Skip to contentSkip to conversation

fstests: generic/563: use fs blocksize to do the writes

PATCH In review Qu Wenruo 5 participants 7 messages Sep 29, 2024 – Oct 10, 2024 Btrfs file system

lore
Share & tools
lore

Qu Wenruo

author
#0 lore raw

PATCH 1/1 fstests: generic/563: use fs blocksize to do the writes

[FALSE ALERTS]
If the system has a page size larger than 4K, and the fs block size matches the page size, test case generic/563 will fail:

    --- tests/generic/563.out	2024-04-25 18:13:45.178550333 +0930
    +++ /home/adam/xfstests-dev/results//generic/563.out.bad	2024-09-30 09:09:16.155312379 +0930
    @@ -3,7 +3,8 @@
     read is in range
     write is in range
     write -> read/write
    -read is in range
    +read has value of 8388608
    +read is NOT in range -33792 .. 33792
     write is in range
    ...

Both Ext4 and btrfs fail with 64K block size and 64K page size

[CAUSE]
The test case writes the 8MiB file using the default block size xfs_io pwrite, which is 4KiB.

Since the fs block size is 64K, such 4KiB write is unaligned inside a block, causing the fs to read out the full page.

Thus the pwrite will cause the fs to read out every page, resulting the above 8MiB+ read value.

[FIX]
Fix the test case by using the fs block size to avoid such unaligned buffered write.

---
 tests/generic/563 | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
tests/generic/563 +6 -4
index 0a8129a6..e8db8acf 100755
@@ -94,6 +94,8 @@ sminor=$((0x`stat -L -c %T $LOOP_DEV`))
 _mkfs_dev $LOOP_DEV >> $seqres.full 2>&1
 _mount $LOOP_DEV $SCRATCH_MNT || _fail "mount failed"
 
+blksize=$(_get_block_size "$SCRATCH_MNT")
+
 drop_io_cgroup=
 grep -q -w io $cgdir/cgroup.subtree_control || drop_io_cgroup=1
 
@@ -103,7 +105,7 @@ echo "+io" > $cgdir/cgroup.subtree_control || _fail "subtree control"
 echo "read/write"
 reset
 switch_cg $cgdir/$seq-cg
-$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite 0 $iosize" -c fsync \
+$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" -c fsync \
 	$SCRATCH_MNT/file >> $seqres.full 2>&1
 switch_cg $cgdir
 $XFS_IO_PROG -c fsync $SCRATCH_MNT/file
@@ -114,9 +116,9 @@ check_cg $cgdir/$seq-cg $iosize $iosize 5% 5%
 echo "write -> read/write"
 reset
 switch_cg $cgdir/$seq-cg
-$XFS_IO_PROG -c "pwrite 0 $iosize" $SCRATCH_MNT/file >> $seqres.full 2>&1
+$XFS_IO_PROG -c "pwrite -b $blksize 0 $iosize" $SCRATCH_MNT/file >> $seqres.full 2>&1
 switch_cg $cgdir/$seq-cg-2
-$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite 0 $iosize" $SCRATCH_MNT/file \
+$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" $SCRATCH_MNT/file \
 	>> $seqres.full 2>&1
 switch_cg $cgdir
 $XFS_IO_PROG -c fsync $SCRATCH_MNT/file
@@ -134,7 +136,7 @@ reset
 switch_cg $cgdir/$seq-cg
 $XFS_IO_PROG -c "pread 0 $iosize" $SCRATCH_MNT/file >> $seqres.full 2>&1
 switch_cg $cgdir/$seq-cg-2
-$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite 0 $iosize" $SCRATCH_MNT/file \
+$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" $SCRATCH_MNT/file \
 	>> $seqres.full 2>&1
 switch_cg $cgdir
 $XFS_IO_PROG -c fsync $SCRATCH_MNT/file
Signature · 1 line
2.46.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 25 lines · Qu Wenruo

[FALSE ALERTS]
If the system has a page size larger than 4K, and the fs block size matches the page size, test case generic/563 will fail:

    --- tests/generic/563.out	2024-04-25 18:13:45.178550333 +0930
    +++ /home/adam/xfstests-dev/results//generic/563.out.bad	2024-09-30 09:09:16.155312379 +0930
    @@ -3,7 +3,8 @@
     read is in range
     write is in range
     write -> read/write
    -read is in range
    +read has value of 8388608
    +read is NOT in range -33792 .. 33792
     write is in range
    ...

