On Wed, 2008-07-09 at 18:07 -0700, Neil Brown wrote:

Dan:  I think this is your code.  In
  __handle_issuing_new_read_requests5
the
                } else if ((s->uptodate < disks - 1) &&
                        test_bit(R5_Insync, &dev->flags)) {

looks wrong.  We at least want a test on s->syncing in there, maybe:
                } else if (((s->uptodate < disks - 1) || s->syncing)
&&
                        test_bit(R5_Insync, &dev->flags)) {

and given that we only compute blocks when a device is failed, (see 15
lines earlier) I think we probably just want
                } else if (test_bit(R5_Insync, &dev->flags)) {

I notice that is was it in linux-next (though the functions are
renamed - it is fetch_block5 there).

Yes, I had realized it was obsolete... missed that it was buggy.

I wonder if there is still time for 2.6.26 .. probably not.  It'll be
released immediately after lwn.net release their weekly edition :-)

Here is a patch against latest mainline.

---snip--->
md: ensure all blocks are uptodate or locked when syncing

From: Dan Williams <redacted>

Remove the dubious attempt to prefer 'compute' over 'read'.  Not only is it
wrong given commit c337869d (md: do not compute parity unless it is on a failed
drive), but it can trigger a BUG_ON in handle_parity_checks5().

Cc: <stable@kernel.org>
Signed-off-by: Dan Williams <redacted>
---

 drivers/md/raid5.c |    7 +------
 1 files changed, 1 insertions(+), 6 deletions(-)

diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index 54c8ee2..3b27df5 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c

@@ -2017,12 +2017,7 @@ static int __handle_issuing_new_read_requests5(struct stripe_head *sh,
 			 */
 			s->uptodate++;
 			return 0; /* uptodate + compute == disks */
-		} else if ((s->uptodate < disks - 1) &&
-			test_bit(R5_Insync, &dev->flags)) {
-			/* Note: we hold off compute operations while checks are
-			 * in flight, but we still prefer 'compute' over 'read'
-			 * hence we only read if (uptodate < * disks-1)
-			 */
+		} else if (test_bit(R5_Insync, &dev->flags)) {
 			set_bit(R5_LOCKED, &dev->flags);
 			set_bit(R5_Wantread, &dev->flags);
 			if (!test_and_set_bit(STRIPE_OP_IO, &sh->ops.pending))