Implement large_dnode pool feature

Justification
-------------

This feature adds support for variable length dnodes. Our motivation is
to eliminate the overhead associated with using spill blocks.  Spill
blocks are used to store system attribute data (i.e. file metadata) that
does not fit in the dnode's bonus buffer. By allowing a larger bonus
buffer area the use of a spill block can be avoided.  Spill blocks
potentially incur an additional read I/O for every dnode in a dnode
block. As a worst case example, reading 32 dnodes from a 16k dnode block
and all of the spill blocks could issue 33 separate reads. Now suppose
those dnodes have size 1024 and therefore don't need spill blocks.  Then
the worst case number of blocks read is reduced to from 33 to two--one
per dnode block. In practice spill blocks may tend to be co-located on
disk with the dnode blocks so the reduction in I/O would not be this
drastic. In a badly fragmented pool, however, the improvement could be
significant.

ZFS-on-Linux systems that make heavy use of extended attributes would
benefit from this feature. In particular, ZFS-on-Linux supports the
xattr=sa dataset property which allows file extended attribute data
to be stored in the dnode bonus buffer as an alternative to the
traditional directory-based format. Workloads such as SELinux and the
Lustre distributed filesystem often store enough xattr data to force
spill bocks when xattr=sa is in effect. Large dnodes may therefore
provide a performance benefit to such systems.

Other use cases that may benefit from this feature include files with
large ACLs and symbolic links with long target names. Furthermore,
this feature may be desirable on other platforms in case future
applications or features are developed that could make use of a
larger bonus buffer area.

Implementation
--------------

The size of a dnode may be a multiple of 512 bytes up to the size of
a dnode block (currently 16384 bytes). A dn_extra_slots field was
added to the current on-disk dnode_phys_t structure to describe the
size of the physical dnode on disk. The 8 bits for this field were
taken from the zero filled dn_pad2 field. The field represents how
many "extra" dnode_phys_t slots a dnode consumes in its dnode block.
This convention results in a value of 0 for 512 byte dnodes which
preserves on-disk format compatibility with older software.

Similarly, the in-memory dnode_t structure has a new dn_num_slots field
to represent the total number of dnode_phys_t slots consumed on disk.
Thus dn->dn_num_slots is 1 greater than the corresponding
dnp->dn_extra_slots. This difference in convention was adopted
because, unlike on-disk structures, backward compatibility is not a
concern for in-memory objects, so we used a more natural way to
represent size for a dnode_t.

The default size for newly created dnodes is determined by the value of
a new "dnodesize" dataset property. By default the property is set to
"legacy" which is compatible with older software. Setting the property
to "auto" will allow the filesystem to choose the most suitable dnode
size. Currently this just sets the default dnode size to 1k, but future
code improvements could dynamically choose a size based on observed
workload patterns. Dnodes of varying sizes can coexist within the same
dataset and even within the same dnode block. For example, to enable
automatically-sized dnodes, run

 # zfs set dnodesize=auto tank/fish

The user can also specify literal values for the dnodesize property.
These are currently limited to powers of two from 1k to 16k. The
power-of-2 limitation is only for simplicity of the user interface.
Internally the implementation can handle any multiple of 512 up to 16k,
and consumers of the DMU API can specify any legal dnode value.

The size of a new dnode is determined at object allocation time and
stored as a new field in the znode in-memory structure. New DMU
interfaces are added to allow the consumer to specify the dnode size
that a newly allocated object should use. Existing interfaces are
unchanged to avoid having to update every call site and to preserve
compatibility with external consumers such as Lustre. The new
interfaces names are given below. The versions of these functions that
don't take a dnodesize parameter now just call the _dnsize() versions
with a dnodesize of 0, which means use the legacy dnode size.

New DMU interfaces:
  dmu_object_alloc_dnsize()
  dmu_object_claim_dnsize()
  dmu_object_reclaim_dnsize()

New ZAP interfaces:
  zap_create_dnsize()
  zap_create_norm_dnsize()
  zap_create_flags_dnsize()
  zap_create_claim_norm_dnsize()
  zap_create_link_dnsize()

The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The
spa_maxdnodesize() function should be used to determine the maximum
bonus length for a pool.

