Merge commit 'refs/top-bases/zfs-branch' into zfs-branch

46 files changed, 2552 insertions, 711 deletions

diff --git a/module/zfs/dmu.c b/module/zfs/dmu.c
index 19bf695f1..8ca5c9d7d 100644
--- a/module/zfs/dmu.c
+++ b/module/zfs/dmu.c
@@ -87,6 +87,7 @@ const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
 	{	zap_byteswap,		TRUE,	"scrub work queue"	},
 	{	zap_byteswap,		TRUE,	"ZFS user/group used"	},
 	{	zap_byteswap,		TRUE,	"ZFS user/group quota"	},
+	{	zap_byteswap,		TRUE,	"snapshot refcount tags"},
 };
 
 int
@@ -195,7 +196,7 @@ dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
 
 	ASSERT(length <= DMU_MAX_ACCESS);
 
-	dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT;
+	dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT;
 	if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz)
 		dbuf_flags |= DB_RF_NOPREFETCH;
 
@@ -212,6 +213,7 @@ dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
 			    os_dsl_dataset->ds_object,
 			    (longlong_t)dn->dn_object, dn->dn_datablksz,
 			    (longlong_t)offset, (longlong_t)length);
+			rw_exit(&dn->dn_struct_rwlock);
 			return (EIO);
 		}
 		nblks = 1;
@@ -234,9 +236,7 @@ dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
 		}
 		/* initiate async i/o */
 		if (rd) {
-			rw_exit(&dn->dn_struct_rwlock);
 			(void) dbuf_read(db, zio, dbuf_flags);
-			rw_enter(&dn->dn_struct_rwlock, RW_READER);
 		}
 		dbp[i] = &db->db;
 	}
@@ -376,56 +376,51 @@ dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len)
 	dnode_rele(dn, FTAG);
 }
 
+/*
+ * Get the next "chunk" of file data to free.  We traverse the file from
+ * the end so that the file gets shorter over time (if we crashes in the
+ * middle, this will leave us in a better state).  We find allocated file
+ * data by simply searching the allocated level 1 indirects.
+ */
 static int
-get_next_chunk(dnode_t *dn, uint64_t *offset, uint64_t limit)
+get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit)
 {
-	uint64_t len = *offset - limit;
-	uint64_t chunk_len = dn->dn_datablksz * DMU_MAX_DELETEBLKCNT;
-	uint64_t subchunk =
+	uint64_t len = *start - limit;
+	uint64_t blkcnt = 0;
+	uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1));
+	uint64_t iblkrange =
 	    dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
 
-	ASSERT(limit <= *offset);
+	ASSERT(limit <= *start);
 
-	if (len <= chunk_len) {
-		*offset = limit;
+	if (len <= iblkrange * maxblks) {
+		*start = limit;
 		return (0);
 	}
+	ASSERT(ISP2(iblkrange));
 
-	ASSERT(ISP2(subchunk));
-
-	while (*offset > limit) {
-		uint64_t initial_offset = P2ROUNDUP(*offset, subchunk);
-		uint64_t delta;
+	while (*start > limit && blkcnt < maxblks) {
 		int err;
 
-		/* skip over allocated data */
+		/* find next allocated L1 indirect */
 		err = dnode_next_offset(dn,
-		    DNODE_FIND_HOLE|DNODE_FIND_BACKWARDS, offset, 1, 1, 0);
-		if (err == ESRCH)
-			*offset = limit;
-		else if (err)
-			return (err);
+		    DNODE_FIND_BACKWARDS, start, 2, 1, 0);
 
-		ASSERT3U(*offset, <=, initial_offset);
-		*offset = P2ALIGN(*offset, subchunk);
-		delta = initial_offset - *offset;
-		if (delta >= chunk_len) {
-			*offset += delta - chunk_len;
+		/* if there are no more, then we are done */
+		if (err == ESRCH) {
+			*start = limit;
 			return (0);
-		}
-		chunk_len -= delta;
-
-		/* skip over unallocated data */
-		err = dnode_next_offset(dn,
-		    DNODE_FIND_BACKWARDS, offset, 1, 1, 0);
-		if (err == ESRCH)
-			*offset = limit;
-		else if (err)
+		} else if (err) {
 			return (err);
+		}
+		blkcnt += 1;
 
-		if (*offset < limit)
-			*offset = limit;
-		ASSERT3U(*offset, <, initial_offset);
+		/* reset offset to end of "next" block back */
+		*start = P2ALIGN(*start, iblkrange);
+		if (*start <= limit)
+			*start = limit;
+		else
+			*start -= 1;
 	}
 	return (0);
 }
@@ -548,7 +543,7 @@ dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 {
 	dnode_t *dn;
 	dmu_buf_t **dbp;
-	int numbufs, i, err;
+	int numbufs, err;
 
 	err = dnode_hold(os->os, object, FTAG, &dn);
 	if (err)
@@ -559,7 +554,7 @@ dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 	 * block.  If we ever do the tail block optimization, we will need to
 	 * handle that here as well.
 	 */
-	if (dn->dn_datablkshift == 0) {
+	if (dn->dn_maxblkid == 0) {
 		int newsz = offset > dn->dn_datablksz ? 0 :
 		    MIN(size, dn->dn_datablksz - offset);
 		bzero((char *)buf + newsz, size - newsz);
@@ -568,6 +563,7 @@ dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 
 	while (size > 0) {
 		uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
+		int i;
 
 		/*
 		 * NB: we could do this block-at-a-time, but it's nice
@@ -803,9 +799,6 @@ dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 		if (tocpy == db->db_size)
 			dmu_buf_fill_done(db, tx);
 
-		if (err)
-			break;
-
 		offset += tocpy;
 		size -= tocpy;
 	}
diff --git a/module/zfs/dmu_objset.c b/module/zfs/dmu_objset.c
index dab6df8bf..01792600b 100644
--- a/module/zfs/dmu_objset.c
+++ b/module/zfs/dmu_objset.c
@@ -679,7 +679,7 @@ dmu_objset_create(const char *name, dmu_objset_type_t type,
 }
 
 int
-dmu_objset_destroy(const char *name)
+dmu_objset_destroy(const char *name, boolean_t defer)
 {
 	objset_t *os;
 	int error;
@@ -696,7 +696,7 @@ dmu_objset_destroy(const char *name)
 		dsl_dataset_t *ds = os->os->os_dsl_dataset;
 		zil_destroy(dmu_objset_zil(os), B_FALSE);
 
-		error = dsl_dataset_destroy(ds, os);
+		error = dsl_dataset_destroy(ds, os, defer);
 		/*
 		 * dsl_dataset_destroy() closes the ds.
 		 */
@@ -1136,7 +1136,7 @@ dmu_objset_userspace_upgrade(objset_t *os)
 	 */
 
 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
-		dmu_tx_t *tx = dmu_tx_create(os);
+		dmu_tx_t *tx;
 		dmu_buf_t *db;
 		int objerr;
 
@@ -1146,6 +1146,7 @@ dmu_objset_userspace_upgrade(objset_t *os)
 		objerr = dmu_bonus_hold(os, obj, FTAG, &db);
 		if (objerr)
 			continue;
+		tx = dmu_tx_create(os);
 		dmu_tx_hold_bonus(tx, obj);
 		objerr = dmu_tx_assign(tx, TXG_WAIT);
 		if (objerr) {
diff --git a/module/zfs/dmu_send.c b/module/zfs/dmu_send.c
index 21e1d91d1..b977a2ff7 100644
--- a/module/zfs/dmu_send.c
+++ b/module/zfs/dmu_send.c
@@ -393,6 +393,7 @@ recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
 	dsl_dataset_t *ds = arg1;
 	struct recvbeginsyncarg *rbsa = arg2;
 	int err;
+	struct dsl_ds_destroyarg dsda = {0};
 
 	/* must be a head ds */
 	if (ds->ds_phys->ds_next_snap_obj != 0)
@@ -402,7 +403,8 @@ recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
 	if (dsl_dir_is_clone(ds->ds_dir))
 		return (EINVAL);
 
-	err = dsl_dataset_destroy_check(ds, rbsa->tag, tx);
+	dsda.ds = ds;
+	err = dsl_dataset_destroy_check(&dsda, rbsa->tag, tx);
 	if (err)
 		return (err);
 
@@ -427,13 +429,16 @@ recv_full_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 	dsl_dir_t *dd = ds->ds_dir;
 	uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
 	uint64_t dsobj;
+	struct dsl_ds_destroyarg dsda = {0};
 
 	/*
 	 * NB: caller must provide an extra hold on the dsl_dir_t, so it
 	 * won't go away when dsl_dataset_destroy_sync() closes the
 	 * dataset.
 	 */
-	dsl_dataset_destroy_sync(ds, rbsa->tag, cr, tx);
+	dsda.ds = ds;
+	dsl_dataset_destroy_sync(&dsda, rbsa->tag, cr, tx);
+	ASSERT3P(dsda.rm_origin, ==, NULL);
 
 	dsobj = dsl_dataset_create_sync_dd(dd, rbsa->origin, flags, tx);
 
@@ -483,7 +488,7 @@ recv_incremental_check(void *arg1, void *arg2, dmu_tx_t *tx)
 
 /* ARGSUSED */
 static void
-recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
+recv_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 {
 	dsl_dataset_t *ohds = arg1;
 	struct recvbeginsyncarg *rbsa = arg2;
@@ -513,27 +518,13 @@ recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 	    dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
 }
 
-/* ARGSUSED */
-static void
-recv_offline_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
-{
-	dsl_dataset_t *ds = arg1;
-
-	dmu_buf_will_dirty(ds->ds_dbuf, tx);
-	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
-
-	spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
-	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "dataset = %lld",
-	    ds->ds_object);
-}
-
 /*
  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
  * succeeds; otherwise we will leak the holds on the datasets.
  */
 int
 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
-    boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *drc)
+    boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc)
 {
 	int err = 0;
 	boolean_t byteswap;
@@ -582,36 +573,8 @@ dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
 	/*
 	 * Process the begin in syncing context.
 	 */
-	if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE) && !online) {
-		/* offline incremental receive */
-		err = dsl_dataset_own(tofs, 0, dmu_recv_tag, &ds);
-		if (err)
-			return (err);
-
-		/*
-		 * Only do the rollback if the most recent snapshot
-		 * matches the incremental source
-		 */
-		if (force) {
-			if (ds->ds_prev == NULL ||
-			    ds->ds_prev->ds_phys->ds_guid !=
-			    rbsa.fromguid) {
-				dsl_dataset_disown(ds, dmu_recv_tag);
-				return (ENODEV);
-			}
-			(void) dsl_dataset_rollback(ds, DMU_OST_NONE);
-		}
-		rbsa.force = B_FALSE;
-		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
-		    recv_incremental_check,
-		    recv_offline_incremental_sync, ds, &rbsa, 1);
-		if (err) {
-			dsl_dataset_disown(ds, dmu_recv_tag);
-			return (err);
-		}
-		drc->drc_logical_ds = drc->drc_real_ds = ds;
-	} else if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
-		/* online incremental receive */
+	if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
+		/* incremental receive */
 
 		/* tmp clone name is: tofs/%tosnap" */
 		(void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
@@ -622,11 +585,18 @@ dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
 		if (err)
 			return (err);
 
+		/* must not have an incremental recv already in progress */
+		if (!mutex_tryenter(&ds->ds_recvlock)) {
+			dsl_dataset_rele(ds, dmu_recv_tag);
+			return (EBUSY);
+		}
+
 		rbsa.force = force;
 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
 		    recv_incremental_check,
-		    recv_online_incremental_sync, ds, &rbsa, 5);
+		    recv_incremental_sync, ds, &rbsa, 5);
 		if (err) {
+			mutex_exit(&ds->ds_recvlock);
 			dsl_dataset_rele(ds, dmu_recv_tag);
 			return (err);
 		}
@@ -931,26 +901,6 @@ restore_free(struct restorearg *ra, objset_t *os,
 	return (err);
 }
 
-void
-dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc)
-{
-	if (drc->drc_newfs || drc->drc_real_ds != drc->drc_logical_ds) {
-		/*
-		 * online incremental or new fs: destroy the fs (which
-		 * may be a clone) that we created
-		 */
-		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
-		if (drc->drc_real_ds != drc->drc_logical_ds)
-			dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
-	} else {
-		/*
-		 * offline incremental: rollback to most recent snapshot.
-		 */
-		(void) dsl_dataset_rollback(drc->drc_real_ds, DMU_OST_NONE);
-		dsl_dataset_disown(drc->drc_real_ds, dmu_recv_tag);
-	}
-}
-
 /*
  * Compute checksum of drr_begin record
  */
@@ -1085,11 +1035,17 @@ out:
 
 	if (ra.err != 0) {
 		/*
-		 * rollback or destroy what we created, so we don't
-		 * leave it in the restoring state.
+		 * destroy what we created, so we don't leave it in the
+		 * inconsistent restoring state.
 		 */
 		txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
-		dmu_recv_abort_cleanup(drc);
+
+		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
+		    B_FALSE);
+		if (drc->drc_real_ds != drc->drc_logical_ds) {
+			mutex_exit(&drc->drc_logical_ds->ds_recvlock);
+			dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
+		}
 	}
 
 	kmem_free(ra.buf, ra.bufsize);
@@ -1156,7 +1112,9 @@ dmu_recv_end(dmu_recv_cookie_t *drc)
 			dsl_dataset_rele(ds, dmu_recv_tag);
 		}
 		/* dsl_dataset_destroy() will disown the ds */
-		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
+		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
+		    B_FALSE);
+		mutex_exit(&drc->drc_logical_ds->ds_recvlock);
 		if (err)
 			return (err);
 	}
@@ -1170,7 +1128,8 @@ dmu_recv_end(dmu_recv_cookie_t *drc)
 	if (err) {
 		if (drc->drc_newfs) {
 			ASSERT(ds == drc->drc_real_ds);
-			(void) dsl_dataset_destroy(ds, dmu_recv_tag);
+			(void) dsl_dataset_destroy(ds, dmu_recv_tag,
+			    B_FALSE);
 			return (err);
 		} else {
 			(void) dsl_dataset_rollback(ds, DMU_OST_NONE);
diff --git a/module/zfs/dmu_tx.c b/module/zfs/dmu_tx.c
index 8735a249a..a3c9e5241 100644
--- a/module/zfs/dmu_tx.c
+++ b/module/zfs/dmu_tx.c
@@ -699,8 +699,7 @@ dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name)
 	}
 
 	err = zap_count_write(&dn->dn_objset->os, dn->dn_object, name, add,
-	    &txh->txh_space_towrite, &txh->txh_space_tooverwrite,
-	    txh->txh_dnode->dn_datablkshift);
+	    &txh->txh_space_towrite, &txh->txh_space_tooverwrite);
 
 	/*
 	 * If the modified blocks are scattered to the four winds,
diff --git a/module/zfs/dnode.c b/module/zfs/dnode.c
index 97e670cfb..c07b2528e 100644
--- a/module/zfs/dnode.c
+++ b/module/zfs/dnode.c
@@ -1308,6 +1308,22 @@ dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx)
 	dmu_tx_willuse_space(tx, space);
 }
 
+/*
+ * This function scans a block at the indicated "level" looking for
+ * a hole or data (depending on 'flags').  If level > 0, then we are
+ * scanning an indirect block looking at its pointers.  If level == 0,
+ * then we are looking at a block of dnodes.  If we don't find what we
+ * are looking for in the block, we return ESRCH.  Otherwise, return
+ * with *offset pointing to the beginning (if searching forwards) or
+ * end (if searching backwards) of the range covered by the block
+ * pointer we matched on (or dnode).
+ *
+ * The basic search algorithm used below by dnode_next_offset() is to
+ * use this function to search up the block tree (widen the search) until
+ * we find something (i.e., we don't return ESRCH) and then search back
+ * down the tree (narrow the search) until we reach our original search
+ * level.
+ */
 static int
 dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset,
 	int lvl, uint64_t blkfill, uint64_t txg)
@@ -1378,6 +1394,7 @@ dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset,
 			error = ESRCH;
 	} else {
 		blkptr_t *bp = data;
+		uint64_t start = *offset;
 		span = (lvl - 1) * epbs + dn->dn_datablkshift;
 		minfill = 0;
 		maxfill = blkfill << ((lvl - 1) * epbs);
@@ -1387,18 +1404,25 @@ dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset,
 		else
 			minfill++;
 
-		for (i = (*offset >> span) & ((1ULL << epbs) - 1);
+		*offset = *offset >> span;
+		for (i = BF64_GET(*offset, 0, epbs);
 		    i >= 0 && i < epb; i += inc) {
 			if (bp[i].blk_fill >= minfill &&
 			    bp[i].blk_fill <= maxfill &&
 			    (hole || bp[i].blk_birth > txg))
 				break;
-			if (inc < 0 && *offset < (1ULL << span))
-				*offset = 0;
-			else
-				*offset += (1ULL << span) * inc;
+			if (inc > 0 || *offset > 0)
+				*offset += inc;
+		}
+		*offset = *offset << span;
+		if (inc < 0) {
+			/* traversing backwards; position offset at the end */
+			ASSERT3U(*offset, <=, start);
+			*offset = MIN(*offset + (1ULL << span) - 1, start);
+		} else if (*offset < start) {
+			*offset = start;
 		}
-		if (i < 0 || i == epb)
+		if (i < 0 || i >= epb)
 			error = ESRCH;
 	}
 
diff --git a/module/zfs/dsl_dataset.c b/module/zfs/dsl_dataset.c
index be8ce684a..926034836 100644
--- a/module/zfs/dsl_dataset.c
+++ b/module/zfs/dsl_dataset.c
@@ -39,6 +39,7 @@
 #include <sys/spa.h>
 #include <sys/zfs_znode.h>
 #include <sys/sunddi.h>
+#include <sys/zvol.h>
 
 static char *dsl_reaper = "the grim reaper";
 
@@ -264,6 +265,7 @@ dsl_dataset_evict(dmu_buf_t *db, void *dsv)
 	ASSERT(!list_link_active(&ds->ds_synced_link));
 
 	mutex_destroy(&ds->ds_lock);
+	mutex_destroy(&ds->ds_recvlock);
 	mutex_destroy(&ds->ds_opening_lock);
 	mutex_destroy(&ds->ds_deadlist.bpl_lock);
 	rw_destroy(&ds->ds_rwlock);
@@ -362,6 +364,7 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 		list_link_init(&ds->ds_synced_link);
 
 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
+		mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 		mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
 		    NULL);
@@ -380,6 +383,7 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 			 * just opened it.
 			 */
 			mutex_destroy(&ds->ds_lock);
+			mutex_destroy(&ds->ds_recvlock);
 			mutex_destroy(&ds->ds_opening_lock);
 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
 			rw_destroy(&ds->ds_rwlock);
@@ -409,8 +413,15 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 					dsl_dataset_rele(origin, FTAG);
 				}
 			}
-		} else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) {
-			err = dsl_dataset_get_snapname(ds);
+		} else {
+			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
+				err = dsl_dataset_get_snapname(ds);
+			if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
+				err = zap_count(
+				    ds->ds_dir->dd_pool->dp_meta_objset,
+				    ds->ds_phys->ds_userrefs_obj,
+				    &ds->ds_userrefs);
+			}
 		}
 
 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
@@ -451,6 +462,7 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 				dsl_dataset_drop_ref(ds->ds_prev, ds);
 			dsl_dir_close(ds->ds_dir, ds);
 			mutex_destroy(&ds->ds_lock);
+			mutex_destroy(&ds->ds_recvlock);
 			mutex_destroy(&ds->ds_opening_lock);
 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
 			rw_destroy(&ds->ds_rwlock);
@@ -848,6 +860,7 @@ struct destroyarg {
 	dsl_sync_task_group_t *dstg;
 	char *snapname;
 	char *failed;
+	boolean_t defer;
 };
 
 static int
@@ -855,23 +868,30 @@ dsl_snapshot_destroy_one(char *name, void *arg)
 {
 	struct destroyarg *da = arg;
 	dsl_dataset_t *ds;
-	char *cp;
 	int err;
-
-	(void) strcat(name, "@");
-	(void) strcat(name, da->snapname);
-	err = dsl_dataset_own(name, DS_MODE_READONLY | DS_MODE_INCONSISTENT,
+	char *dsname;
+	size_t buflen;
+
+	/* alloc a buffer to hold name@snapname, plus the terminating NULL */
+	buflen = strlen(name) + strlen(da->snapname) + 2;
+	dsname = kmem_alloc(buflen, KM_SLEEP);
+	(void) snprintf(dsname, buflen, "%s@%s", name, da->snapname);
+	err = dsl_dataset_own(dsname, DS_MODE_READONLY | DS_MODE_INCONSISTENT,
 	    da->dstg, &ds);
-	cp = strchr(name, '@');
-	*cp = '\0';
+	kmem_free(dsname, buflen);
 	if (err == 0) {
+		struct dsl_ds_destroyarg *dsda;
+
 		dsl_dataset_make_exclusive(ds, da->dstg);
 		if (ds->ds_user_ptr) {
 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
 			ds->ds_user_ptr = NULL;
 		}
+		dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP);
+		dsda->ds = ds;
+		dsda->defer = da->defer;
 		dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
-		    dsl_dataset_destroy_sync, ds, da->dstg, 0);
+		    dsl_dataset_destroy_sync, dsda, da->dstg, 0);
 	} else if (err == ENOENT) {
 		err = 0;
 	} else {
@@ -885,7 +905,7 @@ dsl_snapshot_destroy_one(char *name, void *arg)
  */
 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
 int
-dsl_snapshots_destroy(char *fsname, char *snapname)
+dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer)
 {
 	int err;
 	struct destroyarg da;
@@ -898,6 +918,7 @@ dsl_snapshots_destroy(char *fsname, char *snapname)
 	da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
 	da.snapname = snapname;
 	da.failed = fsname;
+	da.defer = defer;
 
 	err = dmu_objset_find(fsname,
 	    dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
@@ -907,7 +928,9 @@ dsl_snapshots_destroy(char *fsname, char *snapname)
 
 	for (dst = list_head(&da.dstg->dstg_tasks); dst;
 	    dst = list_next(&da.dstg->dstg_tasks, dst)) {
-		dsl_dataset_t *ds = dst->dst_arg1;
+		struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
+		dsl_dataset_t *ds = dsda->ds;
+
 		/*
 		 * Return the file system name that triggered the error
 		 */
@@ -915,7 +938,9 @@ dsl_snapshots_destroy(char *fsname, char *snapname)
 			dsl_dataset_name(ds, fsname);
 			*strchr(fsname, '@') = '\0';
 		}
+		ASSERT3P(dsda->rm_origin, ==, NULL);
 		dsl_dataset_disown(ds, da.dstg);
+		kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
 	}
 
 	dsl_sync_task_group_destroy(da.dstg);
@@ -923,18 +948,100 @@ dsl_snapshots_destroy(char *fsname, char *snapname)
 	return (err);
 }
 
+static boolean_t
+dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
+{
+	boolean_t might_destroy = B_FALSE;
+
+	mutex_enter(&ds->ds_lock);
+	if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
+	    DS_IS_DEFER_DESTROY(ds))
+		might_destroy = B_TRUE;
+	mutex_exit(&ds->ds_lock);
+
+	return (might_destroy);
+}
+
+#ifdef _KERNEL
+static int
+dsl_dataset_zvol_cleanup(dsl_dataset_t *ds, const char *name)
+{
+	int error;
+	objset_t *os;
+
+	error = dmu_objset_open_ds(ds, DMU_OST_ANY, &os);
+	if (error)
+		return (error);
+
+	if (dmu_objset_type(os) == DMU_OST_ZVOL)
+		error = zvol_remove_minor(name);
+	dmu_objset_close(os);
+
+	return (error);
+}
+#endif
+
+/*
+ * If we're removing a clone, and these three conditions are true:
+ *	1) the clone's origin has no other children
+ *	2) the clone's origin has no user references
+ *	3) the clone's origin has been marked for deferred destruction
+ * Then, prepare to remove the origin as part of this sync task group.
+ */
+static int
+dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
+{
+	dsl_dataset_t *ds = dsda->ds;
+	dsl_dataset_t *origin = ds->ds_prev;
+
+	if (dsl_dataset_might_destroy_origin(origin)) {
+		char *name;
+		int namelen;
+		int error;
+
+		namelen = dsl_dataset_namelen(origin) + 1;
+		name = kmem_alloc(namelen, KM_SLEEP);
+		dsl_dataset_name(origin, name);
+#ifdef _KERNEL
+		error = zfs_unmount_snap(name, NULL);
+		if (error) {
+			kmem_free(name, namelen);
+			return (error);
+		}
+		error = dsl_dataset_zvol_cleanup(origin, name);
+		if (error) {
+			kmem_free(name, namelen);
+			return (error);
+		}
+#endif
+		error = dsl_dataset_own(name,
+		    DS_MODE_READONLY | DS_MODE_INCONSISTENT,
+		    tag, &origin);
+		kmem_free(name, namelen);
+		if (error)
+			return (error);
+		dsda->rm_origin = origin;
+		dsl_dataset_make_exclusive(origin, tag);
+	}
+
+	return (0);
+}
+
 /*
  * ds must be opened as OWNER.  On return (whether successful or not),
  * ds will be closed and caller can no longer dereference it.
  */
 int
-dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
+dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
 {
 	int err;
 	dsl_sync_task_group_t *dstg;
 	objset_t *os;
 	dsl_dir_t *dd;
 	uint64_t obj;
+	struct dsl_ds_destroyarg dsda = {0};
+
+	dsda.ds = ds;
 
 	if (dsl_dataset_is_snapshot(ds)) {
 		/* Destroying a snapshot is simpler */
@@ -944,9 +1051,12 @@ dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
 			ds->ds_user_ptr = NULL;
 		}
+		/* NOTE: defer is always B_FALSE for non-snapshots */
+		dsda.defer = defer;
 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
 		    dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
-		    ds, tag, 0);
+		    &dsda, tag, 0);
+		ASSERT3P(dsda.rm_origin, ==, NULL);
 		goto out;
 	}
 
