Re: [PATCH 2/7] Assign bdi in super_block

[Jens Axboe]

On Tue, Sep 15 2009, Jan Kara wrote:
> On Mon 14-09-09 20:36:54, Jens Axboe wrote:
> > On Mon, Sep 14 2009, Trond Myklebust wrote:
> > > On Mon, 2009-09-14 at 15:02 +0200, Jan Kara wrote:
> > > > On Mon 14-09-09 11:36:29, Jens Axboe wrote:
> > > > > We do this automatically in get_sb_bdev() from the set_bdev_super()
> > > > > callback. Filesystems that have their own private backing_dev_info
> > > > > must assign that in ->fill_super().
> > > > > 
> > > > > Note that ->s_bdi assignment is required for proper writeback!
> > > > > 
> > > > > Acked-by: Christoph Hellwig <hch@xxxxxxxxxxxxx>
> > > > > Signed-off-by: Jens Axboe <jens.axboe@xxxxxxxxxx>
> > > >   Hmm, looking at this again, I'm not sure this will work for NFS. It seems
> > > > to set mapping->backing_dev_info to its private backing dev info for
> > > > regular files while it leaves it intact for other inodes (e.g.
> > > > directories). I'm not sure why it does so but it seems its inodes end up on
> > > > two different BDI lists and thus they wouldn't be synced properly. Trond,
> > > > do I read the code properly?
> > > >   Also we definitely need to set *some* bdi in nfs_get_sb as otherwise sync
> > > > won't work for it.
> > > 
> > > There hasn't really been a need for a bdi in NFS other than for the
> > > regular file read and writeback code. The main reason for making it
> > > private was to ensure that we could set a per-superblock readahead limit
> > > that was a decent multiple of the server's preferred read block size.
> > > 
> > > Is there any reason why we couldn't set sb->s_bdi to point to that
> > > private bdi?
> > 
> > No, that should work fine. NFS already works fine with the bdi flusher
> > threads, so you should just point it at that bdi.
>   But will it really work well? I mean if we sync the superblock on the
> client, it will sync only the private BDI. So it won't sync any directory
> inodes because they are on the default_backing_dev_info (NFS leaves
> sb->s_bdev at NULL).

No you are right, it wont be complete. I talked to Chris about this
yesterday, and we agree that the best option is to give NFS the 'btrfs
treatment' bdi wise.

-- 
Jens Axboe

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/