These are a few noteworthy changes to key functions:

* The prototype for dnode_hold_impl() now takes a "slots" parameter.
  When the DNODE_MUST_BE_FREE flag is set, this parameter is used to
  ensure the hole at the specified object offset is large enough to
  hold the dnode being created. The slots parameter is also used
  to ensure a dnode does not span multiple dnode blocks. In both of
  these cases, if a failure occurs, ENOSPC is returned. Keep in mind,
  these failure cases are only possible when using DNODE_MUST_BE_FREE.

  If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0.
  dnode_hold_impl() will check if the requested dnode is already
  consumed as an extra dnode slot by an large dnode, in which case
  it returns ENOENT.

* The function dmu_object_alloc() advances to the next dnode block
  if dnode_hold_impl() returns an error for a requested object.
  This is because the beginning of the next dnode block is the only
  location it can safely assume to either be a hole or a valid
  starting point for a dnode.

* dnode_next_offset_level() and other functions that iterate
  through dnode blocks may no longer use a simple array indexing
  scheme. These now use the current dnode's dn_num_slots field to
  advance to the next dnode in the block. This is to ensure we
  properly skip the current dnode's bonus area and don't interpret it
  as a valid dnode.

zdb
---
The zdb command was updated to display a dnode's size under the
"dnsize" column when the object is dumped.

For ZIL create log records, zdb will now display the slot count for
the object.

ztest
-----
Ztest chooses a random dnodesize for every newly created object. The
random distribution is more heavily weighted toward small dnodes to
better simulate real-world datasets.

Unused bonus buffer space is filled with non-zero values computed from
the object number, dataset id, offset, and generation number.  This
helps ensure that the dnode traversal code properly skips the interior
regions of large dnodes, and that these interior regions are not
overwritten by data belonging to other dnodes. A new test visits each
object in a dataset. It verifies that the actual dnode size matches what
was stored in the ztest block tag when it was created. It also verifies
that the unused bonus buffer space is filled with the expected data
patterns.

ZFS Test Suite
--------------
Added six new large dnode-specific tests, and integrated the dnodesize
property into existing tests for zfs allow and send/recv.

Send/Receive
------------
ZFS send streams for datasets containing large dnodes cannot be received
on pools that don't support the large_dnode feature. A send stream with
large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be
unrecognized by an incompatible receiving pool so that the zfs receive
will fail gracefully.

While not implemented here, it may be possible to generate a
backward-compatible send stream from a dataset containing large
dnodes. The implementation may be tricky, however, because the send
object record for a large dnode would need to be resized to a 512
byte dnode, possibly kicking in a spill block in the process. This
means we would need to construct a new SA layout and possibly
register it in the SA layout object. The SA layout is normally just
sent as an ordinary object record. But if we are constructing new
layouts while generating the send stream we'd have to build the SA
layout object dynamically and send it at the end of the stream.

For sending and receiving between pools that do support large dnodes,
the drr_object send record type is extended with a new field to store
the dnode slot count. This field was repurposed from unused padding
in the structure.

ZIL Replay
----------
The dnode slot count is stored in the uppermost 8 bits of the lr_foid
field. The bits were unused as the object id is currently capped at
48 bits.

Resizing Dnodes
---------------
It should be possible to resize a dnode when it is dirtied if the
current dnodesize dataset property differs from the dnode's size, but
this functionality is not currently implemented. Clearly a dnode can
only grow if there are sufficient contiguous unused slots in the
dnode block, but it should always be possible to shrink a dnode.
Growing dnodes may be useful to reduce fragmentation in a pool with
many spill blocks in use. Shrinking dnodes may be useful to allow
sending a dataset to a pool that doesn't support the large_dnode
feature.

Feature Reference Counting
--------------------------
The reference count for the large_dnode pool feature tracks the
number of datasets that have ever contained a dnode of size larger
than 512 bytes. The first time a large dnode is created in a dataset
the dataset is converted to an extensible dataset. This is a one-way
operation and the only way to decrement the feature count is to
destroy the dataset, even if the dataset no longer contains any large
dnodes. The complexity of reference counting on a per-dnode basis was
too high, so we chose to track it on a per-dataset basis similarly to
the large_block feature.

