $((nr-1)) | while read f; do
> + touch $testdir/file3
> +# _reflink_range $testdir/file2 $((blksz * f)) $testdir/file3 $((blksz * f)) $blksz >> $seqres.full
> + _pwrite_byte 0x62 $((blksz * f)) $blksz $testdir/file3.chk >> $seqres.full
> +done
> +_scratch_cycle_mount
> +_fragment_freesp $testdir/bigfile >> $seqres.full 2>&1
> +filesize=$((blksz * nr))
> +_scratch_cycle_mount
> +
> +echo "Overwrite"
> +cowoff=$((filesize / 4))
> +cowsz=$((filesize / 2))
> +$XFS_IO_PROG -d -f -c "pwrite -S 0x63 -b $cowsz $cowoff $cowsz" $testdir/file3 >> $seqres.full
> +_pwrite_byte 0x63 $cowoff $cowsz $testdir/file3.chk >> $seqres.full
> +
> +# mount / unmount
> +_scratch_cycle_mount
> +
> +echo "Deleting"
> +rm -r $testdir
> +
> +# success, all done
> +status=0
> +exit
> diff --git a/tests/generic/700.out b/tests/generic/700.out
> new file mode 100644
> index 0000000..9f7ce0f
> --- /dev/null
> +++ b/tests/generic/700.out
> @@ -0,0 +1,5 @@
> +QA output created by 700
> +Format and mount
> +Create the original files
> +Overwrite
> +Deleting
> diff --git a/tests/generic/group b/tests/generic/group
> index 2bd40a1..46a796a 100644
> --- a/tests/generic/group
> +++ b/tests/generic/group
> @@ -354,6 +354,7 @@
> 349 blockdev quick rw
> 350 blockdev quick rw
> 351 blockdev quick rw
> +700 auto nfs
> 923 auto quick clone
> 924 auto quick clone
> 925 auto quick clone
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Jun 16, 2016 at 01:29:52PM -0400, J. Bruce Fields wrote:
> On Thu, Jun 16, 2016 at 08:01:47AM -0700, Christoph Hellwig wrote:
> > I've run into an odd bug when testing the clone / reflink functionality,
> > and managed to isolate a testcase not relying on clones. The issue is
> > that after a lot of direct I/O hitting ENOSPC on a NFS mount I run
> > into a silly rename that never gets cleaned up after cleaning up the
> > directory.
>
> I have a redhat bug for what sounds like the same behavior. I spent
> some time looking at it but I don't think I got anywere.
>
> https://bugzilla.redhat.com/show_bug.cgi?id=1132610
>
> (But you probably can't see that right now; I'll see if I can open it.)
I can't see it indeed.
On Fri, Jun 17, 2016 at 04:31:10AM -0700, Christoph Hellwig wrote:
> On Thu, Jun 16, 2016 at 01:29:52PM -0400, J. Bruce Fields wrote:
> > On Thu, Jun 16, 2016 at 08:01:47AM -0700, Christoph Hellwig wrote:
> > > I've run into an odd bug when testing the clone / reflink functionality,
> > > and managed to isolate a testcase not relying on clones. The issue is
> > > that after a lot of direct I/O hitting ENOSPC on a NFS mount I run
> > > into a silly rename that never gets cleaned up after cleaning up the
> > > directory.
> >
> > I have a redhat bug for what sounds like the same behavior. I spent
> > some time looking at it but I don't think I got anywere.
> >
> > https://bugzilla.redhat.com/show_bug.cgi?id=1132610
> >
> > (But you probably can't see that right now; I'll see if I can open it.)
>
> I can't see it indeed.
Fixed. Looking at it, I actually don't think there's much of interest
besides reproducers, which you already have.
--b.