UBIFS node magic number off by single zero-byte

[Goran Cengic]

Hi all,

I've encountered a peculiar issue with the UBIFS failing to mount because of an unclean node.
UBIFS complains that the node is unclean since it can't find the magic number,
however in the bytes dumped I can see that the magic number is there except that it is offset
by a single zero-byte. Here is the log:

[11996.131746] UBIFS error (pid 1635): ubifs_check_node: bad magic 0x10183100, expected 0x6101831
[11996.150772] UBIFS error (pid 1635): ubifs_check_node: bad node at LEB 22:4096
[11996.157958] Not a node, first 24 bytes:
[11996.161647] 00000000: 00 31 18 10 06 08 34 38 e5 cd 26 00 00 00 00 00 00 a0 00 00 00 00 02 00                          .1....48..&.............
[11996.174759] CPU: 0 PID: 1635 Comm: syslogd Tainted: G           O 3.10.62.R1A08-rt62 #1
[11996.182824] [] (unwind_backtrace+0x0/0xe0) from [] (show_stack+0x10/0x14)
[11996.191405] [] (show_stack+0x10/0x14) from [] (ubifs_check_node+0x258/0x280)
[11996.200245] [] (ubifs_check_node+0x258/0x280) from [] (ubifs_read_node+0x19c/0x26c)
[11996.209693] [] (ubifs_read_node+0x19c/0x26c) from [] (ubifs_tnc_read_node+0x44/0x11c)
[11996.219311] [] (ubifs_tnc_read_node+0x44/0x11c) from [] (ubifs_tnc_locate+0x100/0x18c)
[11996.229013] [] (ubifs_tnc_locate+0x100/0x18c) from [] (ubifs_iget+0x60/0x5b8)
[11996.237937] [] (ubifs_iget+0x60/0x5b8) from [] (ubifs_lookup+0x194/0x1c4)
[11996.246511] [] (ubifs_lookup+0x194/0x1c4) from [] (lookup_real+0x30/0x4c)
[11996.255079] [] (lookup_real+0x30/0x4c) from [] (do_last.isra.29+0x510/0xb28)
[11996.263913] [] (do_last.isra.29+0x510/0xb28) from [] (path_openat+0xa8/0x428)
[11996.272832] [] (path_openat+0xa8/0x428) from [] (do_filp_open+0x2c/0x78)
[11996.281317] [] (do_filp_open+0x2c/0x78) from [] (do_sys_open+0xf8/0x184)
[11996.289802] [] (do_sys_open+0xf8/0x184) from [] (ret_fast_syscall+0x0/0x30)
[11996.298543] UBIFS error (pid 1635): ubifs_read_node: expected node type 0
[11996.305365] UBIFS error (pid 1635): ubifs_iget: failed to read inode 94, error -117
[11996.313071] UBIFS error (pid 1635): ubifs_lookup: dead directory entry 'messages', error -117

I see this issue on kernel v3.10 and I've googled it but have come up with nothing. 
I'm unable to reproduce the issue deterministically but it shows up randomly on
a couple of different boards we run in our continuous integration setup and the symptom
is always exactly the same even though the failed process might be different.

I wonder if anyone on the list has seen anything similar?

With kind regards,

	Goran