Re: [PATCH 04/30] xfs: mark inode buffers in cache

[Brian Foster <bfoster@redhat.com>]

On Wed, Jun 03, 2020 at 07:29:18AM +1000, Dave Chinner wrote:
> On Tue, Jun 02, 2020 at 12:45:35PM -0400, Brian Foster wrote:
> > On Tue, Jun 02, 2020 at 07:42:25AM +1000, Dave Chinner wrote:
> > > From: Dave Chinner <dchinner@redhat.com>
> > > 
> > > Inode buffers always have write IO callbacks, so by marking them
> > > directly we can avoid needing to attach ->b_iodone functions to
> > > them. This avoids an indirect call, and makes future modifications
> > > much simpler.
> > > 
> > > This is largely a rearrangement of the code at this point - no IO
> > > completion functionality changes at this point, just how the
> > > code is run is modified.
> > > 
> > 
> > Ok, I was initially thinking this patch looked incomplete in that we
> > continue to set ->b_iodone() on inode buffers even though we'd never
> > call it. Looking ahead, I see that the next few patches continue to
> > clean that up to eventually remove ->b_iodone(), so that addresses that.
> > 
> > My only other curiosity is that while there may not be any functional
> > difference, this technically changes callback behavior in that we set
> > the new flag in some contexts that don't currently attach anything to
> > the buffer, right? E.g., xfs_trans_inode_alloc_buf() sets the flag on
> > inode chunk init, which means we can write out an inode buffer without
> > any attached/flushed inodes.
> 
> Yes, it can happen, and it happens before this patch, too, because
> the AIL can push the buffer log item directly and that does not
> flush dirty inodes to the buffer before it writes back(*).
> 

I was thinking more about cases where there are actually no inodes
attached.

> As it is, xfs_buf_inode_iodone() on a buffer with no inode attached
> if functionally identical to the existing xfs_buf_iodone() callback
> that would otherwise be done. i.e. it just runs the buffer log item
> completion callback. Hence the change here rearranges code, but it
> does not change behaviour at all.
> 

Right. That's indicative from the code, but doesn't help me understand
why the change is made. That's all I'm asking for...

> (*) this is a double-write bug that this patch set does not address.
> i.e. buffer log item flushes the buffer without flushing inodes, IO
> compeletes, then inode flush to the buffer and we do another IO to
> clean them.  This is addressed by a follow-on patchset that tracks
> dirty inodes via ordered cluster buffers, such that pushing the
> buffer always triggers xfs_iflush_cluster() on buffers tagged
> _XBF_INODES...
> 

Ok, interesting (but seems beyond the scope of this series).

> > Is the intent of that to support future
> > changes? If so, a note about that in the commit log would be helpful.
> 
> That's part of it, as you can see from the (*) above. But the commit
> log already says "..., and makes future modifications much simpler."
> Was that insufficient to indicate that it will be used later on?
> 

That's a rather vague hint. ;P I was more hoping for something like:
"While this is largely a refactor of existing functionality, broaden the
scope of the flag to beyond where inodes are explicitly attached because
<some actual reason>. This has the effect of possibly invoking the
callback in cases where it wouldn't have been previously, but this is
not a functional change because the callback is effectively a no-op when
inodes are not attached."

Brian

> Cheers,
> 
> Dave.
> -- 
> Dave Chinner
> david@fromorbit.com
>