On Tue, Mar 22, 2011 at 03:55:55PM -0400, Christoph Hellwig wrote:
> Now that we have reliably tracking of deleted extents in a transaction
> we can easily implement "online" discard support which calls
> blkdev_issue_discard once a transaction commits.
>
> The actual discard is a two stage operation as we first have to mark
> the busy extent as not available for reuse before we can start the
> actual discard. Note that we don't bother supporting discard for
> the non-delaylog mode. While that would be possible with this patch
> performance is awfull, and the optimization in the next patch won't
> work as easily.
>
> Signed-off-by: Christoph Hellwig <hch@xxxxxx>
.....
> @@ -361,13 +362,17 @@ xlog_cil_committed(
> int abort)
> {
> struct xfs_cil_ctx *ctx = args;
> + struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
> struct xfs_busy_extent *busyp, *n;
>
> xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain,
> ctx->start_lsn, abort);
>
> - list_for_each_entry_safe(busyp, n, &ctx->busy_extents, list)
> - xfs_alloc_busy_clear(ctx->cil->xc_log->l_mp, busyp);
> + list_for_each_entry_safe(busyp, n, &ctx->busy_extents, list) {
> + if (!abort)
> + xfs_discard_extent(mp, busyp);
> + xfs_alloc_busy_clear(mp, busyp);
> + }
>
> spin_lock(&ctx->cil->xc_cil_lock);
> list_del(&ctx->committing);
> Index: xfs/fs/xfs/linux-2.6/xfs_discard.c
> ===================================================================
> --- xfs.orig/fs/xfs/linux-2.6/xfs_discard.c 2011-03-21 14:47:03.614474345
> +0100
> +++ xfs/fs/xfs/linux-2.6/xfs_discard.c 2011-03-21 14:51:41.449976366
> +0100
> @@ -191,3 +191,38 @@ xfs_ioc_trim(
> return -XFS_ERROR(EFAULT);
> return 0;
> }
> +
> +int
> +xfs_discard_extent(
> + struct xfs_mount *mp,
> + struct xfs_busy_extent *busyp)
> +{
> + struct xfs_perag *pag;
> + int error = 0;
> + xfs_daddr_t bno;
> + int64_t len;
> + bool done = false;
> +
> + if ((mp->m_flags & XFS_MOUNT_DISCARD) == 0)
> + return 0;
I'd move this check to the callers, otherwise we are going to be
doing lots of function calls in a relatively performance sensitive
loop just to run a single check when discard is not enabled...
> +
> + bno = XFS_AGB_TO_DADDR(mp, busyp->agno, busyp->bno);
> + len = XFS_FSB_TO_BB(mp, busyp->length);
> +
> + pag = xfs_perag_get(mp, busyp->agno);
> + spin_lock(&pag->pagb_lock);
> + if (!busyp->length)
> + done = true;
> + busyp->flags = XFS_ALLOC_BUSY_DISCARDED;
> + spin_unlock(&pag->pagb_lock);
> + xfs_perag_put(pag);
> +
> + if (done)
> + return 0;
> +
> + error = -blkdev_issue_discard(mp->m_ddev_targp->bt_bdev, bno, len,
> + GFP_NOFS, 0);
> + if (error && error != EOPNOTSUPP)
> + xfs_info(mp, "discard failed, error %d", error);
This would be more informative if it also printed the bno and len of
the discard that failed. A couple of tracepoints here (e.g.
discard_extent_issued, discard_extent_failed) could also be useful
for tracking discard operations.
Cheers,
Dave.
--
Dave Chinner
david@xxxxxxxxxxxxx
|