ubifs_jnl_update and extended attribute inode removal

ubifs_jnl_update and extended attribute inode removal

[Subodh Nijsure]

Hello,

Probably not many users of extended attributes on ubifs. I am trying to 
add this feature (patch v3) coming soon.

(xattr == extended attribute)

I have written updated the integck utility and discovered that sometimes 
files have wrong extended attributes. This happens only when some files 
are deleted from the filesystem.

Easy way to describe this is: I create 100 files, assign then extended 
attribute, and remember what the xattr was, it always stays the same. 
However now if I delete some of these files randomly some of the files 
end up having wrong xattr.

So  I have been trying to see how UBIFS handles file removal. I see that 
ubifs_unlink() calls ubifs_jnl_update() as:

         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 1, 0);

However ubifs_jnl_update never removes the  inode that is created to 
store the xattr associated , when reference drops to zero. I certainly 
know how to get access to inode that holds the xattr, what would be the 
right place where I can delete this inode also.  Or does the UBIFS has 
some other logic that will cleanup the xattr orphan inode at later time?

Note, there are implementation of remove_xattr() and that works, but 
that function is only called when some user space program explicitly 
removes proceeds. I am just trying to get some advice from folks who 
understand tnc management far better than I do. Would appreciate the 
help/pointer.

-Subodh

-Subodh

Re: ubifs_jnl_update and extended attribute inode removal

[Artem Bityutskiy]

[-- Attachment #1: Type: text/plain, Size: 861 bytes --]

On Wed, 2012-05-09 at 20:19 -0700, Subodh Nijsure wrote:
> Easy way to describe this is: I create 100 files, assign then extended 
> attribute, and remember what the xattr was, it always stays the same. 
> However now if I delete some of these files randomly some of the files 
> end up having wrong xattr.

I guess this happens only when you do a power cut? I need to look
closer, but quick feed-back is that there is probably a bug in the
journal reply - when we see a deleted inode in the journal - we remove
it from TNC, and we probably forget to look-up for all its xattrs?
Really need to look closer.

Would you prepare a description how I could reproduce this?

Please, go ahead with submitting your older patches, do not cook them
forever in your tree. Then you may start working on other issues.

-- 
Best Regards,
Artem Bityutskiy

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: ubifs_jnl_update and extended attribute inode removal

[Subodh Nijsure]

On 05/09/2012 09:31 PM, Artem Bityutskiy wrote:
> On Wed, 2012-05-09 at 20:19 -0700, Subodh Nijsure wrote:
>> Easy way to describe this is: I create 100 files, assign then extended
>> attribute, and remember what the xattr was, it always stays the same.
>> However now if I delete some of these files randomly some of the files
>> end up having wrong xattr.
> I guess this happens only when you do a power cut? I need to look
> closer, but quick feed-back is that there is probably a bug in the
> journal reply - when we see a deleted inode in the journal - we remove
> it from TNC, and we probably forget to look-up for all its xattrs?
> Really need to look closer.
>
> Would you prepare a description how I could reproduce this?
I re-submitted patches for integck and as well as v3 of extended 
attribute patch set that I have been holding on to.

You can just run integck -p -n 2 <dir-name> and that should  reproduce 
the issue i.e. integck will fail.

If email thread gets too long I could get on the OFTC #mtd irc chat at 
time that works best, to get some guidance.

/Subodh

Re: ubifs_jnl_update and extended attribute inode removal

[Artem Bityutskiy]

[-- Attachment #1: Type: text/plain, Size: 352 bytes --]

On Thu, 2012-05-10 at 16:22 -0700, Subodh Nijsure wrote:
> > Would you prepare a description how I could reproduce this?
> I re-submitted patches for integck and as well as v3 of extended 
> attribute patch set that I have been holding on to.

Hi, did I miss some patches from you or you just disappeared?

-- 
Best Regards,
Artem Bityutskiy

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]