Both Ext4 and btrfs fail with 64K block size and 64K page size

[CAUSE]
The test case writes the 8MiB file using the default block size xfs_io pwrite, which is 4KiB.

Since the fs block size is 64K, such 4KiB write is unaligned inside a block, causing the fs to read out the full page.

Quoted 8 lines

Thus the pwrite will cause the fs to read out every page, resulting the above 8MiB+ read value.

[FIX]
Fix the test case by using the fs block size to avoid such unaligned buffered write.

The rationale and change look good to me. I also tested it on my system with blocksize=pagesize=4k, though I don't have access to one with a 64k page size for the more interesting confirmation.

Quoted 50 lines
---
 tests/generic/563 | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
tests/generic/563 +6 -4
index 0a8129a6..e8db8acf 100755
@@ -94,6 +94,8 @@ sminor=$((0x`stat -L -c %T $LOOP_DEV`))
 _mkfs_dev $LOOP_DEV >> $seqres.full 2>&1
 _mount $LOOP_DEV $SCRATCH_MNT || _fail "mount failed"
 
+blksize=$(_get_block_size "$SCRATCH_MNT")
+
 drop_io_cgroup=
 grep -q -w io $cgdir/cgroup.subtree_control || drop_io_cgroup=1
 
@@ -103,7 +105,7 @@ echo "+io" > $cgdir/cgroup.subtree_control || _fail "subtree control"
 echo "read/write"
 reset
 switch_cg $cgdir/$seq-cg
-$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite 0 $iosize" -c fsync \
+$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" -c fsync \
 	$SCRATCH_MNT/file >> $seqres.full 2>&1
 switch_cg $cgdir
 $XFS_IO_PROG -c fsync $SCRATCH_MNT/file
@@ -114,9 +116,9 @@ check_cg $cgdir/$seq-cg $iosize $iosize 5% 5%
 echo "write -> read/write"
 reset
 switch_cg $cgdir/$seq-cg
-$XFS_IO_PROG -c "pwrite 0 $iosize" $SCRATCH_MNT/file >> $seqres.full 2>&1
+$XFS_IO_PROG -c "pwrite -b $blksize 0 $iosize" $SCRATCH_MNT/file >> $seqres.full 2>&1
 switch_cg $cgdir/$seq-cg-2
-$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite 0 $iosize" $SCRATCH_MNT/file \
+$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" $SCRATCH_MNT/file \
 	>> $seqres.full 2>&1
 switch_cg $cgdir
 $XFS_IO_PROG -c fsync $SCRATCH_MNT/file
@@ -134,7 +136,7 @@ reset
 switch_cg $cgdir/$seq-cg
 $XFS_IO_PROG -c "pread 0 $iosize" $SCRATCH_MNT/file >> $seqres.full 2>&1
 switch_cg $cgdir/$seq-cg-2
-$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite 0 $iosize" $SCRATCH_MNT/file \
+$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" $SCRATCH_MNT/file \
 	>> $seqres.full 2>&1
 switch_cg $cgdir
 $XFS_IO_PROG -c fsync $SCRATCH_MNT/file
Signature · 1 line
2.46.0
#2 lore raw
In reply to Qu Wenruo
Quoted 26 lines · Qu Wenruo
Quoted 1 line

[FALSE ALERTS]
If the system has a page size larger than 4K, and the fs block size matches the page size, test case generic/563 will fail:

     --- tests/generic/563.out	2024-04-25 18:13:45.178550333 +0930
     +++ /home/adam/xfstests-dev/results//generic/563.out.bad	2024-09-30 09:09:16.155312379 +0930
     @@ -3,7 +3,8 @@
      read is in range
      write is in range
      write -> read/write
     -read is in range
     +read has value of 8388608
     +read is NOT in range -33792 .. 33792
      write is in range
     ...

