Offering: HULK hulk inclusion category: bugfix bugzilla: 188996, https://gitee.com/openeuler/kernel/issues/I76JSK
--------------------------------
Our growfs test trigger mount error show as below:
XFS (dm-3): Starting recovery (logdev: internal) XFS (dm-3): Internal error !ino_ok at line 201 of file fs/xfs/libxfs/xfs_dir2.c. Caller xfs_dir_ino_validate+0x54/0xc0 [xfs] CPU: 0 PID: 3719345 Comm: mount Kdump: loaded Not tainted 5.10.0-136.12.0.86.h1036.kasan.eulerosv2r12.aarch64 #1 Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015 Call trace: dump_backtrace+0x0/0x3a4 show_stack+0x34/0x4c dump_stack+0x170/0x1dc xfs_corruption_error+0x104/0x11c [xfs] xfs_dir_ino_validate+0x8c/0xc0 [xfs] __xfs_dir3_data_check+0x5e4/0xb60 [xfs] xfs_dir3_block_verify+0x108/0x190 [xfs] xfs_dir3_block_read_verify+0x10c/0x184 [xfs] xlog_recover_buf_commit_pass2+0x388/0x8e0 [xfs] xlog_recover_items_pass2+0xc8/0x160 [xfs] xlog_recover_commit_trans+0x56c/0x58c [xfs] xlog_recovery_process_trans+0x174/0x180 [xfs] xlog_recover_process_ophdr+0x120/0x210 [xfs] xlog_recover_process_data+0xcc/0x1b4 [xfs] xlog_recover_process+0x124/0x25c [xfs] xlog_do_recovery_pass+0x534/0x864 [xfs] xlog_do_log_recovery+0x98/0xc4 [xfs] xlog_do_recover+0x64/0x2ec [xfs] xlog_recover+0x1c4/0x2f0 [xfs] xfs_log_mount+0x1b8/0x550 [xfs] xfs_mountfs+0x768/0xe40 [xfs] xfs_fc_fill_super+0xb54/0xeb0 [xfs] get_tree_bdev+0x240/0x3e0 xfs_fc_get_tree+0x30/0x40 [xfs] vfs_get_tree+0x5c/0x1a4 do_new_mount+0x1c8/0x220 path_mount+0x2a8/0x3f0 __arm64_sys_mount+0x1cc/0x220 el0_svc_common.constprop.0+0xc0/0x2c4 do_el0_svc+0xb4/0xec el0_svc+0x24/0x3c el0_sync_handler+0x160/0x164 el0_sync+0x160/0x180 XFS (dm-3): Corruption detected. Unmount and run xfs_repair XFS (dm-3): Invalid inode number 0xd00082 XFS (dm-3): Metadata corruption detected at __xfs_dir3_data_check+0xa08/0xb60 [xfs], xfs_dir3_block block 0x100070 XFS (dm-3): Unmount and run xfs_repair XFS (dm-3): First 128 bytes of corrupted metadata buffer: 00000000: 58 44 42 33 f3 c3 ae cf 00 00 00 00 00 10 00 70 XDB3...........p 00000010: 00 00 00 01 00 00 0c 3e 85 a6 68 6c 63 b3 42 30 .......>..hlc.B0 00000020: b0 6b d1 b5 9d eb 55 7d 00 00 00 00 00 10 00 90 .k....U}........ 00000030: 03 60 0b 78 01 40 00 40 00 b0 00 40 00 00 00 00 .`.x.@.@...@.... 00000040: 00 00 00 00 00 10 00 90 01 2e 02 00 00 00 00 40 ...............@ 00000050: 00 00 00 00 00 10 00 81 02 2e 2e 02 00 00 00 50 ...............P 00000060: 00 00 00 00 00 10 00 91 02 66 65 01 00 00 00 60 .........fe....` 00000070: 00 00 00 00 00 10 00 92 02 66 66 01 00 00 00 70 .........ff....p
Consider the following log format, dir3 block has 2 items in the log, and the inode number recorded ondisk in diri3 block exceeds the current file system boundary. When replaying log items, it will skipping replay of first dir3 buffer log item due to the log item LSN being behind the ondisk buffer, but verification is still required. Since the superblock hasn't been replayed yet, the inode number in dir3 block exceeds the file system boundary and causes log recovery to fail.
log record: +---------------+----------------+--------------+-------------------+ | dir3 buf item | growfs sb item | inode item | dir3 buf item .. | +---------------+----------------+--------------+-------------------+ lsn X lsn X+A lsn X+A+B lsn X+A+B+C
metadata block: +-----------+-----------+------------+-----------+-----------+ | sb block | ... | dir3 block | ... | inodes | +-----------+-----------+------------+-----------+-----------+ lsn < X lsn X+A+B+C
Remove buffer read verify during log recovry pass2, clear buffer's XBF_DONE flag, so it can be verified in the next buf read after log recover.
Fixes: 22ed903eee23 ("xfs: verify buffer contents when we skip log replay") Signed-off-by: Long Li leo.lilong@huawei.com --- fs/xfs/xfs_buf_item_recover.c | 9 +++------ 1 file changed, 3 insertions(+), 6 deletions(-)
diff --git a/fs/xfs/xfs_buf_item_recover.c b/fs/xfs/xfs_buf_item_recover.c index 9cb79a494d06..d214e0f9cc09 100644 --- a/fs/xfs/xfs_buf_item_recover.c +++ b/fs/xfs/xfs_buf_item_recover.c @@ -950,13 +950,10 @@ xlog_recover_buf_commit_pass2(
/* * We're skipping replay of this buffer log item due to the log - * item LSN being behind the ondisk buffer. Verify the buffer - * contents since we aren't going to run the write verifier. + * item LSN being behind the ondisk buffer. clear XBF_DONE flag + * of the buffer to prevent buffer from being used without verify. */ - if (bp->b_ops) { - bp->b_ops->verify_read(bp); - error = bp->b_error; - } + bp->b_flags &= ~XBF_DONE; goto out_release; }