xfs
[Top] [All Lists]

Re: [PATCH] generic/033: add xfs delalloc indirect block depletion repro

To: Brian Foster <bfoster@xxxxxxxxxx>
Subject: Re: [PATCH] generic/033: add xfs delalloc indirect block depletion reproducer
From: Eryu Guan <guaneryu@xxxxxxxxx>
Date: Thu, 25 Sep 2014 11:54:16 +0800
Cc: fstests@xxxxxxxxxxxxxxx, xfs@xxxxxxxxxxx
Delivered-to: xfs@xxxxxxxxxxx
Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=date:from:to:cc:subject:message-id:references:mime-version :content-type:content-disposition:in-reply-to:user-agent; bh=UWSfu8VyJxk5gN2IGTYYDqs6jJ/fA5cdxS8ubTaU5G4=; b=aJiylaNZsI/AYro3NlzTTYbX7VlXeyl4R7+BFOEDKqaqUD97lAHEjrBLgJOS3kLSgC aV5nJs2KYeGKYaFPpSdyNz3OJDkLEU9ERpELk8q5wQA5HG5wr7nZg1nseZnNRQv9U4SP tHjJxlX09KNq/6EFO1Aa93CjCwLmdzfwqo9CcmhQGO7gRe8RUTWBbGox8RxLTF/Az9s3 XWZ/NfUO73XtjlQHA4gptBtbqRTrbg5o9p2u7QVeuWRbUsWaMkJRRYAsnUX9cp2mPF2l I8kTOvKPc78BoIdKQWtilOwEd0aQhFICKKH1qzWTUM4yv6m4lQLY8Pp3w8304q9HpWtT e95A==
In-reply-to: <1411584425-52709-1-git-send-email-bfoster@xxxxxxxxxx>
References: <1411584425-52709-1-git-send-email-bfoster@xxxxxxxxxx>
User-agent: Mutt/1.5.23 (2014-03-12)
On Wed, Sep 24, 2014 at 02:47:05PM -0400, Brian Foster wrote:
> XFS allocates extra indirect blocks for delayed allocation extents at
> write time. When delalloc extents are split, the existing indirect block
> reservation was historically divided up evenly among the new extents
> even though the overall requirement for two extents could exceed the
> requirement for the original. Repeated delalloc extent splits ultimately
> leads to extents with 0 indirect blocks and in turn leads to assert
> failures in XFS.
> 
> Add a test to stress indirect block reservation for delayed allocation
> extents. The test converts a single delalloc extent to many and operates
> on the remaining extents to detect or trigger potential problems.
> 
> Signed-off-by: Brian Foster <bfoster@xxxxxxxxxx>
> ---
> 
> Here's a simple reproducer for the indirect block reservation problem
> called out here:
> 
> http://oss.sgi.com/archives/xfs/2014-09/msg00337.html
> 
> It reproduces the assert failures described therein:
> 
> XFS: Assertion failed: startblockval(del.br_startblock) > 0, file: 
> fs/xfs/libxfs/xfs_bmap.c, line: 5281
> 
> Note that this test also unintentionally fails on XFS. The test file
> ends up zero-sized after the remount and thus hexdump doesn't produce
> any output. This doesn't occur on ext4, I suspect due to the fact that
> the range being zeroed is flushed beforehand, though I could be wrong
> about that.

Tested with ext4 and xfs, and ext4 passes/xfs fails the test as
described here.

Reviewed-by: Eryu Guan <eguan@xxxxxxxxxx>

With one nitpick below..