Both Ext4 and btrfs fail with 64K block size and 64K page size

[CAUSE]
The test case writes the 8MiB file using the default block size xfs_io pwrite, which is 4KiB.

Since the fs block size is 64K, such 4KiB write is unaligned inside a block, causing the fs to read out the full page.

Quoted 7 lines

Thus the pwrite will cause the fs to read out every page, resulting the above 8MiB+ read value.

[FIX]
Fix the test case by using the fs block size to avoid such unaligned buffered write.

I ran generic/563 on a Raspberry Pi running 6.4 and with a 64K page size, and got a similar error:

FSTYP -- btrfs
PLATFORM -- Linux/aarch64 fstests-aarch64 6.4.3-arm64-g0ef0e2e48724
#61 SMP Tue Aug 6 16:51:45 BST 2024
MKFS_OPTIONS -- /dev/vdc
MOUNT_OPTIONS -- /dev/vdc /mnt/scratch-dir

generic/563 - output mismatch (see
/root/xfstests/results//generic/563.out.bad)

     --- tests/generic/563.out   2024-08-05 10:33:23.000000000 -0000
     +++ /root/xfstests/results//generic/563.out.bad     2024-10-04 

09:35:51.433413098 -0000
@@ -3,7 +3,8 @@
read is in range
write is in range
write -> read/write
-read is in range
+read has value of 8421376
+read is NOT in range -33792 .. 33792
write is in range
...
(Run 'diff -u /root/xfstests/tests/generic/563.out
/root/xfstests/results//generic/563.out.bad' to see the entire diff)
Ran: generic/563
Failures: generic/563
Failed 1 of 1 tests

The same happens whether the btrfs volume has a sector size of 4K or 64K, and the patch doesn't seem to fix it.

Mark

#3 lore raw
Quoted 64 lines · Mark Harmstone
Quoted 26 lines · Qu Wenruo
Quoted 1 line

[FALSE ALERTS]
If the system has a page size larger than 4K, and the fs block size matches the page size, test case generic/563 will fail:

      --- tests/generic/563.out	2024-04-25 18:13:45.178550333 +0930
      +++ /home/adam/xfstests-dev/results//generic/563.out.bad	2024-09-30 09:09:16.155312379 +0930
      @@ -3,7 +3,8 @@
       read is in range
       write is in range
       write -> read/write
      -read is in range
      +read has value of 8388608
      +read is NOT in range -33792 .. 33792
       write is in range
      ...

Both Ext4 and btrfs fail with 64K block size and 64K page size

[CAUSE]
The test case writes the 8MiB file using the default block size xfs_io pwrite, which is 4KiB.

Since the fs block size is 64K, such 4KiB write is unaligned inside a block, causing the fs to read out the full page.

Quoted 7 lines

Thus the pwrite will cause the fs to read out every page, resulting the above 8MiB+ read value.

[FIX]
Fix the test case by using the fs block size to avoid such unaligned buffered write.

I ran generic/563 on a Raspberry Pi running 6.4 and with a 64K page size, and got a similar error:

FSTYP -- btrfs
PLATFORM -- Linux/aarch64 fstests-aarch64 6.4.3-arm64-g0ef0e2e48724
#61 SMP Tue Aug 6 16:51:45 BST 2024
MKFS_OPTIONS -- /dev/vdc
MOUNT_OPTIONS -- /dev/vdc /mnt/scratch-dir

generic/563 - output mismatch (see
/root/xfstests/results//generic/563.out.bad)

      --- tests/generic/563.out   2024-08-05 10:33:23.000000000 -0000
      +++ /root/xfstests/results//generic/563.out.bad     2024-10-04

09:35:51.433413098 -0000
@@ -3,7 +3,8 @@
read is in range
write is in range
write -> read/write
-read is in range
+read has value of 8421376
+read is NOT in range -33792 .. 33792
write is in range
...
(Run 'diff -u /root/xfstests/tests/generic/563.out
/root/xfstests/results//generic/563.out.bad' to see the entire diff)
Ran: generic/563
Failures: generic/563
Failed 1 of 1 tests

Quoted 2 lines

The same happens whether the btrfs volume has a sector size of 4K or 64K, and the patch doesn't seem to fix it.

