xfs: automatic dfops buffer relogging
authorBrian Foster <bfoster@redhat.com>
Wed, 1 Aug 2018 14:20:32 +0000 (07:20 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Fri, 3 Aug 2018 06:05:13 +0000 (23:05 -0700)
Buffers that are held across deferred operations are explicitly
joined to the dfops structure to ensure appropriate relogging.
While buffers are currently joined explicitly, we can detect the
conditions that require relogging at dfops finish time by inspecting
the transaction item list for held buffers.

Replace the xfs_defer_bjoin() infrastructure with such detection and
automatic relogging of held buffers. This eliminates the need for
the per-dfops buffer list, replaced by an on-stack variant in
xfs_defer_trans_roll().

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
fs/xfs/libxfs/xfs_attr.c
fs/xfs/libxfs/xfs_defer.c
fs/xfs/libxfs/xfs_defer.h
fs/xfs/xfs_dquot.c
fs/xfs/xfs_trans.h

index 3deb5cd..227887b 100644 (file)
@@ -320,7 +320,6 @@ xfs_attr_set(
                 * buffer and run into problems with the write verifier.
                 */
                xfs_trans_bhold(args.trans, leaf_bp);
-               xfs_defer_bjoin(args.trans->t_dfops, leaf_bp);
                xfs_defer_ijoin(args.trans->t_dfops, dp);
                error = xfs_defer_finish(&args.trans);
                if (error)
index 64e1abc..e9b7671 100644 (file)
@@ -14,6 +14,7 @@
 #include "xfs_mount.h"
 #include "xfs_defer.h"
 #include "xfs_trans.h"
+#include "xfs_buf_item.h"
 #include "xfs_trace.h"
 
 /*
@@ -228,6 +229,10 @@ xfs_defer_trans_roll(
        struct xfs_trans                **tp)
 {
        struct xfs_defer_ops            *dop = (*tp)->t_dfops;
+       struct xfs_buf_log_item         *bli;
+       struct xfs_log_item             *lip;
+       struct xfs_buf                  *bplist[XFS_DEFER_OPS_NR_BUFS];
+       int                             bpcount = 0;
        int                             i;
        int                             error;
 
@@ -235,9 +240,24 @@ xfs_defer_trans_roll(
        for (i = 0; i < XFS_DEFER_OPS_NR_INODES && dop->dop_inodes[i]; i++)
                xfs_trans_log_inode(*tp, dop->dop_inodes[i], XFS_ILOG_CORE);
 
-       /* Hold the (previously bjoin'd) buffer locked across the roll. */
-       for (i = 0; i < XFS_DEFER_OPS_NR_BUFS && dop->dop_bufs[i]; i++)
-               xfs_trans_dirty_buf(*tp, dop->dop_bufs[i]);
+       list_for_each_entry(lip, &(*tp)->t_items, li_trans) {
+               switch (lip->li_type) {
+               case XFS_LI_BUF:
+                       bli = container_of(lip, struct xfs_buf_log_item,
+                                          bli_item);
+                       if (bli->bli_flags & XFS_BLI_HOLD) {
+                               if (bpcount >= XFS_DEFER_OPS_NR_BUFS) {
+                                       ASSERT(0);
+                                       return -EFSCORRUPTED;
+                               }
+                               xfs_trans_dirty_buf(*tp, bli->bli_buf);
+                               bplist[bpcount++] = bli->bli_buf;
+                       }
+                       break;
+               default:
+                       break;
+               }
+       }
 
        trace_xfs_defer_trans_roll((*tp)->t_mountp, dop, _RET_IP_);
 
@@ -255,9 +275,9 @@ xfs_defer_trans_roll(
                xfs_trans_ijoin(*tp, dop->dop_inodes[i], 0);
 
        /* Rejoin the buffers and dirty them so the log moves forward. */
-       for (i = 0; i < XFS_DEFER_OPS_NR_BUFS && dop->dop_bufs[i]; i++) {
-               xfs_trans_bjoin(*tp, dop->dop_bufs[i]);
-               xfs_trans_bhold(*tp, dop->dop_bufs[i]);
+       for (i = 0; i < bpcount; i++) {
+               xfs_trans_bjoin(*tp, bplist[i]);
+               xfs_trans_bhold(*tp, bplist[i]);
        }
 
        return error;
@@ -296,30 +316,6 @@ xfs_defer_ijoin(
 }
 
 /*
- * Add this buffer to the deferred op.  Each joined buffer is relogged
- * each time we roll the transaction.
- */
-int
-xfs_defer_bjoin(
-       struct xfs_defer_ops            *dop,
-       struct xfs_buf                  *bp)
-{
-       int                             i;
-
-       for (i = 0; i < XFS_DEFER_OPS_NR_BUFS; i++) {
-               if (dop->dop_bufs[i] == bp)
-                       return 0;
-               else if (dop->dop_bufs[i] == NULL) {
-                       dop->dop_bufs[i] = bp;
-                       return 0;
-               }
-       }
-
-       ASSERT(0);
-       return -EFSCORRUPTED;
-}
-
-/*
  * Reset an already used dfops after finish.
  */
 static void
@@ -331,7 +327,6 @@ xfs_defer_reset(
        ASSERT(!xfs_defer_has_unfinished_work(dop));
 
        memset(dop->dop_inodes, 0, sizeof(dop->dop_inodes));
-       memset(dop->dop_bufs, 0, sizeof(dop->dop_bufs));
 
        /*
         * Low mode state transfers across transaction rolls to mirror dfops
@@ -594,7 +589,6 @@ xfs_defer_move(
        list_splice_init(&src->dop_pending, &dst->dop_pending);
 
        memcpy(dst->dop_inodes, src->dop_inodes, sizeof(dst->dop_inodes));
-       memcpy(dst->dop_bufs, src->dop_bufs, sizeof(dst->dop_bufs));
 
        /*
         * Low free space mode was historically controlled by a dfops field.
index 8908a27..4a8bb83 100644 (file)
@@ -43,7 +43,6 @@ void xfs_defer_cancel(struct xfs_trans *);
 void xfs_defer_init(struct xfs_trans *tp, struct xfs_defer_ops *dop);
 bool xfs_defer_has_unfinished_work(struct xfs_defer_ops *dop);
 int xfs_defer_ijoin(struct xfs_defer_ops *dop, struct xfs_inode *ip);
-int xfs_defer_bjoin(struct xfs_defer_ops *dop, struct xfs_buf *bp);
 void xfs_defer_move(struct xfs_trans *dtp, struct xfs_trans *stp);
 
 /* Description of a deferred type. */
index da5c55c..e119685 100644 (file)
@@ -362,7 +362,6 @@ xfs_dquot_disk_alloc(
         * manually or by committing the transaction.
         */
        xfs_trans_bhold(tp, bp);
-       error = xfs_defer_bjoin(tp->t_dfops, bp);
        if (error) {
                xfs_trans_bhold_release(tp, bp);
                xfs_trans_brelse(tp, bp);
index 7e49322..581456c 100644 (file)
@@ -100,7 +100,6 @@ struct xfs_defer_ops {
 
        /* relog these with each roll */
        struct xfs_inode        *dop_inodes[XFS_DEFER_OPS_NR_INODES];
-       struct xfs_buf          *dop_bufs[XFS_DEFER_OPS_NR_BUFS];
 };
 
 /*