On 23 Mar 2021 at 09:45, Darrick J. Wong wrote:

On Thu, Mar 11, 2021 at 01:28:32PM +0530, Chandan Babu R wrote:

quoted

On 09 Mar 2021 at 10:10, Darrick J. Wong wrote:

quoted

From: Darrick J. Wong <djwong@kernel.org>

Add a couple of regression tests for "xfs: make sure the rt allocator
doesn't run off the end" and "xfs: ensure that fpunch, fcollapse, and
finsert operations are aligned to rt extent size".

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
---
 tests/xfs/759     |  100 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 tests/xfs/759.out |    2 +
 tests/xfs/760     |   68 ++++++++++++++++++++++++++++++++++++
 tests/xfs/760.out |    9 +++++
 tests/xfs/group   |    2 +
 5 files changed, 181 insertions(+)
 create mode 100755 tests/xfs/759
 create mode 100644 tests/xfs/759.out
 create mode 100755 tests/xfs/760
 create mode 100644 tests/xfs/760.out

diff --git a/tests/xfs/759 b/tests/xfs/759
new file mode 100755
index 00000000..8558fe30
--- /dev/null
+++ b/tests/xfs/759

@@ -0,0 +1,100 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0-or-later
+# Copyright (c) 2021 Oracle.  All Rights Reserved.
+#
+# FS QA Test No. 759
+#
+# This is a regression test for commit 2a6ca4baed62 ("xfs: make sure the rt
+# allocator doesn't run off the end") which fixes an overflow error in the
+# _near realtime allocator.  If the rt bitmap ends exactly at the end of a
+# block and the number of rt extents is large enough to allow an allocation
+# request larger than the maximum extent size, it's possible that during a
+# large allocation request, the allocator will fail to constrain maxlen on the
+# second run through the loop, and the rt bitmap range check will run right off
+# the end of the rtbitmap file.  When this happens, xfs triggers a verifier
+# error and returns EFSCORRUPTED.
+
+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
+_supported_fs xfs
+_require_scratch
+_require_realtime
+_require_test_program "punch-alternating"
+
+rm -f $seqres.full
+
+# Format filesystem to get the block size
+_scratch_mkfs > $seqres.full
+_scratch_mount >> $seqres.full
+
+blksz=$(_get_block_size $SCRATCH_MNT)
+rextsize=$($XFS_INFO_PROG $SCRATCH_MNT | grep realtime.*extsz | sed -e 's/^.*extsz=\([0-9]*\).*$/\1/g')
+rextblks=$((rextsize / blksz))
+
+echo "blksz $blksz rextsize $rextsize rextblks $rextblks" >> $seqres.full
+
+_scratch_unmount
+
+# Format filesystem with a realtime volume whose size fits the following:
+# 1. Longer than (XFS MAXEXTLEN * blocksize) bytes.

Shouldn't the multiplier be RT extent size rather than FS block size?

No, because reproducing the bug requires a large enough allocation
request that we can't fit it all in a single data fork extent mapping
and have to call back into the allocator to get more space.  bmbt
mappings are always in units of fsblocks, not rt extents.

Ah, I had not realized that.

The changes look good to me,

Reviewed-by: Chandan Babu R <redacted>

-- 
chandan