For 4K sector size 64K page size btrfs, it needs several kernel patches to proper fix it.
(https://github.com/adam900710/linux/tree/subpage_read)

The above case only shows the 4K sector size case (the new default of mkfs.btrfs, no matter page size now).

For 64K sectorsize with 64K page size case, you need to specify the "-s 64K" mkfs option, apply the patch, only after that the test can pass:

Unpatched:

FSTYP -- btrfs
PLATFORM -- Linux/aarch64 btrfs-aarch64 6.11.0-rc7-custom+ #70 SMP
PREEMPT_DYNAMIC Thu Oct 3 07:25:40 ACST 2024 MKFS_OPTIONS -- -s 64k /dev/mapper/test-scratch1 MOUNT_OPTIONS -- /dev/mapper/test-scratch1 /mnt/scratch

generic/563 4s ... - output mismatch (see /home/adam/xfstests-dev/results//generic/563.out.bad)

    --- tests/generic/563.out	2024-04-25 18:13:45.178550333 +0930
    +++ /home/adam/xfstests-dev/results//generic/563.out.bad	2024-10-04

19:19:19.699153483 +0930
@@ -3,7 +3,8 @@
read is in range
write is in range
write -> read/write
-read is in range
+read has value of 8388608
+read is NOT in range -33792 .. 33792
write is in range
...
(Run 'diff -u /home/adam/xfstests-dev/tests/generic/563.out
/home/adam/xfstests-dev/results//generic/563.out.bad' to see the entire diff)
Ran: generic/563
Failures: generic/563
Failed 1 of 1 tests

Patched:

FSTYP -- btrfs
PLATFORM -- Linux/aarch64 btrfs-aarch64 6.11.0-rc7-custom+ #70 SMP
PREEMPT_DYNAMIC Thu Oct 3 07:25:40 ACST 2024 MKFS_OPTIONS -- -s 64k /dev/mapper/test-scratch1 MOUNT_OPTIONS -- /dev/mapper/test-scratch1 /mnt/scratch

generic/563 4s ... 1s
Ran: generic/563
Passed all 1 tests

You can also do the same using ext4:

Unpatched:

FSTYP -- ext4
PLATFORM -- Linux/aarch64 btrfs-aarch64 6.11.0-rc7-custom+ #70 SMP
PREEMPT_DYNAMIC Thu Oct 3 07:25:40 ACST 2024
MKFS_OPTIONS -- -F -b 64k /dev/mapper/test-scratch1 MOUNT_OPTIONS -- -o acl,user_xattr /dev/mapper/test-scratch1 /mnt/scratch

generic/563 1s ... - output mismatch (see /home/adam/xfstests-dev/results//generic/563.out.bad)

    --- tests/generic/563.out	2024-04-25 18:13:45.178550333 +0930
    +++ /home/adam/xfstests-dev/results//generic/563.out.bad	2024-10-04

19:21:23.377651352 +0930
@@ -3,7 +3,8 @@
read is in range
write is in range
write -> read/write
-read is in range
+read has value of 8388608
+read is NOT in range -33792 .. 33792
write is in range
...
(Run 'diff -u /home/adam/xfstests-dev/tests/generic/563.out
/home/adam/xfstests-dev/results//generic/563.out.bad' to see the entire diff)
Ran: generic/563
Failures: generic/563
Failed 1 of 1 tests

Patched:

FSTYP -- ext4
PLATFORM -- Linux/aarch64 btrfs-aarch64 6.11.0-rc7-custom+ #70 SMP
PREEMPT_DYNAMIC Thu Oct 3 07:25:40 ACST 2024
MKFS_OPTIONS -- -F -b 64k /dev/mapper/test-scratch1 MOUNT_OPTIONS -- -o acl,user_xattr /dev/mapper/test-scratch1 /mnt/scratch

generic/563 1s ... 1s
Ran: generic/563
Passed all 1 tests

Mind to re-verify with the proper mkfs options?

Thanks,
Qu

Quoted 3 lines

Mark

Reviewed-by
#4 lore raw
Quoted 10 lines · Qu Wenruo

For 4K sector size 64K page size btrfs, it needs several kernel patches to proper fix it.
(https://github.com/adam900710/linux/tree/subpage_read)

The above case only shows the 4K sector size case (the new default of mkfs.btrfs, no matter page size now).

