On Wed, Mar 2, 2016 at 4:57 PM, Goldwyn Rodrigues <rgoldwyn@xxxxxxx> wrote:
> On 03/02/2016 08:43 AM, Miklos Szeredi wrote:
>> Not sure how any dentry seen by NFS became associated with overlayfs.
>> Through ->f_path by any chance?
>>
>
> Yes, NFS extracts dentry from filp->f_path.dentry in fs/nfs/inode.c
> nfs_open(). What can it use to evaluate dentry?
Commit 4bacc9c9234c ("overlayfs: Make f_path always point to the
overlay and f_inode to the underlay") broke this. Breakage only
affects regular files. Accessing file->f_path.dentry->d_name (and
"%pD" format etc) is OK for everything. ->d_fsdata is not OK.
I have no idea what the plan was with filesystems that use
f_path.dentry a separate open file for them. David? Al?
My plan was to introduce a file_dentry() helper that MUST be used by
filesystems to get the dentry from the file and that makes sure it's
the right one (check against file_inode()). If not, then we could
call into overlayfs to return the right one, similar to
->d_select_inode(), except we want to have a dentry and we want to
have *a particular dentry* matching file_inode() (the file could have
been copied up in the mean time).
Thanks,
Miklos
Thanks,
Miklos
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
