ntfs ->iput use - was: Re: [PATC] small VFS change for JFFS2

[Anton Altaparmakov <aia21@cam.ac.uk>]

Hi Christoph,

If you remember you complained about ntfs' "fishy use of iput"?

I think that David's explanation below is exactly the reason why I had
to do it IIRC...  So if the vfs is fixed so the below can _never_ happen
anymore then I believe it would be safe to do as you and Al suggest and
stop using iput in a "fishy" way in ntfs...

Best regards,

	Anton

On Mon, 2005-04-18 at 23:08 +1000, David Woodhouse wrote:
> On Mon, 2005-04-18 at 13:46 +0100, Christoph Hellwig wrote:
> > Any, this sounds like you'd want to use ilookup because you don't want
> > to read the inode in the cache anyway, right?
> 
> We use ilookup() in some circumstances -- if the inode has zero nlink
> and hence we definitely don't want to pull it back again if it's gone.
> 
> But sometimes we really do mean to use iget() to bring it into core. And
> it's in that case that I believe Artem has found the problem, because if
> I understand correctly he's still seeing two consecutive calls to
> read_inode() for the same inode, without a clear_inode() in between.
> 
> prune_icache() is removing the inode from i_hash at line 457 of inode.c,
> then being preempted when it drops the inode_lock at line 464, which is
> _before_ it calls dispose_list() to actually get rid of the inode(s) in
> question. So when iget() is called, the VFS ends up calling read_inode()
> again instead of waiting for the original inode to finish going away.

-- 
Anton Altaparmakov <aia21 at cam.ac.uk> (replace at with @)
Unix Support, Computing Service, University of Cambridge, CB2 3QH, UK
Linux NTFS maintainer / IRC: #ntfs on irc.freenode.net
WWW: http://linux-ntfs.sf.net/ & http://www-stu.christs.cam.ac.uk/~aia21/