For 64K sectorsize with 64K page size case, you need to specify the "-s 64K" mkfs option, apply the patch, only after that the test can pass:

Ah, you have to specify an environment variable, because fstests wipes the filesystem every time. I knew I must have been doing something wrong.

Yes, with this change it works for me.

author
#5 lore raw
In reply to Qu Wenruo

Hi Zorro,

Mind to merge this fix?

It got 2 reviews and at least one verification on 64K page systems.

Thanks,
Qu

Quoted 80 lines · Qu Wenruo

[FALSE ALERTS]
If the system has a page size larger than 4K, and the fs block size matches the page size, test case generic/563 will fail:

     --- tests/generic/563.out	2024-04-25 18:13:45.178550333 +0930
     +++ /home/adam/xfstests-dev/results//generic/563.out.bad	2024-09-30 09:09:16.155312379 +0930
     @@ -3,7 +3,8 @@
      read is in range
      write is in range
      write -> read/write
     -read is in range
     +read has value of 8388608
     +read is NOT in range -33792 .. 33792
      write is in range
     ...

Both Ext4 and btrfs fail with 64K block size and 64K page size

[CAUSE]
The test case writes the 8MiB file using the default block size xfs_io pwrite, which is 4KiB.

Since the fs block size is 64K, such 4KiB write is unaligned inside a block, causing the fs to read out the full page.

Thus the pwrite will cause the fs to read out every page, resulting the above 8MiB+ read value.

[FIX]
Fix the test case by using the fs block size to avoid such unaligned buffered write.

---
  tests/generic/563 | 10 ++++++----
  1 file changed, 6 insertions(+), 4 deletions(-)
tests/generic/563 +6 -4
index 0a8129a6..e8db8acf 100755
@@ -94,6 +94,8 @@ sminor=$((0x`stat -L -c %T $LOOP_DEV`))
  _mkfs_dev $LOOP_DEV >> $seqres.full 2>&1
  _mount $LOOP_DEV $SCRATCH_MNT || _fail "mount failed"
 
+blksize=$(_get_block_size "$SCRATCH_MNT")
+
  drop_io_cgroup=
  grep -q -w io $cgdir/cgroup.subtree_control || drop_io_cgroup=1
 
@@ -103,7 +105,7 @@ echo "+io" > $cgdir/cgroup.subtree_control || _fail "subtree control"
  echo "read/write"
  reset
  switch_cg $cgdir/$seq-cg
-$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite 0 $iosize" -c fsync \
+$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" -c fsync \
  	$SCRATCH_MNT/file >> $seqres.full 2>&1
  switch_cg $cgdir
  $XFS_IO_PROG -c fsync $SCRATCH_MNT/file
@@ -114,9 +116,9 @@ check_cg $cgdir/$seq-cg $iosize $iosize 5% 5%
  echo "write -> read/write"
  reset
  switch_cg $cgdir/$seq-cg
-$XFS_IO_PROG -c "pwrite 0 $iosize" $SCRATCH_MNT/file >> $seqres.full 2>&1
+$XFS_IO_PROG -c "pwrite -b $blksize 0 $iosize" $SCRATCH_MNT/file >> $seqres.full 2>&1
  switch_cg $cgdir/$seq-cg-2
-$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite 0 $iosize" $SCRATCH_MNT/file \
+$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" $SCRATCH_MNT/file \
  	>> $seqres.full 2>&1
  switch_cg $cgdir
  $XFS_IO_PROG -c fsync $SCRATCH_MNT/file
@@ -134,7 +136,7 @@ reset
  switch_cg $cgdir/$seq-cg
  $XFS_IO_PROG -c "pread 0 $iosize" $SCRATCH_MNT/file >> $seqres.full 2>&1
  switch_cg $cgdir/$seq-cg-2
-$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite 0 $iosize" $SCRATCH_MNT/file \
+$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" $SCRATCH_MNT/file \
  	>> $seqres.full 2>&1
  switch_cg $cgdir
  $XFS_IO_PROG -c fsync $SCRATCH_MNT/file
#6 lore raw
Quoted 5 lines · Qu Wenruo

Hi Zorro,

Mind to merge this fix?