Signed-off-by: Ned Bass <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Closes #3542

25 files changed, 679 insertions, 154 deletions

diff --git a/module/zcommon/zfs_prop.c b/module/zcommon/zfs_prop.c
index 76d564b43..1dbeab084 100644
--- a/module/zcommon/zfs_prop.c
+++ b/module/zcommon/zfs_prop.c
@@ -211,6 +211,17 @@ zfs_prop_init(void)
 		{ NULL }
 	};
 
+	static zprop_index_t dnsize_table[] = {
+		{ "legacy",	ZFS_DNSIZE_LEGACY },
+		{ "auto",	ZFS_DNSIZE_AUTO },
+		{ "1k",		ZFS_DNSIZE_1K },
+		{ "2k",		ZFS_DNSIZE_2K },
+		{ "4k",		ZFS_DNSIZE_4K },
+		{ "8k",		ZFS_DNSIZE_8K },
+		{ "16k",	ZFS_DNSIZE_16K },
+		{ NULL }
+	};
+
 	static zprop_index_t redundant_metadata_table[] = {
 		{ "all",	ZFS_REDUNDANT_METADATA_ALL },
 		{ "most",	ZFS_REDUNDANT_METADATA_MOST },
@@ -271,6 +282,9 @@ zfs_prop_init(void)
 	zprop_register_index(ZFS_PROP_XATTR, "xattr", ZFS_XATTR_DIR,
 	    PROP_INHERIT, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_SNAPSHOT,
 	    "on | off | dir | sa", "XATTR", xattr_table);
+	zprop_register_index(ZFS_PROP_DNODESIZE, "dnodesize",
+	    ZFS_DNSIZE_LEGACY, PROP_INHERIT, ZFS_TYPE_FILESYSTEM,
+	    "legacy | auto | 1k | 2k | 4k | 8k | 16k", "DNSIZE", dnsize_table);
 
 	/* inherit index (boolean) properties */
 	zprop_register_index(ZFS_PROP_ATIME, "atime", 1, PROP_INHERIT,
diff --git a/module/zcommon/zpool_prop.c b/module/zcommon/zpool_prop.c
index 910c56dcc..4a5836e5b 100644
--- a/module/zcommon/zpool_prop.c
+++ b/module/zcommon/zpool_prop.c
@@ -135,6 +135,8 @@ zpool_prop_init(void)
 	    PROP_TYPE_NUMBER, PROP_READONLY, ZFS_TYPE_POOL, "MAXBLOCKSIZE");
 	zprop_register_hidden(ZPOOL_PROP_TNAME, "tname", PROP_TYPE_STRING,
 	    PROP_ONETIME, ZFS_TYPE_POOL, "TNAME");
+	zprop_register_hidden(ZPOOL_PROP_MAXDNODESIZE, "maxdnodesize",
+	    PROP_TYPE_NUMBER, PROP_READONLY, ZFS_TYPE_POOL, "MAXDNODESIZE");
 }
 
 /*
diff --git a/module/zfs/dbuf.c b/module/zfs/dbuf.c
index 126748994..4bbbd0525 100644
--- a/module/zfs/dbuf.c
+++ b/module/zfs/dbuf.c
@@ -478,7 +478,6 @@ dbuf_verify(dmu_buf_impl_t *db)
 		ASSERT3U(db->db.db_offset, ==, DMU_BONUS_BLKID);
 	} else if (db->db_blkid == DMU_SPILL_BLKID) {
 		ASSERT(dn != NULL);
-		ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen);
 		ASSERT0(db->db.db_offset);
 	} else {
 		ASSERT3U(db->db.db_offset, ==, db->db_blkid * db->db.db_size);
@@ -730,13 +729,18 @@ dbuf_read_impl(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags)
 	ASSERT(db->db_buf == NULL);
 
 	if (db->db_blkid == DMU_BONUS_BLKID) {
+		/*
+		 * The bonus length stored in the dnode may be less than
+		 * the maximum available space in the bonus buffer.
+		 */
 		int bonuslen = MIN(dn->dn_bonuslen, dn->dn_phys->dn_bonuslen);
+		int max_bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots);
 