@@ -1027,13 +1137,45 @@ dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
 		ds->ds_user_ptr = NULL;
 	}
-	dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
-	dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
-	    dsl_dataset_destroy_sync, ds, tag, 0);
-	dsl_sync_task_create(dstg, dsl_dir_destroy_check,
-	    dsl_dir_destroy_sync, dd, FTAG, 0);
-	err = dsl_sync_task_group_wait(dstg);
-	dsl_sync_task_group_destroy(dstg);
+
+	/*
+	 * If we're removing a clone, we might also need to remove its
+	 * origin.
+	 */
+	do {
+		dsda.need_prep = B_FALSE;
+		if (dsl_dir_is_clone(dd)) {
+			err = dsl_dataset_origin_rm_prep(&dsda, tag);
+			if (err) {
+				dsl_dir_close(dd, FTAG);
+				goto out;
+			}
+		}
+
+		dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
+		dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
+		    dsl_dataset_destroy_sync, &dsda, tag, 0);
+		dsl_sync_task_create(dstg, dsl_dir_destroy_check,
+		    dsl_dir_destroy_sync, dd, FTAG, 0);
+		err = dsl_sync_task_group_wait(dstg);
+		dsl_sync_task_group_destroy(dstg);
+
+		/*
+		 * We could be racing against 'zfs release' or 'zfs destroy -d'
+		 * on the origin snap, in which case we can get EBUSY if we
+		 * needed to destroy the origin snap but were not ready to
+		 * do so.
+		 */
+		if (dsda.need_prep) {
+			ASSERT(err == EBUSY);
+			ASSERT(dsl_dir_is_clone(dd));
+			ASSERT(dsda.rm_origin == NULL);
+		}
+	} while (dsda.need_prep);
+
+	if (dsda.rm_origin != NULL)
+		dsl_dataset_disown(dsda.rm_origin, tag);
+
 	/* if it is successful, dsl_dir_destroy_sync will close the dd */
 	if (err)
 		dsl_dir_close(dd, FTAG);
@@ -1214,7 +1356,8 @@ dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx)
 	/*
 	 * We can only roll back to emptyness if it is a ZPL objset.
 	 */
-	if (*ost != DMU_OST_ZFS && ds->ds_phys->ds_prev_snap_txg == 0)
+	if (*ost != DMU_OST_ZFS &&
+	    ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL)
 		return (EINVAL);
 
 	/*
@@ -1319,6 +1462,7 @@ dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 	} else {
 		objset_impl_t *osi;
 
+		ASSERT(*ost != DMU_OST_ZVOL);
 		ASSERT3U(ds->ds_phys->ds_used_bytes, ==, 0);
 		ASSERT3U(ds->ds_phys->ds_compressed_bytes, ==, 0);
 		ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, ==, 0);
@@ -1388,18 +1532,63 @@ dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 	    cr, "dataset = %llu", ds->ds_object);
 }
 
+static int
+dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
+    dmu_tx_t *tx)
+{
+	dsl_dataset_t *ds = dsda->ds;
+	dsl_dataset_t *ds_prev = ds->ds_prev;
+
+	if (dsl_dataset_might_destroy_origin(ds_prev)) {
+		struct dsl_ds_destroyarg ndsda = {0};
+
+		/*
+		 * If we're not prepared to remove the origin, don't remove
+		 * the clone either.
+		 */
+		if (dsda->rm_origin == NULL) {
+			dsda->need_prep = B_TRUE;
+			return (EBUSY);
+		}
+
+		ndsda.ds = ds_prev;
+		ndsda.is_origin_rm = B_TRUE;
+		return (dsl_dataset_destroy_check(&ndsda, tag, tx));
+	}
+
+	/*
+	 * If we're not going to remove the origin after all,
+	 * undo the open context setup.
+	 */
+	if (dsda->rm_origin != NULL) {
+		dsl_dataset_disown(dsda->rm_origin, tag);
+		dsda->rm_origin = NULL;
+	}
+
+	return (0);
+}
+
 /* ARGSUSED */
 int
 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
 {
-	dsl_dataset_t *ds = arg1;
+	struct dsl_ds_destroyarg *dsda = arg1;
+	dsl_dataset_t *ds = dsda->ds;
 
 	/* we have an owner hold, so noone else can destroy us */
 	ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
 
-	/* Can't delete a branch point. */
-	if (ds->ds_phys->ds_num_children > 1)
-		return (EEXIST);
+	/*
+	 * Only allow deferred destroy on pools that support it.
+	 * NOTE: deferred destroy is only supported on snapshots.
+	 */
+	if (dsda->defer) {
+		if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
+		    SPA_VERSION_USERREFS)
+			return (ENOTSUP);
+		ASSERT(dsl_dataset_is_snapshot(ds));
+		return (0);
+	}
 
 	/*
 	 * Can't delete a head dataset if there are snapshots of it.
@@ -1417,6 +1606,31 @@ dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
 		return (EAGAIN);
 
+	if (dsl_dataset_is_snapshot(ds)) {
+		/*
+		 * If this snapshot has an elevated user reference count,
+		 * we can't destroy it yet.
+		 */
+		if (ds->ds_userrefs > 0 && !dsda->releasing)
+			return (EBUSY);
+
+		mutex_enter(&ds->ds_lock);
+		/*
+		 * Can't delete a branch point. However, if we're destroying
+		 * a clone and removing its origin due to it having a user
+		 * hold count of 0 and having been marked for deferred destroy,
+		 * it's OK for the origin to have a single clone.
+		 */
+		if (ds->ds_phys->ds_num_children >
+		    (dsda->is_origin_rm ? 2 : 1)) {
+			mutex_exit(&ds->ds_lock);
+			return (EEXIST);
+		}
+		mutex_exit(&ds->ds_lock);
+	} else if (dsl_dir_is_clone(ds->ds_dir)) {
+		return (dsl_dataset_origin_check(dsda, arg2, tx));
+	}
+
 	/* XXX we should do some i/o error checking... */
 	return (0);
 }
@@ -1464,7 +1678,8 @@ dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
 void
 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
 {
-	dsl_dataset_t *ds = arg1;
+	struct dsl_ds_destroyarg *dsda = arg1;
+	dsl_dataset_t *ds = dsda->ds;
 	zio_t *zio;
 	int err;
 	int after_branch_point = FALSE;
@@ -1474,11 +1689,20 @@ dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
 	uint64_t obj;
 
 	ASSERT(ds->ds_owner);
-	ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
+	ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
 	ASSERT(ds->ds_prev == NULL ||
 	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
 	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
 
+	if (dsda->defer) {
+		ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
+		if (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1) {
+			dmu_buf_will_dirty(ds->ds_dbuf, tx);
+			ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
+			return;
+		}
+	}
+
 	/* signal any waiters that this dataset is going away */
 	mutex_enter(&ds->ds_lock);
 	ds->ds_owner = dsl_reaper;
@@ -1524,6 +1748,20 @@ dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
 			/* This clone is toast. */
 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
 			ds_prev->ds_phys->ds_num_children--;
+
+			/*
+			 * If the clone's origin has no other clones, no
+			 * user holds, and has been marked for deferred
+			 * deletion, then we should have done the necessary
+			 * destroy setup for it.
+			 */
+			if (ds_prev->ds_phys->ds_num_children == 1 &&
+			    ds_prev->ds_userrefs == 0 &&
+			    DS_IS_DEFER_DESTROY(ds_prev)) {
+				ASSERT3P(dsda->rm_origin, !=, NULL);
+			} else {
+				ASSERT3P(dsda->rm_origin, ==, NULL);
+			}
 		} else if (!after_branch_point) {
 			ds_prev->ds_phys->ds_next_snap_obj =
 			    ds->ds_phys->ds_next_snap_obj;
@@ -1736,10 +1974,32 @@ dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
 	}
 	if (ds->ds_phys->ds_props_obj != 0)
 		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
+	if (ds->ds_phys->ds_userrefs_obj != 0)
+		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
 	dsl_dir_close(ds->ds_dir, ds);
 	ds->ds_dir = NULL;
 	dsl_dataset_drain_refs(ds, tag);
 	VERIFY(0 == dmu_object_free(mos, obj, tx));
+
+	if (dsda->rm_origin) {
+		/*
+		 * Remove the origin of the clone we just destroyed.
+		 */
+		dsl_dataset_t *origin = ds->ds_prev;
+		struct dsl_ds_destroyarg ndsda = {0};
+
+		ASSERT3P(origin, ==, dsda->rm_origin);
+		if (origin->ds_user_ptr) {
+			origin->ds_user_evict_func(origin, origin->ds_user_ptr);
+			origin->ds_user_ptr = NULL;
+		}
+
+		dsl_dataset_rele(origin, tag);
+		ds->ds_prev = NULL;
+
+		ndsda.ds = origin;
+		dsl_dataset_destroy_sync(&ndsda, tag, cr, tx);
+	}
 }
 
 static int
@@ -1954,6 +2214,9 @@ dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
 	    ds->ds_reserved);
 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
 	    ds->ds_phys->ds_guid);
+	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, ds->ds_userrefs);
+	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
+	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
 
 	if (ds->ds_phys->ds_next_snap_obj) {
 		/*
@@ -3023,7 +3286,7 @@ dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 
 	ds->ds_quota = new_quota;
 
-	dsl_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
+	dsl_dir_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
 
 	spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
 	    tx, cr, "%lld dataset = %llu ",
@@ -3118,7 +3381,7 @@ dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
 
 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
 	mutex_exit(&ds->ds_dir->dd_lock);
-	dsl_prop_set_uint64_sync(ds->ds_dir, "refreservation",
+	dsl_dir_prop_set_uint64_sync(ds->ds_dir, "refreservation",
 	    new_reservation, cr, tx);
 
 	spa_history_internal_log(LOG_DS_REFRESERV,
@@ -3142,3 +3405,421 @@ dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
 	dsl_dataset_rele(ds, FTAG);
 	return (err);
 }
+
+static int
+dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
+{
+	dsl_dataset_t *ds = arg1;
+	char *htag = arg2;
+	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
+	int error = 0;
+
+	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
+		return (ENOTSUP);
+
+	if (!dsl_dataset_is_snapshot(ds))
+		return (EINVAL);
+
+	if (strlen(htag) >= ZAP_MAXNAMELEN)
+		return (ENAMETOOLONG);
+
+	/* tags must be unique */
+	mutex_enter(&ds->ds_lock);
+	if (ds->ds_phys->ds_userrefs_obj) {
+		error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
+		    8, 1, tx);
+		if (error == 0)
+			error = EEXIST;
+		else if (error == ENOENT)
+			error = 0;
+	}
+	mutex_exit(&ds->ds_lock);
+
+	return (error);
+}
+
+static void
+dsl_dataset_user_hold_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
+{
+	dsl_dataset_t *ds = arg1;
+	char *htag = arg2;
+	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
+	time_t now = gethrestime_sec();
+	uint64_t zapobj;
+
+	mutex_enter(&ds->ds_lock);
+	if (ds->ds_phys->ds_userrefs_obj == 0) {
+		/*
+		 * This is the first user hold for this dataset.  Create
+		 * the userrefs zap object.
+		 */
+		dmu_buf_will_dirty(ds->ds_dbuf, tx);
+		zapobj = ds->ds_phys->ds_userrefs_obj =
+		    zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
+	} else {
+		zapobj = ds->ds_phys->ds_userrefs_obj;
+	}
+	ds->ds_userrefs++;
+	mutex_exit(&ds->ds_lock);
+
+	VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
+
+	spa_history_internal_log(LOG_DS_USER_HOLD,
+	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "<%s> dataset = %llu",
+	    htag, ds->ds_object);
+}
+
+struct dsl_ds_holdarg {
+	dsl_sync_task_group_t *dstg;
+	char *htag;
+	char *snapname;
+	boolean_t recursive;
+	char failed[MAXPATHLEN];
+};
+
+static int
+dsl_dataset_user_hold_one(char *dsname, void *arg)
+{
+	struct dsl_ds_holdarg *ha = arg;
+	dsl_dataset_t *ds;
+	int error;
+	char *name;
+	size_t buflen;
+
+	/* alloc a buffer to hold dsname@snapname plus terminating NULL */
+	buflen = strlen(dsname) + strlen(ha->snapname) + 2;
+	name = kmem_alloc(buflen, KM_SLEEP);
+	(void) snprintf(name, buflen, "%s@%s", dsname, ha->snapname);
+	error = dsl_dataset_hold(name, ha->dstg, &ds);
+	kmem_free(name, buflen);
+	if (error == 0) {
+		dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
+		    dsl_dataset_user_hold_sync, ds, ha->htag, 0);
+	} else if (error == ENOENT && ha->recursive) {
+		error = 0;
+	} else {
+		(void) strcpy(ha->failed, dsname);
+	}
+	return (error);
+}
+
+int
+dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
+    boolean_t recursive)
+{
+	struct dsl_ds_holdarg *ha;
+	dsl_sync_task_t *dst;
+	spa_t *spa;
+	int error;
+
+	ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
+
+	(void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
+
+	error = spa_open(dsname, &spa, FTAG);
+	if (error) {
+		kmem_free(ha, sizeof (struct dsl_ds_holdarg));
+		return (error);
+	}
+
+	ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
+	ha->htag = htag;
+	ha->snapname = snapname;
+	ha->recursive = recursive;
+	if (recursive) {
+		error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
+		    ha, DS_FIND_CHILDREN);
+	} else {
+		error = dsl_dataset_user_hold_one(dsname, ha);
+	}
+	if (error == 0)
+		error = dsl_sync_task_group_wait(ha->dstg);
+
+	for (dst = list_head(&ha->dstg->dstg_tasks); dst;
+	    dst = list_next(&ha->dstg->dstg_tasks, dst)) {
+		dsl_dataset_t *ds = dst->dst_arg1;
+
+		if (dst->dst_err) {
+			dsl_dataset_name(ds, ha->failed);
+			*strchr(ha->failed, '@') = '\0';
+		}
+		dsl_dataset_rele(ds, ha->dstg);
+	}
+
+	if (error)
+		(void) strcpy(dsname, ha->failed);
+
+	dsl_sync_task_group_destroy(ha->dstg);
+	kmem_free(ha, sizeof (struct dsl_ds_holdarg));
+	spa_close(spa, FTAG);
+	return (error);
+}
+
+struct dsl_ds_releasearg {
+	dsl_dataset_t *ds;
+	const char *htag;
+	boolean_t own;		/* do we own or just hold ds? */
+};
+
+static int
+dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
+    boolean_t *might_destroy)
+{
+	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
+	uint64_t zapobj;
+	uint64_t tmp;
+	int error;
+
+	*might_destroy = B_FALSE;
+
+	mutex_enter(&ds->ds_lock);
+	zapobj = ds->ds_phys->ds_userrefs_obj;
+	if (zapobj == 0) {
+		/* The tag can't possibly exist */
+		mutex_exit(&ds->ds_lock);
+		return (ESRCH);
+	}
+
+	/* Make sure the tag exists */
+	error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
+	if (error) {
+		mutex_exit(&ds->ds_lock);
+		if (error == ENOENT)
+			error = ESRCH;
+		return (error);
+	}
+
+	if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
+	    DS_IS_DEFER_DESTROY(ds))
+		*might_destroy = B_TRUE;
+
+	mutex_exit(&ds->ds_lock);
+	return (0);
+}
+
+static int
+dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
+{
+	struct dsl_ds_releasearg *ra = arg1;
+	dsl_dataset_t *ds = ra->ds;
+	boolean_t might_destroy;
+	int error;
+
+	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
+		return (ENOTSUP);
+
+	error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
+	if (error)
+		return (error);
+
+	if (might_destroy) {
+		struct dsl_ds_destroyarg dsda = {0};
+
+		if (dmu_tx_is_syncing(tx)) {
+			/*
+			 * If we're not prepared to remove the snapshot,
+			 * we can't allow the release to happen right now.
+			 */
+			if (!ra->own)
+				return (EBUSY);
+			if (ds->ds_user_ptr) {
+				ds->ds_user_evict_func(ds, ds->ds_user_ptr);
+				ds->ds_user_ptr = NULL;
+			}
+		}
+		dsda.ds = ds;
+		dsda.releasing = B_TRUE;
+		return (dsl_dataset_destroy_check(&dsda, tag, tx));
+	}
+
+	return (0);
+}
+
+static void
+dsl_dataset_user_release_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
+{
+	struct dsl_ds_releasearg *ra = arg1;
+	dsl_dataset_t *ds = ra->ds;
+	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
+	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
+	uint64_t zapobj;
+	uint64_t dsobj = ds->ds_object;
+	uint64_t refs;
+
+	mutex_enter(&ds->ds_lock);
+	ds->ds_userrefs--;
+	refs = ds->ds_userrefs;
+	mutex_exit(&ds->ds_lock);
+	zapobj = ds->ds_phys->ds_userrefs_obj;
+	VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
+	if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
+	    DS_IS_DEFER_DESTROY(ds)) {
+		struct dsl_ds_destroyarg dsda = {0};
+
+		ASSERT(ra->own);
+		dsda.ds = ds;
+		dsda.releasing = B_TRUE;
+		/* We already did the destroy_check */
+		dsl_dataset_destroy_sync(&dsda, tag, cr, tx);
+	}
+
+	spa_history_internal_log(LOG_DS_USER_RELEASE,
+	    spa, tx, cr, "<%s> %lld dataset = %llu",
+	    ra->htag, (longlong_t)refs, dsobj);
+}
+
+static int
+dsl_dataset_user_release_one(char *dsname, void *arg)
+{
+	struct dsl_ds_holdarg *ha = arg;
+	struct dsl_ds_releasearg *ra;
+	dsl_dataset_t *ds;
+	int error;
+	void *dtag = ha->dstg;
+	char *name;
+	size_t buflen;
+	boolean_t own = B_FALSE;
+	boolean_t might_destroy;
+
+	if (strlen(ha->htag) >= ZAP_MAXNAMELEN)
+		return (ENAMETOOLONG);
+
+	/* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
+	buflen = strlen(dsname) + strlen(ha->snapname) + 2;
+	name = kmem_alloc(buflen, KM_SLEEP);
+	(void) snprintf(name, buflen, "%s@%s", dsname, ha->snapname);
+	error = dsl_dataset_hold(name, dtag, &ds);
+	kmem_free(name, buflen);
+	if (error == ENOENT && ha->recursive)
+		return (0);
+	(void) strcpy(ha->failed, dsname);
+	if (error)
+		return (error);
+
+	ASSERT(dsl_dataset_is_snapshot(ds));
+
+	error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
+	if (error) {
+		dsl_dataset_rele(ds, dtag);
+		return (error);
+	}
+
+	if (might_destroy) {
+#ifdef _KERNEL
+		error = zfs_unmount_snap(name, NULL);
+		if (error) {
+			dsl_dataset_rele(ds, dtag);
+			return (error);
+		}
+		error = dsl_dataset_zvol_cleanup(ds, name);
+		if (error) {
+			dsl_dataset_rele(ds, dtag);
+			return (error);
+		}
+#endif
+		if (!dsl_dataset_tryown(ds,
+		    DS_MODE_READONLY | DS_MODE_INCONSISTENT, dtag)) {
+			dsl_dataset_rele(ds, dtag);
+			return (EBUSY);
+		} else {
+			own = B_TRUE;
+			dsl_dataset_make_exclusive(ds, dtag);
+		}
+	}
+
+	ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
+	ra->ds = ds;
+	ra->htag = ha->htag;
+	ra->own = own;
+	dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
+	    dsl_dataset_user_release_sync, ra, dtag, 0);
+
+	return (0);
+}
+
+int
+dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
+    boolean_t recursive)
+{
+	struct dsl_ds_holdarg *ha;
+	dsl_sync_task_t *dst;
+	spa_t *spa;
+	int error;
+
+	ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
+
+	(void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
+
+	error = spa_open(dsname, &spa, FTAG);
+	if (error) {
+		kmem_free(ha, sizeof (struct dsl_ds_holdarg));
+		return (error);
+	}
+
+	ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
+	ha->htag = htag;
+	ha->snapname = snapname;
+	ha->recursive = recursive;
+	if (recursive) {
+		error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
+		    ha, DS_FIND_CHILDREN);
+	} else {
+		error = dsl_dataset_user_release_one(dsname, ha);
+	}
+	if (error == 0)
+		error = dsl_sync_task_group_wait(ha->dstg);
+
+	for (dst = list_head(&ha->dstg->dstg_tasks); dst;
+	    dst = list_next(&ha->dstg->dstg_tasks, dst)) {
+		struct dsl_ds_releasearg *ra = dst->dst_arg1;
+		dsl_dataset_t *ds = ra->ds;
+
+		if (dst->dst_err)
+			dsl_dataset_name(ds, ha->failed);
+
+		if (ra->own)
+			dsl_dataset_disown(ds, ha->dstg);
+		else
+			dsl_dataset_rele(ds, ha->dstg);
+
+		kmem_free(ra, sizeof (struct dsl_ds_releasearg));
+	}
+
+	if (error)
+		(void) strcpy(dsname, ha->failed);
+
+	dsl_sync_task_group_destroy(ha->dstg);
+	kmem_free(ha, sizeof (struct dsl_ds_holdarg));
+	spa_close(spa, FTAG);
+	return (error);
+}
+
+int
+dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
+{
+	dsl_dataset_t *ds;
+	int err;
+
+	err = dsl_dataset_hold(dsname, FTAG, &ds);
+	if (err)
+		return (err);
+
+	VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
+	if (ds->ds_phys->ds_userrefs_obj != 0) {
+		zap_attribute_t *za;
+		zap_cursor_t zc;
+
+		za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
+		for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
+		    ds->ds_phys->ds_userrefs_obj);
+		    zap_cursor_retrieve(&zc, za) == 0;
+		    zap_cursor_advance(&zc)) {
+			VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
+			    za->za_first_integer));
+		}
+		zap_cursor_fini(&zc);
+		kmem_free(za, sizeof (zap_attribute_t));
+	}
+	dsl_dataset_rele(ds, FTAG);
+	return (0);
+}
diff --git a/module/zfs/dsl_deleg.c b/module/zfs/dsl_deleg.c
index 0c522fbfa..22605053e 100644
--- a/module/zfs/dsl_deleg.c
+++ b/module/zfs/dsl_deleg.c
@@ -19,7 +19,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -66,8 +66,6 @@
  * The ZAP OBJ is referred to as the jump object.
  */
 
-
-
 #include <sys/dmu.h>
 #include <sys/dmu_objset.h>
 #include <sys/dmu_tx.h>
@@ -540,7 +538,7 @@ dsl_deleg_access(const char *dsname, const char *perm, cred_t *cr)
 	dsl_pool_t *dp;
 	void *cookie;
 	int	error;
-	char	checkflag = ZFS_DELEG_LOCAL;
+	char	checkflag;
 	objset_t *mos;
 	avl_tree_t permsets;
 	perm_set_t *setnode;
@@ -563,6 +561,16 @@ dsl_deleg_access(const char *dsname, const char *perm, cred_t *cr)
 		return (EPERM);
 	}
 
