Re: [PATCH] nfs: fix congestion control

[Trond Myklebust <trond.myklebust@fys.uio.no>]

On Thu, 2007-01-18 at 14:27 +0100, Peter Zijlstra wrote:
> On Wed, 2007-01-17 at 16:54 -0500, Trond Myklebust wrote:
> > On Wed, 2007-01-17 at 22:52 +0100, Peter Zijlstra wrote:
> > 
> > > > 
> > > > > Index: linux-2.6-git/fs/inode.c
> > > > > ===================================================================
> > > > > --- linux-2.6-git.orig/fs/inode.c	2007-01-12 08:03:47.000000000 +0100
> > > > > +++ linux-2.6-git/fs/inode.c	2007-01-12 08:53:26.000000000 +0100
> > > > > @@ -81,6 +81,7 @@ static struct hlist_head *inode_hashtabl
> > > > >   * the i_state of an inode while it is in use..
> > > > >   */
> > > > >  DEFINE_SPINLOCK(inode_lock);
> > > > > +EXPORT_SYMBOL_GPL(inode_lock);
> > > > 
> > > > Hmmm... Commits to all NFS servers will be globally serialized via the 
> > > > inode_lock?
> > > 
> > > Hmm, right, thats not good indeed, I can pull the call to
> > > nfs_commit_list() out of that loop.
> > 
> > There is no reason to modify any of the commit stuff. Please just drop
> > that code.
> 
> I though you agreed to flushing commit pages when hitting the dirty page
> limit.

After the dirty page has been written to unstable storage, it marks the
inode using I_DIRTY_DATASYNC, which should then ensure that the VFS
calls write_inode() on the next pass through __sync_single_inode.

> Or would you rather see a notifier chain from congestion_wait() calling
> into the various NFS mounts?

I'd rather like to see fs/fs-writeback.c do this correctly (assuming
that it is incorrect now).

Trond