UBIFS file-system corruption (missing inode) after power-cut on 4.14.96

[Lars Persson <lists@xxxxxxx>]

Hi

We have recently encountered multiple cases of corrupted UBIFS volumes that
were triggered by a power-cut during startup. It may be a regression in the
4.14 stable branch.

The symptom is seen when accessing a file in the corrupted FS:
  UBIFS error (ubi0:20 pid 513): ubifs_iget: failed to read inode 348, error -2
  UBIFS error (ubi0:20 pid 513): ubifs_lookup: dead directory entry
'tampering.conf', error -2
  UBIFS warning (ubi0:20 pid 513): ubifs_ro_mode.part.0: switched to
read-only mode, error -2
  [<80506a51>] (dump_stack) from [<80297e2f>] (ubifs_lookup+0x29b/0x300)
  [<80297e2f>] (ubifs_lookup) from [<80226169>] (lookup_slow+0x69/0xe4)

We enabled chk_fs and reproduced it:
  UBIFS error (ubi0:20 pid 120): read_add_inode: inode 352 not found in index
  UBIFS error (ubi0:20 pid 120): check_leaf: error -2 while processing
entry node and trying to find inode node 352
  UBIFS (ubi0:20): dump of node at LEB 29:94656
        magic          0x6101831
        crc            0xda5d6bee
        node_type      2 (direntry node)
        group_type     1 (in node group)
        sqnum          4079
        len            70
        key            (66, direntry, 0x4974b0f)
        inum           352
        type           0
        nlen           13
        name           emotiond.conf
  UBIFS error (ubi0:20 pid 120): dbg_walk_index: leaf checking
function returned error -2, for leaf at LEB 29:94656
  UBIFS (ubi0:20): dump of znode at LEB 37:57680
  znode be0bf000, LEB 37:57680 len 128 parent be0af800 iip 1 level 0
child_cnt 5 flags 0
  zbranches:
        0: LNC   (null) LEB 32:53248 len 81 key (66, direntry, 0x363e97f)
        1: LNC   (null) LEB 29:94208 len 63 key (66, direntry, 0x3f88c71)
        2: LNC   (null) LEB 12:92160 len 74 key (66, direntry, 0x477308e)
        3: LNC   (null) LEB 29:94656 len 70 key (66, direntry, 0x4974b0f)
        4: LNC be084600 LEB 814:49152 len 63 key (66, direntry, 0x49bb6f7)
  UBIFS error (ubi0:20 pid 120): dbg_check_filesystem: file-system
check failed with error -2

The setup of our system is:
An overlayfs stack for /etc with:
- The lower file-system is a read-only squashfs on ubi
- The upper file-system is an ubifs

The ubi partition resides on an SLC NAND from Toshiba (TH58NVG2S3HBAI4).

The missing inode is always triggered on those two particular files
emotiond.conf and tampering.conf that share the same write pattern at startup:

 cp file file.tmp
 echo some data > file.tmp
 mv file.tmp file
 fsync file

Do not ask me about the logic of this script :> It overwrites the result of cp
and does not implement a proper atomic move.. Anyway the end result must not be
a file-system that is corrupt and mounts as RO.

BR,
 Lars

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/