[PATCH v2 0/3] block: use pwrite_zeroes_alignment when writing first sector

[PATCH v2 0/3] block: use pwrite_zeroes_alignment when writing first sector

[Stefan Hajnoczi]

v2:
- Simplify condition to if (!s->needs_alignment) in patch 1 [Vladimir]

This series fixes a bug I introduced in commit 5634622bcb33 ("file-posix: allow
BLKZEROOUT with -t writeback"). The Linux fallocate(2) and ioctl(BLKZEROOUT)
syscalls require logical block size alignment of the offset and length, even
when the file is opened in buffered I/O mode where read/write operations do not
require alignment.

The fix is to populate the pwrite_zeroes_alignment block limits field and to
use that limit in create_file_fallback_zero_first_sector().

One issue I want to raise is that pwrite_zeroes_alignment is an "optimal
alignment" hint. Hence create_file_fallback_zero_first_sector() had to be
modified to honor the limit explicitly. The block layer doesn't automatically
apply padding in order to align requests. This is different from how QEMU's
block layer pwrite/pread works, where it does automatically apply padding and
read/modify/write as necessary. If you want consistency, please let me know.

Stefan Hajnoczi (3):
  file-posix: populate pwrite_zeroes_alignment
  block: use pwrite_zeroes_alignment when writing first sector
  iotests: add Linux loop device image creation test

 include/system/block-backend-io.h             |  1 +
 block.c                                       |  3 +-
 block/block-backend.c                         | 11 ++++
 block/file-posix.c                            | 16 +++++
 tests/qemu-iotests/tests/loop-create-file     | 59 +++++++++++++++++++
 tests/qemu-iotests/tests/loop-create-file.out |  8 +++
 6 files changed, 97 insertions(+), 1 deletion(-)
 create mode 100755 tests/qemu-iotests/tests/loop-create-file
 create mode 100644 tests/qemu-iotests/tests/loop-create-file.out

-- 
2.51.0

[PATCH v2 1/3] file-posix: populate pwrite_zeroes_alignment

[Stefan Hajnoczi]

Linux block devices require write zeroes alignment whereas files do not.

It may come as a surprise that block devices opened in buffered I/O mode
require the alignment for write zeroes requests although normal
read/write requests do not.

Therefore it is necessary to populate the pwrite_zeroes_alignment field.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 block/file-posix.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/block/file-posix.c b/block/file-posix.c
index 8c738674ce..827ffa77a5 100644
--- a/block/file-posix.c
+++ b/block/file-posix.c
@@ -1602,6 +1602,22 @@ static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
 
             bs->bl.pdiscard_alignment = dalign;
         }
+
+#ifdef __linux__
+        /*
+         * Linux requires logical block size alignment for write zeroes even
+         * when normal reads/writes do not require alignment.
+         */
+        if (!s->needs_alignment) {
+            ret = probe_logical_blocksize(s->fd,
+                                          &bs->bl.pwrite_zeroes_alignment);
+            if (ret < 0) {
+                error_setg_errno(errp, -ret,
+                                 "Failed to probe logical block size");
+                return;
+            }
+        }
+#endif /* __linux__ */
     }
 
     raw_refresh_zoned_limits(bs, &st, errp);
-- 
2.51.0

Re: [PATCH v2 1/3] file-posix: populate pwrite_zeroes_alignment

[Vladimir Sementsov-Ogievskiy]

On 07.10.25 17:16, Stefan Hajnoczi wrote:
> Linux block devices require write zeroes alignment whereas files do not.
> 
> It may come as a surprise that block devices opened in buffered I/O mode
> require the alignment for write zeroes requests although normal
> read/write requests do not.
> 
> Therefore it is necessary to populate the pwrite_zeroes_alignment field.
> 
> Signed-off-by: Stefan Hajnoczi<stefanha@redhat.com>


Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>

-- 
Best regards,
Vladimir

[PATCH v2 2/3] block: use pwrite_zeroes_alignment when writing first sector

[Stefan Hajnoczi]

Since commit 5634622bcb33 ("file-posix: allow BLKZEROOUT with -t
writeback"), qemu-img create errors out on a Linux loop block device
with a 4 KB sector size:

  # dd if=/dev/zero of=blockfile bs=1M count=1024
  # losetup --sector-size 4096 /dev/loop0 blockfile
  # qemu-img create -f raw /dev/loop0 1G
  Formatting '/dev/loop0', fmt=raw size=1073741824
  qemu-img: /dev/loop0: Failed to clear the new image's first sector: Invalid argument

Use the pwrite_zeroes_alignment block limit to avoid misaligned
fallocate(2) or ioctl(BLKZEROOUT) in the block/file-posix.c block
driver.

