xfs: allow queued realtime intents to drain before scrubbing

When a writer thread executes a chain of log intent items for the
realtime volume, the ILOCKs taken during each step are for each rt
metadata file, not the entire rt volume itself.  Although scrub takes
all rt metadata ILOCKs, this isn't sufficient to guard against scrub
checking the rt volume while that writer thread is in the middle of
finishing a chain because there's no higher level locking primitive
guarding the realtime volume.

When there's a collision, cross-referencing between data structures
(e.g. rtrmapbt and rtrefcountbt) yields false corruption events; if
repair is running, this results in incorrect repairs, which is
catastrophic.

Fix this by adding to the mount structure the same drain that we use to
protect scrub against concurrent AG updates, but this time for the
realtime volume.

[Contains a few cleanups from hch]

Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>

diff --git a/include/xfs_mount.h b/include/xfs_mount.h
index 8ef5fcc2cc5b51b42c9a46356de975bcb4ef2308..f78f0d050cc3fd40b72d125608ccc8c22b21bc6f 100644 (file)
--- a/include/xfs_mount.h
+++ b/include/xfs_mount.h
@@ -333,6 +333,18 @@ struct xfs_defer_drain { /* empty */ };
 static inline void xfs_perag_intent_hold(struct xfs_perag *pag) {}
 static inline void xfs_perag_intent_rele(struct xfs_perag *pag) {}
 
+struct xfs_rtgroup;
+
+#define xfs_rtgroup_intent_get(mp, rgno) \
+       xfs_rtgroup_get((mp), xfs_rtb_to_rgno((mp), (rgno)))
+#define xfs_rtgroup_intent_put(rtg)            xfs_rtgroup_put(rtg)
+
+static inline void xfs_rtgroup_intent_hold(struct xfs_rtgroup *rtg) { }
+static inline void xfs_rtgroup_intent_rele(struct xfs_rtgroup *rtg) { }
+
+#define xfs_drain_free(dr)             ((void)0)
+#define xfs_drain_init(dr)             ((void)0)
+
 static inline void libxfs_buftarg_drain(struct xfs_buftarg *btp)
 {
        cache_purge(btp->bcache);

diff --git a/libxfs/defer_item.c b/libxfs/defer_item.c
index 658e6d608f5c5e8df9981dcefe9c04a39c4ab0e1..e748ce8b948b173fe88fb1af47f82429d8af8f51 100644 (file)
--- a/libxfs/defer_item.c
+++ b/libxfs/defer_item.c
@@ -87,11 +87,8 @@ xfs_extent_free_defer_add(
        struct xfs_mount                *mp = tp->t_mountp;
 
        if (xfs_efi_is_realtime(xefi)) {
-               xfs_rgnumber_t          rgno;
-
-               rgno = xfs_rtb_to_rgno(mp, xefi->xefi_startblock);
-               xefi->xefi_rtg = xfs_rtgroup_get(mp, rgno);
-
+               xefi->xefi_rtg = xfs_rtgroup_intent_get(mp,
+                                               xefi->xefi_startblock);
                *dfpp = xfs_defer_add(tp, &xefi->xefi_list,
                                &xfs_rtextent_free_defer_type);
                return;
@@ -203,7 +200,7 @@ xfs_rtextent_free_cancel_item(
 {
        struct xfs_extent_free_item     *xefi = xefi_entry(item);
 
-       xfs_rtgroup_put(xefi->xefi_rtg);
+       xfs_rtgroup_intent_put(xefi->xefi_rtg);
        kmem_cache_free(xfs_extfree_item_cache, xefi);
 }
 
@@ -337,13 +334,12 @@ xfs_rmap_defer_add(
         * section updates.
         */
        if (ri->ri_realtime) {
-               xfs_rgnumber_t  rgno;
-
-               rgno = xfs_rtb_to_rgno(mp, ri->ri_bmap.br_startblock);
-               ri->ri_rtg = xfs_rtgroup_get(mp, rgno);
+               ri->ri_rtg = xfs_rtgroup_intent_get(mp,
+                                               ri->ri_bmap.br_startblock);
                xfs_defer_add(tp, &ri->ri_list, &xfs_rtrmap_update_defer_type);
        } else {
-               ri->ri_pag = xfs_perag_intent_get(mp, ri->ri_bmap.br_startblock);
+               ri->ri_pag = xfs_perag_intent_get(mp,
+                                               ri->ri_bmap.br_startblock);
                xfs_defer_add(tp, &ri->ri_list, &xfs_rmap_update_defer_type);
        }
 }
@@ -444,7 +440,7 @@ xfs_rtrmap_update_cancel_item(
 {
        struct xfs_rmap_intent          *ri = ri_entry(item);
 
