on 2022/12/12 13:56, Ziyang Zhang wrote:
> Sometimes "$((128 * dblksz / 40))" dirents cannot make sure that
> S_IFDIR.FMT_BTREE could become btree format for its DATA fork.
>
> Actually we just observed it can fail after apply our inode
> extent-to-btree workaround. The root cause is that the kernel may be
> too good at allocating consecutive blocks so that the data fork is
> still in extents format.
>
> Therefore instead of using a fixed number, let's make sure the number
> of extents is large enough than (inode size - inode core size) /
> sizeof(xfs_bmbt_rec_t).
After this patch, xfs/083 and xfs/155 failed on my envrionment(6.1.0+
kernel).
the 083 fail as below:
1 fuzzing xfs with FUZZ_ARGS=-3 -n 32 and FSCK_PASSES=10
2 + create scratch fs
3 meta-data=/dev/sdb9 isize=512 agcount=4,
agsize=529878 blks
4 = sectsz=512 attr=2, projid32bit=1
5 = crc=1 finobt=1, sparse=1,
rmapbt=0
6 = reflink=0 bigtime=1
inobtcount=1 nrext64=0
7 data = bsize=4096 blocks=2119510,
imaxpct=25
8 = sunit=0 swidth=0 blks
9 naming =version 2 bsize=4096 ascii-ci=0, ftype=1
10 log =internal log bsize=4096 blocks=16384, version=2
11 = sectsz=512 sunit=0 blks,
lazy-count=1
12 realtime =none extsz=4096 blocks=0, rtextents=0
13 + populate fs image
14 MOUNT_OPTIONS = -o usrquota,grpquota,prjquota
15 + fill root ino chunk
16 + extents file
17 wrote 4096/4096 bytes at offset 0
18 4 KiB, 1 ops; 0.0187 sec (212.891 KiB/sec and 53.2226 ops/sec)
19 + btree extents file
20 wrote 2097152/2097152 bytes at offset 0
21 2 MiB, 2 ops; 0.0637 sec (31.370 MiB/sec and 31.3701 ops/sec)
22 + inline dir
23 + block dir
24 + leaf dir
25 + leafn dir
26 + node dir
27 + btree dir
28 + inline symlink
29 + extents symlink
30 + special
31 + local attr
32 + leaf attr
33 + node attr
34 + btree attr
35 + attr extents with a remote less-than-a-block value
36 + attr extents with a remote one-block value
37 + empty file
38 + freesp btree
39 wrote 4194304/4194304 bytes at offset 0
40 4 MiB, 4 ops; 0.0941 sec (42.470 MiB/sec and 42.4696 ops/sec)
41 + inobt btree
42 + real files
43 FILL FS
44 src_sz 2052 fs_sz 8342940 nr 203
45 failed to create ino 8578 dformat expected btree saw extents
46 failed to create ino 8578 dformat expected btree saw extents
47 (see /var/lib/xfstests/results//xfs/083.full for details)
It seems this logic can't ensure to creat a btree format dir and it
is a extent format dir. Or I miss something?
Best Regards
Yang Xu
>
> Reviewed-by: Zorro Lang <zlang@xxxxxxxxxx>
> Reviewed-by: Allison Henderson <allison.henderson@xxxxxxxxxx>
> Suggested-by: "Darrick J. Wong" <djwong@xxxxxxxxxx>
> Signed-off-by: Gao Xiang <hsiangkao@xxxxxxxxxxxxxxxxx>
> Signed-off-by: Ziyang Zhang <ZiyangZhang@xxxxxxxxxxxxxxxxx>
> ---
> common/populate | 34 +++++++++++++++++++++++++++++++++-
> common/xfs | 9 +++++++++
> 2 files changed, 42 insertions(+), 1 deletion(-)
>
> diff --git a/common/populate b/common/populate
> index 6e004997..8f7f2113 100644
> --- a/common/populate
> +++ b/common/populate
> @@ -71,6 +71,37 @@ __populate_create_dir() {
> done
> }
>
> +# Create a large directory and ensure that it's a btree format
> +__populate_xfs_create_btree_dir() {
> + local name="$1"
> + local isize="$2"
> + local missing="$3"
> + local icore_size="$(_xfs_get_inode_core_bytes $SCRATCH_MNT)"
> + # We need enough extents to guarantee that the data fork is in
> + # btree format. Cycling the mount to use xfs_db is too slow, so
> + # watch for when the extent count exceeds the space after the
> + # inode core.
> + local max_nextents="$(((isize - icore_size) / 16))"
> + local nr=0
> +
> + mkdir -p "${name}"
> + while true; do
> + local creat=mkdir
> + test "$((nr % 20))" -eq 0 && creat=touch
> + $creat "${name}/$(printf "%.08d" "$nr")"
> + if [ "$((nr % 40))" -eq 0 ]; then
> + local nextents="$(_xfs_get_fsxattr nextents $name)"
> + [ $nextents -gt $max_nextents ] && break
> + fi
> + nr=$((nr+1))
> + done
> +
> + test -z "${missing}" && return
> + seq 1 2 "${nr}" | while read d; do
> + rm -rf "${name}/$(printf "%.08d" "$d")"
> + done
> +}
> +
> # Add a bunch of attrs to a file
> __populate_create_attr() {
> name="$1"
> @@ -176,6 +207,7 @@ _scratch_xfs_populate() {
>
> blksz="$(stat -f -c '%s' "${SCRATCH_MNT}")"
> dblksz="$(_xfs_get_dir_blocksize "$SCRATCH_MNT")"
> + isize="$(_xfs_get_inode_size "$SCRATCH_MNT")"
> crc="$(_xfs_has_feature "$SCRATCH_MNT" crc -v)"
> if [ $crc -eq 1 ]; then
> leaf_hdr_size=64
> @@ -226,7 +258,7 @@ _scratch_xfs_populate() {
>
> # - BTREE
> echo "+ btree dir"
> - __populate_create_dir "${SCRATCH_MNT}/S_IFDIR.FMT_BTREE" "$((128 * dblksz / 40))" true
> + __populate_xfs_create_btree_dir "${SCRATCH_MNT}/S_IFDIR.FMT_BTREE" "$isize" true
>
> # Symlinks
> # - FMT_LOCAL
> diff --git a/common/xfs b/common/xfs
> index 674384a9..7aaa63c7 100644
> --- a/common/xfs
> +++ b/common/xfs
> @@ -1487,6 +1487,15 @@ _require_xfsrestore_xflag()
> _notrun 'xfsrestore does not support -x flag.'
> }
>
> +# Number of bytes reserved for a full inode record, which includes the
> +# immediate fork areas.
> +_xfs_get_inode_size()
> +{
> + local mntpoint="$1"
> +
> + $XFS_INFO_PROG "$mntpoint" | sed -n '/meta-data=.*isize/s/^.*isize=\([0-9]*\).*$/\1/p'
> +}
> +
> # Number of bytes reserved for only the inode record, excluding the
> # immediate fork areas.
> _xfs_get_inode_core_bytes()
