On Tue, Jul 19, 2022 at 02:37:13PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <djwong@xxxxxxxxxx>
>
> When running this test on an XFS filesystem with a 64k block size, I
> see this error:
>
> generic/275 - output mismatch (see /var/tmp/fstests/generic/275.out.bad)
> --- tests/generic/275.out 2021-05-13 11:47:55.694860280 -0700
> +++ /var/tmp/fstests/generic/275.out.bad 2022-07-19 10:38:41.840000000 -0700
> @@ -2,4 +2,7 @@
> ------------------------------
> write until ENOSPC test
> ------------------------------
> +du: cannot access '/opt/tmp1': No such file or directory
> +stat: cannot statx '/opt/tmp1': No such file or directory
> +/tmp/fstests/tests/generic/275: line 74: [: -lt: unary operator expected
> done
> ...
> (Run 'diff -u /tmp/fstests/tests/generic/275.out /var/tmp/fstests/generic/275.out.bad' to see the entire diff)
>
> The 275.full file indicates that the test was unable to recreate the
> $SCRATCH_MNT/tmp1 file after we freed all but the last 256K of free
> space in the filesystem. I mounted the scratch fs, and df reported
> exactly 256K of free space available, which means there are 4 blocks
> left in the filesystem for user programs to use.
>
> Unfortunately for this test, xfs_create requires sufficient free blocks
> in the filesystem to handle full inode btree splits and the maximal
> directory expansion for a new dirent. In other words, there must be
> enough free space to handle the worst case space consumption. That
> quantity is 26 blocks, hence the last dd in the test fails with ENOSPC,
> which makes the test fail.
>
> Fix all this by creating the file that we use to test the low-space file
> write *before* we drain the free space down to 256K.
>
> Signed-off-by: Darrick J. Wong <djwong@xxxxxxxxxx>
> ---
> tests/generic/275 | 15 ++++++++++++---
> 1 file changed, 12 insertions(+), 3 deletions(-)
>
>
> diff --git a/tests/generic/275 b/tests/generic/275
> index 6189edca..f3b05409 100755
> --- a/tests/generic/275
> +++ b/tests/generic/275
> @@ -37,6 +37,15 @@ _scratch_unmount 2>/dev/null
> _scratch_mkfs_sized $((2 * 1024 * 1024 * 1024)) >>$seqres.full 2>&1
> _scratch_mount
>
> +# Certain filesystems such as XFS require sufficient free blocks to handle the
> +# worst-case directory expansion as a result of a creat() call. If the fs
> +# block size is very large (e.g. 64k) then the number of blocks required for
> +# the creat() call can represent far more free space than the 256K left at the
> +# end of this test. Therefore, create the file that the last dd will write to
> +# now when we know there's enough free blocks.
> +later_file=$SCRATCH_MNT/later
> +touch $later_file
Make sense to me,
Reviewed-by: Zorro Lang <zlang@xxxxxxxxxx>
> +
> # this file will get removed to create 256k of free space after ENOSPC
> # conditions are created.
> dd if=/dev/zero of=$SCRATCH_MNT/tmp1 bs=256K count=1 >>$seqres.full 2>&1
> @@ -63,12 +72,12 @@ _freespace=`$DF_PROG -k $SCRATCH_MNT | tail -n 1 | awk '{print $5}'`
>
> # Try to write more than available space in chunks that will allow at least one
> # full write to succeed.
> -dd if=/dev/zero of=$SCRATCH_MNT/tmp1 bs=128k count=8 >>$seqres.full 2>&1
> +dd if=/dev/zero of=$later_file bs=128k count=8 >>$seqres.full 2>&1
> echo "Bytes written until ENOSPC:" >>$seqres.full
> -du $SCRATCH_MNT/tmp1 >>$seqres.full
> +du $later_file >>$seqres.full
>
> # And at least some of it should succeed.
> -_filesize=`_get_filesize $SCRATCH_MNT/tmp1`
> +_filesize=`_get_filesize $later_file`
> [ $_filesize -lt $((128 * 1024)) ] && \
> _fail "Partial write until enospc failed; wrote $_filesize bytes."
>
>