 		ASSERT3U(bonuslen, <=, db->db.db_size);
-		db->db.db_data = zio_buf_alloc(DN_MAX_BONUSLEN);
-		arc_space_consume(DN_MAX_BONUSLEN, ARC_SPACE_OTHER);
-		if (bonuslen < DN_MAX_BONUSLEN)
-			bzero(db->db.db_data, DN_MAX_BONUSLEN);
+		db->db.db_data = zio_buf_alloc(max_bonuslen);
+		arc_space_consume(max_bonuslen, ARC_SPACE_OTHER);
+		if (bonuslen < max_bonuslen)
+			bzero(db->db.db_data, max_bonuslen);
 		if (bonuslen)
 			bcopy(DN_BONUS(dn->dn_phys), db->db.db_data, bonuslen);
 		DB_DNODE_EXIT(db);
@@ -962,9 +966,11 @@ dbuf_fix_old_data(dmu_buf_impl_t *db, uint64_t txg)
 	ASSERT(dr->dr_txg >= txg - 2);
 	if (db->db_blkid == DMU_BONUS_BLKID) {
 		/* Note that the data bufs here are zio_bufs */
-		dr->dt.dl.dr_data = zio_buf_alloc(DN_MAX_BONUSLEN);
-		arc_space_consume(DN_MAX_BONUSLEN, ARC_SPACE_OTHER);
-		bcopy(db->db.db_data, dr->dt.dl.dr_data, DN_MAX_BONUSLEN);
+		dnode_t *dn = DB_DNODE(db);
+		int bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots);
+		dr->dt.dl.dr_data = zio_buf_alloc(bonuslen);
+		arc_space_consume(bonuslen, ARC_SPACE_OTHER);
+		bcopy(db->db.db_data, dr->dt.dl.dr_data, bonuslen);
 	} else if (refcount_count(&db->db_holds) > db->db_dirtycnt) {
 		int size = db->db.db_size;
 		arc_buf_contents_t type = DBUF_GET_BUFC_TYPE(db);
@@ -1858,8 +1864,10 @@ dbuf_clear(dmu_buf_impl_t *db)
 	if (db->db_state == DB_CACHED) {
 		ASSERT(db->db.db_data != NULL);
 		if (db->db_blkid == DMU_BONUS_BLKID) {
-			zio_buf_free(db->db.db_data, DN_MAX_BONUSLEN);
-			arc_space_return(DN_MAX_BONUSLEN, ARC_SPACE_OTHER);
+			int slots = DB_DNODE(db)->dn_num_slots;
+			int bonuslen = DN_SLOTS_TO_BONUSLEN(slots);
+			zio_buf_free(db->db.db_data, bonuslen);
+			arc_space_return(bonuslen, ARC_SPACE_OTHER);
 		}
 		db->db.db_data = NULL;
 		db->db_state = DB_UNCACHED;
@@ -1929,7 +1937,7 @@ dbuf_findbp(dnode_t *dn, int level, uint64_t blkid, int fail_sparse,
 		mutex_enter(&dn->dn_mtx);
 		if (dn->dn_have_spill &&
 		    (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR))
-			*bpp = &dn->dn_phys->dn_spill;
+			*bpp = DN_SPILL_BLKPTR(dn->dn_phys);
 		else
 			*bpp = NULL;
 		dbuf_add_ref(dn->dn_dbuf, NULL);
@@ -2018,7 +2026,7 @@ dbuf_create(dnode_t *dn, uint8_t level, uint64_t blkid,
 
 	if (blkid == DMU_BONUS_BLKID) {
 		ASSERT3P(parent, ==, dn->dn_dbuf);
-		db->db.db_size = DN_MAX_BONUSLEN -
+		db->db.db_size = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots) -
 		    (dn->dn_nblkptr-1) * sizeof (blkptr_t);
 		ASSERT3U(db->db.db_size, >=, dn->dn_bonuslen);
 		db->db.db_offset = DMU_BONUS_BLKID;
@@ -2810,7 +2818,7 @@ dbuf_check_blkptr(dnode_t *dn, dmu_buf_impl_t *db)
 		return;
 
 	if (db->db_blkid == DMU_SPILL_BLKID) {
-		db->db_blkptr = &dn->dn_phys->dn_spill;
+		db->db_blkptr = DN_SPILL_BLKPTR(dn->dn_phys);
 		BP_ZERO(db->db_blkptr);
 		return;
 	}
@@ -2950,13 +2958,16 @@ dbuf_sync_leaf(dbuf_dirty_record_t *dr, dmu_tx_t *tx)
 