Fixes: 5634622bcb33 ("file-posix: allow BLKZEROOUT with -t writeback")
Reported-by: Jean-Louis Dupond <jean-louis@dupond.be>
Buglink: https://gitlab.com/qemu-project/qemu/-/issues/3127
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 include/system/block-backend-io.h |  1 +
 block.c                           |  3 ++-
 block/block-backend.c             | 11 +++++++++++
 3 files changed, 14 insertions(+), 1 deletion(-)

diff --git a/include/system/block-backend-io.h b/include/system/block-backend-io.h
index ba8dfcc7d0..6d5ac476fc 100644
--- a/include/system/block-backend-io.h
+++ b/include/system/block-backend-io.h
@@ -116,6 +116,7 @@ BlockAIOCB *blk_abort_aio_request(BlockBackend *blk,
                                   void *opaque, int ret);
 
 uint32_t blk_get_request_alignment(BlockBackend *blk);
+uint32_t blk_get_pwrite_zeroes_alignment(BlockBackend *blk);
 uint32_t blk_get_max_transfer(BlockBackend *blk);
 uint64_t blk_get_max_hw_transfer(BlockBackend *blk);
 
diff --git a/block.c b/block.c
index 8848e9a7ed..be77e03904 100644
--- a/block.c
+++ b/block.c
@@ -606,12 +606,13 @@ create_file_fallback_zero_first_sector(BlockBackend *blk,
                                        int64_t current_size,
                                        Error **errp)
 {
+    uint32_t alignment = blk_get_pwrite_zeroes_alignment(blk);
     int64_t bytes_to_clear;
     int ret;
 
     GLOBAL_STATE_CODE();
 
-    bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
+    bytes_to_clear = MIN(current_size, MAX(BDRV_SECTOR_SIZE, alignment));
     if (bytes_to_clear) {
         ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
         if (ret < 0) {
diff --git a/block/block-backend.c b/block/block-backend.c
index f8d6ba65c1..239d6eca37 100644
--- a/block/block-backend.c
+++ b/block/block-backend.c
@@ -2305,6 +2305,17 @@ uint32_t blk_get_request_alignment(BlockBackend *blk)
     return bs ? bs->bl.request_alignment : BDRV_SECTOR_SIZE;
 }
 
+/* Returns the optimal write zeroes alignment, in bytes; guaranteed nonzero */
+uint32_t blk_get_pwrite_zeroes_alignment(BlockBackend *blk)
+{
+    BlockDriverState *bs = blk_bs(blk);
+    IO_CODE();
+    if (!bs) {
+        return BDRV_SECTOR_SIZE;
+    }
+    return bs->bl.pwrite_zeroes_alignment ?: bs->bl.request_alignment;
+}
+
 /* Returns the maximum hardware transfer length, in bytes; guaranteed nonzero */
 uint64_t blk_get_max_hw_transfer(BlockBackend *blk)
 {
-- 
2.51.0

[PATCH v2 3/3] iotests: add Linux loop device image creation test

[Stefan Hajnoczi]

This qemu-iotests test case is based on the reproducer that Jean-Louis
Dupond <jean-louis@dupond.be> shared in
https://gitlab.com/qemu-project/qemu/-/issues/3127.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 tests/qemu-iotests/tests/loop-create-file     | 59 +++++++++++++++++++
 tests/qemu-iotests/tests/loop-create-file.out |  8 +++
 2 files changed, 67 insertions(+)
 create mode 100755 tests/qemu-iotests/tests/loop-create-file
 create mode 100644 tests/qemu-iotests/tests/loop-create-file.out