+	if (dsl_dataset_is_snapshot(ds)) {
+		/*
+		 * Snapshots are treated as descendents only,
+		 * local permissions do not apply.
+		 */
+		checkflag = ZFS_DELEG_DESCENDENT;
+	} else {
+		checkflag = ZFS_DELEG_LOCAL;
+	}
+
 	avl_create(&permsets, perm_set_compare, sizeof (perm_set_t),
 	    offsetof(perm_set_t, p_node));
 
diff --git a/module/zfs/dsl_dir.c b/module/zfs/dsl_dir.c
index e3be20d80..eb2767fe1 100644
--- a/module/zfs/dsl_dir.c
+++ b/module/zfs/dsl_dir.c
@@ -96,7 +96,6 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
 #endif
 	if (dd == NULL) {
 		dsl_dir_t *winner;
-		int err;
 
 		dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
 		dd->dd_object = ddobj;
diff --git a/module/zfs/dsl_prop.c b/module/zfs/dsl_prop.c
index 526fb5cc4..4708565b1 100644
--- a/module/zfs/dsl_prop.c
+++ b/module/zfs/dsl_prop.c
@@ -442,7 +442,7 @@ dsl_props_set_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 }
 
 void
-dsl_prop_set_uint64_sync(dsl_dir_t *dd, const char *name, uint64_t val,
+dsl_dir_prop_set_uint64_sync(dsl_dir_t *dd, const char *name, uint64_t val,
     cred_t *cr, dmu_tx_t *tx)
 {
 	objset_t *mos = dd->dd_pool->dp_meta_objset;
diff --git a/module/zfs/dsl_scrub.c b/module/zfs/dsl_scrub.c
index d1d1e42d3..3d59be3f9 100644
--- a/module/zfs/dsl_scrub.c
+++ b/module/zfs/dsl_scrub.c
@@ -1025,6 +1025,8 @@ dsl_pool_scrub_clean_cb(dsl_pool_t *dp,
 int
 dsl_pool_scrub_clean(dsl_pool_t *dp)
 {
+	spa_t *spa = dp->dp_spa;
+
 	/*
 	 * Purge all vdev caches.  We do this here rather than in sync
 	 * context because this requires a writer lock on the spa_config
@@ -1032,11 +1034,11 @@ dsl_pool_scrub_clean(dsl_pool_t *dp)
 	 * spa_scrub_reopen flag indicates that vdev_open() should not
 	 * attempt to start another scrub.
 	 */
-	spa_config_enter(dp->dp_spa, SCL_ALL, FTAG, RW_WRITER);
-	dp->dp_spa->spa_scrub_reopen = B_TRUE;
-	vdev_reopen(dp->dp_spa->spa_root_vdev);
-	dp->dp_spa->spa_scrub_reopen = B_FALSE;
-	spa_config_exit(dp->dp_spa, SCL_ALL, FTAG);
+	spa_vdev_state_enter(spa);
+	spa->spa_scrub_reopen = B_TRUE;
+	vdev_reopen(spa->spa_root_vdev);
+	spa->spa_scrub_reopen = B_FALSE;
+	(void) spa_vdev_state_exit(spa, NULL, 0);
 
 	return (dsl_pool_scrub_setup(dp, SCRUB_FUNC_CLEAN));
 }
diff --git a/module/zfs/include/sys/dmu.h b/module/zfs/include/sys/dmu.h
index 989b6ff24..b15da8391 100644
--- a/module/zfs/include/sys/dmu.h
+++ b/module/zfs/include/sys/dmu.h
@@ -117,6 +117,7 @@ typedef enum dmu_object_type {
 	DMU_OT_SCRUB_QUEUE,		/* ZAP */
 	DMU_OT_USERGROUP_USED,		/* ZAP */
 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
+	DMU_OT_USERREFS,		/* ZAP */
 	DMU_OT_NUMTYPES
 } dmu_object_type_t;
 
@@ -174,8 +175,8 @@ int dmu_objset_evict_dbufs(objset_t *os);
 int dmu_objset_create(const char *name, dmu_objset_type_t type,
     objset_t *clone_parent, uint64_t flags,
     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
-int dmu_objset_destroy(const char *name);
-int dmu_snapshots_destroy(char *fsname, char *snapname);
+int dmu_objset_destroy(const char *name, boolean_t defer);
+int dmu_snapshots_destroy(char *fsname, char *snapname, boolean_t defer);
 int dmu_objset_rollback(objset_t *os);
 int dmu_objset_snapshot(char *fsname, char *snapname, struct nvlist *props,
     boolean_t recursive);
@@ -666,10 +667,9 @@ typedef struct dmu_recv_cookie {
 } dmu_recv_cookie_t;
 
 int dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *,
-    boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *);
+    boolean_t force, objset_t *origin, dmu_recv_cookie_t *);
 int dmu_recv_stream(dmu_recv_cookie_t *drc, struct vnode *vp, offset_t *voffp);
 int dmu_recv_end(dmu_recv_cookie_t *drc);
-void dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc);
 
 /* CRC64 table */
 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
diff --git a/module/zfs/include/sys/dmu_impl.h b/module/zfs/include/sys/dmu_impl.h
index 96ce688e1..3868a5816 100644
--- a/module/zfs/include/sys/dmu_impl.h
+++ b/module/zfs/include/sys/dmu_impl.h
@@ -19,7 +19,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -211,10 +211,11 @@ extern "C" {
  * ds_lock
  *    protects:
  *    	ds_user_ptr
- *    	ds_user_evice_func
+ *    	ds_user_evict_func
  *    	ds_open_refcount
  *    	ds_snapname
  *    	ds_phys accounting
+ *	ds_phys userrefs zapobj
  *	ds_reserved
  *    held from:
  *    	dsl_dataset_*
diff --git a/module/zfs/include/sys/dmu_objset.h b/module/zfs/include/sys/dmu_objset.h
index 82cb6ad7d..052cb8dd9 100644
--- a/module/zfs/include/sys/dmu_objset.h
+++ b/module/zfs/include/sys/dmu_objset.h
@@ -117,7 +117,7 @@ void dmu_objset_close(objset_t *os);
 int dmu_objset_create(const char *name, dmu_objset_type_t type,
     objset_t *clone_parent, uint64_t flags,
     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
-int dmu_objset_destroy(const char *name);
+int dmu_objset_destroy(const char *name, boolean_t defer);
 int dmu_objset_rollback(objset_t *os);
 int dmu_objset_snapshot(char *fsname, char *snapname, nvlist_t *props,
     boolean_t recursive);
diff --git a/module/zfs/include/sys/dsl_dataset.h b/module/zfs/include/sys/dsl_dataset.h
index a1c2896e3..b51036d38 100644
--- a/module/zfs/include/sys/dsl_dataset.h
+++ b/module/zfs/include/sys/dsl_dataset.h
@@ -63,6 +63,14 @@ typedef void dsl_dataset_evict_func_t(struct dsl_dataset *, void *);
 #define	DS_FLAG_UNIQUE_ACCURATE	(1ULL<<2)
 
 /*
+ * DS_FLAG_DEFER_DESTROY is set after 'zfs destroy -d' has been called
+ * on a dataset. This allows the dataset to be destroyed using 'zfs release'.
+ */
+#define	DS_FLAG_DEFER_DESTROY	(1ULL<<3)
+#define	DS_IS_DEFER_DESTROY(ds)	\
+	((ds)->ds_phys->ds_flags & DS_FLAG_DEFER_DESTROY)
+
+/*
  * DS_FLAG_CI_DATASET is set if the dataset contains a file system whose
  * name lookups should be performed case-insensitively.
  */
@@ -93,7 +101,8 @@ typedef struct dsl_dataset_phys {
 	blkptr_t ds_bp;
 	uint64_t ds_next_clones_obj;	/* DMU_OT_DSL_CLONES */
 	uint64_t ds_props_obj;		/* DMU_OT_DSL_PROPS for snaps */
-	uint64_t ds_pad[6]; /* pad out to 320 bytes for good measure */
+	uint64_t ds_userrefs_obj;	/* DMU_OT_USERREFS */
+	uint64_t ds_pad[5]; /* pad out to 320 bytes for good measure */
 } dsl_dataset_phys_t;
 
 typedef struct dsl_dataset {
@@ -111,6 +120,9 @@ typedef struct dsl_dataset {
 	/* has internal locking: */
 	bplist_t ds_deadlist;
 
+	/* to protect against multiple concurrent incremental recv */
+	kmutex_t ds_recvlock;
+
 	/* protected by lock on pool's dp_dirty_datasets list */
 	txg_node_t ds_dirty_link;
 	list_node_t ds_synced_link;
@@ -122,6 +134,7 @@ typedef struct dsl_dataset {
 	kmutex_t ds_lock;
 	void *ds_user_ptr;
 	dsl_dataset_evict_func_t *ds_user_evict_func;
+	uint64_t ds_userrefs;
 
 	/*
 	 * ds_owner is protected by the ds_rwlock and the ds_lock
@@ -143,6 +156,15 @@ typedef struct dsl_dataset {
 	char ds_snapname[MAXNAMELEN];
 } dsl_dataset_t;
 
+struct dsl_ds_destroyarg {
+	dsl_dataset_t *ds;		/* ds to destroy */
+	dsl_dataset_t *rm_origin;	/* also remove our origin? */
+	boolean_t is_origin_rm;		/* set if removing origin snap */
+	boolean_t defer;		/* destroy -d requested? */
+	boolean_t releasing;		/* destroying due to release? */
+	boolean_t need_prep;		/* do we need to retry due to EBUSY? */
+};
+
 #define	dsl_dataset_is_snapshot(ds)	\
 	((ds)->ds_phys->ds_num_children != 0)
 
@@ -167,8 +189,8 @@ uint64_t dsl_dataset_create_sync(dsl_dir_t *pds, const char *lastname,
     dsl_dataset_t *origin, uint64_t flags, cred_t *, dmu_tx_t *);
 uint64_t dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
     uint64_t flags, dmu_tx_t *tx);
-int dsl_dataset_destroy(dsl_dataset_t *ds, void *tag);
-int dsl_snapshots_destroy(char *fsname, char *snapname);
+int dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer);
+int dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer);
 dsl_checkfunc_t dsl_dataset_destroy_check;
 dsl_syncfunc_t dsl_dataset_destroy_sync;
 dsl_checkfunc_t dsl_dataset_snapshot_check;
@@ -178,6 +200,11 @@ int dsl_dataset_rename(char *name, const char *newname, boolean_t recursive);
 int dsl_dataset_promote(const char *name);
 int dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
     boolean_t force);
+int dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
+    boolean_t recursive);
+int dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
+    boolean_t recursive);
+int dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp);
 
 void *dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
     void *p, dsl_dataset_evict_func_t func);
diff --git a/module/zfs/include/sys/dsl_deleg.h b/module/zfs/include/sys/dsl_deleg.h
index b064c9228..a26a3f705 100644
--- a/module/zfs/include/sys/dsl_deleg.h
+++ b/module/zfs/include/sys/dsl_deleg.h
@@ -53,6 +53,8 @@ extern "C" {
 #define	ZFS_DELEG_PERM_GROUPQUOTA	"groupquota"
 #define	ZFS_DELEG_PERM_USERUSED		"userused"
 #define	ZFS_DELEG_PERM_GROUPUSED	"groupused"
+#define	ZFS_DELEG_PERM_HOLD		"hold"
+#define	ZFS_DELEG_PERM_RELEASE		"release"
 
 /*
  * Note: the names of properties that are marked delegatable are also
diff --git a/module/zfs/include/sys/dsl_prop.h b/module/zfs/include/sys/dsl_prop.h
index 26018a46d..5afaa1f0d 100644
--- a/module/zfs/include/sys/dsl_prop.h
+++ b/module/zfs/include/sys/dsl_prop.h
@@ -69,7 +69,7 @@ dsl_syncfunc_t dsl_props_set_sync;
 int dsl_prop_set(const char *ddname, const char *propname,
     int intsz, int numints, const void *buf);
 int dsl_props_set(const char *dsname, nvlist_t *nvl);
-void dsl_prop_set_uint64_sync(dsl_dir_t *dd, const char *name, uint64_t val,
+void dsl_dir_prop_set_uint64_sync(dsl_dir_t *dd, const char *name, uint64_t val,
     cred_t *cr, dmu_tx_t *tx);
 
 void dsl_prop_nvlist_add_uint64(nvlist_t *nv, zfs_prop_t prop, uint64_t value);
diff --git a/module/zfs/include/sys/spa.h b/module/zfs/include/sys/spa.h
index f306c7fa3..30554ae0e 100644
--- a/module/zfs/include/sys/spa.h
+++ b/module/zfs/include/sys/spa.h
@@ -500,8 +500,9 @@ extern int spa_history_get(spa_t *spa, uint64_t *offset, uint64_t *len_read,
     char *his_buf);
 extern int spa_history_log(spa_t *spa, const char *his_buf,
     history_log_type_t what);
-void spa_history_internal_log(history_internal_events_t event, spa_t *spa,
-    dmu_tx_t *tx, cred_t *cr, const char *fmt, ...);
+extern void spa_history_internal_log(history_internal_events_t event,
+    spa_t *spa, dmu_tx_t *tx, cred_t *cr, const char *fmt, ...);
+extern void spa_history_log_version(spa_t *spa, history_internal_events_t evt);
 
 /* error handling */
 struct zbookmark;
diff --git a/module/zfs/include/sys/spa_impl.h b/module/zfs/include/sys/spa_impl.h
index 12999ee9e..84da68488 100644
--- a/module/zfs/include/sys/spa_impl.h
+++ b/module/zfs/include/sys/spa_impl.h
@@ -105,6 +105,7 @@ struct spa {
 	int		spa_inject_ref;		/* injection references */
 	uint8_t		spa_sync_on;		/* sync threads are running */
 	spa_load_state_t spa_load_state;	/* current load operation */
+	boolean_t	spa_load_verbatim;	/* load the given config? */
 	taskq_t		*spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES];
 	dsl_pool_t	*spa_dsl_pool;
 	metaslab_class_t *spa_normal_class;	/* normal data class */
diff --git a/module/zfs/include/sys/vdev.h b/module/zfs/include/sys/vdev.h
index 71b9b12d6..7e53f62d2 100644
--- a/module/zfs/include/sys/vdev.h
+++ b/module/zfs/include/sys/vdev.h
@@ -47,6 +47,7 @@ typedef enum vdev_dtl_type {
 extern boolean_t zfs_nocacheflush;
 
 extern int vdev_open(vdev_t *);
+extern void vdev_open_children(vdev_t *vd);
 extern int vdev_validate(vdev_t *);
 extern void vdev_close(vdev_t *);
 extern int vdev_create(vdev_t *, uint64_t txg, boolean_t isreplace);
diff --git a/module/zfs/include/sys/vdev_impl.h b/module/zfs/include/sys/vdev_impl.h
index 8240b66ac..23780430d 100644
--- a/module/zfs/include/sys/vdev_impl.h
+++ b/module/zfs/include/sys/vdev_impl.h
@@ -127,6 +127,8 @@ struct vdev {
 	space_map_t	vdev_dtl[DTL_TYPES]; /* in-core dirty time logs	*/
 	vdev_stat_t	vdev_stat;	/* virtual device statistics	*/
 	boolean_t	vdev_expanding;	/* expand the vdev?		*/
+	int		vdev_open_error; /* error on last open		*/
+	kthread_t	*vdev_open_thread; /* thread opening children	*/
 
 	/*
 	 * Top-level vdev state.
diff --git a/module/zfs/include/sys/zap.h b/module/zfs/include/sys/zap.h
index aa18ea832..e0775adc7 100644
--- a/module/zfs/include/sys/zap.h
+++ b/module/zfs/include/sys/zap.h
@@ -182,8 +182,7 @@ int zap_lookup_norm(objset_t *ds, uint64_t zapobj, const char *name,
     boolean_t *normalization_conflictp);
 
 int zap_count_write(objset_t *os, uint64_t zapobj, const char *name,
-    int add, uint64_t *towrite, uint64_t *tooverwrite,
-    uint64_t dn_datablkshift);
+    int add, uint64_t *towrite, uint64_t *tooverwrite);
 
 /*
  * Create an attribute with the given name and value.
diff --git a/module/zfs/include/sys/zfs_acl.h b/module/zfs/include/sys/zfs_acl.h
index f5e5aa7f4..3488962e2 100644
--- a/module/zfs/include/sys/zfs_acl.h
+++ b/module/zfs/include/sys/zfs_acl.h
@@ -203,6 +203,7 @@ void zfs_oldace_byteswap(ace_t *, int);
 void zfs_ace_byteswap(void *, size_t, boolean_t);
 extern boolean_t zfs_has_access(struct znode *zp, cred_t *cr);
 extern int zfs_zaccess(struct znode *, int, int, boolean_t, cred_t *);
+int zfs_fastaccesschk_execute(struct znode *, cred_t *);
 extern int zfs_zaccess_rwx(struct znode *, mode_t, int, cred_t *);
 extern int zfs_zaccess_unix(struct znode *, mode_t, cred_t *);
 extern int zfs_acl_access(struct znode *, int, cred_t *);
diff --git a/module/zfs/include/sys/zfs_ioctl.h b/module/zfs/include/sys/zfs_ioctl.h
index 1e9f35155..3a3e6e711 100644
--- a/module/zfs/include/sys/zfs_ioctl.h
+++ b/module/zfs/include/sys/zfs_ioctl.h
@@ -165,6 +165,7 @@ typedef struct zfs_cmd {
 	dmu_objset_stats_t zc_objset_stats;
 	struct drr_begin zc_begin_record;
 	zinject_record_t zc_inject_record;
+	boolean_t	zc_defer_destroy;
 } zfs_cmd_t;
 
 typedef struct zfs_useracct {
diff --git a/module/zfs/include/sys/zfs_vfsops.h b/module/zfs/include/sys/zfs_vfsops.h
index b8ed7b27f..28555232b 100644
--- a/module/zfs/include/sys/zfs_vfsops.h
+++ b/module/zfs/include/sys/zfs_vfsops.h
@@ -73,7 +73,6 @@ struct zfsvfs {
 	boolean_t	z_vscan;	/* virus scan on/off */
 	boolean_t	z_use_fuids;	/* version allows fuids */
 	boolean_t	z_replay;	/* set during ZIL replay */
-	kmutex_t	z_online_recv_lock; /* held while recv in progress */
 	uint64_t	z_version;	/* ZPL version */
 	uint64_t	z_shares_dir;	/* hidden shares dir */
 	kmutex_t	z_lock;
diff --git a/module/zfs/include/sys/zfs_znode.h b/module/zfs/include/sys/zfs_znode.h
index 5d06bcd28..f5ee2fc7b 100644
--- a/module/zfs/include/sys/zfs_znode.h
+++ b/module/zfs/include/sys/zfs_znode.h
@@ -77,6 +77,7 @@ extern "C" {
 #define	ZFS_ACL_DEFAULTED	0x20		/* ACL should be defaulted */
 #define	ZFS_ACL_AUTO_INHERIT	0x40		/* ACL should be inherited */
 #define	ZFS_BONUS_SCANSTAMP	0x80		/* Scanstamp in bonus area */
+#define	ZFS_NO_EXECS_DENIED	0x100		/* exec was given to everyone */
 
 /*
  * Is ID ephemeral?
@@ -200,6 +201,7 @@ typedef struct znode {
 	uint64_t	z_gen;		/* generation (same as zp_gen) */
 	uint32_t	z_sync_cnt;	/* synchronous open count */
 	kmutex_t	z_acl_lock;	/* acl data lock */
+	zfs_acl_t	*z_acl_cached;	/* cached acl */
 	list_node_t	z_link_node;	/* all znodes in fs link */
 	/*
 	 * These are dmu managed fields.
diff --git a/module/zfs/include/sys/zio.h b/module/zfs/include/sys/zio.h
index 5c51717c1..e47d8f468 100644
--- a/module/zfs/include/sys/zio.h
+++ b/module/zfs/include/sys/zio.h
@@ -143,6 +143,8 @@ enum zio_compress {
 #define	ZIO_FLAG_GODFATHER		0x080000
 
 #define	ZIO_FLAG_TRYHARD		0x100000
+#define	ZIO_FLAG_NODATA			0x200000
+#define	ZIO_FLAG_OPTIONAL		0x400000
 
 #define	ZIO_FLAG_GANG_INHERIT		\
 	(ZIO_FLAG_CANFAIL |		\
@@ -161,7 +163,9 @@ enum zio_compress {
 	ZIO_FLAG_IO_REPAIR |		\
 	ZIO_FLAG_IO_RETRY |		\
 	ZIO_FLAG_PROBE |		\
-	ZIO_FLAG_TRYHARD)
+	ZIO_FLAG_TRYHARD |		\
+	ZIO_FLAG_NODATA |		\
+	ZIO_FLAG_OPTIONAL)
 
 #define	ZIO_FLAG_AGG_INHERIT		\
 	(ZIO_FLAG_DONT_AGGREGATE |	\
diff --git a/module/zfs/rrwlock.c b/module/zfs/rrwlock.c
index 1bb8a285b..4cef53f95 100644
--- a/module/zfs/rrwlock.c
+++ b/module/zfs/rrwlock.c
@@ -19,12 +19,10 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
-
-
 #include <sys/refcount.h>
 #include <sys/rrwlock.h>
 
@@ -159,6 +157,14 @@ static void
 rrw_enter_read(rrwlock_t *rrl, void *tag)
 {
 	mutex_enter(&rrl->rr_lock);
+#if !defined(DEBUG) && defined(_KERNEL)
+	if (!rrl->rr_writer && !rrl->rr_writer_wanted) {
+		rrl->rr_anon_rcount.rc_count++;
+		mutex_exit(&rrl->rr_lock);
+		return;
+	}
+	DTRACE_PROBE(zfs__rrwfastpath__rdmiss);
+#endif
 	ASSERT(rrl->rr_writer != curthread);
 	ASSERT(refcount_count(&rrl->rr_anon_rcount) >= 0);
 
@@ -208,19 +214,28 @@ void
 rrw_exit(rrwlock_t *rrl, void *tag)
 {
 	mutex_enter(&rrl->rr_lock);
+#if !defined(DEBUG) && defined(_KERNEL)
+	if (!rrl->rr_writer && rrl->rr_linked_rcount.rc_count == 0) {
+		rrl->rr_anon_rcount.rc_count--;
+		if (rrl->rr_anon_rcount.rc_count == 0)
+			cv_broadcast(&rrl->rr_cv);
+		mutex_exit(&rrl->rr_lock);
+		return;
+	}
+	DTRACE_PROBE(zfs__rrwfastpath__exitmiss);
+#endif
 	ASSERT(!refcount_is_zero(&rrl->rr_anon_rcount) ||
 	    !refcount_is_zero(&rrl->rr_linked_rcount) ||
 	    rrl->rr_writer != NULL);
 
 	if (rrl->rr_writer == NULL) {
-		if (rrn_find_and_remove(rrl)) {
-			if (refcount_remove(&rrl->rr_linked_rcount, tag) == 0)
-				cv_broadcast(&rrl->rr_cv);
-
-		} else {
-			if (refcount_remove(&rrl->rr_anon_rcount, tag) == 0)
-				cv_broadcast(&rrl->rr_cv);
-		}
+		int64_t count;
+		if (rrn_find_and_remove(rrl))
+			count = refcount_remove(&rrl->rr_linked_rcount, tag);
+		else
+			count = refcount_remove(&rrl->rr_anon_rcount, tag);
+		if (count == 0)
+			cv_broadcast(&rrl->rr_cv);
 	} else {
 		ASSERT(rrl->rr_writer == curthread);
 		ASSERT(refcount_is_zero(&rrl->rr_anon_rcount) &&
diff --git a/module/zfs/spa.c b/module/zfs/spa.c
index e13ef765a..d147b8e91 100644
--- a/module/zfs/spa.c
+++ b/module/zfs/spa.c
@@ -1587,9 +1587,12 @@ spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
 		/*
 		 * If the config cache is stale, or we have uninitialized
 		 * metaslabs (see spa_vdev_add()), then update the config.
+		 *
+		 * If spa_load_verbatim is true, trust the current
+		 * in-core spa_config and update the disk labels.
 		 */
 		if (config_cache_txg != spa->spa_config_txg ||
-		    state == SPA_LOAD_IMPORT)
+		    state == SPA_LOAD_IMPORT || spa->spa_load_verbatim)
 			need_update = B_TRUE;
 
 		for (c = 0; c < rvd->vdev_children; c++)
@@ -2285,6 +2288,7 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 
 	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
 		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
+	spa_history_log_version(spa, LOG_POOL_CREATE);
 
 	spa->spa_minref = refcount_count(&spa->spa_refcount);
 
@@ -2420,6 +2424,7 @@ spa_import_rootpool(char *devpath, char *devid)
 
 	spa = spa_add(pname, NULL);
 	spa->spa_is_root = B_TRUE;
+	spa->spa_load_verbatim = B_TRUE;
 
 	/*
 	 * Build up a vdev tree based on the boot device's label config.
@@ -2475,6 +2480,7 @@ spa_import_rootpool(char *devpath, char *devid)
 
 	VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
 	error = 0;
+	spa_history_log_version(spa, LOG_POOL_IMPORT);
 out:
 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
 	vdev_free(rvd);
@@ -2507,6 +2513,8 @@ spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
 	spa = spa_add(pool, altroot);
 
+	spa->spa_load_verbatim = B_TRUE;
+
 	VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
 
 	if (props != NULL)
@@ -2515,6 +2523,7 @@ spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
 	spa_config_sync(spa, B_FALSE, B_TRUE);
 
 	mutex_exit(&spa_namespace_lock);
+	spa_history_log_version(spa, LOG_POOL_IMPORT);
 
 	return (0);
 }
@@ -2640,7 +2649,7 @@ spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
 		/*
 		 * Update the config cache to include the newly-imported pool.
 		 */
-		spa_config_update_common(spa, SPA_CONFIG_UPDATE_POOL, B_FALSE);
+		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
 	}
 
