On Sat, Nov 03, 2012 at 10:47:41AM +1100, Dave Chinner wrote:
> I think that's irrelevant here - there will *never* be an IO waiter
> at this point in time. This processing is in log buffer IO
> completion context, so the buffers are still pinned in memory. Hence
> anyone trying to do IO on it will be waiting in xfs_buf_wait_unpin()
> and never get to xfs_buf_iowait(). And because xfs_buf_wait_unpin()
> is called with the buffer lock held, we'll never do the failure
> handling in xfs_buf_item_unpin until the buffer IO is completed and
> it is unlocked.
How do we manage to submit it synchronously then? The inode and dquot
reclaim xfs_bwrite calls already wait for an unpin first, so I don't
think these are the problem. The only other call on a live fs seems
to xfs_qm_shake -> xfs_buf_delwri_submit, but that one does wait
for the complete() call on b_iowait. I suspect we are hitting that
and due to it skipping the wait if b_ioerror is set and waiting on
multiple buffers that complete together might hide the issue.
__xfs_buf_delwri_submit for the wait == true case also seems to be
the only place that actually skips the ispinned check.