diff --git a/tests/qemu-iotests/tests/loop-create-file b/tests/qemu-iotests/tests/loop-create-file
new file mode 100755
index 0000000000..5ec75b046b
--- /dev/null
+++ b/tests/qemu-iotests/tests/loop-create-file
@@ -0,0 +1,59 @@
+#!/usr/bin/env bash
+# group: quick
+#
+# SPDX-License-Identifier: GPL-2.0-or-later
+#
+# Copyright Red Hat, Inc.
+#
+# Test Linux loop device image creation
+#
+# This test verifies #3127 "qemu-img create fails on loop device with sector size 4096"
+# https://gitlab.com/qemu-project/qemu/-/issues/3127
+
+seq="$(basename $0)"
+echo "QA output created by $seq"
+
+status=1	# failure is the default!
+
+_cleanup() {
+    if [ -n "$loopdev" ]; then
+        sudo losetup --detach "$loopdev"
+    fi
+
+    _cleanup_test_img
+}
+
+trap "_cleanup; exit \$status" 0 1 2 3 15
+
+# get standard environment, filters and checks
+cd ..
+. ./common.rc
+. ./common.filter
+
+_supported_fmt raw
+_supported_proto file
+_supported_os Linux
+
+if ! sudo -n losetup &>/dev/null; then
+    _notrun "sudo losetup not available"
+fi
+
+echo
+echo "=== Create image on a 4 KB sector size loop device ==="
+echo
+
+_make_test_img -f $IMGFMT 1M
+
+loopdev=$(sudo losetup --sector-size 4096 --find --show "$TEST_IMG")
+if [ -z "$loopdev" ]; then
+    _fail
+fi
+
+sudo $QEMU_IMG_PROG create -f raw "$loopdev" 1M | \
+    sed -e "s#/dev/loop[0-9]\\+#LOOPDEV#g"
+
+# success, all done
+echo
+echo '*** done'
+rm -f $seq.full
+status=0
diff --git a/tests/qemu-iotests/tests/loop-create-file.out b/tests/qemu-iotests/tests/loop-create-file.out
new file mode 100644
index 0000000000..32d4155695
--- /dev/null
+++ b/tests/qemu-iotests/tests/loop-create-file.out
@@ -0,0 +1,8 @@
+QA output created by loop-create-file
+
+=== Create image on a 4 KB sector size loop device ===
+
+Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=1048576
+Formatting 'LOOPDEV', fmt=raw size=1048576
+
+*** done
-- 
2.51.0

Re: [PATCH v2 3/3] iotests: add Linux loop device image creation test

[Vladimir Sementsov-Ogievskiy]

On 07.10.25 17:17, Stefan Hajnoczi wrote:
> This qemu-iotests test case is based on the reproducer that Jean-Louis
> Dupond<jean-louis@dupond.be> shared in
> https://gitlab.com/qemu-project/qemu/-/issues/3127.
> 
> Signed-off-by: Stefan Hajnoczi<stefanha@redhat.com>

Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Tested-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>

-- 
Best regards,
Vladimir

Re: [PATCH v2 0/3] block: use pwrite_zeroes_alignment when writing first sector

[Fiona Ebner]

Am 07.10.25 um 6:07 PM schrieb Stefan Hajnoczi:
> v2:
> - Simplify condition to if (!s->needs_alignment) in patch 1 [Vladimir]
> 
> This series fixes a bug I introduced in commit 5634622bcb33 ("file-posix: allow
> BLKZEROOUT with -t writeback"). The Linux fallocate(2) and ioctl(BLKZEROOUT)
> syscalls require logical block size alignment of the offset and length, even
> when the file is opened in buffered I/O mode where read/write operations do not
> require alignment.
> 
> The fix is to populate the pwrite_zeroes_alignment block limits field and to
> use that limit in create_file_fallback_zero_first_sector().
> 
> One issue I want to raise is that pwrite_zeroes_alignment is an "optimal
> alignment" hint. Hence create_file_fallback_zero_first_sector() had to be
> modified to honor the limit explicitly. The block layer doesn't automatically
> apply padding in order to align requests. This is different from how QEMU's
> block layer pwrite/pread works, where it does automatically apply padding and
> read/modify/write as necessary. If you want consistency, please let me know.
> 
> Stefan Hajnoczi (3):
>   file-posix: populate pwrite_zeroes_alignment
>   block: use pwrite_zeroes_alignment when writing first sector
>   iotests: add Linux loop device image creation test
> 
>  include/system/block-backend-io.h             |  1 +
>  block.c                                       |  3 +-
>  block/block-backend.c                         | 11 ++++
>  block/file-posix.c                            | 16 +++++
>  tests/qemu-iotests/tests/loop-create-file     | 59 +++++++++++++++++++
>  tests/qemu-iotests/tests/loop-create-file.out |  8 +++
>  6 files changed, 97 insertions(+), 1 deletion(-)
>  create mode 100755 tests/qemu-iotests/tests/loop-create-file
>  create mode 100644 tests/qemu-iotests/tests/loop-create-file.out
> 

Thank you for the fix!