 	/*
@@ -2650,6 +2659,7 @@ spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
 	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
 
 	mutex_exit(&spa_namespace_lock);
+	spa_history_log_version(spa, LOG_POOL_IMPORT);
 
 	return (0);
 }
@@ -3007,7 +3017,6 @@ spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
 	ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
 	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
 	vdev_ops_t *pvops;
-	dmu_tx_t *tx;
 	char *oldvdpath, *newvdpath;
 	int newvd_isspare;
 	int error;
@@ -3163,17 +3172,11 @@ spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
 
 	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
 
-	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
-	if (dmu_tx_assign(tx, TXG_WAIT) == 0) {
-		spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, tx,
-		    CRED(),  "%s vdev=%s %s vdev=%s",
-		    replacing && newvd_isspare ? "spare in" :
-		    replacing ? "replace" : "attach", newvdpath,
-		    replacing ? "for" : "to", oldvdpath);
-		dmu_tx_commit(tx);
-	} else {
-		dmu_tx_abort(tx);
-	}
+	spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, NULL,
+	    CRED(),  "%s vdev=%s %s vdev=%s",
+	    replacing && newvd_isspare ? "spare in" :
+	    replacing ? "replace" : "attach", newvdpath,
+	    replacing ? "for" : "to", oldvdpath);
 
 	spa_strfree(oldvdpath);
 	spa_strfree(newvdpath);
@@ -3773,19 +3776,11 @@ spa_async_thread(spa_t *spa)
 		 * then log an internal history event.
 		 */
 		if (space_update) {
-			dmu_tx_t *tx;
-
-			tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
-			if (dmu_tx_assign(tx, TXG_WAIT) == 0) {
-				spa_history_internal_log(LOG_POOL_VDEV_ONLINE,
-				    spa, tx, CRED(),
-				    "pool '%s' size: %llu(+%llu)",
-				    spa_name(spa), spa_get_space(spa),
-				    space_update);
-				dmu_tx_commit(tx);
-			} else {
-				dmu_tx_abort(tx);
-			}
+			spa_history_internal_log(LOG_POOL_VDEV_ONLINE,
+			    spa, NULL, CRED(),
+			    "pool '%s' size: %llu(+%llu)",
+			    spa_name(spa), spa_get_space(spa),
+			    space_update);
 		}
 	}
 
diff --git a/module/zfs/spa_config.c b/module/zfs/spa_config.c
index 7103e179b..b2063bba1 100644
--- a/module/zfs/spa_config.c
+++ b/module/zfs/spa_config.c
@@ -209,7 +209,7 @@ spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
 
 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
 
-	if (rootdir == NULL)
+	if (rootdir == NULL || !(spa_mode_global & FWRITE))
 		return;
 
 	/*
@@ -394,23 +394,12 @@ spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 }
 
 /*
- * For a pool that's not currently a booting rootpool, update all disk labels,
- * generate a fresh config based on the current in-core state, and sync the
- * global config cache.
- */
-void
-spa_config_update(spa_t *spa, int what)
-{
-	spa_config_update_common(spa, what, FALSE);
-}
-
-/*
  * Update all disk labels, generate a fresh config based on the current
  * in-core state, and sync the global config cache (do not sync the config
  * cache if this is a booting rootpool).
  */
 void
-spa_config_update_common(spa_t *spa, int what, boolean_t isroot)
+spa_config_update(spa_t *spa, int what)
 {
 	vdev_t *rvd = spa->spa_root_vdev;
 	uint64_t txg;
@@ -447,9 +436,9 @@ spa_config_update_common(spa_t *spa, int what, boolean_t isroot)
 	/*
 	 * Update the global config cache to reflect the new mosconfig.
 	 */
-	if (!isroot)
+	if (!spa->spa_is_root)
 		spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
 
 	if (what == SPA_CONFIG_UPDATE_POOL)
-		spa_config_update_common(spa, SPA_CONFIG_UPDATE_VDEVS, isroot);
+		spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
 }
diff --git a/module/zfs/spa_history.c b/module/zfs/spa_history.c
index 03dd198c0..b719f7b80 100644
--- a/module/zfs/spa_history.c
+++ b/module/zfs/spa_history.c
@@ -394,13 +394,12 @@ spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
 	return (err);
 }
 
-void
-spa_history_internal_log(history_internal_events_t event, spa_t *spa,
-    dmu_tx_t *tx, cred_t *cr, const char *fmt, ...)
+static void
+log_internal(history_internal_events_t event, spa_t *spa,
+    dmu_tx_t *tx, cred_t *cr, const char *fmt, va_list adx)
 {
 	history_arg_t *hap;
 	char *str;
-	va_list adx;
 
 	/*
 	 * If this is part of creating a pool, not everything is
@@ -412,9 +411,7 @@ spa_history_internal_log(history_internal_events_t event, spa_t *spa,
 	hap = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
 	str = kmem_alloc(HIS_MAX_RECORD_LEN, KM_SLEEP);
 
-	va_start(adx, fmt);
 	(void) vsnprintf(str, HIS_MAX_RECORD_LEN, fmt, adx);
-	va_end(adx);
 
 	hap->ha_log_type = LOG_INTERNAL;
 	hap->ha_history_str = str;
@@ -429,3 +426,48 @@ spa_history_internal_log(history_internal_events_t event, spa_t *spa,
 	}
 	/* spa_history_log_sync() will free hap and str */
 }
+
+void
+spa_history_internal_log(history_internal_events_t event, spa_t *spa,
+    dmu_tx_t *tx, cred_t *cr, const char *fmt, ...)
+{
+	dmu_tx_t *htx = tx;
+	va_list adx;
+
+	/* create a tx if we didn't get one */
+	if (tx == NULL) {
+		htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
+		if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
+			dmu_tx_abort(htx);
+			return;
+		}
+	}
+
+	va_start(adx, fmt);
+	log_internal(event, spa, htx, cr, fmt, adx);
+	va_end(adx);
+
+	/* if we didn't get a tx from the caller, commit the one we made */
+	if (tx == NULL)
+		dmu_tx_commit(htx);
+}
+
+void
+spa_history_log_version(spa_t *spa, history_internal_events_t event)
+{
+#ifdef _KERNEL
+	uint64_t current_vers = spa_version(spa);
+
+	if (current_vers >= SPA_VERSION_ZPOOL_HISTORY) {
+		spa_history_internal_log(event, spa, NULL, CRED(),
+		    "pool spa %llu; zfs spa %llu; zpl %d; uts %s %s %s %s",
+		    (u_longlong_t)current_vers, SPA_VERSION, ZPL_VERSION,
+		    utsname.nodename, utsname.release, utsname.version,
+		    utsname.machine);
+	}
+	cmn_err(CE_CONT, "!%s version %llu pool %s using %llu",
+	    event == LOG_POOL_IMPORT ? "imported" :
+	    event == LOG_POOL_CREATE ? "created" : "accessed",
+	    (u_longlong_t)current_vers, spa_name(spa), SPA_VERSION);
+#endif
+}
diff --git a/module/zfs/spa_misc.c b/module/zfs/spa_misc.c
index 84434d530..ef74a443d 100644
--- a/module/zfs/spa_misc.c
+++ b/module/zfs/spa_misc.c
@@ -316,10 +316,13 @@ spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
 void
 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
 {
+	int wlocks_held = 0;
 	int i;
 
 	for (i = 0; i < SCL_LOCKS; i++) {
 		spa_config_lock_t *scl = &spa->spa_config_lock[i];
+		if (scl->scl_writer == curthread)
+			wlocks_held |= (1 << i);
 		if (!(locks & (1 << i)))
 			continue;
 		mutex_enter(&scl->scl_lock);
@@ -339,6 +342,7 @@ spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
 		(void) refcount_add(&scl->scl_count, tag);
 		mutex_exit(&scl->scl_lock);
 	}
+	ASSERT(wlocks_held <= locks);
 }
 
 void
diff --git a/module/zfs/vdev.c b/module/zfs/vdev.c
index 63f7a83f3..54e154390 100644
--- a/module/zfs/vdev.c
+++ b/module/zfs/vdev.c
@@ -412,22 +412,26 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
 		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
 		    &nparity) == 0) {
 			/*
-			 * Currently, we can only support 2 parity devices.
+			 * Currently, we can only support 3 parity devices.
 			 */
-			if (nparity == 0 || nparity > 2)
+			if (nparity == 0 || nparity > 3)
 				return (EINVAL);
 			/*
-			 * Older versions can only support 1 parity device.
+			 * Previous versions could only support 1 or 2 parity
+			 * device.
 			 */
-			if (nparity == 2 &&
-			    spa_version(spa) < SPA_VERSION_RAID6)
+			if (nparity > 1 &&
+			    spa_version(spa) < SPA_VERSION_RAIDZ2)
+				return (ENOTSUP);
+			if (nparity > 2 &&
+			    spa_version(spa) < SPA_VERSION_RAIDZ3)
 				return (ENOTSUP);
 		} else {
 			/*
 			 * We require the parity to be specified for SPAs that
 			 * support multiple parity levels.
 			 */
-			if (spa_version(spa) >= SPA_VERSION_RAID6)
+			if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
 				return (EINVAL);
 			/*
 			 * Otherwise, we default to 1 parity device for RAID-Z.
@@ -1004,6 +1008,32 @@ vdev_probe(vdev_t *vd, zio_t *zio)
 	return (NULL);
 }
 
+static void
+vdev_open_child(void *arg)
+{
+	vdev_t *vd = arg;
+
+	vd->vdev_open_thread = curthread;
+	vd->vdev_open_error = vdev_open(vd);
+	vd->vdev_open_thread = NULL;
+}
+
+void
+vdev_open_children(vdev_t *vd)
+{
+	taskq_t *tq;
+	int children = vd->vdev_children;
+
+	tq = taskq_create("vdev_open", children, minclsyspri,
+	    children, children, TASKQ_PREPOPULATE);
+
+	for (int c = 0; c < children; c++)
+		VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
+		    TQ_SLEEP) != NULL);
+
+	taskq_destroy(tq);
+}
+
 /*
  * Prepare a virtual device for access.
  */
@@ -1017,8 +1047,8 @@ vdev_open(vdev_t *vd)
 	uint64_t ashift = 0;
 	int c;
 
-	ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
-
+	ASSERT(vd->vdev_open_thread == curthread ||
+	    spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
 	ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
 	    vd->vdev_state == VDEV_STATE_CANT_OPEN ||
 	    vd->vdev_state == VDEV_STATE_OFFLINE);
@@ -1230,7 +1260,12 @@ vdev_validate(vdev_t *vd)
 
 		nvlist_free(label);
 
-		if (spa->spa_load_state == SPA_LOAD_OPEN &&
+		/*
+		 * If spa->spa_load_verbatim is true, no need to check the
+		 * state of the pool.
+		 */
+		if (!spa->spa_load_verbatim &&
+		    spa->spa_load_state == SPA_LOAD_OPEN &&
 		    state != POOL_STATE_ACTIVE)
 			return (EBADF);
 
diff --git a/module/zfs/vdev_label.c b/module/zfs/vdev_label.c
index e33d2660e..ec49d3d3c 100644
--- a/module/zfs/vdev_label.c
+++ b/module/zfs/vdev_label.c
@@ -246,8 +246,10 @@ vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
 		 * into a crufty old storage pool.
 		 */
 		ASSERT(vd->vdev_nparity == 1 ||
-		    (vd->vdev_nparity == 2 &&
-		    spa_version(spa) >= SPA_VERSION_RAID6));
+		    (vd->vdev_nparity <= 2 &&
+		    spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
+		    (vd->vdev_nparity <= 3 &&
+		    spa_version(spa) >= SPA_VERSION_RAIDZ3));
 
 		/*
 		 * Note that we'll add the nparity tag even on storage pools
@@ -509,7 +511,7 @@ vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
 	int error;
 	uint64_t spare_guid = 0, l2cache_guid;
 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
-	int c, l, n;
+	int c, l;
 	vdev_t *pvd;
 
 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
@@ -647,8 +649,8 @@ vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
 	/*
 	 * Initialize uberblock template.
 	 */
-	ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
-	bzero(ub, VDEV_UBERBLOCK_SIZE(vd));
+	ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
+	bzero(ub, VDEV_UBERBLOCK_RING);
 	*ub = spa->spa_uberblock;
 	ub->ub_txg = 0;
 
@@ -677,11 +679,9 @@ retry:
 		    offsetof(vdev_label_t, vl_pad2),
 		    VDEV_PAD_SIZE, NULL, NULL, flags);
 
-		for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
-			vdev_label_write(zio, vd, l, ub,
-			    VDEV_UBERBLOCK_OFFSET(vd, n),
-			    VDEV_UBERBLOCK_SIZE(vd), NULL, NULL, flags);
-		}
+		vdev_label_write(zio, vd, l, ub,
+		    offsetof(vdev_label_t, vl_uberblock),
+		    VDEV_UBERBLOCK_RING, NULL, NULL, flags);
 	}
 
 	error = zio_wait(zio);
@@ -693,7 +693,7 @@ retry:
 
 	nvlist_free(label);
 	zio_buf_free(pad2, VDEV_PAD_SIZE);
-	zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd));
+	zio_buf_free(ub, VDEV_UBERBLOCK_RING);
 	zio_buf_free(vp, sizeof (vdev_phys_t));
 
 	/*
diff --git a/module/zfs/vdev_mirror.c b/module/zfs/vdev_mirror.c
index c38cd6800..58e3a5328 100644
--- a/module/zfs/vdev_mirror.c
+++ b/module/zfs/vdev_mirror.c
@@ -124,21 +124,21 @@ vdev_mirror_map_alloc(zio_t *zio)
 static int
 vdev_mirror_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift)
 {
-	vdev_t *cvd;
-	uint64_t c;
 	int numerrors = 0;
-	int ret, lasterror = 0;
+	int lasterror = 0;
 
 	if (vd->vdev_children == 0) {
 		vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
 		return (EINVAL);
 	}
 
-	for (c = 0; c < vd->vdev_children; c++) {
-		cvd = vd->vdev_child[c];
+	vdev_open_children(vd);
 
-		if ((ret = vdev_open(cvd)) != 0) {
-			lasterror = ret;
+	for (int c = 0; c < vd->vdev_children; c++) {
+		vdev_t *cvd = vd->vdev_child[c];
+
+		if (cvd->vdev_open_error) {
+			lasterror = cvd->vdev_open_error;
 			numerrors++;
 			continue;
 		}
@@ -158,9 +158,7 @@ vdev_mirror_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift)
 static void
 vdev_mirror_close(vdev_t *vd)
 {
-	uint64_t c;
-
-	for (c = 0; c < vd->vdev_children; c++)
+	for (int c = 0; c < vd->vdev_children; c++)
 		vdev_close(vd->vdev_child[c]);
 }
 
diff --git a/module/zfs/vdev_queue.c b/module/zfs/vdev_queue.c
index a1cf74fe1..9958a0e78 100644
--- a/module/zfs/vdev_queue.c
+++ b/module/zfs/vdev_queue.c
@@ -24,7 +24,7 @@
  */
 
 #include <sys/zfs_context.h>
-#include <sys/spa.h>
+#include <sys/spa_impl.h>
 #include <sys/vdev_impl.h>
 #include <sys/zio.h>
 #include <sys/avl.h>
@@ -48,11 +48,14 @@ int zfs_vdev_time_shift = 6;
 int zfs_vdev_ramp_rate = 2;
 
 /*
- * To reduce IOPs, we aggregate small adjacent i/os into one large i/o.
- * For read i/os, we also aggregate across small adjacency gaps.
+ * To reduce IOPs, we aggregate small adjacent I/Os into one large I/O.
+ * For read I/Os, we also aggregate across small adjacency gaps; for writes
+ * we include spans of optional I/Os to aid aggregation at the disk even when
+ * they aren't able to help us aggregate at this level.
  */
 int zfs_vdev_aggregation_limit = SPA_MAXBLOCKSIZE;
 int zfs_vdev_read_gap_limit = 32 << 10;
+int zfs_vdev_write_gap_limit = 4 << 10;
 
 /*
  * Virtual device vector for disk I/O scheduling.
@@ -172,12 +175,14 @@ vdev_queue_agg_io_done(zio_t *aio)
 static zio_t *
 vdev_queue_io_to_issue(vdev_queue_t *vq, uint64_t pending_limit)
 {
-	zio_t *fio, *lio, *aio, *dio, *nio;
+	zio_t *fio, *lio, *aio, *dio, *nio, *mio;
 	avl_tree_t *t;
 	int flags;
 	uint64_t maxspan = zfs_vdev_aggregation_limit;
 	uint64_t maxgap;
+	int stretch;
 
+again:
 	ASSERT(MUTEX_HELD(&vq->vq_lock));
 
 	if (avl_numnodes(&vq->vq_pending_tree) >= pending_limit ||
@@ -192,21 +197,88 @@ vdev_queue_io_to_issue(vdev_queue_t *vq, uint64_t pending_limit)
 
 	if (!(flags & ZIO_FLAG_DONT_AGGREGATE)) {
 		/*
-		 * We can aggregate I/Os that are adjacent and of the
-		 * same flavor, as expressed by the AGG_INHERIT flags.
-		 * The latter is necessary so that certain attributes
-		 * of the I/O, such as whether it's a normal I/O or a
-		 * scrub/resilver, can be preserved in the aggregate.
+		 * We can aggregate I/Os that are sufficiently adjacent and of
+		 * the same flavor, as expressed by the AGG_INHERIT flags.
+		 * The latter requirement is necessary so that certain
+		 * attributes of the I/O, such as whether it's a normal I/O
+		 * or a scrub/resilver, can be preserved in the aggregate.
+		 * We can include optional I/Os, but don't allow them
+		 * to begin a range as they add no benefit in that situation.
+		 */
+
+		/*
+		 * We keep track of the last non-optional I/O.
+		 */
+		mio = (fio->io_flags & ZIO_FLAG_OPTIONAL) ? NULL : fio;
+
+		/*
+		 * Walk backwards through sufficiently contiguous I/Os
+		 * recording the last non-option I/O.
 		 */
 		while ((dio = AVL_PREV(t, fio)) != NULL &&
 		    (dio->io_flags & ZIO_FLAG_AGG_INHERIT) == flags &&
-		    IO_SPAN(dio, lio) <= maxspan && IO_GAP(dio, fio) <= maxgap)
+		    IO_SPAN(dio, lio) <= maxspan &&
+		    IO_GAP(dio, fio) <= maxgap) {
 			fio = dio;
+			if (mio == NULL && !(fio->io_flags & ZIO_FLAG_OPTIONAL))
+				mio = fio;
+		}
 
+		/*
+		 * Skip any initial optional I/Os.
+		 */
+		while ((fio->io_flags & ZIO_FLAG_OPTIONAL) && fio != lio) {
+			fio = AVL_NEXT(t, fio);
+			ASSERT(fio != NULL);
+		}
+
+		/*
+		 * Walk forward through sufficiently contiguous I/Os.
+		 */
 		while ((dio = AVL_NEXT(t, lio)) != NULL &&
 		    (dio->io_flags & ZIO_FLAG_AGG_INHERIT) == flags &&
-		    IO_SPAN(fio, dio) <= maxspan && IO_GAP(lio, dio) <= maxgap)
+		    IO_SPAN(fio, dio) <= maxspan &&
+		    IO_GAP(lio, dio) <= maxgap) {
 			lio = dio;
+			if (!(lio->io_flags & ZIO_FLAG_OPTIONAL))
+				mio = lio;
+		}
+
+		/*
+		 * Now that we've established the range of the I/O aggregation
+		 * we must decide what to do with trailing optional I/Os.
+		 * For reads, there's nothing to do. While we are unable to
+		 * aggregate further, it's possible that a trailing optional
+		 * I/O would allow the underlying device to aggregate with
+		 * subsequent I/Os. We must therefore determine if the next
+		 * non-optional I/O is close enough to make aggregation
+		 * worthwhile.
+		 */
+		stretch = B_FALSE;
+		if (t != &vq->vq_read_tree && mio != NULL) {
+			nio = lio;
+			while ((dio = AVL_NEXT(t, nio)) != NULL &&
+			    IO_GAP(nio, dio) == 0 &&
+			    IO_GAP(mio, dio) <= zfs_vdev_write_gap_limit) {
+				nio = dio;
+				if (!(nio->io_flags & ZIO_FLAG_OPTIONAL)) {
+					stretch = B_TRUE;
+					break;
+				}
+			}
+		}
+
+		if (stretch) {
+			/* This may be a no-op. */
+			VERIFY((dio = AVL_NEXT(t, lio)) != NULL);
+			dio->io_flags &= ~ZIO_FLAG_OPTIONAL;
+		} else {
+			while (lio != mio && lio != fio) {
+				ASSERT(lio->io_flags & ZIO_FLAG_OPTIONAL);
+				lio = AVL_PREV(t, lio);
+				ASSERT(lio != NULL);
+			}
+		}
 	}
 
 	if (fio != lio) {
@@ -225,10 +297,15 @@ vdev_queue_io_to_issue(vdev_queue_t *vq, uint64_t pending_limit)
 			ASSERT(dio->io_type == aio->io_type);
 			ASSERT(dio->io_vdev_tree == t);
 
-			if (dio->io_type == ZIO_TYPE_WRITE)
+			if (dio->io_flags & ZIO_FLAG_NODATA) {
+				ASSERT(dio->io_type == ZIO_TYPE_WRITE);
+				bzero((char *)aio->io_data + (dio->io_offset -
+				    aio->io_offset), dio->io_size);
+			} else if (dio->io_type == ZIO_TYPE_WRITE) {
 				bcopy(dio->io_data, (char *)aio->io_data +
 				    (dio->io_offset - aio->io_offset),
 				    dio->io_size);
+			}
 
 			zio_add_child(dio, aio);
 			vdev_queue_io_remove(vq, dio);
@@ -244,6 +321,20 @@ vdev_queue_io_to_issue(vdev_queue_t *vq, uint64_t pending_limit)
 	ASSERT(fio->io_vdev_tree == t);
 	vdev_queue_io_remove(vq, fio);
 
