Index: arc.c
===================================================================
RCS file: /zoo/pjd/repo/src/sys/contrib/opensolaris/uts/common/fs/zfs/arc.c,v
retrieving revision 1.11
diff -u -p -r1.11 arc.c
--- arc.c	10 Sep 2007 18:12:27 -0000	1.11
+++ arc.c	10 Nov 2007 13:25:36 -0000
@@ -373,7 +373,7 @@ struct arc_buf_hdr {
 static arc_buf_t *arc_eviction_list;
 static kmutex_t arc_eviction_mtx;
 static arc_buf_hdr_t arc_eviction_hdr;
-static void arc_get_data_buf(arc_buf_t *buf);
+static int arc_get_data_buf(arc_buf_t *buf, kmutex_t *hash_lock);
 static void arc_access(arc_buf_hdr_t *buf, kmutex_t *hash_lock);
 
 #define	GHOST_STATE(state)	\
@@ -460,19 +460,33 @@ buf_hash(spa_t *spa, dva_t *dva, uint64_
 	((buf)->b_birth == birth) && ((buf)->b_spa == spa)
 
 static arc_buf_hdr_t *
-buf_hash_find(spa_t *spa, dva_t *dva, uint64_t birth, kmutex_t **lockp)
+buf_hash_lookup(spa_t *spa, dva_t *dva, uint64_t birth)
 {
 	uint64_t idx = BUF_HASH_INDEX(spa, dva, birth);
 	kmutex_t *hash_lock = BUF_HASH_LOCK(idx);
 	arc_buf_hdr_t *buf;
 
-	mutex_enter(hash_lock);
+	ASSERT(MUTEX_HELD(BUF_HASH_LOCK(idx)));
 	for (buf = buf_hash_table.ht_table[idx]; buf != NULL;
 	    buf = buf->b_hash_next) {
-		if (BUF_EQUAL(spa, dva, birth, buf)) {
-			*lockp = hash_lock;
+		if (BUF_EQUAL(spa, dva, birth, buf))
 			return (buf);
-		}
+	}
+	return (NULL);
+}
+
+static arc_buf_hdr_t *
+buf_hash_find(spa_t *spa, dva_t *dva, uint64_t birth, kmutex_t **lockp)
+{
+	uint64_t idx = BUF_HASH_INDEX(spa, dva, birth);
+	kmutex_t *hash_lock = BUF_HASH_LOCK(idx);
+	arc_buf_hdr_t *buf;
+
+	mutex_enter(hash_lock);
+	buf = buf_hash_lookup(spa, dva, birth);
+	if (buf != NULL) {
+		*lockp = hash_lock;
+		return (buf);
 	}
 	mutex_exit(hash_lock);
 	*lockp = NULL;
@@ -878,7 +892,7 @@ arc_buf_alloc(spa_t *spa, int size, void
 	buf->b_private = NULL;
 	buf->b_next = NULL;
 	hdr->b_buf = buf;
-	arc_get_data_buf(buf);
+	arc_get_data_buf(buf, NULL);
 	hdr->b_datacnt = 1;
 	hdr->b_flags = 0;
 	ASSERT(refcount_is_zero(&hdr->b_refcnt));
@@ -901,7 +915,7 @@ arc_buf_clone(arc_buf_t *from)
 	buf->b_private = NULL;
 	buf->b_next = hdr->b_buf;
 	hdr->b_buf = buf;
-	arc_get_data_buf(buf);
+	arc_get_data_buf(buf, NULL);
 	bcopy(from->b_data, buf->b_data, size);
 	hdr->b_datacnt += 1;
 	return (buf);
@@ -1665,12 +1679,13 @@ arc_evict_needed()
  * MFU's resident set is consuming more space than it has been allotted.  In
  * this situation, we must victimize our own cache, the MFU, for this insertion.
  */
-static void
-arc_get_data_buf(arc_buf_t *buf)
+static boolean_t
+arc_get_data_buf(arc_buf_t *buf, kmutex_t *hash_lock)
 {
 	arc_state_t		*state = buf->b_hdr->b_state;
 	uint64_t		size = buf->b_hdr->b_size;
 	arc_buf_contents_t	type = buf->b_hdr->b_type;
+	boolean_t		relookup = 0;
 
 	arc_adapt(size, state);
 
@@ -1679,12 +1694,18 @@ arc_get_data_buf(arc_buf_t *buf)
 	 * just allocate a new buffer.
 	 */
 	if (!arc_evict_needed()) {
+		if (hash_lock != NULL) {
+			relookup = 1;
+			mutex_exit(hash_lock);
+		}
 		if (type == ARC_BUFC_METADATA) {
 			buf->b_data = zio_buf_alloc(size);
 		} else {
 			ASSERT(type == ARC_BUFC_DATA);
 			buf->b_data = zio_data_buf_alloc(size);
 		}
+		if (hash_lock != NULL)
+			mutex_enter(hash_lock);
 		atomic_add_64(&arc_size, size);
 		goto out;
 	}
@@ -1707,12 +1728,18 @@ arc_get_data_buf(arc_buf_t *buf)
 		state =  (mfu_space > arc_mfu->arcs_size) ? arc_mru : arc_mfu;
 	}
 	if ((buf->b_data = arc_evict(state, size, TRUE, type)) == NULL) {
+		if (hash_lock != NULL) {
+			relookup = 1;
+			mutex_exit(hash_lock);
+		}
 		if (type == ARC_BUFC_METADATA) {
 			buf->b_data = zio_buf_alloc(size);
 		} else {
 			ASSERT(type == ARC_BUFC_DATA);
 			buf->b_data = zio_data_buf_alloc(size);
 		}
+		if (hash_lock != NULL)
+			mutex_enter(hash_lock);
 		atomic_add_64(&arc_size, size);
 		ARCSTAT_BUMP(arcstat_recycle_miss);
 	}
@@ -1738,6 +1765,7 @@ out:
 		    arc_anon->arcs_size + arc_mru->arcs_size > arc_p)
 			arc_p = MIN(arc_c, arc_p + size);
 	}
+	return (relookup);
 }
 
 /*
@@ -2151,7 +2179,10 @@ top:
 			buf->b_private = NULL;
 			buf->b_next = NULL;
 			hdr->b_buf = buf;
-			arc_get_data_buf(buf);
+			if (arc_get_data_buf(buf, hash_lock)) {
+				ASSERT(hdr == buf_hash_lookup(spa, BP_IDENTITY(bp),
+				    bp->blk_birth));
+			}
 			ASSERT(hdr->b_datacnt == 0);
 			hdr->b_datacnt = 1;