JFFS2 error messages

[C. Scott Ananian]

2008/6/3 Bill Mccormick <billmcc at nortel.com>:
> A couple of my XOs are reporting what look like FS error messages on boot:
>
> [91.463670] JFFS2 notice:  (664) check_node_data:  wrong data CRC in data
> node at 0x1ec215f0: read 0x3e7c7e03, calculated 0xf7e1d50c
> ...
>
> is this a known problem?

According to Dave Woodhouse, who wrote JFFS2, these notices typically
mean that at some point you've hard powered off, and as a result JFFS
has some uncommitted data lying around on flash.  It's almost always
harmless, a part of the journal which was never committed: "either it
was a new write which hadn't yet been synced, or it was a GC write
which just doesn't achieve anything now."  However, these messages
*could* indicate an actual problem -- "we never came up with a good
heuristic for when _not_ to complain".  Woodhouse suggests that in the
future "perhaps we should write a 'yes, I know there's a CRC failure'
node _after_ the offending node, when we reboot and find it" since
directly rewriting the node is not an option due to the mechanics of
NAND flash; that would help confine these messages to immediately
after a hard reboot.  At present, you'll keep seeing a "bad CRC"
message every time that particular JFFS node is accessed until it is
eventually GC'ed.

Anyone want to volunteer to turn the above into a FAQ which would be
discoverable on our wiki, so that future wonderers don't have to pry
the information from the head of dwmw2?
 --scott

-- 
 ( http://cscott.net/ )