Re: Status of buffered write path (deadlock fixes)

[Steven Whitehouse <swhiteho@redhat.com>]

Hi,

On Tue, 2006-12-12 at 03:39 +1100, Nick Piggin wrote:
> Steven Whitehouse wrote:
> 
> >>Hmm, doesn't look like we can do this either because at least GFS2
> >>uses BH_New for its own special things.
> >>
> > 
> > What makes you say that? As far as I know we are not doing anything we
> > shouldn't with this flag, and if we are, then I'm quite happy to
> > consider fixing it up so that we don't,
> 
> Bad wording. Many other filesystems seem to only make use of buffer_new
> between prepare and commit_write.
> 
> gfs2 seems to at least test it in a lot of places, so it is hard to know
> whether we can change the current semantics or not. I didn't mean that
> gfs2 is doing anything wrong.
> 
> So can we clear it in commit_write?
> 
Are you perhaps looking at an older copy of the GFS2 code? We set it (or
clear it) in bmap.c:gfs2_block_map() as appropriate. It also seems to be
cleared when we release buffers (which may or may not be actually
required. I suspect not, but its harmless anyway). There is only one
test that I can find for it which is in bmap.c:gfs2_extent_map() where
its value is then later ignored in the only caller relevant to "normal"
files, which is ops_vm.c:alloc_page_backing(). So I think you should be
quite safe to clear it in commit_write().

Steve.