Skip to contentSkip to conversation

block/file-posix: Fix xfs_write_zeroes()

PATCH In review Max Reitz 4 participants 7 messages Aug 22, 2019 – Aug 23, 2019 Block layer

lore
Share & tools
lore

Max Reitz

author
#0 lore raw

cover block/file-posix: Fix xfs_write_zeroes()

Lukàš ran over a nasty regression in our xfs_write_zeroes() function (sorry, my fault) made apparent by a recent patch from Anton that makes qcow2 images heavily exercise the offending code path.

This series fixes the bug and adds a test to prevent it from reoccurring.

Max Reitz (2):
block/file-posix: Fix xfs_write_zeroes()
iotests: Test reverse sub-cluster qcow2 writes

 block/file-posix.c         | 16 ++++++---
 tests/qemu-iotests/265     | 67 ++++++++++++++++++++++++++++++++++++++
 tests/qemu-iotests/265.out |  6 ++++
 tests/qemu-iotests/group   |  1 +
 4 files changed, 85 insertions(+), 5 deletions(-)
 create mode 100755 tests/qemu-iotests/265
 create mode 100644 tests/qemu-iotests/265.out
Signature · 1 line
2.21.0
author
#1 lore raw

PATCH 1/2 block/file-posix: Fix xfs_write_zeroes()

Calling ftruncate() in xfs_write_zeroes() is dangerous because it may yield and then discard data that parallel write requests have written past the old EOF in the meantime. We must not use it here.

Instead, return -ENOTSUP and let the more generic fallocate code handle writing zeroes past the EOF.

Fixes 50ba5b2d994853b38fed10e0841b119da0f8b8e5
---
 block/file-posix.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)
block/file-posix.c +11 -5
index fbeb0068db..b49e0784a4 100644
@@ -1472,10 +1472,13 @@ static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
     }
 
     if (offset + bytes > len) {
-        /* XFS_IOC_ZERO_RANGE does not increase the file length */
-        if (ftruncate(s->fd, offset + bytes) < 0) {
-            return -errno;
-        }
+        /*
+         * XFS_IOC_ZERO_RANGE does not increase the file length, but
+         * the caller probably wants us to.
+         * Calling ftruncate() would not be safe, so let the generic
+         * implementation handle this case.
+         */
+        return -ENOTSUP;
     }
 
     memset(&fl, 0, sizeof(fl));
@@ -1580,7 +1583,10 @@ static int handle_aiocb_write_zeroes(void *opaque)
 
 #ifdef CONFIG_XFS
     if (s->is_xfs) {
-        return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
+        int ret = xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
+        if (ret != -ENOTSUP) {
+            return ret;
+        }
     }
 #endif
 
Signature · 1 line
2.21.0
author
#2 lore raw

PATCH 2/2 iotests: Test reverse sub-cluster qcow2 writes

In reply to Max Reitz

This exercises the regression introduced in commit 50ba5b2d994853b38fed10e0841b119da0f8b8e5. On my machine, it has close to a 50 % false-negative rate, but that should still be sufficient to test the fix.

---
 tests/qemu-iotests/265     | 67 ++++++++++++++++++++++++++++++++++++++
 tests/qemu-iotests/265.out |  6 ++++
 tests/qemu-iotests/group   |  1 +
 3 files changed, 74 insertions(+)
 create mode 100755 tests/qemu-iotests/265
 create mode 100644 tests/qemu-iotests/265.out
tests/qemu-iotests/265 new file +67 -0
new file mode 100755
index 0000000000..dce6f77be3
@@ -0,0 +1,67 @@
+#!/usr/bin/env bash
+#
+# Test reverse-ordered qcow2 writes on a sub-cluster level
+#
+# Copyright (C) 2019 Red Hat, Inc.
+#
+# 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; either version 2 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will 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, see <http://www.gnu.org/licenses/>.
+#
+
+seq=$(basename $0)
+echo "QA output created by $seq"
+
+status=1	# failure is the default!
+
+_cleanup()
+{
+    _cleanup_test_img
+}
+trap "_cleanup; exit \$status" 0 1 2 3 15
+
+# get standard environment, filters and checks
+. ./common.rc
+. ./common.filter
+
+# qcow2-specific test
+_supported_fmt qcow2
+_supported_proto file
+_supported_os Linux
+
+echo '--- Writing to the image ---'
+
+# Reduce cluster size so we get more and quicker I/O
+IMGOPTS='cluster_size=4096' _make_test_img 1M
+(for ((kb = 1024 - 4; kb >= 0; kb -= 4)); do \
+     echo "aio_write -P 42 $((kb + 1))k 2k"; \
+ done) \
+ | $QEMU_IO "$TEST_IMG" > /dev/null
+
+echo '--- Verifying its content ---'
+
+(for ((kb = 0; kb < 1024; kb += 4)); do \
+    echo "read -P 0 ${kb}k 1k"; \
+    echo "read -P 42 $((kb + 1))k 2k"; \
+    echo "read -P 0 $((kb + 3))k 1k"; \
+ done) \
+ | $QEMU_IO "$TEST_IMG" | _filter_qemu_io | grep 'verification'
+
+# Status of qemu-io
+if [ ${PIPESTATUS[1]} = 0 ]; then
+    echo 'Content verified.'
+fi
+
+# success, all done
+echo "*** done"
+rm -f $seq.full
+status=0
tests/qemu-iotests/265.out new file +6 -0
new file mode 100644
index 0000000000..6eac620f25
@@ -0,0 +1,6 @@
+QA output created by 265
+--- Writing to the image ---
+Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=1048576
+--- Verifying its content ---
+Content verified.
+*** done
tests/qemu-iotests/group +1 -0
index d95d556414..0c129c1644 100644
@@ -274,3 +274,4 @@
 257 rw
 258 rw quick
 262 rw quick migration
+265 rw auto quick
Signature · 1 line
2.21.0
#3 lore raw
In reply to Max Reitz
Quoted 12 lines · Max Reitz

Lukàš ran over a nasty regression in our xfs_write_zeroes() function (sorry, my fault) made apparent by a recent patch from Anton that makes qcow2 images heavily exercise the offending code path.

