On Tue, Apr 15, 2025 at 05:38:22PM +0800, Chao Yu wrote:
> As Han Qi reported in bugzilla [1], commit 5c26d2f1d3f5 ("unicode: Don't
> special case ignorable code points") changes logic to handle ignorable
> code points, when we handle filename which has special unicode charactor
> like red heart ❤️, utf8_casefold() may return different length than in the
> past, result in that f2fs cacluates hash code incorreclty, eventually,
> filename w/ special unicode charactor will not be found w/ the wrong hash
> code.
>
> f2fs has introduces a linear lookup fallback w/ commit 91b587ba79e1
> ("f2fs: Introduce linear search for dentries"), once hash-based lookup
> fails, it tries linear lookup as a fallback.
>
> This patch can help to check whether linear lookup fallback works well
> or not.
>
> Firstly, we use below testcase to check how hash code changes:
>
> mkfs.f2fs -f -O casefold -C utf8 /dev/vda
> mount /dev/vda /mnt/f2fs
> mkdir /mnt/f2fs/dir/
> f2fs_io setflags casefold /mnt/f2fs/dir
> touch /mnt/f2fs/dir/$'\u2764\ufe0f'
> umount /mnt/f2fs
> dump.f2fs -i 4 -d 3 /dev/vda
> dump.f2fs -b $blkaddr_of_dir -d 3 /dev/vda
>
> w/o ("unicode: Don't special case ignorable code points")
> [dump_dirent: 991] bitmap pos[0x2] name[❤️] len[0x6] hash[0x19dd7132]
> ino[0x5] type[0x1]
> [dump_dirent: 998] name(hex)[0xe2 0x9d 0xa4 0xef 0xb8 0x8f 0x0]
>
> w/ ("unicode: Don't special case ignorable code points")
> [dump_dirent: 991] bitmap pos[0x2] name[❤️] len[0x6] hash[0x9a2ea068]
> ino[0x5] type[0x1]
> [dump_dirent: 998] name(hex)[0xe2 0x9d 0xa4 0xef 0xb8 0x8f 0x0]
>
> Correct hash code: 0x19dd7132
> Wrong hash code: 0x9a2ea068
>
> So, we can test as below:
>
> 1. create file w/ red heart as its filename
> 2. inject wrong hash code to the file
> 3. disable linear lookup, expect lookup failure
> 4. enable linear lookup, expect lookup succeed
>
> [1] https://bugzilla.kernel.org/show_bug.cgi?id=219586
>
> Cc: Jaegeuk Kim <[email protected]>
> Cc: Daniel Lee <[email protected]>
> Signed-off-by: Chao Yu <[email protected]>
> ---
Sorry for this late review, this verssion looks good to me, just one review
point
as below ..
> v2:
> - tune linear_lookup feature before fault injection
> - check status after tuning linear_lookup feature
> - clean up codes
> tests/f2fs/012 | 73 ++++++++++++++++++++++++++++++++++++++++++++++
> tests/f2fs/012.out | 2 ++
> 2 files changed, 75 insertions(+)
> create mode 100755 tests/f2fs/012
> create mode 100644 tests/f2fs/012.out
>
> diff --git a/tests/f2fs/012 b/tests/f2fs/012
> new file mode 100755
> index 00000000..38a63f2e
> --- /dev/null
> +++ b/tests/f2fs/012
> @@ -0,0 +1,73 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2025 Chao Yu. All Rights Reserved.
> +#
> +# FS QA Test No. f2fs/012
> +#
> +# This testcase checks whether linear lookup fallback works well
> +# or not as below:
> +# 1.create file w/ red heart as its filename
> +# 2.inject wrong hash code to the file
> +# 3.disable linear lookup, expect lookup failure
> +# 4.enable linear lookup, expect lookup succeed
> +#
> +. ./common/preamble
> +_begin_fstest auto quick casefold
> +
> +_fixed_by_kernel_commit 91b587ba79e1 \
> + "f2fs: Introduce linear search for dentries"
> +
> +export LC_ALL=C.UTF-8
> +_require_scratch
> +_require_command "$F2FS_IO_PROG" f2fs_io
> +_require_command "$F2FS_INJECT_PROG" inject.f2fs
> +
> +check_lookup()
> +{
> + local nolinear_lookup=$1
> + local dir=$SCRATCH_MNT/dir
> + # red heart charactor in unicode format
> + local redheart=$dir/$'\u2764\ufe0f'
> +
> + _scratch_mkfs -O casefold -C utf8 >> $seqres.full
> + _scratch_mount
> +
> + mkdir $dir
> + $F2FS_IO_PROG setflags casefold $dir >> $seqres.full
> + touch $redheart
> + ino=`stat -c '%i' $redheart`
> + _scratch_unmount
> +
> + # should tune nolinear_lookup feature before fault injection
> + $F2FS_FSCK_PROG --nolinear-lookup=$nolinear_lookup $SCRATCH_DEV >>
> $seqres.full
> +
> + # check whether linear_lookup is changed as expected
> + if [ $nolinear_lookup == "1" ]; then
> + $F2FS_FSCK_PROG $SCRATCH_DEV | grep -q "linear_lookup
> \[disable\]" || \
> + _fail "linear_lookup is not disabled"
> + else
> + $F2FS_FSCK_PROG $SCRATCH_DEV | grep -q "linear_lookup
> \[enable\]" || \
> + _fail "linear_lookup is not enabled"
> + fi
> +
> + $F2FS_INJECT_PROG --dent --mb d_hash --nid $ino --val 0x9a2ea068
> $SCRATCH_DEV >> $seqres.full
> +
> + _scratch_mount
> + if [ $nolinear_lookup == "1" ]; then
> + [ -f $redheart ] && _fail "red heart file should not exist"
> + else
> + [ -f $redheart ] || _fail "red heart file should exist"
> + fi
> + _scratch_unmount
> +
> + # to avoid _check_filesystems failure in the end of test
> + _repair_scratch_fs >> $seqres.full
If you want to make sure "fsck.f2fs can repair above corruption", then this
comment isn't right.
If you just want to avoid _check_filesystems failure, you can replace
_require_scratch with
_require_scratch_nocheck, then remove this line.
Thanks,
Zorro
> +}
> +
> +check_lookup 1
> +check_lookup 0
> +
> +echo "Silence is golden"
> +
> +status=0
> +exit
> diff --git a/tests/f2fs/012.out b/tests/f2fs/012.out
> new file mode 100644
> index 00000000..3afeca2f
> --- /dev/null
> +++ b/tests/f2fs/012.out
> @@ -0,0 +1,2 @@
> +QA output created by 012
> +Silence is golden
> --
> 2.49.0
>
_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