Tested-by: Fiona Ebner <f.ebner@proxmox.com>
Reviewed-by: Fiona Ebner <f.ebner@proxmox.com>

Re: [PATCH v2 0/3] block: use pwrite_zeroes_alignment when writing first sector

[Fiona Ebner]

Am 19.11.25 um 4:40 PM schrieb Fiona Ebner:
> Am 07.10.25 um 6:07 PM schrieb Stefan Hajnoczi:
>> v2:
>> - Simplify condition to if (!s->needs_alignment) in patch 1 [Vladimir]
>>
>> This series fixes a bug I introduced in commit 5634622bcb33 ("file-posix: allow
>> BLKZEROOUT with -t writeback"). The Linux fallocate(2) and ioctl(BLKZEROOUT)
>> syscalls require logical block size alignment of the offset and length, even
>> when the file is opened in buffered I/O mode where read/write operations do not
>> require alignment.
>>
>> The fix is to populate the pwrite_zeroes_alignment block limits field and to
>> use that limit in create_file_fallback_zero_first_sector().
>>
>> One issue I want to raise is that pwrite_zeroes_alignment is an "optimal
>> alignment" hint. Hence create_file_fallback_zero_first_sector() had to be
>> modified to honor the limit explicitly. The block layer doesn't automatically
>> apply padding in order to align requests. This is different from how QEMU's
>> block layer pwrite/pread works, where it does automatically apply padding and
>> read/modify/write as necessary. If you want consistency, please let me know.
>>
>> Stefan Hajnoczi (3):
>>   file-posix: populate pwrite_zeroes_alignment
>>   block: use pwrite_zeroes_alignment when writing first sector
>>   iotests: add Linux loop device image creation test
>>
>>  include/system/block-backend-io.h             |  1 +
>>  block.c                                       |  3 +-
>>  block/block-backend.c                         | 11 ++++
>>  block/file-posix.c                            | 16 +++++
>>  tests/qemu-iotests/tests/loop-create-file     | 59 +++++++++++++++++++
>>  tests/qemu-iotests/tests/loop-create-file.out |  8 +++
>>  6 files changed, 97 insertions(+), 1 deletion(-)
>>  create mode 100755 tests/qemu-iotests/tests/loop-create-file
>>  create mode 100644 tests/qemu-iotests/tests/loop-create-file.out
>>
> 
> Thank you for the fix!
> 
> Tested-by: Fiona Ebner <f.ebner@proxmox.com>
> Reviewed-by: Fiona Ebner <f.ebner@proxmox.com>

Oh, and CC qemu-stable, because 5634622bcb33 ("file-posix: allow
BLKZEROOUT with -t writeback") is in 10.1.

Re: [PATCH v2 0/3] block: use pwrite_zeroes_alignment when writing first sector

[Kevin Wolf]

Am 07.10.2025 um 16:16 hat Stefan Hajnoczi geschrieben:
> v2:
> - Simplify condition to if (!s->needs_alignment) in patch 1 [Vladimir]
> 
> This series fixes a bug I introduced in commit 5634622bcb33 ("file-posix: allow
> BLKZEROOUT with -t writeback"). The Linux fallocate(2) and ioctl(BLKZEROOUT)
> syscalls require logical block size alignment of the offset and length, even
> when the file is opened in buffered I/O mode where read/write operations do not
> require alignment.
> 
> The fix is to populate the pwrite_zeroes_alignment block limits field and to
> use that limit in create_file_fallback_zero_first_sector().
> 
> One issue I want to raise is that pwrite_zeroes_alignment is an "optimal
> alignment" hint. Hence create_file_fallback_zero_first_sector() had to be
> modified to honor the limit explicitly. The block layer doesn't automatically
> apply padding in order to align requests. This is different from how QEMU's
> block layer pwrite/pread works, where it does automatically apply padding and
> read/modify/write as necessary. If you want consistency, please let me know.

Thanks, applied to the block branch.

Kevin