[LU-10837] no bitmap check if block bitmap is uninitialized

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.12.0, Lustre 2.10.4
• Labels: patch
• Severity: 3
• Assignee: Wang Shilong (Inactive)
• Reporter: Wang Shilong (Inactive)

Description

Following commit:

LU-9410 ldiskfs: no check mb bitmap if flex_bg enabled

Tried to expose a problem that we should be careful with block bitmap uninitialized.

However, it only tried to skip 0 in @free_gdp, at some customer site, we observed

cases

Mar 13 13:59:29 oss2c105 kernel: LDISKFS-fs error (device sfa0002): ldiskfs_mb_check_ondisk_bitmap:3605: comm ll_ost_io01_057: on-disk bitmap for group 420934corrupted: 32768 blocks free in bitmap, 559 - in gd\x0a
Mar 13 13:59:29 oss2c105 kernel: LDISKFS-fs error (device sfa0002): ldiskfs_mb_check_ondisk_bitmap:3605: comm ll_ost_io00_056: on-disk bitmap for group 420938corrupted: 32768 blocks free in bitmap, 680 - in gd\x0a

And also, I did a very simple reformat and tried to dumpe2fs OST:

...

Group 158: (Blocks 5177344-5210111) [INODE_UNINIT, BLOCK_UNINIT, ITABLE_ZEROED]

  Checksum 0x0617, unused inodes 1920

  Block bitmap at 1182 (bg #0 + 1182), Inode bitmap at 1342 (bg #0 + 1342)

  Inode table at 40512-40751 (bg #1 + 7744)

  32768 free blocks, 1920 free inodes, 0 directories, 1920 unused inodes

  Free blocks: 5177344-5210111

  Free inodes: 303361-305280

Group 159: (Blocks 5210112-5242

....

With more check, I did not see any 0 counters...

See comments in ext4_free_clusters_after_init:

    /* Return the number of free blocks in a block group.  It is used when

     * the block bitmap is uninitialized, so we can't just count the bits

     * in the bitmap. */

    So extra check we enhanced here is wrong if this block group

    bitmap is uninitialized, since we only check bitmaps here.

 

    Further, Looking at EXT4_BG_BLOCK_UNINIT clear codes, Kernel

    will reinit free_clusters_count when tried to clear the flag, so

    extra check for uninited block bitmaps dosen't make much sense.

 

    Let's skip uninited block bitmap check if EXT4_BG_BLOCK_UNINIT

    is set, whatever free count group desc recorded is untrustable somehow.

Issue Links

• is related to: LU-9410 on-disk bitmap corrupted