Re: [PATCH] UBI: add ubi_err() to report the failure of leb read

[Richard Weinberger <richard@nod.at>]

Am 16.12.2014 um 10:52 schrieb hujianyang:
> On 2014/12/16 17:21, Richard Weinberger wrote:
>> Am 16.12.2014 um 09:02 schrieb hujianyang:
>>> Hi,
>>>
>>> I met a problem that I was failed to mount a UBIFS partition.
>>>
>>> [   38.442770] UBI error: ubi_io_read: error -74 (ECC error) while reading 26624 bytes from PEB 54:104448, read 26624 bytes
>>> [   38.852461] UBI error: ubi_io_read: error -74 (ECC error) while reading 77824 bytes from PEB 346:53248, read 77824 bytes
>>> [   38.864142] UBIFS error (pid 1444): ubifs_recover_leb: corruption -3
>>> [   38.870487] UBIFS error (pid 1444): ubifs_scanned_corruption: corruption at LEB 928:55280
>>> [   38.878625] UBIFS error (pid 1444): ubifs_scanned_corruption: first 8192 bytes from LEB 928:55280
>>> [   38.892117] UBIFS error (pid 1444): ubifs_recover_leb: LEB 928 scanning failed
>>> mount: mounting ubi1:bak on /HFFS2: failed: Structure needs cleaning
>>>
>>> I think it is caused by an ECC error of nand flash. Do we have some methods
>>> to mount this partition? Data losing is acceptable.
>>
>> I don't think that UBIFS has such a mount option.
> 
> Er, I don't know it either. How about a mount option like --force?

Then every single embedded vendor will use this flag to keep the broken MTD/UBI/UBIFS setups running
as long as possible no mater of how corrupted the data is. :-)
IIRC UBIFS will either mount and work correctly as expected or fail hard.

>> You can dump the raw data and inspect the corrupted data.
>> Maybe you can fix it by hand.
> 
> Yes, I want a try~! If we have to introduce a new feature or new mount
> option. So would you like to help me? Do you think it's a valuable
> work?

I'm not a fan of such a mount option.
What we really need is a fsck.ubifs and a ubifs dump tool to fix and recover
broken UBIFS images.

Thanks,
//richard