| Message ID | 20210707104745.98822-1-jefflexu@linux.alibaba.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [RFC] misc: skip fsx tests when op length not congruent with file allocation unit | expand |
On Wed, Jul 07, 2021 at 06:47:45PM +0800, Jeffle Xu wrote: > > Yes it works. For example in this RFC patch, generic/075 could be > skipped once 'bigalloc' is detected for ext4. > > I could fix all related test cases in this way, if it is acceptable to > the community. Looks fine to me. > > Signed-off-by: Jeffle Xu <jefflexu@linux.alibaba.com> > --- > common/rc | 20 ++++++++++++++++++++ > tests/generic/075 | 3 +++ > 2 files changed, 23 insertions(+) > > diff --git a/common/rc b/common/rc > index f58a542..92245e0 100644 > --- a/common/rc > +++ b/common/rc > @@ -4041,6 +4041,23 @@ _sysfs_dev() > min=$(echo "ibase=16; $min" | bc) > echo /sys/dev/block/$maj:$min > } > + > +_ext4_get_file_block_size() Another ext4 specific helper in common/rc, perhaps it's time to create its own file for ext4, i.e. common/ext4, and move all ext4 specific helpers in common/rc there. > +{ > + local path="$1" > + path="$(readlink -m "$path")" > + local dev=`$DF_PROG $dir | tail -1 | $AWK_PROG '{print $1}'` $dir is not defined, you mean $path here? > + > + # The alloc unit size equals block size for ext4 without 'bigalloc'. > + if ! (tune2fs -l $dev | grep -q 'Cluster size'); then $TUNE2FS_PROG Perhaps you could call tune2fs only once and save its "Cluster size" line for later use. Thanks, Eryu > + _get_block_size "$path" > + return > + fi > + > + # Otherwise, call tune2fs acquiring cluster size. > + tune2fs -l $dev | sed -n -e 's/^.*Cluster size:\s*\([0-9]*\).*$/\1/p' > + > +} > > # Get the minimum block size of a file. Usually this is the > # minimum fs block size, but some filesystems (ocfs2) do block > @@ -4059,6 +4076,9 @@ _get_file_block_size() > "xfs") > _xfs_get_file_block_size $1 > ;; > + "ext4") > + _ext4_get_file_block_size $1 > + ;; > *) > _get_block_size $1 > ;; > diff --git a/tests/generic/075 b/tests/generic/075 > index 7467bb7..d269a7c 100755 > --- a/tests/generic/075 > +++ b/tests/generic/075 > @@ -109,6 +109,9 @@ _process_args() > _supported_fs generic > _require_test > > +blksz=$(_get_block_size $TEST_DIR) > +_require_congruent_file_oplen $TEST_DIR $blksz > + > size10=`expr 10 \* 1024 \* 1024` # 10 megabytes > filelen=$size10 > numops1=1000 > -- > 1.8.3.1
diff --git a/common/rc b/common/rc index f58a542..92245e0 100644 --- a/common/rc +++ b/common/rc @@ -4041,6 +4041,23 @@ _sysfs_dev() min=$(echo "ibase=16; $min" | bc) echo /sys/dev/block/$maj:$min } + +_ext4_get_file_block_size() +{ + local path="$1" + path="$(readlink -m "$path")" + local dev=`$DF_PROG $dir | tail -1 | $AWK_PROG '{print $1}'` + + # The alloc unit size equals block size for ext4 without 'bigalloc'. + if ! (tune2fs -l $dev | grep -q 'Cluster size'); then + _get_block_size "$path" + return + fi + + # Otherwise, call tune2fs acquiring cluster size. + tune2fs -l $dev | sed -n -e 's/^.*Cluster size:\s*\([0-9]*\).*$/\1/p' + +} # Get the minimum block size of a file. Usually this is the # minimum fs block size, but some filesystems (ocfs2) do block @@ -4059,6 +4076,9 @@ _get_file_block_size() "xfs") _xfs_get_file_block_size $1 ;; + "ext4") + _ext4_get_file_block_size $1 + ;; *) _get_block_size $1 ;; diff --git a/tests/generic/075 b/tests/generic/075 index 7467bb7..d269a7c 100755 --- a/tests/generic/075 +++ b/tests/generic/075 @@ -109,6 +109,9 @@ _process_args() _supported_fs generic _require_test +blksz=$(_get_block_size $TEST_DIR) +_require_congruent_file_oplen $TEST_DIR $blksz + size10=`expr 10 \* 1024 \* 1024` # 10 megabytes filelen=$size10 numops1=1000
Yes it works. For example in this RFC patch, generic/075 could be skipped once 'bigalloc' is detected for ext4. I could fix all related test cases in this way, if it is acceptable to the community. Signed-off-by: Jeffle Xu <jefflexu@linux.alibaba.com> --- common/rc | 20 ++++++++++++++++++++ tests/generic/075 | 3 +++ 2 files changed, 23 insertions(+)