> 
> In any event, this calls out a separate bug in XFS where if appending
> data is chucked from cache by zero range before written back (eof is
> page aligned), we lose the on-disk inode size update and the inode size
> changes unexpectedly across the remount (assuming nothing else changes
> the size, of course).
> 
> Brian
> 
>  tests/generic/033     | 88 
> +++++++++++++++++++++++++++++++++++++++++++++++++++
>  tests/generic/033.out |  4 +++
>  tests/generic/group   |  1 +
>  3 files changed, 93 insertions(+)
>  create mode 100755 tests/generic/033
>  create mode 100644 tests/generic/033.out
> 
> diff --git a/tests/generic/033 b/tests/generic/033
> new file mode 100755
> index 0000000..41198b7
> --- /dev/null
> +++ b/tests/generic/033
> @@ -0,0 +1,88 @@
> +#! /bin/bash
> +# FS QA Test No. 033
> +#
> +# This test stresses indirect block reservation for delayed allocation 
> extents.
> +# XFS reserves extra blocks for deferred allocation of delalloc extents. 
> These
> +# reserved blocks can be divided among more extents than anticipated if the
> +# original extent for which the blocks were reserved is split into multiple
> +# delalloc extents. If this scenario repeats, eventually some extents are 
> left
> +# without any indirect block reservation whatsoever. This leads to assert
> +# failures and possibly other problems in XFS.
> +#
> +#-----------------------------------------------------------------------
> +# Copyright (c) 2014 Red Hat, Inc.  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 -f $tmp.*
> +}
> +
> +# get standard environment, filters and checks
> +. ./common/rc
> +. ./common/filter
> +
> +# real QA test starts here
> +rm -f $seqres.full
> +
> +# Modify as appropriate.
> +_supported_fs generic
> +_supported_os Linux
> +_require_scratch
> +_require_xfs_io_command "fzero"
> +
> +_scratch_mkfs >/dev/null 2>&1
> +_scratch_mount
> +
> +file=$SCRATCH_MNT/file.$seq
> +bytes=$((64 * 1024))
> +
> +# create sequential delayed allocation
> +$XFS_IO_PROG -f -c "pwrite 0 $bytes" $file | _filter_xfs_io \
> +     >> $seqres.full 2>&1

The output of xfs_io is redirected to $seqres.full, so it's not
necessary to be filtered, for debug purpose.

And the following two xfs_io calls.

Thanks,
Eryu

> +
> +# Zero every other 4k range to split the larger delalloc extent into many 
> more
> +# smaller extents. Use zero instead of hole punch because the former does not
> +# force writeback (and hence delalloc conversion). It can simply discard
> +# delalloc blocks and convert the ranges to unwritten.
> +endoff=$((bytes - 4096))
> +for i in $(seq 0 8192 $endoff); do
> +     $XFS_IO_PROG -c "fzero -k $i 4k" $file | _filter_xfs_io \
> +             >> $seqres.full 2>&1
> +done
> +
> +# now zero the opposite set to remove remaining delalloc extents
> +for i in $(seq 4096 8192 $endoff); do
> +     $XFS_IO_PROG -c "fzero -k $i 4k" $file | _filter_xfs_io \
> +             >> $seqres.full 2>&1
> +done
> +
> +_scratch_remount
> +hexdump $file
> +
> +status=0
> +exit
> diff --git a/tests/generic/033.out b/tests/generic/033.out
> new file mode 100644
> index 0000000..419d831
> --- /dev/null
> +++ b/tests/generic/033.out
> @@ -0,0 +1,4 @@
> +QA output created by 033
> +0000000 0000 0000 0000 0000 0000 0000 0000 0000
> +*
> +0010000
> diff --git a/tests/generic/group b/tests/generic/group
> index 8e0c22a..1227408 100644
> --- a/tests/generic/group
> +++ b/tests/generic/group
> @@ -32,6 +32,7 @@
>  027 auto enospc
>  028 auto quick
>  032 auto quick rw
> +033 auto quick rw
>  053 acl repair auto quick
>  062 attr udf auto quick
>  068 other auto freeze dangerous stress
> -- 
> 1.8.3.1
> 
> _______________________________________________
> xfs mailing list
> xfs@xxxxxxxxxxx
> http://oss.sgi.com/mailman/listinfo/xfs

<Prev in Thread] Current Thread [Next in Thread>