+	/*
+	 * If the I/O is or was optional and therefore has no data, we need to
+	 * simply discard it. We need to drop the vdev queue's lock to avoid a
+	 * deadlock that we could encounter since this I/O will complete
+	 * immediately.
+	 */
+	if (fio->io_flags & ZIO_FLAG_NODATA) {
+		mutex_exit(&vq->vq_lock);
+		zio_vdev_io_bypass(fio);
+		zio_execute(fio);
+		mutex_enter(&vq->vq_lock);
+		goto again;
+	}
+
 	avl_add(&vq->vq_pending_tree, fio);
 
 	return (fio);
diff --git a/module/zfs/vdev_raidz.c b/module/zfs/vdev_raidz.c
index 23c6c6600..6d6f166ca 100644
--- a/module/zfs/vdev_raidz.c
+++ b/module/zfs/vdev_raidz.c
@@ -35,12 +35,27 @@
 /*
  * Virtual device vector for RAID-Z.
  *
- * This vdev supports both single and double parity. For single parity, we
- * use a simple XOR of all the data columns. For double parity, we use both
- * the simple XOR as well as a technique described in "The mathematics of
- * RAID-6" by H. Peter Anvin. This technique defines a Galois field, GF(2^8),
- * over the integers expressable in a single byte. Briefly, the operations on
- * the field are defined as follows:
+ * This vdev supports single, double, and triple parity. For single parity,
+ * we use a simple XOR of all the data columns. For double or triple parity,
+ * we use a special case of Reed-Solomon coding. This extends the
+ * technique described in "The mathematics of RAID-6" by H. Peter Anvin by
+ * drawing on the system described in "A Tutorial on Reed-Solomon Coding for
+ * Fault-Tolerance in RAID-like Systems" by James S. Plank on which the
+ * former is also based. The latter is designed to provide higher performance
+ * for writes.
+ *
+ * Note that the Plank paper claimed to support arbitrary N+M, but was then
+ * amended six years later identifying a critical flaw that invalidates its
+ * claims. Nevertheless, the technique can be adapted to work for up to
+ * triple parity. For additional parity, the amendment "Note: Correction to
+ * the 1997 Tutorial on Reed-Solomon Coding" by James S. Plank and Ying Ding
+ * is viable, but the additional complexity means that write performance will
+ * suffer.
+ *
+ * All of the methods above operate on a Galois field, defined over the
+ * integers mod 2^N. In our case we choose N=8 for GF(8) so that all elements
+ * can be expressed with a single byte. Briefly, the operations on the
+ * field are defined as follows:
  *
  *   o addition (+) is represented by a bitwise XOR
  *   o subtraction (-) is therefore identical to addition: A + B = A - B
@@ -55,22 +70,32 @@
  *	(A * 2)_0 = A_7
  *
  * In C, multiplying by 2 is therefore ((a << 1) ^ ((a & 0x80) ? 0x1d : 0)).
+ * As an aside, this multiplication is derived from the error correcting
+ * primitive polynomial x^8 + x^4 + x^3 + x^2 + 1.
  *
  * Observe that any number in the field (except for 0) can be expressed as a
  * power of 2 -- a generator for the field. We store a table of the powers of
  * 2 and logs base 2 for quick look ups, and exploit the fact that A * B can
  * be rewritten as 2^(log_2(A) + log_2(B)) (where '+' is normal addition rather
- * than field addition). The inverse of a field element A (A^-1) is A^254.
+ * than field addition). The inverse of a field element A (A^-1) is therefore
+ * A ^ (255 - 1) = A^254.
  *
- * The two parity columns, P and Q, over several data columns, D_0, ... D_n-1,
- * can be expressed by field operations:
+ * The up-to-three parity columns, P, Q, R over several data columns,
+ * D_0, ... D_n-1, can be expressed by field operations:
  *
  *	P = D_0 + D_1 + ... + D_n-2 + D_n-1
  *	Q = 2^n-1 * D_0 + 2^n-2 * D_1 + ... + 2^1 * D_n-2 + 2^0 * D_n-1
  *	  = ((...((D_0) * 2 + D_1) * 2 + ...) * 2 + D_n-2) * 2 + D_n-1
+ *	R = 4^n-1 * D_0 + 4^n-2 * D_1 + ... + 4^1 * D_n-2 + 4^0 * D_n-1
+ *	  = ((...((D_0) * 4 + D_1) * 4 + ...) * 4 + D_n-2) * 4 + D_n-1
  *
- * See the reconstruction code below for how P and Q can used individually or
- * in concert to recover missing data columns.
+ * We chose 1, 2, and 4 as our generators because 1 corresponds to the trival
+ * XOR operation, and 2 and 4 can be computed quickly and generate linearly-
+ * independent coefficients. (There are no additional coefficients that have
+ * this property which is why the uncorrected Plank method breaks down.)
+ *
+ * See the reconstruction code below for how P, Q and R can used individually
+ * or in concert to recover missing data columns.
  */
 
 typedef struct raidz_col {
@@ -84,21 +109,49 @@ typedef struct raidz_col {
 } raidz_col_t;
 
 typedef struct raidz_map {
-	uint64_t rm_cols;		/* Column count */
+	uint64_t rm_cols;		/* Regular column count */
+	uint64_t rm_scols;		/* Count including skipped columns */
 	uint64_t rm_bigcols;		/* Number of oversized columns */
 	uint64_t rm_asize;		/* Actual total I/O size */
 	uint64_t rm_missingdata;	/* Count of missing data devices */
 	uint64_t rm_missingparity;	/* Count of missing parity devices */
 	uint64_t rm_firstdatacol;	/* First data column/parity count */
+	uint64_t rm_skipped;		/* Skipped sectors for padding */
 	raidz_col_t rm_col[1];		/* Flexible array of I/O columns */
 } raidz_map_t;
 
 #define	VDEV_RAIDZ_P		0
 #define	VDEV_RAIDZ_Q		1
+#define	VDEV_RAIDZ_R		2
+#define	VDEV_RAIDZ_MAXPARITY	3
+
+#define	VDEV_RAIDZ_MUL_2(x)	(((x) << 1) ^ (((x) & 0x80) ? 0x1d : 0))
+#define	VDEV_RAIDZ_MUL_4(x)	(VDEV_RAIDZ_MUL_2(VDEV_RAIDZ_MUL_2(x)))
+
+/*
+ * We provide a mechanism to perform the field multiplication operation on a
+ * 64-bit value all at once rather than a byte at a time. This works by
+ * creating a mask from the top bit in each byte and using that to
+ * conditionally apply the XOR of 0x1d.
+ */
+#define	VDEV_RAIDZ_64MUL_2(x, mask) \
+{ \
+	(mask) = (x) & 0x8080808080808080ULL; \
+	(mask) = ((mask) << 1) - ((mask) >> 7); \
+	(x) = (((x) << 1) & 0xfefefefefefefefeULL) ^ \
+	    ((mask) & 0x1d1d1d1d1d1d1d1d); \
+}
 
-#define	VDEV_RAIDZ_MAXPARITY	2
+#define	VDEV_RAIDZ_64MUL_4(x, mask) \
+{ \
+	VDEV_RAIDZ_64MUL_2((x), mask); \
+	VDEV_RAIDZ_64MUL_2((x), mask); \
+}
 
-#define	VDEV_RAIDZ_MUL_2(a)	(((a) << 1) ^ (((a) & 0x80) ? 0x1d : 0))
+/*
+ * Force reconstruction to use the general purpose method.
+ */
+int vdev_raidz_default_to_general;
 
 /*
  * These two tables represent powers and logs of 2 in the Galois field defined
@@ -201,7 +254,7 @@ vdev_raidz_map_free(zio_t *zio)
 	for (c = 0; c < rm->rm_firstdatacol; c++)
 		zio_buf_free(rm->rm_col[c].rc_data, rm->rm_col[c].rc_size);
 
-	kmem_free(rm, offsetof(raidz_map_t, rm_col[rm->rm_cols]));
+	kmem_free(rm, offsetof(raidz_map_t, rm_col[rm->rm_scols]));
 }
 
 static raidz_map_t *
@@ -213,24 +266,35 @@ vdev_raidz_map_alloc(zio_t *zio, uint64_t unit_shift, uint64_t dcols,
 	uint64_t s = zio->io_size >> unit_shift;
 	uint64_t f = b % dcols;
 	uint64_t o = (b / dcols) << unit_shift;
-	uint64_t q, r, c, bc, col, acols, coff, devidx;
+	uint64_t q, r, c, bc, col, acols, scols, coff, devidx, asize, tot;
 
 	q = s / (dcols - nparity);
 	r = s - q * (dcols - nparity);
 	bc = (r == 0 ? 0 : r + nparity);
+	tot = s + nparity * (q + (r == 0 ? 0 : 1));
+
+	if (q == 0) {
+		acols = bc;
+		scols = MIN(dcols, roundup(bc, nparity + 1));
+	} else {
+		acols = dcols;
+		scols = dcols;
+	}
 
-	acols = (q == 0 ? bc : dcols);
+	ASSERT3U(acols, <=, scols);
 
-	rm = kmem_alloc(offsetof(raidz_map_t, rm_col[acols]), KM_SLEEP);
+	rm = kmem_alloc(offsetof(raidz_map_t, rm_col[scols]), KM_SLEEP);
 
 	rm->rm_cols = acols;
+	rm->rm_scols = scols;
 	rm->rm_bigcols = bc;
-	rm->rm_asize = 0;
 	rm->rm_missingdata = 0;
 	rm->rm_missingparity = 0;
 	rm->rm_firstdatacol = nparity;
 
-	for (c = 0; c < acols; c++) {
+	asize = 0;
+
+	for (c = 0; c < scols; c++) {
 		col = f + c;
 		coff = o;
 		if (col >= dcols) {
@@ -239,15 +303,26 @@ vdev_raidz_map_alloc(zio_t *zio, uint64_t unit_shift, uint64_t dcols,
 		}
 		rm->rm_col[c].rc_devidx = col;
 		rm->rm_col[c].rc_offset = coff;
-		rm->rm_col[c].rc_size = (q + (c < bc)) << unit_shift;
 		rm->rm_col[c].rc_data = NULL;
 		rm->rm_col[c].rc_error = 0;
 		rm->rm_col[c].rc_tried = 0;
 		rm->rm_col[c].rc_skipped = 0;
-		rm->rm_asize += rm->rm_col[c].rc_size;
+
+		if (c >= acols)
+			rm->rm_col[c].rc_size = 0;
+		else if (c < bc)
+			rm->rm_col[c].rc_size = (q + 1) << unit_shift;
+		else
+			rm->rm_col[c].rc_size = q << unit_shift;
+
+		asize += rm->rm_col[c].rc_size;
 	}
 
-	rm->rm_asize = roundup(rm->rm_asize, (nparity + 1) << unit_shift);
+	ASSERT3U(asize, ==, tot << unit_shift);
+	rm->rm_asize = roundup(asize, (nparity + 1) << unit_shift);
+	rm->rm_skipped = roundup(tot, nparity + 1) - tot;
+	ASSERT3U(rm->rm_asize - asize, ==, rm->rm_skipped << unit_shift);
+	ASSERT3U(rm->rm_skipped, <=, nparity);
 
 	for (c = 0; c < rm->rm_firstdatacol; c++)
 		rm->rm_col[c].rc_data = zio_buf_alloc(rm->rm_col[c].rc_size);
@@ -305,12 +380,12 @@ vdev_raidz_generate_parity_p(raidz_map_t *rm)
 
 		if (c == rm->rm_firstdatacol) {
 			ASSERT(ccount == pcount);
-			for (i = 0; i < ccount; i++, p++, src++) {
+			for (i = 0; i < ccount; i++, src++, p++) {
 				*p = *src;
 			}
 		} else {
 			ASSERT(ccount <= pcount);
-			for (i = 0; i < ccount; i++, p++, src++) {
+			for (i = 0; i < ccount; i++, src++, p++) {
 				*p ^= *src;
 			}
 		}
@@ -320,10 +395,10 @@ vdev_raidz_generate_parity_p(raidz_map_t *rm)
 static void
 vdev_raidz_generate_parity_pq(raidz_map_t *rm)
 {
-	uint64_t *q, *p, *src, pcount, ccount, mask, i;
+	uint64_t *p, *q, *src, pcnt, ccnt, mask, i;
 	int c;
 
-	pcount = rm->rm_col[VDEV_RAIDZ_P].rc_size / sizeof (src[0]);
+	pcnt = rm->rm_col[VDEV_RAIDZ_P].rc_size / sizeof (src[0]);
 	ASSERT(rm->rm_col[VDEV_RAIDZ_P].rc_size ==
 	    rm->rm_col[VDEV_RAIDZ_Q].rc_size);
 
@@ -331,55 +406,138 @@ vdev_raidz_generate_parity_pq(raidz_map_t *rm)
 		src = rm->rm_col[c].rc_data;
 		p = rm->rm_col[VDEV_RAIDZ_P].rc_data;
 		q = rm->rm_col[VDEV_RAIDZ_Q].rc_data;
-		ccount = rm->rm_col[c].rc_size / sizeof (src[0]);
+
+		ccnt = rm->rm_col[c].rc_size / sizeof (src[0]);
 
 		if (c == rm->rm_firstdatacol) {
-			ASSERT(ccount == pcount || ccount == 0);
-			for (i = 0; i < ccount; i++, p++, q++, src++) {
-				*q = *src;
+			ASSERT(ccnt == pcnt || ccnt == 0);
+			for (i = 0; i < ccnt; i++, src++, p++, q++) {
 				*p = *src;
+				*q = *src;
 			}
-			for (; i < pcount; i++, p++, q++, src++) {
-				*q = 0;
+			for (; i < pcnt; i++, src++, p++, q++) {
 				*p = 0;
+				*q = 0;
 			}
 		} else {
-			ASSERT(ccount <= pcount);
+			ASSERT(ccnt <= pcnt);
 
 			/*
-			 * Rather than multiplying each byte individually (as
-			 * described above), we are able to handle 8 at once
-			 * by generating a mask based on the high bit in each
-			 * byte and using that to conditionally XOR in 0x1d.
+			 * Apply the algorithm described above by multiplying
+			 * the previous result and adding in the new value.
 			 */
-			for (i = 0; i < ccount; i++, p++, q++, src++) {
-				mask = *q & 0x8080808080808080ULL;
-				mask = (mask << 1) - (mask >> 7);
-				*q = ((*q << 1) & 0xfefefefefefefefeULL) ^
-				    (mask & 0x1d1d1d1d1d1d1d1dULL);
+			for (i = 0; i < ccnt; i++, src++, p++, q++) {
+				*p ^= *src;
+
+				VDEV_RAIDZ_64MUL_2(*q, mask);
 				*q ^= *src;
+			}
+
+			/*
+			 * Treat short columns as though they are full of 0s.
+			 * Note that there's therefore nothing needed for P.
+			 */
+			for (; i < pcnt; i++, q++) {
+				VDEV_RAIDZ_64MUL_2(*q, mask);
+			}
+		}
+	}
+}
+
+static void
+vdev_raidz_generate_parity_pqr(raidz_map_t *rm)
+{
+	uint64_t *p, *q, *r, *src, pcnt, ccnt, mask, i;
+	int c;
+
+	pcnt = rm->rm_col[VDEV_RAIDZ_P].rc_size / sizeof (src[0]);
+	ASSERT(rm->rm_col[VDEV_RAIDZ_P].rc_size ==
+	    rm->rm_col[VDEV_RAIDZ_Q].rc_size);
+	ASSERT(rm->rm_col[VDEV_RAIDZ_P].rc_size ==
+	    rm->rm_col[VDEV_RAIDZ_R].rc_size);
+
+	for (c = rm->rm_firstdatacol; c < rm->rm_cols; c++) {
+		src = rm->rm_col[c].rc_data;
+		p = rm->rm_col[VDEV_RAIDZ_P].rc_data;
+		q = rm->rm_col[VDEV_RAIDZ_Q].rc_data;
+		r = rm->rm_col[VDEV_RAIDZ_R].rc_data;
+
+		ccnt = rm->rm_col[c].rc_size / sizeof (src[0]);
+
+		if (c == rm->rm_firstdatacol) {
+			ASSERT(ccnt == pcnt || ccnt == 0);
+			for (i = 0; i < ccnt; i++, src++, p++, q++, r++) {
+				*p = *src;
+				*q = *src;
+				*r = *src;
+			}
+			for (; i < pcnt; i++, src++, p++, q++, r++) {
+				*p = 0;
+				*q = 0;
+				*r = 0;
+			}
+		} else {
+			ASSERT(ccnt <= pcnt);
+
+			/*
+			 * Apply the algorithm described above by multiplying
+			 * the previous result and adding in the new value.
+			 */
+			for (i = 0; i < ccnt; i++, src++, p++, q++, r++) {
 				*p ^= *src;
+
+				VDEV_RAIDZ_64MUL_2(*q, mask);
+				*q ^= *src;
+
+				VDEV_RAIDZ_64MUL_4(*r, mask);
+				*r ^= *src;
 			}
 
 			/*
 			 * Treat short columns as though they are full of 0s.
+			 * Note that there's therefore nothing needed for P.
 			 */
-			for (; i < pcount; i++, q++) {
-				mask = *q & 0x8080808080808080ULL;
-				mask = (mask << 1) - (mask >> 7);
-				*q = ((*q << 1) & 0xfefefefefefefefeULL) ^
-				    (mask & 0x1d1d1d1d1d1d1d1dULL);
+			for (; i < pcnt; i++, q++, r++) {
+				VDEV_RAIDZ_64MUL_2(*q, mask);
+				VDEV_RAIDZ_64MUL_4(*r, mask);
 			}
 		}
 	}
 }
 
+/*
+ * Generate RAID parity in the first virtual columns according to the number of
+ * parity columns available.
+ */
 static void
-vdev_raidz_reconstruct_p(raidz_map_t *rm, int x)
+vdev_raidz_generate_parity(raidz_map_t *rm)
+{
+	switch (rm->rm_firstdatacol) {
+	case 1:
+		vdev_raidz_generate_parity_p(rm);
+		break;
+	case 2:
+		vdev_raidz_generate_parity_pq(rm);
+		break;
+	case 3:
+		vdev_raidz_generate_parity_pqr(rm);
+		break;
+	default:
+		cmn_err(CE_PANIC, "invalid RAID-Z configuration");
+	}
+}
+
+static int
+vdev_raidz_reconstruct_p(raidz_map_t *rm, int *tgts, int ntgts)
 {
 	uint64_t *dst, *src, xcount, ccount, count, i;
+	int x = tgts[0];
 	int c;
 
+	ASSERT(ntgts == 1);
+	ASSERT(x >= rm->rm_firstdatacol);
+	ASSERT(x < rm->rm_cols);
+
 	xcount = rm->rm_col[x].rc_size / sizeof (src[0]);
 	ASSERT(xcount <= rm->rm_col[VDEV_RAIDZ_P].rc_size / sizeof (src[0]));
 	ASSERT(xcount > 0);
@@ -404,15 +562,20 @@ vdev_raidz_reconstruct_p(raidz_map_t *rm, int x)
 			*dst ^= *src;
 		}
 	}
+
+	return (1 << VDEV_RAIDZ_P);
 }
 
-static void
-vdev_raidz_reconstruct_q(raidz_map_t *rm, int x)
+static int
+vdev_raidz_reconstruct_q(raidz_map_t *rm, int *tgts, int ntgts)
 {
 	uint64_t *dst, *src, xcount, ccount, count, mask, i;
 	uint8_t *b;
+	int x = tgts[0];
 	int c, j, exp;
 
+	ASSERT(ntgts == 1);
+
 	xcount = rm->rm_col[x].rc_size / sizeof (src[0]);
 	ASSERT(xcount <= rm->rm_col[VDEV_RAIDZ_Q].rc_size / sizeof (src[0]));
 
@@ -436,23 +599,13 @@ vdev_raidz_reconstruct_q(raidz_map_t *rm, int x)
 			}
 
 		} else {
-			/*
-			 * For an explanation of this, see the comment in
-			 * vdev_raidz_generate_parity_pq() above.
-			 */
 			for (i = 0; i < count; i++, dst++, src++) {
-				mask = *dst & 0x8080808080808080ULL;
-				mask = (mask << 1) - (mask >> 7);
-				*dst = ((*dst << 1) & 0xfefefefefefefefeULL) ^
-				    (mask & 0x1d1d1d1d1d1d1d1dULL);
+				VDEV_RAIDZ_64MUL_2(*dst, mask);
 				*dst ^= *src;
 			}
 
 			for (; i < xcount; i++, dst++) {
-				mask = *dst & 0x8080808080808080ULL;
-				mask = (mask << 1) - (mask >> 7);
-				*dst = ((*dst << 1) & 0xfefefefefefefefeULL) ^
-				    (mask & 0x1d1d1d1d1d1d1d1dULL);
+				VDEV_RAIDZ_64MUL_2(*dst, mask);
 			}
 		}
 	}
@@ -467,15 +620,20 @@ vdev_raidz_reconstruct_q(raidz_map_t *rm, int x)
 			*b = vdev_raidz_exp2(*b, exp);
 		}
 	}
+
+	return (1 << VDEV_RAIDZ_Q);
 }
 
-static void
-vdev_raidz_reconstruct_pq(raidz_map_t *rm, int x, int y)
+static int
+vdev_raidz_reconstruct_pq(raidz_map_t *rm, int *tgts, int ntgts)
 {
 	uint8_t *p, *q, *pxy, *qxy, *xd, *yd, tmp, a, b, aexp, bexp;
 	void *pdata, *qdata;
 	uint64_t xsize, ysize, i;
+	int x = tgts[0];
+	int y = tgts[1];
 