This series fixes the bug and adds a test to prevent it from reoccurring.

Max Reitz (2):
block/file-posix: Fix xfs_write_zeroes()
iotests: Test reverse sub-cluster qcow2 writes

Quoted 8 lines
 block/file-posix.c         | 16 ++++++---
 tests/qemu-iotests/265     | 67 ++++++++++++++++++++++++++++++++++++++
 tests/qemu-iotests/265.out |  6 ++++
 tests/qemu-iotests/group   |  1 +
 4 files changed, 85 insertions(+), 5 deletions(-)
 create mode 100755 tests/qemu-iotests/265
 create mode 100644 tests/qemu-iotests/265.out

What about just killing libxfs support and only use fallocate? FALLOC_FL_ZERO_RANGE was added in Linux 3.15 (2014) and the only platform we probably support with such an old kernel is of course RHEL/CentOS 7 which has had it backported.

Paolo

author
#4 lore raw
Quoted 22 lines · Paolo Bonzini
Quoted 12 lines · Max Reitz

Lukàš ran over a nasty regression in our xfs_write_zeroes() function (sorry, my fault) made apparent by a recent patch from Anton that makes qcow2 images heavily exercise the offending code path.

This series fixes the bug and adds a test to prevent it from reoccurring.

Max Reitz (2):
block/file-posix: Fix xfs_write_zeroes()
iotests: Test reverse sub-cluster qcow2 writes

Quoted 8 lines
 block/file-posix.c         | 16 ++++++---
 tests/qemu-iotests/265     | 67 ++++++++++++++++++++++++++++++++++++++
 tests/qemu-iotests/265.out |  6 ++++
 tests/qemu-iotests/group   |  1 +
 4 files changed, 85 insertions(+), 5 deletions(-)
 create mode 100755 tests/qemu-iotests/265
 create mode 100644 tests/qemu-iotests/265.out
Quoted 4 lines

What about just killing libxfs support and only use fallocate? FALLOC_FL_ZERO_RANGE was added in Linux 3.15 (2014) and the only platform we probably support with such an old kernel is of course RHEL/CentOS 7 which has had it backported.

Works just as well for me. :-)

Max

Tested-by
#5 lore raw

Dne 22. 08. 19 v 19:09 Max Reitz napsal(a):

Quoted 23 lines · Paolo Bonzini
Quoted 22 lines · Max Reitz
Quoted 20 lines

Lukàš ran over a nasty regression in our xfs_write_zeroes() function (sorry, my fault) made apparent by a recent patch from Anton that makes qcow2 images heavily exercise the offending code path.

This series fixes the bug and adds a test to prevent it from reoccurring.

Max Reitz (2):
block/file-posix: Fix xfs_write_zeroes()
iotests: Test reverse sub-cluster qcow2 writes

 block/file-posix.c         | 16 ++++++---
 tests/qemu-iotests/265     | 67 ++++++++++++++++++++++++++++++++++++++
 tests/qemu-iotests/265.out |  6 ++++
 tests/qemu-iotests/group   |  1 +
 4 files changed, 85 insertions(+), 5 deletions(-)
 create mode 100755 tests/qemu-iotests/265
 create mode 100644 tests/qemu-iotests/265.out
Quoted 9 lines
Quoted 4 lines

What about just killing libxfs support and only use fallocate? FALLOC_FL_ZERO_RANGE was added in Linux 3.15 (2014) and the only platform we probably support with such an old kernel is of course RHEL/CentOS 7 which has had it backported.

Works just as well for me. :-)

Max

Whatever you decide, anyway this patch fixes my arm installation issue and the attached qemu iotest is also passing on my system.

Thank you,
Lukáš

Reviewed-by on the v1 series
#6 lore raw
In reply to Max Reitz
Quoted 12 lines · Max Reitz

Lukàš ran over a nasty regression in our xfs_write_zeroes() function (sorry, my fault) made apparent by a recent patch from Anton that makes qcow2 images heavily exercise the offending code path.

This series fixes the bug and adds a test to prevent it from reoccurring.

Max Reitz (2):
block/file-posix: Fix xfs_write_zeroes()
iotests: Test reverse sub-cluster qcow2 writes

Quoted 8 lines
  block/file-posix.c         | 16 ++++++---
  tests/qemu-iotests/265     | 67 ++++++++++++++++++++++++++++++++++++++
  tests/qemu-iotests/265.out |  6 ++++
  tests/qemu-iotests/group   |  1 +
  4 files changed, 85 insertions(+), 5 deletions(-)

create mode 100755 tests/qemu-iotests/265
create mode 100644 tests/qemu-iotests/265.out

Nice! thanks Max