[U-Boot] UBIFS fails on SheevaPlug

[Vikram Narayanan <vikram186@gmail.com>]

Hello Andreas,

On 10/31/2012 10:20 PM, Andreas Bie?mann wrote:
> Dear Vikram Narayanan,
>
> On 31.10.2012 17:27, Vikram Narayanan wrote:
>> Hello Andreas,
>>
>> On 10/30/2012 4:32 AM, Andreas Bie?mann wrote:
>>> Dear Vikram Narayanan,
>>>
>>> first of all you are right. u-boot ubifs implementation will never
>>> recover the ubifs on media, cause it is mounted read only.
>>>
>>
>> <snip>
>>
>>> calls sget() (line 1043)
>>> which in turn calls kzalloc() (line 67)
>>> which may return -ENOMEM
>>
>> I agree. But in Dimax's case this isn't. Right?
>
> I dunno cause he has an old version which do not print the error code
> (patch f75325e1927398f7e18e15f186b1fc52174cc19f is missing).
>
>>> But u-boot will manage to get the data out of the unordered ubifs (if no
>>> error like this ENOMEM occur). That is the same process as in kernel if
>>> it is mounted read-only (recovery deferred).
>>
>> I can't comment on this, unless I know the specifics.
>
> Please read mount_ubifs() in fs/ubifs/super.c:582
>
> Even if it is mounted read only and recovery is needed it will
> ubifs_lpt_init(), ubifs_replay_journal(), ubifs_mount_orphans(), ...
> If recovery is needed it will ubifs_recover_size() and if it is mounted
> read only it will print at last 'recovery deferred' (in the other case
> it would be still repaired here cause the steps before do it).
> So at least a read only fs which needs recovery can be successfully
> mounted (if no other error occurs; the most likely error is a ENOMEM
> cause the ubifs does a lot of k(z)alloc).

Thanks. I'll look at this.

>>> So if the kernel can manage to mount the unordered ubifs u-boot should
>>> do so. If it can't (but the kernel can) there is an error that should be
>>> fixed.
>>>
>>
>> But in the kernel, the read-only isn't hardcoded. So, the kernel code
>> can try to recover and even update the corrupted data back to the media
>> and mount it. (It's my guess. The kernel may/mayn't do this way).
>
> Well bootargs can have 'ro' so the root partition will be mounted read
> only, wouldn't it? Wouldn't this 'ro' root partition not mounted by the
> kernel then?

Got your point here. Then no feature could be claimed as 'missing' in 
the u-boot's ubifs compared to the kernel's, except for the well known 
fact that u-boot's ubi layer is old code.

Regards,
Vikram