+	ASSERT(ntgts == 2);
 	ASSERT(x < y);
 	ASSERT(x >= rm->rm_firstdatacol);
 	ASSERT(y < rm->rm_cols);
@@ -553,15 +711,554 @@ vdev_raidz_reconstruct_pq(raidz_map_t *rm, int x, int y)
 	 */
 	rm->rm_col[VDEV_RAIDZ_P].rc_data = pdata;
 	rm->rm_col[VDEV_RAIDZ_Q].rc_data = qdata;
+
+	return ((1 << VDEV_RAIDZ_P) | (1 << VDEV_RAIDZ_Q));
+}
+
+/* BEGIN CSTYLED */
+/*
+ * In the general case of reconstruction, we must solve the system of linear
+ * equations defined by the coeffecients used to generate parity as well as
+ * the contents of the data and parity disks. This can be expressed with
+ * vectors for the original data (D) and the actual data (d) and parity (p)
+ * and a matrix composed of the identity matrix (I) and a dispersal matrix (V):
+ *
+ *            __   __                     __     __
+ *            |     |         __     __   |  p_0  |
+ *            |  V  |         |  D_0  |   | p_m-1 |
+ *            |     |    x    |   :   | = |  d_0  |
+ *            |  I  |         | D_n-1 |   |   :   |
+ *            |     |         ~~     ~~   | d_n-1 |
+ *            ~~   ~~                     ~~     ~~
+ *
+ * I is simply a square identity matrix of size n, and V is a vandermonde
+ * matrix defined by the coeffecients we chose for the various parity columns
+ * (1, 2, 4). Note that these values were chosen both for simplicity, speedy
+ * computation as well as linear separability.
+ *
+ *      __               __               __     __
+ *      |   1   ..  1 1 1 |               |  p_0  |
+ *      | 2^n-1 ..  4 2 1 |   __     __   |   :   |
+ *      | 4^n-1 .. 16 4 1 |   |  D_0  |   | p_m-1 |
+ *      |   1   ..  0 0 0 |   |  D_1  |   |  d_0  |
+ *      |   0   ..  0 0 0 | x |  D_2  | = |  d_1  |
+ *      |   :       : : : |   |   :   |   |  d_2  |
+ *      |   0   ..  1 0 0 |   | D_n-1 |   |   :   |
+ *      |   0   ..  0 1 0 |   ~~     ~~   |   :   |
+ *      |   0   ..  0 0 1 |               | d_n-1 |
+ *      ~~               ~~               ~~     ~~
+ *
+ * Note that I, V, d, and p are known. To compute D, we must invert the
+ * matrix and use the known data and parity values to reconstruct the unknown
+ * data values. We begin by removing the rows in V|I and d|p that correspond
+ * to failed or missing columns; we then make V|I square (n x n) and d|p
+ * sized n by removing rows corresponding to unused parity from the bottom up
+ * to generate (V|I)' and (d|p)'. We can then generate the inverse of (V|I)'
+ * using Gauss-Jordan elimination. In the example below we use m=3 parity
+ * columns, n=8 data columns, with errors in d_1, d_2, and p_1:
+ *           __                               __
+ *           |  1   1   1   1   1   1   1   1  |
+ *           | 128  64  32  16  8   4   2   1  | <-----+-+-- missing disks
+ *           |  19 205 116  29  64  16  4   1  |      / /
+ *           |  1   0   0   0   0   0   0   0  |     / /
+ *           |  0   1   0   0   0   0   0   0  | <--' /
+ *  (V|I)  = |  0   0   1   0   0   0   0   0  | <---'
+ *           |  0   0   0   1   0   0   0   0  |
+ *           |  0   0   0   0   1   0   0   0  |
+ *           |  0   0   0   0   0   1   0   0  |
+ *           |  0   0   0   0   0   0   1   0  |
+ *           |  0   0   0   0   0   0   0   1  |
+ *           ~~                               ~~
+ *           __                               __
+ *           |  1   1   1   1   1   1   1   1  |
+ *           | 128  64  32  16  8   4   2   1  |
+ *           |  19 205 116  29  64  16  4   1  |
+ *           |  1   0   0   0   0   0   0   0  |
+ *           |  0   1   0   0   0   0   0   0  |
+ *  (V|I)' = |  0   0   1   0   0   0   0   0  |
+ *           |  0   0   0   1   0   0   0   0  |
+ *           |  0   0   0   0   1   0   0   0  |
+ *           |  0   0   0   0   0   1   0   0  |
+ *           |  0   0   0   0   0   0   1   0  |
+ *           |  0   0   0   0   0   0   0   1  |
+ *           ~~                               ~~
+ *
+ * Here we employ Gauss-Jordan elimination to find the inverse of (V|I)'. We
+ * have carefully chosen the seed values 1, 2, and 4 to ensure that this
+ * matrix is not singular.
+ * __                                                                 __
+ * |  1   1   1   1   1   1   1   1     1   0   0   0   0   0   0   0  |
+ * |  19 205 116  29  64  16  4   1     0   1   0   0   0   0   0   0  |
+ * |  1   0   0   0   0   0   0   0     0   0   1   0   0   0   0   0  |
+ * |  0   0   0   1   0   0   0   0     0   0   0   1   0   0   0   0  |
+ * |  0   0   0   0   1   0   0   0     0   0   0   0   1   0   0   0  |
+ * |  0   0   0   0   0   1   0   0     0   0   0   0   0   1   0   0  |
+ * |  0   0   0   0   0   0   1   0     0   0   0   0   0   0   1   0  |
+ * |  0   0   0   0   0   0   0   1     0   0   0   0   0   0   0   1  |
+ * ~~                                                                 ~~
+ * __                                                                 __
+ * |  1   0   0   0   0   0   0   0     0   0   1   0   0   0   0   0  |
+ * |  1   1   1   1   1   1   1   1     1   0   0   0   0   0   0   0  |
+ * |  19 205 116  29  64  16  4   1     0   1   0   0   0   0   0   0  |
+ * |  0   0   0   1   0   0   0   0     0   0   0   1   0   0   0   0  |
+ * |  0   0   0   0   1   0   0   0     0   0   0   0   1   0   0   0  |
+ * |  0   0   0   0   0   1   0   0     0   0   0   0   0   1   0   0  |
+ * |  0   0   0   0   0   0   1   0     0   0   0   0   0   0   1   0  |
+ * |  0   0   0   0   0   0   0   1     0   0   0   0   0   0   0   1  |
+ * ~~                                                                 ~~
+ * __                                                                 __
+ * |  1   0   0   0   0   0   0   0     0   0   1   0   0   0   0   0  |
+ * |  0   1   1   0   0   0   0   0     1   0   1   1   1   1   1   1  |
+ * |  0  205 116  0   0   0   0   0     0   1   19  29  64  16  4   1  |
+ * |  0   0   0   1   0   0   0   0     0   0   0   1   0   0   0   0  |
+ * |  0   0   0   0   1   0   0   0     0   0   0   0   1   0   0   0  |
+ * |  0   0   0   0   0   1   0   0     0   0   0   0   0   1   0   0  |
+ * |  0   0   0   0   0   0   1   0     0   0   0   0   0   0   1   0  |
+ * |  0   0   0   0   0   0   0   1     0   0   0   0   0   0   0   1  |
+ * ~~                                                                 ~~
+ * __                                                                 __
+ * |  1   0   0   0   0   0   0   0     0   0   1   0   0   0   0   0  |
+ * |  0   1   1   0   0   0   0   0     1   0   1   1   1   1   1   1  |
+ * |  0   0  185  0   0   0   0   0    205  1  222 208 141 221 201 204 |
+ * |  0   0   0   1   0   0   0   0     0   0   0   1   0   0   0   0  |
+ * |  0   0   0   0   1   0   0   0     0   0   0   0   1   0   0   0  |
+ * |  0   0   0   0   0   1   0   0     0   0   0   0   0   1   0   0  |
+ * |  0   0   0   0   0   0   1   0     0   0   0   0   0   0   1   0  |
+ * |  0   0   0   0   0   0   0   1     0   0   0   0   0   0   0   1  |
+ * ~~                                                                 ~~
+ * __                                                                 __
+ * |  1   0   0   0   0   0   0   0     0   0   1   0   0   0   0   0  |
+ * |  0   1   1   0   0   0   0   0     1   0   1   1   1   1   1   1  |
+ * |  0   0   1   0   0   0   0   0    166 100  4   40 158 168 216 209 |
+ * |  0   0   0   1   0   0   0   0     0   0   0   1   0   0   0   0  |
+ * |  0   0   0   0   1   0   0   0     0   0   0   0   1   0   0   0  |
+ * |  0   0   0   0   0   1   0   0     0   0   0   0   0   1   0   0  |
+ * |  0   0   0   0   0   0   1   0     0   0   0   0   0   0   1   0  |
+ * |  0   0   0   0   0   0   0   1     0   0   0   0   0   0   0   1  |
+ * ~~                                                                 ~~
+ * __                                                                 __
+ * |  1   0   0   0   0   0   0   0     0   0   1   0   0   0   0   0  |
+ * |  0   1   0   0   0   0   0   0    167 100  5   41 159 169 217 208 |
+ * |  0   0   1   0   0   0   0   0    166 100  4   40 158 168 216 209 |
+ * |  0   0   0   1   0   0   0   0     0   0   0   1   0   0   0   0  |
+ * |  0   0   0   0   1   0   0   0     0   0   0   0   1   0   0   0  |
+ * |  0   0   0   0   0   1   0   0     0   0   0   0   0   1   0   0  |
+ * |  0   0   0   0   0   0   1   0     0   0   0   0   0   0   1   0  |
+ * |  0   0   0   0   0   0   0   1     0   0   0   0   0   0   0   1  |
+ * ~~                                                                 ~~
+ *                   __                               __
+ *                   |  0   0   1   0   0   0   0   0  |
+ *                   | 167 100  5   41 159 169 217 208 |
+ *                   | 166 100  4   40 158 168 216 209 |
+ *       (V|I)'^-1 = |  0   0   0   1   0   0   0   0  |
+ *                   |  0   0   0   0   1   0   0   0  |
+ *                   |  0   0   0   0   0   1   0   0  |
+ *                   |  0   0   0   0   0   0   1   0  |
+ *                   |  0   0   0   0   0   0   0   1  |
+ *                   ~~                               ~~
+ *
+ * We can then simply compute D = (V|I)'^-1 x (d|p)' to discover the values
+ * of the missing data.
+ *
+ * As is apparent from the example above, the only non-trivial rows in the
+ * inverse matrix correspond to the data disks that we're trying to
+ * reconstruct. Indeed, those are the only rows we need as the others would
+ * only be useful for reconstructing data known or assumed to be valid. For
+ * that reason, we only build the coefficients in the rows that correspond to
+ * targeted columns.
+ */
+/* END CSTYLED */
+
+static void
+vdev_raidz_matrix_init(raidz_map_t *rm, int n, int nmap, int *map,
+    uint8_t **rows)
+{
+	int i, j;
+	int pow;
+
+	ASSERT(n == rm->rm_cols - rm->rm_firstdatacol);
+
+	/*
+	 * Fill in the missing rows of interest.
+	 */
+	for (i = 0; i < nmap; i++) {
+		ASSERT3S(0, <=, map[i]);
+		ASSERT3S(map[i], <=, 2);
+
+		pow = map[i] * n;
+		if (pow > 255)
+			pow -= 255;
+		ASSERT(pow <= 255);
+
+		for (j = 0; j < n; j++) {
+			pow -= map[i];
+			if (pow < 0)
+				pow += 255;
+			rows[i][j] = vdev_raidz_pow2[pow];
+		}
+	}
+}
+
+static void
+vdev_raidz_matrix_invert(raidz_map_t *rm, int n, int nmissing, int *missing,
+    uint8_t **rows, uint8_t **invrows, const uint8_t *used)
+{
+	int i, j, ii, jj;
+	uint8_t log;
+
+	/*
+	 * Assert that the first nmissing entries from the array of used
+	 * columns correspond to parity columns and that subsequent entries
+	 * correspond to data columns.
+	 */
+	for (i = 0; i < nmissing; i++) {
+		ASSERT3S(used[i], <, rm->rm_firstdatacol);
+	}
+	for (; i < n; i++) {
+		ASSERT3S(used[i], >=, rm->rm_firstdatacol);
+	}
+
+	/*
+	 * First initialize the storage where we'll compute the inverse rows.
+	 */
+	for (i = 0; i < nmissing; i++) {
+		for (j = 0; j < n; j++) {
+			invrows[i][j] = (i == j) ? 1 : 0;
+		}
+	}
+
+	/*
+	 * Subtract all trivial rows from the rows of consequence.
+	 */
+	for (i = 0; i < nmissing; i++) {
+		for (j = nmissing; j < n; j++) {
+			ASSERT3U(used[j], >=, rm->rm_firstdatacol);
+			jj = used[j] - rm->rm_firstdatacol;
+			ASSERT3S(jj, <, n);
+			invrows[i][j] = rows[i][jj];
+			rows[i][jj] = 0;
+		}
+	}
+
+	/*
+	 * For each of the rows of interest, we must normalize it and subtract
+	 * a multiple of it from the other rows.
+	 */
+	for (i = 0; i < nmissing; i++) {
+		for (j = 0; j < missing[i]; j++) {
+			ASSERT3U(rows[i][j], ==, 0);
+		}
+		ASSERT3U(rows[i][missing[i]], !=, 0);
+
+		/*
+		 * Compute the inverse of the first element and multiply each
+		 * element in the row by that value.
+		 */
+		log = 255 - vdev_raidz_log2[rows[i][missing[i]]];
+
+		for (j = 0; j < n; j++) {
+			rows[i][j] = vdev_raidz_exp2(rows[i][j], log);
+			invrows[i][j] = vdev_raidz_exp2(invrows[i][j], log);
+		}
+
+		for (ii = 0; ii < nmissing; ii++) {
+			if (i == ii)
+				continue;
+
+			ASSERT3U(rows[ii][missing[i]], !=, 0);
+
+			log = vdev_raidz_log2[rows[ii][missing[i]]];
+
+			for (j = 0; j < n; j++) {
+				rows[ii][j] ^=
+				    vdev_raidz_exp2(rows[i][j], log);
+				invrows[ii][j] ^=
+				    vdev_raidz_exp2(invrows[i][j], log);
+			}
+		}
+	}
+
+	/*
+	 * Verify that the data that is left in the rows are properly part of
+	 * an identity matrix.
+	 */
+	for (i = 0; i < nmissing; i++) {
+		for (j = 0; j < n; j++) {
+			if (j == missing[i]) {
+				ASSERT3U(rows[i][j], ==, 1);
+			} else {
+				ASSERT3U(rows[i][j], ==, 0);
+			}
+		}
+	}
+}
+
+static void
+vdev_raidz_matrix_reconstruct(raidz_map_t *rm, int n, int nmissing,
+    int *missing, uint8_t **invrows, const uint8_t *used)
+{
+	int i, j, x, cc, c;
+	uint8_t *src;
+	uint64_t ccount;
+	uint8_t *dst[VDEV_RAIDZ_MAXPARITY];
+	uint64_t dcount[VDEV_RAIDZ_MAXPARITY];
+	uint8_t log, val;
+	int ll;
+	uint8_t *invlog[VDEV_RAIDZ_MAXPARITY];
+	uint8_t *p, *pp;
+	size_t psize;
+
+	psize = sizeof (invlog[0][0]) * n * nmissing;
+	p = kmem_alloc(psize, KM_SLEEP);
+
+	for (pp = p, i = 0; i < nmissing; i++) {
+		invlog[i] = pp;
+		pp += n;
+	}
+
+	for (i = 0; i < nmissing; i++) {
+		for (j = 0; j < n; j++) {
+			ASSERT3U(invrows[i][j], !=, 0);
+			invlog[i][j] = vdev_raidz_log2[invrows[i][j]];
+		}
+	}
+
+	for (i = 0; i < n; i++) {
+		c = used[i];
+		ASSERT3U(c, <, rm->rm_cols);
+
+		src = rm->rm_col[c].rc_data;
+		ccount = rm->rm_col[c].rc_size;
+		for (j = 0; j < nmissing; j++) {
+			cc = missing[j] + rm->rm_firstdatacol;
+			ASSERT3U(cc, >=, rm->rm_firstdatacol);
+			ASSERT3U(cc, <, rm->rm_cols);
+			ASSERT3U(cc, !=, c);
+
+			dst[j] = rm->rm_col[cc].rc_data;
+			dcount[j] = rm->rm_col[cc].rc_size;
+		}
+
+		ASSERT(ccount >= rm->rm_col[missing[0]].rc_size || i > 0);
+
+		for (x = 0; x < ccount; x++, src++) {
+			if (*src != 0)
+				log = vdev_raidz_log2[*src];
+
+			for (cc = 0; cc < nmissing; cc++) {
+				if (x >= dcount[cc])
+					continue;
+
+				if (*src == 0) {
+					val = 0;
+				} else {
+					if ((ll = log + invlog[cc][i]) >= 255)
+						ll -= 255;
+					val = vdev_raidz_pow2[ll];
+				}
+
+				if (i == 0)
+					dst[cc][x] = val;
+				else
+					dst[cc][x] ^= val;
+			}
+		}
+	}
+
+	kmem_free(p, psize);
+}
+
+static int
+vdev_raidz_reconstruct_general(raidz_map_t *rm, int *tgts, int ntgts)
+{
+	int n, i, c, t, tt;
+	int nmissing_rows;
+	int missing_rows[VDEV_RAIDZ_MAXPARITY];
+	int parity_map[VDEV_RAIDZ_MAXPARITY];
+
+	uint8_t *p, *pp;
+	size_t psize;
+
+	uint8_t *rows[VDEV_RAIDZ_MAXPARITY];
+	uint8_t *invrows[VDEV_RAIDZ_MAXPARITY];
+	uint8_t *used;
+
+	int code = 0;
+
+
+	n = rm->rm_cols - rm->rm_firstdatacol;
+
+	/*
+	 * Figure out which data columns are missing.
+	 */
+	nmissing_rows = 0;
+	for (t = 0; t < ntgts; t++) {
+		if (tgts[t] >= rm->rm_firstdatacol) {
+			missing_rows[nmissing_rows++] =
+			    tgts[t] - rm->rm_firstdatacol;
+		}
+	}
+
+	/*
+	 * Figure out which parity columns to use to help generate the missing
+	 * data columns.
+	 */
+	for (tt = 0, c = 0, i = 0; i < nmissing_rows; c++) {
+		ASSERT(tt < ntgts);
+		ASSERT(c < rm->rm_firstdatacol);
+
+		/*
+		 * Skip any targeted parity columns.
+		 */
+		if (c == tgts[tt]) {
+			tt++;
+			continue;
+		}
+
+		code |= 1 << c;
+
+		parity_map[i] = c;
+		i++;
+	}
+
+	ASSERT(code != 0);
+	ASSERT3U(code, <, 1 << VDEV_RAIDZ_MAXPARITY);
+
+	psize = (sizeof (rows[0][0]) + sizeof (invrows[0][0])) *
+	    nmissing_rows * n + sizeof (used[0]) * n;
+	p = kmem_alloc(psize, KM_SLEEP);
+
+	for (pp = p, i = 0; i < nmissing_rows; i++) {
+		rows[i] = pp;
+		pp += n;
+		invrows[i] = pp;
+		pp += n;
+	}
+	used = pp;
+
+	for (i = 0; i < nmissing_rows; i++) {
+		used[i] = parity_map[i];
+	}
+
+	for (tt = 0, c = rm->rm_firstdatacol; c < rm->rm_cols; c++) {
+		if (tt < nmissing_rows &&
+		    c == missing_rows[tt] + rm->rm_firstdatacol) {
+			tt++;
+			continue;
+		}
+
+		ASSERT3S(i, <, n);
+		used[i] = c;
+		i++;
+	}
+
+	/*
+	 * Initialize the interesting rows of the matrix.
+	 */
+	vdev_raidz_matrix_init(rm, n, nmissing_rows, parity_map, rows);
+
+	/*
+	 * Invert the matrix.
+	 */
+	vdev_raidz_matrix_invert(rm, n, nmissing_rows, missing_rows, rows,
+	    invrows, used);
+
+	/*
+	 * Reconstruct the missing data using the generated matrix.
+	 */
+	vdev_raidz_matrix_reconstruct(rm, n, nmissing_rows, missing_rows,
+	    invrows, used);
+
+	kmem_free(p, psize);
+
+	return (code);
 }
 
+static int
+vdev_raidz_reconstruct(raidz_map_t *rm, int *t, int nt)
+{
+	int tgts[VDEV_RAIDZ_MAXPARITY], *dt;
+	int ntgts;
+	int i, c;
+	int code;
+	int nbadparity, nbaddata;
+	int parity_valid[VDEV_RAIDZ_MAXPARITY];
+
+	/*
+	 * The tgts list must already be sorted.
+	 */
+	for (i = 1; i < nt; i++) {
+		ASSERT(t[i] > t[i - 1]);
+	}
+
+	nbadparity = rm->rm_firstdatacol;
+	nbaddata = rm->rm_cols - nbadparity;
+	ntgts = 0;
+	for (i = 0, c = 0; c < rm->rm_cols; c++) {
+		if (c < rm->rm_firstdatacol)
+			parity_valid[c] = B_FALSE;
+
+		if (i < nt && c == t[i]) {
+			tgts[ntgts++] = c;
+			i++;
+		} else if (rm->rm_col[c].rc_error != 0) {
+			tgts[ntgts++] = c;
+		} else if (c >= rm->rm_firstdatacol) {
+			nbaddata--;
+		} else {
+			parity_valid[c] = B_TRUE;
+			nbadparity--;
+		}
+	}
+
+	ASSERT(ntgts >= nt);
+	ASSERT(nbaddata >= 0);
+	ASSERT(nbaddata + nbadparity == ntgts);
+
+	dt = &tgts[nbadparity];
+
+	/*
+	 * See if we can use any of our optimized reconstruction routines.
+	 */
+	if (!vdev_raidz_default_to_general) {
+		switch (nbaddata) {
+		case 1:
+			if (parity_valid[VDEV_RAIDZ_P])
+				return (vdev_raidz_reconstruct_p(rm, dt, 1));
+
+			ASSERT(rm->rm_firstdatacol > 1);
+
+			if (parity_valid[VDEV_RAIDZ_Q])
+				return (vdev_raidz_reconstruct_q(rm, dt, 1));
+
+			ASSERT(rm->rm_firstdatacol > 2);
+			break;
+
+		case 2:
+			ASSERT(rm->rm_firstdatacol > 1);
+
+			if (parity_valid[VDEV_RAIDZ_P] &&
+			    parity_valid[VDEV_RAIDZ_Q])
+				return (vdev_raidz_reconstruct_pq(rm, dt, 2));
+
+			ASSERT(rm->rm_firstdatacol > 2);
+
+			break;
+		}
+	}
+
+	code = vdev_raidz_reconstruct_general(rm, tgts, ntgts);
+	ASSERT(code < (1 << VDEV_RAIDZ_MAXPARITY));
+	ASSERT(code > 0);
+	return (code);
+}
 
 static int
 vdev_raidz_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift)
 {
 	vdev_t *cvd;
 	uint64_t nparity = vd->vdev_nparity;
-	int c, error;
+	int c;
 	int lasterror = 0;
 	int numerrors = 0;
 
@@ -573,11 +1270,13 @@ vdev_raidz_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift)
 		return (EINVAL);
 	}
 
+	vdev_open_children(vd);
+
 	for (c = 0; c < vd->vdev_children; c++) {
 		cvd = vd->vdev_child[c];
 
-		if ((error = vdev_open(cvd)) != 0) {
-			lasterror = error;
+		if (cvd->vdev_open_error != 0) {
+			lasterror = cvd->vdev_open_error;
 			numerrors++;
 			continue;
 		}
@@ -639,7 +1338,7 @@ vdev_raidz_io_start(zio_t *zio)
 	blkptr_t *bp = zio->io_bp;
 	raidz_map_t *rm;
 	raidz_col_t *rc;
-	int c;
+	int c, i;
 
 	rm = vdev_raidz_map_alloc(zio, tvd->vdev_ashift, vd->vdev_children,
 	    vd->vdev_nparity);
@@ -647,13 +1346,7 @@ vdev_raidz_io_start(zio_t *zio)
 	ASSERT3U(rm->rm_asize, ==, vdev_psize_to_asize(vd, zio->io_size));
 
 	if (zio->io_type == ZIO_TYPE_WRITE) {
-		/*
-		 * Generate RAID parity in the first virtual columns.
-		 */
-		if (rm->rm_firstdatacol == 1)
-			vdev_raidz_generate_parity_p(rm);
-		else
-			vdev_raidz_generate_parity_pq(rm);
+		vdev_raidz_generate_parity(rm);
 
 		for (c = 0; c < rm->rm_cols; c++) {
 			rc = &rm->rm_col[c];
@@ -664,6 +1357,23 @@ vdev_raidz_io_start(zio_t *zio)
 			    vdev_raidz_child_done, rc));
 		}
 
+		/*
+		 * Generate optional I/Os for any skipped sectors to improve
+		 * aggregation contiguity.
+		 */
+		for (c = rm->rm_bigcols, i = 0; i < rm->rm_skipped; c++, i++) {
+			ASSERT(c <= rm->rm_scols);
+			if (c == rm->rm_scols)
+				c = 0;
+			rc = &rm->rm_col[c];
+			cvd = vd->vdev_child[rc->rc_devidx];
+			zio_nowait(zio_vdev_child_io(zio, NULL, cvd,
+			    rc->rc_offset + rc->rc_size, NULL,
+			    1 << tvd->vdev_ashift,
+			    zio->io_type, zio->io_priority,
+			    ZIO_FLAG_NODATA | ZIO_FLAG_OPTIONAL, NULL, NULL));
+		}
+
 		return (ZIO_PIPELINE_CONTINUE);
 	}
 
@@ -671,8 +1381,7 @@ vdev_raidz_io_start(zio_t *zio)
 
 	/*
 	 * Iterate over the columns in reverse order so that we hit the parity
-	 * last -- any errors along the way will force us to read the parity
-	 * data.
+	 * last -- any errors along the way will force us to read the parity.
 	 */
 	for (c = rm->rm_cols - 1; c >= 0; c--) {
 		rc = &rm->rm_col[c];
@@ -748,10 +1457,7 @@ raidz_parity_verify(zio_t *zio, raidz_map_t *rm)
 		bcopy(rc->rc_data, orig[c], rc->rc_size);
 	}
 
-	if (rm->rm_firstdatacol == 1)
-		vdev_raidz_generate_parity_p(rm);
-	else
-		vdev_raidz_generate_parity_pq(rm);
+	vdev_raidz_generate_parity(rm);
 
 	for (c = 0; c < rm->rm_firstdatacol; c++) {
 		rc = &rm->rm_col[c];
@@ -768,9 +1474,10 @@ raidz_parity_verify(zio_t *zio, raidz_map_t *rm)
 	return (ret);
 }
 
-static uint64_t raidz_corrected_p;
-static uint64_t raidz_corrected_q;
-static uint64_t raidz_corrected_pq;
+/*
+ * Keep statistics on all the ways that we used parity to correct data.
+ */
+static uint64_t raidz_corrected[1 << VDEV_RAIDZ_MAXPARITY];
 
 static int
 vdev_raidz_worst_error(raidz_map_t *rm)
@@ -783,19 +1490,176 @@ vdev_raidz_worst_error(raidz_map_t *rm)
 	return (error);
 }
 
+/*
+ * Iterate over all combinations of bad data and attempt a reconstruction.
+ * Note that the algorithm below is non-optimal because it doesn't take into
+ * account how reconstruction is actually performed. For example, with
+ * triple-parity RAID-Z the reconstruction procedure is the same if column 4
+ * is targeted as invalid as if columns 1 and 4 are targeted since in both
+ * cases we'd only use parity information in column 0.
+ */
+static int
+vdev_raidz_combrec(zio_t *zio, int total_errors, int data_errors)
+{
+	raidz_map_t *rm = zio->io_vsd;
+	raidz_col_t *rc;
+	void *orig[VDEV_RAIDZ_MAXPARITY];
+	int tstore[VDEV_RAIDZ_MAXPARITY + 2];
+	int *tgts = &tstore[1];
+	int current, next, i, c, n;
+	int code, ret = 0;
+
+	ASSERT(total_errors < rm->rm_firstdatacol);
+
+	/*
+	 * This simplifies one edge condition.
+	 */
+	tgts[-1] = -1;
+
+	for (n = 1; n <= rm->rm_firstdatacol - total_errors; n++) {
+		/*
+		 * Initialize the targets array by finding the first n columns
+		 * that contain no error.
+		 *
+		 * If there were no data errors, we need to ensure that we're
+		 * always explicitly attempting to reconstruct at least one
+		 * data column. To do this, we simply push the highest target
+		 * up into the data columns.
+		 */
+		for (c = 0, i = 0; i < n; i++) {
+			if (i == n - 1 && data_errors == 0 &&
+			    c < rm->rm_firstdatacol) {
+				c = rm->rm_firstdatacol;
+			}
+
+			while (rm->rm_col[c].rc_error != 0) {
+				c++;
+				ASSERT3S(c, <, rm->rm_cols);
+			}
+
+			tgts[i] = c++;
+		}
+
+		/*
+		 * Setting tgts[n] simplifies the other edge condition.
+		 */
+		tgts[n] = rm->rm_cols;
+
+		/*
+		 * These buffers were allocated in previous iterations.
+		 */
+		for (i = 0; i < n - 1; i++) {
+			ASSERT(orig[i] != NULL);
+		}
+
+		orig[n - 1] = zio_buf_alloc(rm->rm_col[0].rc_size);
+
+		current = 0;
+		next = tgts[current];
+
+		while (current != n) {
+			tgts[current] = next;
+			current = 0;
+
+			/*
+			 * Save off the original data that we're going to
+			 * attempt to reconstruct.
+			 */
+			for (i = 0; i < n; i++) {
+				ASSERT(orig[i] != NULL);
+				c = tgts[i];
+				ASSERT3S(c, >=, 0);
+				ASSERT3S(c, <, rm->rm_cols);
+				rc = &rm->rm_col[c];
+				bcopy(rc->rc_data, orig[i], rc->rc_size);
+			}
+
+			/*
+			 * Attempt a reconstruction and exit the outer loop on
+			 * success.
+			 */
+			code = vdev_raidz_reconstruct(rm, tgts, n);
+			if (zio_checksum_error(zio) == 0) {
+				atomic_inc_64(&raidz_corrected[code]);
+
+				for (i = 0; i < n; i++) {
+					c = tgts[i];
+					rc = &rm->rm_col[c];
+					ASSERT(rc->rc_error == 0);
+					if (rc->rc_tried)
+						raidz_checksum_error(zio, rc);
+					rc->rc_error = ECKSUM;
+				}
+
+				ret = code;
+				goto done;
+			}
+
+			/*
+			 * Restore the original data.
+			 */
+			for (i = 0; i < n; i++) {
+				c = tgts[i];
+				rc = &rm->rm_col[c];
+				bcopy(orig[i], rc->rc_data, rc->rc_size);
+			}
+
+			do {
+				/*
+				 * Find the next valid column after the current
+				 * position..
+				 */
+				for (next = tgts[current] + 1;
+				    next < rm->rm_cols &&
+				    rm->rm_col[next].rc_error != 0; next++)
+					continue;
+
+				ASSERT(next <= tgts[current + 1]);
+
+				/*
+				 * If that spot is available, we're done here.
+				 */
+				if (next != tgts[current + 1])
+					break;
+
+				/*
+				 * Otherwise, find the next valid column after
+				 * the previous position.
+				 */
+				for (c = tgts[current - 1] + 1;
+				    rm->rm_col[c].rc_error != 0; c++)
+					continue;
+
+				tgts[current] = c;
+				current++;
+
+			} while (current != n);
+		}
+	}
+	n--;
+done:
+	for (i = 0; i < n; i++) {
+		zio_buf_free(orig[i], rm->rm_col[0].rc_size);
+	}
+
+	return (ret);
+}
+
 static void
 vdev_raidz_io_done(zio_t *zio)
 {
 	vdev_t *vd = zio->io_vd;
 	vdev_t *cvd;
 	raidz_map_t *rm = zio->io_vsd;
-	raidz_col_t *rc = NULL, *rc1;
+	raidz_col_t *rc = NULL;
 	int unexpected_errors = 0;
 	int parity_errors = 0;
 	int parity_untried = 0;
 	int data_errors = 0;
 	int total_errors = 0;
-	int n, c, c1;
+	int n, c;
+	int tgts[VDEV_RAIDZ_MAXPARITY];
+	int code;
 