-       xfs_rtgroup_put(ri->ri_rtg);
+       xfs_rtgroup_intent_put(ri->ri_rtg);
        kmem_cache_free(xfs_rmap_intent_cache, ri);
 }
 
@@ -667,10 +663,8 @@ xfs_bmap_update_get_group(
 {
        if (xfs_ifork_is_realtime(bi->bi_owner, bi->bi_whichfork)) {
                if (xfs_has_rtgroups(mp)) {
-                       xfs_rgnumber_t  rgno;
-
-                       rgno = xfs_rtb_to_rgno(mp, bi->bi_bmap.br_startblock);
-                       bi->bi_rtg = xfs_rtgroup_get(mp, rgno);
+                       bi->bi_rtg = xfs_rtgroup_intent_get(mp,
+                                               bi->bi_bmap.br_startblock);
                } else {
                        bi->bi_rtg = NULL;
                }
@@ -706,8 +700,9 @@ xfs_bmap_update_put_group(
        struct xfs_bmap_intent  *bi)
 {
        if (xfs_ifork_is_realtime(bi->bi_owner, bi->bi_whichfork)) {
-               if (xfs_has_rtgroups(bi->bi_owner->i_mount))
-                       xfs_rtgroup_put(bi->bi_rtg);
+               if (xfs_has_rtgroups(bi->bi_owner->i_mount)) {
+                       xfs_rtgroup_intent_put(bi->bi_rtg);
+               }
                return;
        }
 

diff --git a/libxfs/xfs_rtgroup.c b/libxfs/xfs_rtgroup.c
index 89a02f74170d25188905d311f2b98538f61ab01c..f58b271029ec997a7fc251aeb0fd12fe747c3b8c 100644 (file)
--- a/libxfs/xfs_rtgroup.c
+++ b/libxfs/xfs_rtgroup.c
@@ -164,6 +164,7 @@ xfs_initialize_rtgroups(
                /* Place kernel structure only init below this point. */
                spin_lock_init(&rtg->rtg_state_lock);
                init_waitqueue_head(&rtg->rtg_active_wq);
+               xfs_defer_drain_init(&rtg->rtg_intents_drain);
 #endif /* __KERNEL__ */
 
                /* Active ref owned by mount indicates rtgroup is online. */
@@ -203,6 +204,9 @@ xfs_free_unused_rtgroup_range(
                spin_unlock(&mp->m_rtgroup_lock);
                if (!rtg)
                        break;
+#ifdef __KERNEL__
+               xfs_defer_drain_free(&rtg->rtg_intents_drain);
+#endif
                kfree(rtg);
        }
 }
@@ -236,6 +240,9 @@ xfs_free_rtgroups(
                spin_unlock(&mp->m_rtgroup_lock);
                ASSERT(rtg);
                XFS_IS_CORRUPT(mp, atomic_read(&rtg->rtg_ref) != 0);
+#ifdef __KERNEL__
+               xfs_defer_drain_free(&rtg->rtg_intents_drain);
+#endif
 
                /* drop the mount's active reference */
                xfs_rtgroup_rele(rtg);

diff --git a/libxfs/xfs_rtgroup.h b/libxfs/xfs_rtgroup.h
index acd88ae161ce4e6abd6de358ccb9eb3990ade16d..f0abe53ef6cbb1a5905ece8fa3c808d2e66b6f3a 100644 (file)
--- a/libxfs/xfs_rtgroup.h
+++ b/libxfs/xfs_rtgroup.h
@@ -48,6 +48,15 @@ struct xfs_rtgroup {
 #ifdef __KERNEL__
        /* -- kernel only structures below this line -- */
        spinlock_t              rtg_state_lock;
+
+       /*
+        * We use xfs_drain to track the number of deferred log intent items
+        * that have been queued (but not yet processed) so that waiters (e.g.
+        * scrub) will not lock resources when other threads are in the middle
+        * of processing a chain of intent items only to find momentary
+        * inconsistencies.
+        */
+       struct xfs_defer_drain  rtg_intents_drain;
 #endif /* __KERNEL__ */
 };