xfs
[Top] [All Lists]

Re: [PATCH 13/19] xfs: vectorise directory data operations

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 13/19] xfs: vectorise directory data operations
From: Ben Myers <bpm@xxxxxxx>
Date: Thu, 24 Oct 2013 16:41:12 -0500
Cc: xfs@xxxxxxxxxxx
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <20131024213117.GX2797@dastard>
References: <1381789085-21923-1-git-send-email-david@xxxxxxxxxxxxx> <1381789085-21923-14-git-send-email-david@xxxxxxxxxxxxx> <20131024183909.GV1935@xxxxxxx> <20131024213117.GX2797@dastard>
User-agent: Mutt/1.5.20 (2009-06-14)
On Fri, Oct 25, 2013 at 08:31:17AM +1100, Dave Chinner wrote:
> On Thu, Oct 24, 2013 at 01:39:09PM -0500, Ben Myers wrote:
> > On Tue, Oct 15, 2013 at 09:17:59AM +1100, Dave Chinner wrote:
> > > From: Dave Chinner <dchinner@xxxxxxxxxx>
> > > 
> > > Following from the initial patches to vectorise the shortform
> > > directory encode/decode operations, convert half the data block
> > > operations to use the vector. The rest will be done in a second
> > > patch.
> > > 
> > > This further reduces the size of the built binary:
> > > 
> > >    text    data     bss     dec     hex filename
> > >  794490   96802    1096  892388   d9de4 fs/xfs/xfs.o.orig
> > >  792986   96802    1096  890884   d9804 fs/xfs/xfs.o.p1
> > >  792350   96802    1096  890248   d9588 fs/xfs/xfs.o.p2
> > >  789293   96802    1096  887191   d8997 fs/xfs/xfs.o.p3
> > > 
> > > Signed-off-by: Dave Chinner <dchinner@xxxxxxxxxx>
> > > Reviewed-by: Christoph Hellwig <hch@xxxxxx>
> > 
> > Generally looks pretty good, I have a question below...
> > 
> > >  const struct xfs_dir_ops xfs_dir2_ftype_ops = {
> > > @@ -223,6 +415,18 @@ const struct xfs_dir_ops xfs_dir2_ftype_ops = {
> > >   .sf_put_ino = xfs_dir3_sfe_put_ino,
> > >   .sf_get_parent_ino = xfs_dir2_sf_get_parent_ino,
> > >   .sf_put_parent_ino = xfs_dir2_sf_put_parent_ino,
> > > +
> > > + .data_entsize = xfs_dir3_data_entsize,
> > > + .data_get_ftype = xfs_dir3_data_get_ftype,
> > > + .data_put_ftype = xfs_dir3_data_put_ftype,
> > > + .data_entry_tag_p = xfs_dir3_data_entry_tag_p,
> > > +
> > > + .data_dot_offset = xfs_dir2_data_dot_offset,
> > > + .data_dotdot_offset = xfs_dir2_data_dotdot_offset,
> > > + .data_first_offset = xfs_dir2_data_first_offset,
> > > + .data_dot_entry_p = xfs_dir2_data_dot_entry_p,
> > > + .data_dotdot_entry_p = xfs_dir2_data_dotdot_entry_p,
> > > + .data_first_entry_p = xfs_dir2_data_first_entry_p,
> > >  };
> > 
> > I think there may be a problem here.  Although the dirv2 functions for
> > ., .., and first entry offset account for the v2 header size, they
> > appear not to be accounting for the modified entry size due to the file
> > type field.  Am I missing something?
> 
> The ftype field is handled by the alignment roundup. i.e. namelen is
> 1 or two bytes, plus ftype is 2 or 3 bytes, roundup is to 8 bytes.
> Hence adding a byte for the ftype field is not a problem for these
> first entries because of their small, fixed size.

It should either be explicitly correct (and I think it is today), or we need a
comment to explain why it's not.  I would prefer the former.

<Prev in Thread] Current Thread [Next in Thread>