 	ASSERT(zio->io_bp != NULL);  /* XXX need to add code to enforce this */
 
@@ -859,8 +1723,7 @@ vdev_raidz_io_done(zio_t *zio)
 	 * any errors.
 	 */
 	if (total_errors <= rm->rm_firstdatacol - parity_untried) {
-		switch (data_errors) {
-		case 0:
+		if (data_errors == 0) {
 			if (zio_checksum_error(zio) == 0) {
 				/*
 				 * If we read parity information (unnecessarily
@@ -880,9 +1743,7 @@ vdev_raidz_io_done(zio_t *zio)
 				}
 				goto done;
 			}
-			break;
-
-		case 1:
+		} else {
 			/*
 			 * We either attempt to read all the parity columns or
 			 * none of them. If we didn't try to read parity, we
@@ -894,45 +1755,38 @@ vdev_raidz_io_done(zio_t *zio)
 			ASSERT(parity_errors < rm->rm_firstdatacol);
 
 			/*
-			 * Find the column that reported the error.
+			 * Identify the data columns that reported an error.
 			 */
+			n = 0;
 			for (c = rm->rm_firstdatacol; c < rm->rm_cols; c++) {
 				rc = &rm->rm_col[c];
-				if (rc->rc_error != 0)
-					break;
+				if (rc->rc_error != 0) {
+					ASSERT(n < VDEV_RAIDZ_MAXPARITY);
+					tgts[n++] = c;
+				}
 			}
-			ASSERT(c != rm->rm_cols);
-			ASSERT(!rc->rc_skipped || rc->rc_error == ENXIO ||
-			    rc->rc_error == ESTALE);
 
-			if (rm->rm_col[VDEV_RAIDZ_P].rc_error == 0) {
-				vdev_raidz_reconstruct_p(rm, c);
-			} else {
-				ASSERT(rm->rm_firstdatacol > 1);
-				vdev_raidz_reconstruct_q(rm, c);
-			}
+			ASSERT(rm->rm_firstdatacol >= n);
+
+			code = vdev_raidz_reconstruct(rm, tgts, n);
 
 			if (zio_checksum_error(zio) == 0) {
-				if (rm->rm_col[VDEV_RAIDZ_P].rc_error == 0)
-					atomic_inc_64(&raidz_corrected_p);
-				else
-					atomic_inc_64(&raidz_corrected_q);
+				atomic_inc_64(&raidz_corrected[code]);
 
 				/*
-				 * If there's more than one parity disk that
-				 * was successfully read, confirm that the
-				 * other parity disk produced the correct data.
-				 * This routine is suboptimal in that it
-				 * regenerates both the parity we wish to test
-				 * as well as the parity we just used to
-				 * perform the reconstruction, but this should
-				 * be a relatively uncommon case, and can be
-				 * optimized if it becomes a problem.
-				 * We also regenerate parity when resilvering
-				 * so we can write it out to the failed device
-				 * later.
+				 * If we read more parity disks than were used
+				 * for reconstruction, confirm that the other
+				 * parity disks produced correct data. This
+				 * routine is suboptimal in that it regenerates
+				 * the parity that we already used in addition
+				 * to the parity that we're attempting to
+				 * verify, but this should be a relatively
+				 * uncommon case, and can be optimized if it
+				 * becomes a problem. Note that we regenerate
+				 * parity when resilvering so we can write it
+				 * out to failed devices later.
 				 */
-				if (parity_errors < rm->rm_firstdatacol - 1 ||
+				if (parity_errors < rm->rm_firstdatacol - n ||
 				    (zio->io_flags & ZIO_FLAG_RESILVER)) {
 					n = raidz_parity_verify(zio, rm);
 					unexpected_errors += n;
@@ -942,46 +1796,6 @@ vdev_raidz_io_done(zio_t *zio)
 
 				goto done;
 			}
-			break;
-
-		case 2:
-			/*
-			 * Two data column errors require double parity.
-			 */
-			ASSERT(rm->rm_firstdatacol == 2);
-
-			/*
-			 * Find the two columns that reported errors.
-			 */
-			for (c = rm->rm_firstdatacol; c < rm->rm_cols; c++) {
-				rc = &rm->rm_col[c];
-				if (rc->rc_error != 0)
-					break;
-			}
-			ASSERT(c != rm->rm_cols);
-			ASSERT(!rc->rc_skipped || rc->rc_error == ENXIO ||
-			    rc->rc_error == ESTALE);
-
-			for (c1 = c++; c < rm->rm_cols; c++) {
-				rc = &rm->rm_col[c];
-				if (rc->rc_error != 0)
-					break;
-			}
-			ASSERT(c != rm->rm_cols);
-			ASSERT(!rc->rc_skipped || rc->rc_error == ENXIO ||
-			    rc->rc_error == ESTALE);
-
-			vdev_raidz_reconstruct_pq(rm, c1, c);
-
-			if (zio_checksum_error(zio) == 0) {
-				atomic_inc_64(&raidz_corrected_pq);
-				goto done;
-			}
-			break;
-
-		default:
-			ASSERT(rm->rm_firstdatacol <= 2);
-			ASSERT(0);
 		}
 	}
 
@@ -1020,8 +1834,10 @@ vdev_raidz_io_done(zio_t *zio)
 	 * errors we detected, and we've attempted to read all columns. There
 	 * must, therefore, be one or more additional problems -- silent errors
 	 * resulting in invalid data rather than explicit I/O errors resulting
-	 * in absent data. Before we attempt combinatorial reconstruction make
-	 * sure we have a chance of coming up with the right answer.
+	 * in absent data. We check if there is enough additional data to
+	 * possibly reconstruct the data and then perform combinatorial
+	 * reconstruction over all possible combinations. If that fails,
+	 * we're cooked.
 	 */
 	if (total_errors >= rm->rm_firstdatacol) {
 		zio->io_error = vdev_raidz_worst_error(rm);
@@ -1032,133 +1848,30 @@ vdev_raidz_io_done(zio_t *zio)
 		 */
 		if (total_errors == rm->rm_firstdatacol)
 			zio->io_error = zio_worst_error(zio->io_error, ECKSUM);
-		goto done;
-	}
 
-	if (rm->rm_col[VDEV_RAIDZ_P].rc_error == 0) {
+	} else if ((code = vdev_raidz_combrec(zio, total_errors,
+	    data_errors)) != 0) {
 		/*
-		 * Attempt to reconstruct the data from parity P.
+		 * If we didn't use all the available parity for the
+		 * combinatorial reconstruction, verify that the remaining
+		 * parity is correct.
 		 */
-		for (c = rm->rm_firstdatacol; c < rm->rm_cols; c++) {
-			void *orig;
-			rc = &rm->rm_col[c];
-
-			orig = zio_buf_alloc(rc->rc_size);
-			bcopy(rc->rc_data, orig, rc->rc_size);
-			vdev_raidz_reconstruct_p(rm, c);
-
-			if (zio_checksum_error(zio) == 0) {
-				zio_buf_free(orig, rc->rc_size);
-				atomic_inc_64(&raidz_corrected_p);
-
-				/*
-				 * If this child didn't know that it returned
-				 * bad data, inform it.
-				 */
-				if (rc->rc_tried && rc->rc_error == 0)
-					raidz_checksum_error(zio, rc);
-				rc->rc_error = ECKSUM;
-				goto done;
-			}
-
-			bcopy(orig, rc->rc_data, rc->rc_size);
-			zio_buf_free(orig, rc->rc_size);
-		}
-	}
-
-	if (rm->rm_firstdatacol > 1 && rm->rm_col[VDEV_RAIDZ_Q].rc_error == 0) {
+		if (code != (1 << rm->rm_firstdatacol) - 1)
+			(void) raidz_parity_verify(zio, rm);
+	} else {
 		/*
-		 * Attempt to reconstruct the data from parity Q.
+		 * All combinations failed to checksum. Generate checksum
+		 * ereports for all children.
 		 */
-		for (c = rm->rm_firstdatacol; c < rm->rm_cols; c++) {
-			void *orig;
-			rc = &rm->rm_col[c];
-
-			orig = zio_buf_alloc(rc->rc_size);
-			bcopy(rc->rc_data, orig, rc->rc_size);
-			vdev_raidz_reconstruct_q(rm, c);
-
-			if (zio_checksum_error(zio) == 0) {
-				zio_buf_free(orig, rc->rc_size);
-				atomic_inc_64(&raidz_corrected_q);
-
-				/*
-				 * If this child didn't know that it returned
-				 * bad data, inform it.
-				 */
-				if (rc->rc_tried && rc->rc_error == 0)
-					raidz_checksum_error(zio, rc);
-				rc->rc_error = ECKSUM;
-				goto done;
-			}
-
-			bcopy(orig, rc->rc_data, rc->rc_size);
-			zio_buf_free(orig, rc->rc_size);
-		}
-	}
+		zio->io_error = ECKSUM;
 
-	if (rm->rm_firstdatacol > 1 &&
-	    rm->rm_col[VDEV_RAIDZ_P].rc_error == 0 &&
-	    rm->rm_col[VDEV_RAIDZ_Q].rc_error == 0) {
-		/*
-		 * Attempt to reconstruct the data from both P and Q.
-		 */
-		for (c = rm->rm_firstdatacol; c < rm->rm_cols - 1; c++) {
-			void *orig, *orig1;
-			rc = &rm->rm_col[c];
-
-			orig = zio_buf_alloc(rc->rc_size);
-			bcopy(rc->rc_data, orig, rc->rc_size);
-
-			for (c1 = c + 1; c1 < rm->rm_cols; c1++) {
-				rc1 = &rm->rm_col[c1];
-
-				orig1 = zio_buf_alloc(rc1->rc_size);
-				bcopy(rc1->rc_data, orig1, rc1->rc_size);
-
-				vdev_raidz_reconstruct_pq(rm, c, c1);
-
-				if (zio_checksum_error(zio) == 0) {
-					zio_buf_free(orig, rc->rc_size);
-					zio_buf_free(orig1, rc1->rc_size);
-					atomic_inc_64(&raidz_corrected_pq);
-
-					/*
-					 * If these children didn't know they
-					 * returned bad data, inform them.
-					 */
-					if (rc->rc_tried && rc->rc_error == 0)
-						raidz_checksum_error(zio, rc);
-					if (rc1->rc_tried && rc1->rc_error == 0)
-						raidz_checksum_error(zio, rc1);
-
-					rc->rc_error = ECKSUM;
-					rc1->rc_error = ECKSUM;
-
-					goto done;
-				}
-
-				bcopy(orig1, rc1->rc_data, rc1->rc_size);
-				zio_buf_free(orig1, rc1->rc_size);
+		if (!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
+			for (c = 0; c < rm->rm_cols; c++) {
+				rc = &rm->rm_col[c];
+				zfs_ereport_post(FM_EREPORT_ZFS_CHECKSUM,
+				    zio->io_spa, vd->vdev_child[rc->rc_devidx],
+				    zio, rc->rc_offset, rc->rc_size);
 			}
-
-			bcopy(orig, rc->rc_data, rc->rc_size);
-			zio_buf_free(orig, rc->rc_size);
-		}
-	}
-
-	/*
-	 * All combinations failed to checksum. Generate checksum ereports for
-	 * all children.
-	 */
-	zio->io_error = ECKSUM;
-
-	if (!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
-		for (c = 0; c < rm->rm_cols; c++) {
-			rc = &rm->rm_col[c];
-			zfs_ereport_post(FM_EREPORT_ZFS_CHECKSUM,
-			    zio->io_spa, vd->vdev_child[rc->rc_devidx], zio,
-			    rc->rc_offset, rc->rc_size);
 		}
 	}
 
diff --git a/module/zfs/vdev_root.c b/module/zfs/vdev_root.c
index 88383f002..524c8e606 100644
--- a/module/zfs/vdev_root.c
+++ b/module/zfs/vdev_root.c
@@ -19,7 +19,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -52,7 +52,6 @@ too_many_errors(vdev_t *vd, int numerrors)
 static int
 vdev_root_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift)
 {
-	int c;
 	int lasterror = 0;
 	int numerrors = 0;
 
@@ -61,15 +60,14 @@ vdev_root_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift)
 		return (EINVAL);
 	}
 
-	for (c = 0; c < vd->vdev_children; c++) {
+	vdev_open_children(vd);
+
+	for (int c = 0; c < vd->vdev_children; c++) {
 		vdev_t *cvd = vd->vdev_child[c];
-		int error;
 
-		if ((error = vdev_open(cvd)) != 0 &&
-		    !cvd->vdev_islog) {
-			lasterror = error;
+		if (cvd->vdev_open_error && !cvd->vdev_islog) {
+			lasterror = cvd->vdev_open_error;
 			numerrors++;
-			continue;
 		}
 	}
 
@@ -87,9 +85,7 @@ vdev_root_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift)
 static void
 vdev_root_close(vdev_t *vd)
 {
-	int c;
-
-	for (c = 0; c < vd->vdev_children; c++)
+	for (int c = 0; c < vd->vdev_children; c++)
 		vdev_close(vd->vdev_child[c]);
 }
 
diff --git a/module/zfs/zap_micro.c b/module/zfs/zap_micro.c
index 891c948da..7e278dd2b 100644
--- a/module/zfs/zap_micro.c
+++ b/module/zfs/zap_micro.c
@@ -1107,7 +1107,7 @@ zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
 
 int
 zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
-    uint64_t *towrite, uint64_t *tooverwrite, uint64_t dn_datablkshift)
+    uint64_t *towrite, uint64_t *tooverwrite)
 {
 	zap_t *zap;
 	int err = 0;
@@ -1152,28 +1152,28 @@ zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
 			*towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE;
 		}
 	} else {
-		if (!add) {
-			if (dmu_buf_freeable(zap->zap_dbuf))
-				*tooverwrite += SPA_MAXBLOCKSIZE;
-			else
-				*towrite += SPA_MAXBLOCKSIZE;
-		} else {
-			/*
-			 * We are here if we are adding and (name != NULL).
-			 * It is hard to find out if this add will promote this
-			 * microzap to fatzap. Hence, we assume the worst case
-			 * and account for the blocks assuming this microzap
-			 * would be promoted to a fatzap.
-			 *
-			 * 1 block overwritten  : header block
-			 * 4 new blocks written : 2 new split leaf, 2 grown
-			 *			ptrtbl blocks
-			 */
-			if (dmu_buf_freeable(zap->zap_dbuf))
-				*tooverwrite += 1 << dn_datablkshift;
-			else
-				*towrite += 1 << dn_datablkshift;
-			*towrite += 4 << dn_datablkshift;
+		/*
+		 * We are here if (name != NULL) and this is a micro-zap.
+		 * We account for the header block depending on whether it
+		 * is freeable.
+		 *
+		 * Incase of an add-operation it is hard to find out
+		 * if this add will promote this microzap to fatzap.
+		 * Hence, we consider the worst case and account for the
+		 * blocks assuming this microzap would be promoted to a
+		 * fatzap.
+		 *
+		 * 1 block overwritten  : header block
+		 * 4 new blocks written : 2 new split leaf, 2 grown
+		 *			ptrtbl blocks
+		 */
+		if (dmu_buf_freeable(zap->zap_dbuf))
+			*tooverwrite += SPA_MAXBLOCKSIZE;
+		else
+			*towrite += SPA_MAXBLOCKSIZE;
+
+		if (add) {
+			*towrite += 4 * SPA_MAXBLOCKSIZE;
 		}
 	}
 
diff --git a/module/zfs/zfs_acl.c b/module/zfs/zfs_acl.c
index 734bd8395..12ffe9f30 100644
--- a/module/zfs/zfs_acl.c
+++ b/module/zfs/zfs_acl.c
@@ -93,6 +93,8 @@
 #define	ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
     ZFS_ACL_OBJ_ACE)
 
+#define	ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
+
 static uint16_t
 zfs_ace_v0_get_type(void *acep)
 {
@@ -781,6 +783,7 @@ zfs_mode_compute(znode_t *zp, zfs_acl_t *aclp)
 	uint64_t	who;
 	uint16_t	iflags, type;
 	uint32_t	access_mask;
+	boolean_t	an_exec_denied = B_FALSE;
 
 	mode = (zp->z_phys->zp_mode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
 
@@ -905,8 +908,32 @@ zfs_mode_compute(znode_t *zp, zfs_acl_t *aclp)
 					}
 				}
 			}
+		} else {
+			/*
+			 * Only care if this IDENTIFIER_GROUP or
+			 * USER ACE denies execute access to someone,
+			 * mode is not affected
+			 */
+			if ((access_mask & ACE_EXECUTE) && type == DENY)
+				an_exec_denied = B_TRUE;
 		}
 	}
+
+	/*
+	 * Failure to allow is effectively a deny, so execute permission
+	 * is denied if it was never mentioned or if we explicitly
+	 * weren't allowed it.
+	 */
+	if (!an_exec_denied &&
+	    ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
+	    (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
+		an_exec_denied = B_TRUE;
+
+	if (an_exec_denied)
+		zp->z_phys->zp_flags &= ~ZFS_NO_EXECS_DENIED;
+	else
+		zp->z_phys->zp_flags |= ZFS_NO_EXECS_DENIED;
+
 	return (mode);
 }
 
@@ -946,7 +973,8 @@ zfs_acl_node_read_internal(znode_t *zp, boolean_t will_modify)
 }
 
 /*
- * Read an external acl object.
+ * Read an external acl object.  If the intent is to modify, always
+ * create a new acl and leave any cached acl in place.
  */
 static int
 zfs_acl_node_read(znode_t *zp, zfs_acl_t **aclpp, boolean_t will_modify)
@@ -960,8 +988,15 @@ zfs_acl_node_read(znode_t *zp, zfs_acl_t **aclpp, boolean_t will_modify)
 
 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
 
+	if (zp->z_acl_cached && !will_modify) {
+		*aclpp = zp->z_acl_cached;
+		return (0);
+	}
+
 	if (zp->z_phys->zp_acl.z_acl_extern_obj == 0) {
 		*aclpp = zfs_acl_node_read_internal(zp, will_modify);
+		if (!will_modify)
+			zp->z_acl_cached = *aclpp;
 		return (0);
 	}
 
@@ -995,6 +1030,8 @@ zfs_acl_node_read(znode_t *zp, zfs_acl_t **aclpp, boolean_t will_modify)
 	}
 
 	*aclpp = aclp;
+	if (!will_modify)
+		zp->z_acl_cached = aclp;
 	return (0);
 }
 
@@ -1019,11 +1056,16 @@ zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
 
 	dmu_buf_will_dirty(zp->z_dbuf, tx);
 
+	if (zp->z_acl_cached) {
+		zfs_acl_free(zp->z_acl_cached);
+		zp->z_acl_cached = NULL;
+	}
+
 	zphys->zp_mode = zfs_mode_compute(zp, aclp);
 
 	/*
-	 * Decide which opbject type to use.  If we are forced to
-	 * use old ACL format than transform ACL into zfs_oldace_t
+	 * Decide which object type to use.  If we are forced to
+	 * use old ACL format then transform ACL into zfs_oldace_t
 	 * layout.
 	 */
 	if (!zfsvfs->z_use_fuids) {
@@ -1869,7 +1911,6 @@ zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
 			mutex_exit(&dzp->z_acl_lock);
 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
 			    vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
-			zfs_acl_free(paclp);
 		} else {
 			acl_ids->z_aclp =
 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
@@ -1998,8 +2039,6 @@ zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
 
 	mutex_exit(&zp->z_acl_lock);
 
-	zfs_acl_free(aclp);
-
 	return (0);
 }
 
@@ -2095,11 +2134,6 @@ zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
 		aclp->z_hints |= (zp->z_phys->zp_flags & V4_ACL_WIDE_FLAGS);
 	}
 top:
-	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr)) {
-		zfs_acl_free(aclp);
-		return (error);
-	}
-
 	mutex_enter(&zp->z_lock);
 	mutex_enter(&zp->z_acl_lock);
 
@@ -2145,6 +2179,7 @@ top:
 
 	error = zfs_aclset_common(zp, aclp, cr, tx);
 	ASSERT(error == 0);
+	zp->z_acl_cached = aclp;
 
 	if (fuid_dirtied)
 		zfs_fuid_sync(zfsvfs, tx);
@@ -2154,7 +2189,6 @@ top:
 
 	if (fuidp)
 		zfs_fuid_info_free(fuidp);
-	zfs_acl_free(aclp);
 	dmu_tx_commit(tx);
 done:
 	mutex_exit(&zp->z_acl_lock);
@@ -2301,7 +2335,6 @@ zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
 					checkit = B_TRUE;
 				break;
 			} else {
-				zfs_acl_free(aclp);
 				mutex_exit(&zp->z_acl_lock);
 				return (EIO);
 			}
@@ -2321,7 +2354,6 @@ zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
 				    uint32_t, mask_matched);
 				if (anyaccess) {
 					mutex_exit(&zp->z_acl_lock);
-					zfs_acl_free(aclp);
 					return (0);
 				}
 			}
@@ -2334,7 +2366,6 @@ zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
 	}
 
 	mutex_exit(&zp->z_acl_lock);
-	zfs_acl_free(aclp);
 
 	/* Put the found 'denies' back on the working mode */
 	if (deny_mask) {
@@ -2366,8 +2397,7 @@ zfs_has_access(znode_t *zp, cred_t *cr)
 		    secpolicy_vnode_access(cr, ZTOV(zp), owner, VREAD) == 0 ||
 		    secpolicy_vnode_access(cr, ZTOV(zp), owner, VWRITE) == 0 ||
 		    secpolicy_vnode_access(cr, ZTOV(zp), owner, VEXEC) == 0 ||
-		    secpolicy_vnode_chown(cr, B_TRUE) == 0 ||
-		    secpolicy_vnode_chown(cr, B_FALSE) == 0 ||
+		    secpolicy_vnode_chown(cr, owner) == 0 ||
 		    secpolicy_vnode_setdac(cr, owner) == 0 ||
 		    secpolicy_vnode_remove(cr) == 0);
 	}
@@ -2421,6 +2451,78 @@ zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
 	    check_privs, B_FALSE, cr));
 }
 