It got 2 reviews and at least one verification on 64K page systems.

Sure, it'll be merged in next release. Thanks for fix this large block size test error. More filesystems support bigger blocksize, e.g. xfs supports blocksize > pagesize. It's great to fix more errors like that :)

Thanks,
Zorro

Quoted 85 lines

Thanks,
Qu

Quoted 78 lines · Qu Wenruo

[FALSE ALERTS]
If the system has a page size larger than 4K, and the fs block size matches the page size, test case generic/563 will fail:

     --- tests/generic/563.out	2024-04-25 18:13:45.178550333 +0930
     +++ /home/adam/xfstests-dev/results//generic/563.out.bad	2024-09-30 09:09:16.155312379 +0930
     @@ -3,7 +3,8 @@
      read is in range
      write is in range
      write -> read/write
     -read is in range
     +read has value of 8388608
     +read is NOT in range -33792 .. 33792
      write is in range
     ...

Both Ext4 and btrfs fail with 64K block size and 64K page size

[CAUSE]
The test case writes the 8MiB file using the default block size xfs_io pwrite, which is 4KiB.

Since the fs block size is 64K, such 4KiB write is unaligned inside a block, causing the fs to read out the full page.

Thus the pwrite will cause the fs to read out every page, resulting the above 8MiB+ read value.

[FIX]
Fix the test case by using the fs block size to avoid such unaligned buffered write.

---
  tests/generic/563 | 10 ++++++----
  1 file changed, 6 insertions(+), 4 deletions(-)
tests/generic/563 +6 -4
index 0a8129a6..e8db8acf 100755
@@ -94,6 +94,8 @@ sminor=$((0x`stat -L -c %T $LOOP_DEV`))
  _mkfs_dev $LOOP_DEV >> $seqres.full 2>&1
  _mount $LOOP_DEV $SCRATCH_MNT || _fail "mount failed"
+blksize=$(_get_block_size "$SCRATCH_MNT")
+
  drop_io_cgroup=
  grep -q -w io $cgdir/cgroup.subtree_control || drop_io_cgroup=1
@@ -103,7 +105,7 @@ echo "+io" > $cgdir/cgroup.subtree_control || _fail "subtree control"
  echo "read/write"
  reset
  switch_cg $cgdir/$seq-cg
-$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite 0 $iosize" -c fsync \
+$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" -c fsync \
  	$SCRATCH_MNT/file >> $seqres.full 2>&1
  switch_cg $cgdir
  $XFS_IO_PROG -c fsync $SCRATCH_MNT/file
@@ -114,9 +116,9 @@ check_cg $cgdir/$seq-cg $iosize $iosize 5% 5%
  echo "write -> read/write"
  reset
  switch_cg $cgdir/$seq-cg
-$XFS_IO_PROG -c "pwrite 0 $iosize" $SCRATCH_MNT/file >> $seqres.full 2>&1
+$XFS_IO_PROG -c "pwrite -b $blksize 0 $iosize" $SCRATCH_MNT/file >> $seqres.full 2>&1
  switch_cg $cgdir/$seq-cg-2
-$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite 0 $iosize" $SCRATCH_MNT/file \
+$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" $SCRATCH_MNT/file \
  	>> $seqres.full 2>&1
  switch_cg $cgdir
  $XFS_IO_PROG -c fsync $SCRATCH_MNT/file
@@ -134,7 +136,7 @@ reset
  switch_cg $cgdir/$seq-cg
  $XFS_IO_PROG -c "pread 0 $iosize" $SCRATCH_MNT/file >> $seqres.full 2>&1
  switch_cg $cgdir/$seq-cg-2
-$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite 0 $iosize" $SCRATCH_MNT/file \
+$XFS_IO_PROG -c "pread 0 $iosize" -c "pwrite -b $blksize 0 $iosize" $SCRATCH_MNT/file \
  	>> $seqres.full 2>&1
  switch_cg $cgdir
  $XFS_IO_PROG -c fsync $SCRATCH_MNT/file