-------- Original Message --------
Subject: Re: [XFSTEST Add Fallocate Punch Hole Tests 3/3 v4] Add ENOSPC Hole Punch Test
Date: Wed, 01 Jun 2011 16:02:10 -0700
From: Allison Henderson <[email protected]>
To: xfs-oss <[email protected]>, Dave Chinner <[email protected]>
On 5/25/2011 7:34 PM, Allison Henderson wrote:
> This patch adds a test to 252 that tests that a hole can be punched even when the
> disk is full. Reserved blocks should be used to allow a punch hole to proceed even
> when there is not enough blocks to further fragment the file. To test this, the
> file system is fragmented by punching holes in regular intervals and filling
> the file system between punches. This will eventually force the file system to use
> reserved blocks to proceed with the punch hole operation.
>
> Signed-off-by: Allison Henderson<[email protected]>
> ---
> :100755 100755 5efa243... b5204fe... M 252
> :100644 100644 ddf63b0... fc6123c... M common.punch
> 252 | 12 +++++++
> common.punch | 95 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> 2 files changed, 107 insertions(+), 0 deletions(-)
>
> diff --git a/252 b/252
> index 5efa243..b5204fe 100755
> --- a/252
> +++ b/252
> @@ -49,6 +49,7 @@ _supported_os Linux
>
> _require_xfs_io_falloc_punch
> _require_xfs_io_fiemap
> +_require_scratch
>
> testfile=$TEST_DIR/252.$$
>
> @@ -64,4 +65,15 @@ _test_generic_punch -k falloc fpunch fpunch fiemap _filter_fiemap $testfile -F
> # Delayed allocation multi punch hole tests
> _test_generic_punch -d -k falloc fpunch fpunch fiemap _filter_fiemap $testfile -F
>
> +# Test full filesystem hole punching.
> +# Make a small file system to fill
> +umount $SCRATCH_DEV&> /dev/null
> +_scratch_mkfs_sized $(( 1024 * 1024 * 1024 ))&> /dev/null
> +_scratch_mount
> +# Test must be able to write files with non-root permissions
> +chmod 777 $SCRATCH_MNT
> +
> +block_size=`stat -f $SCRATCH_DEV | grep "Block size" | cut -d " " -f3`
> +_test_full_fs_punch $(( $block_size * 2 )) $block_size 500 $SCRATCH_MNT/252.$$
> +
> status=0 ; exit
> diff --git a/common.punch b/common.punch
> index ddf63b0..fc6123c 100644
> --- a/common.punch
> +++ b/common.punch
> @@ -481,5 +481,100 @@ _test_generic_punch()
> -c "$zero_cmd 128 128" \
> -c "$map_cmd -v" $testfile | $filter_cmd
> [ $? -ne 0 ]&& die_now
> +}
> +
> +# _fill_fs()
> +#
> +# Fills a file system by repeatedly creating files in the given folder
> +# starting with the given file size. Files are reduced in size when
> +# they can no longer fit untill no more files can be created.
> +#
> +# This routine is used by _test_full_fs_punch to test that a hole may
> +# still be punched when the disk is full by borrowing reserved blocks.
> +# All files are created as a non root user to prevent reserved blocks
> +# from being consumed.
> +#
> +_fill_fs() {
> + local file_size=$1
> + local dir=$2
> + local file_count=1
> +
> + if [ $# -ne 2 ]
> + then
> + echo "USAGE: $0 filesize dir"
> + exit 1
> + fi
> +
> + mkdir -p $dir&> /dev/null
> + if [[ $? != 0 ]] ; then
> + return 0
> + fi
> + chmod 777 $dir
> +
> + rc=0
> + while [ $file_size -gt 0 -a $rc == 0 ]
> + do
> + # This part must not be done as root or
> + # reserved blocks will be consumed
> + sudo -u nobody $XFS_IO_PROG -F -f -c "pwrite 0 $file_size" $dir/$file_count.bin&> /dev/null
Hi all,
This is the ENOSPC test that we used on the ext4 punch hole, but modified to use the xfsprogs facilities.
I notice the test takes a lot longer to run after doing this. If I replace the above command with the original code:
sudo -u nobody dd if=/dev/zero of=$dir/$file_count.bin bs=$file_size count=1 &> /dev/null
it runs a lot faster (takes off almost 15 minutes). Is there anything we can do to improve the xfsprogs command? Thx!
Allison Henderson
> + rc=$?
> +
> + # If there was no room to make the file,
> + # and the file size can still be made smaller,
> + # then divide it in half, and keep going
> + if [ $file_size -gt 1 -a $rc != 0 ]
> + then
> + file_size=$(( $file_size / 2 ))
> + rc=0
> + fi
> + file_count=$(( $file_count + 1 ))
> +
> + done
> +}
>
> +# _test_full_fs_punch()
> +#
> +# This function will test that a hole may be punched
> +# even when the file system is full. Reserved blocks
> +# should be used to allow a punch hole to proceed even
> +# when there is not enough blocks to further fragment the
> +# file. To test this, this function will fragment the file
> +# system by punching holes in regular intervals and filling
> +# the file system between punches.
> +#
> +_test_full_fs_punch()
> +{
> + hole_len=$1 # The length of the holes to punch
> + hole_interval=$2 # The interval between the holes
> + iterations=$3 # The number of holes to punch
> + file_name=$4 # File to punch holes in
> + file_len=$(( $(( $hole_len + $hole_interval )) * $iterations ))
> + path=`dirname $file_name`
> + hole_offset=0
> +
> + rm -f $file_name&> /dev/null
> +
> + $XFS_IO_PROG -F -f -c "pwrite 0 $file_len" \
> + -c "fsync" $file_name&> /dev/null
> + chmod 666 $file_name
> +
> + _fill_fs $(( 1024 * 1024 * 1024 )) $path/fill
> +
> + for (( i=0; i<$iterations; i++ ))
> + do
> + # This part must not be done as root in order to
> + # test that reserved blocks are used when needed
> + sudo -u nobody $XFS_IO_PROG -F -f -c "fpunch $hole_offset $hole_len" $file_name
> + rc=$?
> + if [[ $? != 0 ]] ; then
> + echo Punch hole failed
> + break
> + fi
> +
> + hole_offset=$(( $hole_offset + $hole_len + $hole_interval ))
> +
> + _fill_fs $hole_len $path/fill.$i
> +
> + done
> }
> +