+int
+zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
+{
+	boolean_t owner = B_FALSE;
+	boolean_t groupmbr = B_FALSE;
+	boolean_t is_attr;
+	uid_t fowner;
+	uid_t gowner;
+	uid_t uid = crgetuid(cr);
+	int error;
+
+	if (zdp->z_phys->zp_flags & ZFS_AV_QUARANTINED)
+		return (EACCES);
+
+	is_attr = ((zdp->z_phys->zp_flags & ZFS_XATTR) &&
+	    (ZTOV(zdp)->v_type == VDIR));
+	if (is_attr)
+		goto slow;
+
+	mutex_enter(&zdp->z_acl_lock);
+
+	if (zdp->z_phys->zp_flags & ZFS_NO_EXECS_DENIED) {
+		mutex_exit(&zdp->z_acl_lock);
+		return (0);
+	}
+
+	if (FUID_INDEX(zdp->z_phys->zp_uid) != 0 ||
+	    FUID_INDEX(zdp->z_phys->zp_gid) != 0) {
+		mutex_exit(&zdp->z_acl_lock);
+		goto slow;
+	}
+
+	fowner = (uid_t)zdp->z_phys->zp_uid;
+	gowner = (uid_t)zdp->z_phys->zp_gid;
+
+	if (uid == fowner) {
+		owner = B_TRUE;
+		if (zdp->z_phys->zp_mode & S_IXUSR) {
+			mutex_exit(&zdp->z_acl_lock);
+			return (0);
+		} else {
+			mutex_exit(&zdp->z_acl_lock);
+			goto slow;
+		}
+	}
+	if (groupmember(gowner, cr)) {
+		groupmbr = B_TRUE;
+		if (zdp->z_phys->zp_mode & S_IXGRP) {
+			mutex_exit(&zdp->z_acl_lock);
+			return (0);
+		} else {
+			mutex_exit(&zdp->z_acl_lock);
+			goto slow;
+		}
+	}
+	if (!owner && !groupmbr) {
+		if (zdp->z_phys->zp_mode & S_IXOTH) {
+			mutex_exit(&zdp->z_acl_lock);
+			return (0);
+		}
+	}
+
+	mutex_exit(&zdp->z_acl_lock);
+
+slow:
+	DTRACE_PROBE(zfs__fastpath__execute__access__miss);
+	ZFS_ENTER(zdp->z_zfsvfs);
+	error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr);
+	ZFS_EXIT(zdp->z_zfsvfs);
+	return (error);
+}
+
 /*
  * Determine whether Access should be granted/denied, invoking least
  * priv subsytem when a deny is determined.
@@ -2515,7 +2617,7 @@ zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
 			    owner, checkmode);
 
 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
-			error = secpolicy_vnode_chown(cr, B_TRUE);
+			error = secpolicy_vnode_chown(cr, owner);
 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
 			error = secpolicy_vnode_setdac(cr, owner);
 
@@ -2524,7 +2626,7 @@ zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
 			error = secpolicy_vnode_remove(cr);
 
 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
-			error = secpolicy_vnode_chown(cr, B_FALSE);
+			error = secpolicy_vnode_chown(cr, owner);
 		}
 		if (error == 0) {
 			/*
diff --git a/module/zfs/zfs_ctldir.c b/module/zfs/zfs_ctldir.c
index 27c2c51a3..c6c719871 100644
--- a/module/zfs/zfs_ctldir.c
+++ b/module/zfs/zfs_ctldir.c
@@ -700,7 +700,7 @@ zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
 		if (err)
 			avl_add(&sdp->sd_snaps, sep);
 		else
-			err = dmu_objset_destroy(snapname);
+			err = dmu_objset_destroy(snapname, B_FALSE);
 	} else {
 		err = ENOENT;
 	}
diff --git a/module/zfs/zfs_fuid.c b/module/zfs/zfs_fuid.c
index 25b54f05a..240965218 100644
--- a/module/zfs/zfs_fuid.c
+++ b/module/zfs/zfs_fuid.c
@@ -353,6 +353,7 @@ retry:
 		rw_exit(&zfsvfs->z_fuid_lock);
 		return (retidx);
 	} else {
+		rw_exit(&zfsvfs->z_fuid_lock);
 		return (-1);
 	}
 }
diff --git a/module/zfs/zfs_ioctl.c b/module/zfs/zfs_ioctl.c
index 0b90ce789..b039414db 100644
--- a/module/zfs/zfs_ioctl.c
+++ b/module/zfs/zfs_ioctl.c
@@ -763,6 +763,20 @@ zfs_secpolicy_userspace_upgrade(zfs_cmd_t *zc, cred_t *cr)
 	return (zfs_secpolicy_setprop(zc->zc_name, ZFS_PROP_VERSION, cr));
 }
 
+static int
+zfs_secpolicy_hold(zfs_cmd_t *zc, cred_t *cr)
+{
+	return (zfs_secpolicy_write_perms(zc->zc_name,
+	    ZFS_DELEG_PERM_HOLD, cr));
+}
+
+static int
+zfs_secpolicy_release(zfs_cmd_t *zc, cred_t *cr)
+{
+	return (zfs_secpolicy_write_perms(zc->zc_name,
+	    ZFS_DELEG_PERM_RELEASE, cr));
+}
+
 /*
  * Returns the nvlist as specified by the user in the zfs_cmd_t.
  */
@@ -2470,7 +2484,7 @@ zfs_ioc_create(zfs_cmd_t *zc)
 	 */
 	if (error == 0) {
 		if ((error = zfs_set_prop_nvlist(zc->zc_name, nvprops)) != 0)
-			(void) dmu_objset_destroy(zc->zc_name);
+			(void) dmu_objset_destroy(zc->zc_name, B_FALSE);
 	}
 	nvlist_free(nvprops);
 	return (error);
@@ -2557,8 +2571,9 @@ zfs_unmount_snap(char *name, void *arg)
 
 /*
  * inputs:
- * zc_name	name of filesystem
- * zc_value	short name of snapshot
+ * zc_name		name of filesystem
+ * zc_value		short name of snapshot
+ * zc_defer_destroy	mark for deferred destroy
  *
  * outputs:	none
  */
@@ -2573,13 +2588,15 @@ zfs_ioc_destroy_snaps(zfs_cmd_t *zc)
 	    zfs_unmount_snap, zc->zc_value, DS_FIND_CHILDREN);
 	if (err)
 		return (err);
-	return (dmu_snapshots_destroy(zc->zc_name, zc->zc_value));
+	return (dmu_snapshots_destroy(zc->zc_name, zc->zc_value,
+	    zc->zc_defer_destroy));
 }
 
 /*
  * inputs:
  * zc_name		name of dataset to destroy
  * zc_objset_type	type of objset
+ * zc_defer_destroy	mark for deferred destroy
  *
  * outputs:		none
  */
@@ -2592,7 +2609,7 @@ zfs_ioc_destroy(zfs_cmd_t *zc)
 			return (err);
 	}
 
-	return (dmu_objset_destroy(zc->zc_name));
+	return (dmu_objset_destroy(zc->zc_name, zc->zc_defer_destroy));
 }
 
 /*
@@ -2712,7 +2729,6 @@ zfs_ioc_recv(zfs_cmd_t *zc)
 	file_t *fp;
 	objset_t *os;
 	dmu_recv_cookie_t drc;
-	zfsvfs_t *zfsvfs = NULL;
 	boolean_t force = (boolean_t)zc->zc_guid;
 	int error, fd;
 	offset_t off;
@@ -2744,25 +2760,12 @@ zfs_ioc_recv(zfs_cmd_t *zc)
 		return (EBADF);
 	}
 
-	if (getzfsvfs(tofs, &zfsvfs) == 0) {
-		if (!mutex_tryenter(&zfsvfs->z_online_recv_lock)) {
-			VFS_RELE(zfsvfs->z_vfs);
-			zfsvfs = NULL;
-			error = EBUSY;
-			goto out;
-		}
+	if (props && dmu_objset_open(tofs, DMU_OST_ANY,
+	    DS_MODE_USER | DS_MODE_READONLY, &os) == 0) {
 		/*
 		 * If new properties are supplied, they are to completely
 		 * replace the existing ones, so stash away the existing ones.
 		 */
-		if (props)
-			(void) dsl_prop_get_all(zfsvfs->z_os, &origprops, TRUE);
-	} else if (props && dmu_objset_open(tofs, DMU_OST_ANY,
-	    DS_MODE_USER | DS_MODE_READONLY, &os) == 0) {
-		/*
-		 * Get the props even if there was no zfsvfs (zvol or
-		 * unmounted zpl).
-		 */
 		(void) dsl_prop_get_all(os, &origprops, TRUE);
 
 		dmu_objset_close(os);
@@ -2776,7 +2779,7 @@ zfs_ioc_recv(zfs_cmd_t *zc)
 	}
 
 	error = dmu_recv_begin(tofs, tosnap, &zc->zc_begin_record,
-	    force, origin, zfsvfs != NULL, &drc);
+	    force, origin, &drc);
 	if (origin)
 		dmu_objset_close(origin);
 	if (error)
@@ -2797,25 +2800,33 @@ zfs_ioc_recv(zfs_cmd_t *zc)
 	off = fp->f_offset;
 	error = dmu_recv_stream(&drc, fp->f_vnode, &off);
 
-	if (error == 0 && zfsvfs) {
-		char *osname;
-		int mode;
+	if (error == 0) {
+		zfsvfs_t *zfsvfs = NULL;
 
-		/* online recv */
-		osname = kmem_alloc(MAXNAMELEN, KM_SLEEP);
-		error = zfs_suspend_fs(zfsvfs, osname, &mode);
-		if (error == 0) {
-			int resume_err;
+		if (getzfsvfs(tofs, &zfsvfs) == 0) {
+			/* online recv */
+			int end_err;
+			char *osname;
+			int mode;
 
-			error = dmu_recv_end(&drc);
-			resume_err = zfs_resume_fs(zfsvfs, osname, mode);
-			error = error ? error : resume_err;
+			osname = kmem_alloc(MAXNAMELEN, KM_SLEEP);
+			error = zfs_suspend_fs(zfsvfs, osname, &mode);
+			/*
+			 * If the suspend fails, then the recv_end will
+			 * likely also fail, and clean up after itself.
+			 */
+			end_err = dmu_recv_end(&drc);
+			if (error == 0) {
+				int resume_err =
+				    zfs_resume_fs(zfsvfs, osname, mode);
+				error = error ? error : resume_err;
+			}
+			error = error ? error : end_err;
+			VFS_RELE(zfsvfs->z_vfs);
+			kmem_free(osname, MAXNAMELEN);
 		} else {
-			dmu_recv_abort_cleanup(&drc);
+			error = dmu_recv_end(&drc);
 		}
-		kmem_free(osname, MAXNAMELEN);
-	} else if (error == 0) {
-		error = dmu_recv_end(&drc);
 	}
 
 	zc->zc_cookie = off - fp->f_offset;
@@ -2830,10 +2841,6 @@ zfs_ioc_recv(zfs_cmd_t *zc)
 		(void) zfs_set_prop_nvlist(tofs, origprops);
 	}
 out:
-	if (zfsvfs) {
-		mutex_exit(&zfsvfs->z_online_recv_lock);
-		VFS_RELE(zfsvfs->z_vfs);
-	}
 	nvlist_free(props);
 	nvlist_free(origprops);
 	releasef(fd);
@@ -3436,6 +3443,69 @@ zfs_ioc_smb_acl(zfs_cmd_t *zc)
 }
 
 /*
+ * inputs:
+ * zc_name	name of filesystem
+ * zc_value	short name of snap
+ * zc_string	user-supplied tag for this reference
+ * zc_cookie	recursive flag
+ *
+ * outputs:		none
+ */
+static int
+zfs_ioc_hold(zfs_cmd_t *zc)
+{
+	boolean_t recursive = zc->zc_cookie;
+
+	if (snapshot_namecheck(zc->zc_value, NULL, NULL) != 0)
+		return (EINVAL);
+
+	return (dsl_dataset_user_hold(zc->zc_name, zc->zc_value,
+	    zc->zc_string, recursive));
+}
+
+/*
+ * inputs:
+ * zc_name	name of dataset from which we're releasing a user reference
+ * zc_value	short name of snap
+ * zc_string	user-supplied tag for this reference
+ * zc_cookie	recursive flag
+ *
+ * outputs:		none
+ */
+static int
+zfs_ioc_release(zfs_cmd_t *zc)
+{
+	boolean_t recursive = zc->zc_cookie;
+
+	if (snapshot_namecheck(zc->zc_value, NULL, NULL) != 0)
+		return (EINVAL);
+
+	return (dsl_dataset_user_release(zc->zc_name, zc->zc_value,
+	    zc->zc_string, recursive));
+}
+
+/*
+ * inputs:
+ * zc_name		name of filesystem
+ *
+ * outputs:
+ * zc_nvlist_src{_size}	nvlist of snapshot holds
+ */
+static int
+zfs_ioc_get_holds(zfs_cmd_t *zc)
+{
+	nvlist_t *nvp;
+	int error;
+
+	if ((error = dsl_dataset_get_holds(zc->zc_name, &nvp)) == 0) {
+		error = put_nvlist(zc, nvp);
+		nvlist_free(nvp);
+	}
+
+	return (error);
+}
+
+/*
  * pool create, destroy, and export don't log the history as part of
  * zfsdev_ioctl, but rather zfs_ioc_pool_create, and zfs_ioc_pool_export
  * do the logging of those commands.
@@ -3515,8 +3585,8 @@ static zfs_ioc_vec_t zfs_ioc_vec[] = {
 	    B_TRUE },
 	{ zfs_ioc_dsobj_to_dsname, zfs_secpolicy_config, POOL_NAME, B_FALSE,
 	    B_FALSE },
-	{ zfs_ioc_obj_to_path, zfs_secpolicy_config, NO_NAME, B_FALSE,
-	    B_FALSE },
+	{ zfs_ioc_obj_to_path, zfs_secpolicy_config, DATASET_NAME, B_FALSE,
+	    B_TRUE },
 	{ zfs_ioc_pool_set_props, zfs_secpolicy_config,	POOL_NAME, B_TRUE,
 	    B_TRUE },
 	{ zfs_ioc_pool_get_props, zfs_secpolicy_read, POOL_NAME, B_FALSE,
@@ -3538,6 +3608,11 @@ static zfs_ioc_vec_t zfs_ioc_vec[] = {
 	    DATASET_NAME, B_FALSE, B_FALSE },
 	{ zfs_ioc_userspace_upgrade, zfs_secpolicy_userspace_upgrade,
 	    DATASET_NAME, B_FALSE, B_TRUE },
+	{ zfs_ioc_hold, zfs_secpolicy_hold, DATASET_NAME, B_TRUE, B_TRUE },
+	{ zfs_ioc_release, zfs_secpolicy_release, DATASET_NAME, B_TRUE,
+	    B_TRUE },
+	{ zfs_ioc_get_holds, zfs_secpolicy_read, DATASET_NAME, B_FALSE,
+	    B_TRUE }
 };
 
 int
diff --git a/module/zfs/zfs_vfsops.c b/module/zfs/zfs_vfsops.c
index 8a859b575..d03f92ba0 100644
--- a/module/zfs/zfs_vfsops.c
+++ b/module/zfs/zfs_vfsops.c
@@ -935,7 +935,6 @@ zfsvfs_create(const char *osname, int mode, zfsvfs_t **zvp)
 		goto out;
 
 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
-	mutex_init(&zfsvfs->z_online_recv_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
 	    offsetof(znode_t, z_link_node));
@@ -1051,7 +1050,6 @@ zfsvfs_free(zfsvfs_t *zfsvfs)
 	zfs_fuid_destroy(zfsvfs);
 
 	mutex_destroy(&zfsvfs->z_znodes_lock);
-	mutex_destroy(&zfsvfs->z_online_recv_lock);
 	mutex_destroy(&zfsvfs->z_lock);
 	list_destroy(&zfsvfs->z_all_znodes);
 	rrw_destroy(&zfsvfs->z_teardown_lock);
diff --git a/module/zfs/zfs_vnops.c b/module/zfs/zfs_vnops.c
index 88d4e52c1..8eb4665ae 100644
--- a/module/zfs/zfs_vnops.c
+++ b/module/zfs/zfs_vnops.c
@@ -208,6 +208,12 @@ zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
 	znode_t	*zp = VTOZ(vp);
 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 
+	/*
+	 * Clean up any locks held by this process on the vp.
+	 */
+	cleanlocks(vp, ddi_get_pid(), 0);
+	cleanshares(vp, ddi_get_pid());
+
 	ZFS_ENTER(zfsvfs);
 	ZFS_VERIFY_ZP(zp);
 
@@ -215,12 +221,6 @@ zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
 		atomic_dec_32(&zp->z_sync_cnt);
 
-	/*
-	 * Clean up any locks held by this process on the vp.
-	 */
-	cleanlocks(vp, ddi_get_pid(), 0);
-	cleanshares(vp, ddi_get_pid());
-
 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
 	    ZTOV(zp)->v_type == VREG &&
 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
@@ -855,6 +855,10 @@ zfs_get_done(dmu_buf_t *db, void *vzgd)
 	kmem_free(zgd, sizeof (zgd_t));
 }
 
+#ifdef DEBUG
+static int zil_fault_io = 0;
+#endif
+
 /*
  * Get data to generate a TX_WRITE intent log record.
  */
@@ -936,7 +940,21 @@ zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
 		zgd->zgd_rl = rl;
 		zgd->zgd_zilog = zfsvfs->z_log;
 		zgd->zgd_bp = &lr->lr_blkptr;
-		VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db));
+#ifdef DEBUG
+		if (zil_fault_io) {
+			error = EIO;
+			zil_fault_io = 0;
+		} else {
+			error = dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db);
+		}
+#else
+		error = dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db);
+#endif
+		if (error != 0) {
+			kmem_free(zgd, sizeof (zgd_t));
+			goto out;
+		}
+
 		ASSERT(boff == db->db_offset);
 		lr->lr_blkoff = off - boff;
 		error = dmu_sync(zio, db, &lr->lr_blkptr,
@@ -988,6 +1006,27 @@ zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
 }
 
 /*
+ * If vnode is for a device return a specfs vnode instead.
+ */
+static int
+specvp_check(vnode_t **vpp, cred_t *cr)
+{
+	int error = 0;
+
+	if (IS_DEVVP(*vpp)) {
+		struct vnode *svp;
+
+		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
+		VN_RELE(*vpp);
+		if (svp == NULL)
+			error = ENOSYS;
+		*vpp = svp;
+	}
+	return (error);
+}
+
+
+/*
  * Lookup an entry in a directory, or an extended attribute directory.
  * If it exists, return a held vnode reference for it.
  *
@@ -1017,7 +1056,46 @@ zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
 {
 	znode_t *zdp = VTOZ(dvp);
 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
-	int	error;
+	int	error = 0;
+
+	/* fast path */
+	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
+
+		if (dvp->v_type != VDIR) {
+			return (ENOTDIR);
+		} else if (zdp->z_dbuf == NULL) {
+			return (EIO);
+		}
+
+		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
+			error = zfs_fastaccesschk_execute(zdp, cr);
+			if (!error) {
+				*vpp = dvp;
+				VN_HOLD(*vpp);
+				return (0);
+			}
+			return (error);
+		} else {
+			vnode_t *tvp = dnlc_lookup(dvp, nm);
+
+			if (tvp) {
+				error = zfs_fastaccesschk_execute(zdp, cr);
+				if (error) {
+					VN_RELE(tvp);
+					return (error);
+				}
+				if (tvp == DNLC_NO_VNODE) {
+					VN_RELE(tvp);
+					return (ENOENT);
+				} else {
+					*vpp = tvp;
+					return (specvp_check(vpp, cr));
+				}
+			}
+		}
+	}
+
+	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
 
 	ZFS_ENTER(zfsvfs);
 	ZFS_VERIFY_ZP(zdp);
@@ -1082,21 +1160,8 @@ zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
 	}
 
 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
-	if (error == 0) {
-		/*
-		 * Convert device special files
-		 */
-		if (IS_DEVVP(*vpp)) {
-			vnode_t	*svp;
-
-			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
-			VN_RELE(*vpp);
-			if (svp == NULL)
-				error = ENOSYS;
-			else
-				*vpp = svp;
-		}
-	}
+	if (error == 0)
+		error = specvp_check(vpp, cr);
 
 	ZFS_EXIT(zfsvfs);
 	return (error);
@@ -1235,6 +1300,7 @@ top:
 		    &acl_ids)) != 0)
 			goto out;
 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
+			zfs_acl_ids_free(&acl_ids);
 			error = EDQUOT;
 			goto out;
 		}
@@ -1332,19 +1398,7 @@ out:
 			VN_RELE(ZTOV(zp));
 	} else {
 		*vpp = ZTOV(zp);
-		/*
-		 * If vnode is for a device return a specfs vnode instead.
-		 */
-		if (IS_DEVVP(*vpp)) {
-			struct vnode *svp;
-
-			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
-			VN_RELE(*vpp);
-			if (svp == NULL) {
-				error = ENOSYS;
-			}
-			*vpp = svp;
-		}
+		error = specvp_check(vpp, cr);
 	}
 
 	ZFS_EXIT(zfsvfs);
@@ -1653,6 +1707,7 @@ top:
 		return (error);
 	}
 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
+		zfs_acl_ids_free(&acl_ids);
 		zfs_dirent_unlock(dl);
 		ZFS_EXIT(zfsvfs);
 		return (EDQUOT);
@@ -2456,6 +2511,7 @@ zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
 top:
 	attrzp = NULL;
 
+	/* Can this be moved to before the top label? */
 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
 		ZFS_EXIT(zfsvfs);
 		return (EROFS);
@@ -2765,6 +2821,8 @@ top:
 		zp->z_phys->zp_mode = new_mode;
 		err = zfs_aclset_common(zp, aclp, cr, tx);
 		ASSERT3U(err, ==, 0);
+		zp->z_acl_cached = aclp;
+		aclp = NULL;
 		mutex_exit(&zp->z_acl_lock);
 	}
 
@@ -2856,10 +2914,8 @@ out:
 	if (attrzp)
 		VN_RELE(ZTOV(attrzp));
 
-	if (aclp) {
+	if (aclp)
 		zfs_acl_free(aclp);
-		aclp = NULL;
-	}
 
 	if (fuidp) {
 		zfs_fuid_info_free(fuidp);
@@ -3724,8 +3780,8 @@ top:
 	if (err == 0) {
 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
-		dmu_tx_commit(tx);
 	}
+	dmu_tx_commit(tx);
 
 out:
 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
diff --git a/module/zfs/zfs_znode.c b/module/zfs/zfs_znode.c
index 3f866bde5..ee27195a4 100644
--- a/module/zfs/zfs_znode.c
+++ b/module/zfs/zfs_znode.c
@@ -133,6 +133,7 @@ zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
 
 	zp->z_dbuf = NULL;
 	zp->z_dirlocks = NULL;
+	zp->z_acl_cached = NULL;
 	return (0);
 }
 
@@ -155,6 +156,7 @@ zfs_znode_cache_destructor(void *buf, void *arg)
 
 	ASSERT(zp->z_dbuf == NULL);
 	ASSERT(zp->z_dirlocks == NULL);
+	ASSERT(zp->z_acl_cached == NULL);
 }
 
 #ifdef	ZNODE_STATS
@@ -199,6 +201,18 @@ zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
 	nzp->z_phys = ozp->z_phys;
 	nzp->z_dbuf = ozp->z_dbuf;
 
+	/*
+	 * Release any cached ACL, since it *may* have
+	 * zfs_acl_node_t's that directly references an
+	 * embedded ACL in the zp_acl of the old znode_phys_t
+	 *
+	 * It will be recached the next time the ACL is needed.
+	 */
+	if (ozp->z_acl_cached) {
+		zfs_acl_free(ozp->z_acl_cached);
+		ozp->z_acl_cached = NULL;
+	}
+
 	/* Update back pointers. */
 	(void) dmu_buf_update_user(nzp->z_dbuf, ozp, nzp, &nzp->z_phys,
 	    znode_evict_error);
@@ -1081,6 +1095,11 @@ zfs_znode_free(znode_t *zp)
 	list_remove(&zfsvfs->z_all_znodes, zp);
 	mutex_exit(&zfsvfs->z_znodes_lock);
 
+	if (zp->z_acl_cached) {
+		zfs_acl_free(zp->z_acl_cached);
+		zp->z_acl_cached = NULL;
+	}
+
 	kmem_cache_free(znode_cache, zp);
 
 	VFS_RELE(zfsvfs->z_vfs);
diff --git a/module/zfs/zil.c b/module/zfs/zil.c
index 53d9d9bf7..db3822f5a 100644
--- a/module/zfs/zil.c
+++ b/module/zfs/zil.c
@@ -714,14 +714,15 @@ zil_lwb_write_done(zio_t *zio)
 	lwb->lwb_buf = NULL;
 	if (zio->io_error)
 		zilog->zl_log_error = B_TRUE;
-	mutex_exit(&zilog->zl_lock);
 
 	/*
 	 * Now that we've written this log block, we have a stable pointer
 	 * to the next block in the chain, so it's OK to let the txg in
-	 * which we allocated the next block sync.
+	 * which we allocated the next block sync. We still have the
+	 * zl_lock to ensure zil_sync doesn't kmem free the lwb.
 	 */
 	txg_rele_to_sync(&lwb->lwb_txgh);
+	mutex_exit(&zilog->zl_lock);
 }
 
 /*
@@ -925,6 +926,10 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
 			}
 			error = zilog->zl_get_data(
 			    itx->itx_private, lr, dbuf, lwb->lwb_zio);
+			if (error == EIO) {
+				txg_wait_synced(zilog->zl_dmu_pool, txg);
+				return (lwb);
+			}
 			if (error) {
 				ASSERT(error == ENOENT || error == EEXIST ||
 				    error == EALREADY);