 		ASSERT(*datap != NULL);
 		ASSERT0(db->db_level);
-		ASSERT3U(dn->dn_phys->dn_bonuslen, <=, DN_MAX_BONUSLEN);
+		ASSERT3U(dn->dn_phys->dn_bonuslen, <=,
+		    DN_SLOTS_TO_BONUSLEN(dn->dn_phys->dn_extra_slots + 1));
 		bcopy(*datap, DN_BONUS(dn->dn_phys), dn->dn_phys->dn_bonuslen);
 		DB_DNODE_EXIT(db);
 
 		if (*datap != db->db.db_data) {
-			zio_buf_free(*datap, DN_MAX_BONUSLEN);
-			arc_space_return(DN_MAX_BONUSLEN, ARC_SPACE_OTHER);
+			int slots = DB_DNODE(db)->dn_num_slots;
+			int bonuslen = DN_SLOTS_TO_BONUSLEN(slots);
+			zio_buf_free(*datap, bonuslen);
+			arc_space_return(bonuslen, ARC_SPACE_OTHER);
 		}
 		db->db_data_pending = NULL;
 		drp = &db->db_last_dirty;
@@ -3107,7 +3118,7 @@ dbuf_write_ready(zio_t *zio, arc_buf_t *buf, void *vdb)
 	if (db->db_blkid == DMU_SPILL_BLKID) {
 		ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR);
 		ASSERT(!(BP_IS_HOLE(bp)) &&
-		    db->db_blkptr == &dn->dn_phys->dn_spill);
+		    db->db_blkptr == DN_SPILL_BLKPTR(dn->dn_phys));
 	}
 #endif
 
@@ -3119,11 +3130,16 @@ dbuf_write_ready(zio_t *zio, arc_buf_t *buf, void *vdb)
 		mutex_exit(&dn->dn_mtx);
 
 		if (dn->dn_type == DMU_OT_DNODE) {
-			dnode_phys_t *dnp = db->db.db_data;
-			for (i = db->db.db_size >> DNODE_SHIFT; i > 0;
-			    i--, dnp++) {
-				if (dnp->dn_type != DMU_OT_NONE)
+			i = 0;
+			while (i < db->db.db_size) {
+				dnode_phys_t *dnp = db->db.db_data + i;
+
+				i += DNODE_MIN_SIZE;
+				if (dnp->dn_type != DMU_OT_NONE) {
 					fill++;
+					i += dnp->dn_extra_slots *
+					    DNODE_MIN_SIZE;
+				}
 			}
 		} else {
 			if (BP_IS_HOLE(bp)) {
@@ -3270,7 +3286,7 @@ dbuf_write_done(zio_t *zio, arc_buf_t *buf, void *vdb)
 		dn = DB_DNODE(db);
 		ASSERT(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR);
 		ASSERT(!(BP_IS_HOLE(db->db_blkptr)) &&
-		    db->db_blkptr == &dn->dn_phys->dn_spill);
+		    db->db_blkptr == DN_SPILL_BLKPTR(dn->dn_phys));
 		DB_DNODE_EXIT(db);
 	}
 #endif
diff --git a/module/zfs/dmu.c b/module/zfs/dmu.c
index a423264c6..e1dfb41ff 100644
--- a/module/zfs/dmu.c
+++ b/module/zfs/dmu.c
@@ -180,7 +180,7 @@ dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
 int
 dmu_bonus_max(void)
 {
-	return (DN_MAX_BONUSLEN);
+	return (DN_OLD_MAX_BONUSLEN);
 }
 
 int
@@ -1853,6 +1853,7 @@ __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
 	doi->doi_type = dn->dn_type;
 	doi->doi_bonus_type = dn->dn_bonustype;
 	doi->doi_bonus_size = dn->dn_bonuslen;
+	doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT;
 	doi->doi_indirection = dn->dn_nlevels;
 	doi->doi_checksum = dn->dn_checksum;
 	doi->doi_compress = dn->dn_compress;
@@ -1924,9 +1925,21 @@ dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize,
 	dn = DB_DNODE(db);
 
 	*blksize = dn->dn_datablksz;
-	/* add 1 for dnode space */
+	/* add in number of slots used for the dnode itself */
 	*nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
-	    SPA_MINBLOCKSHIFT) + 1;
+	    SPA_MINBLOCKSHIFT) + dn->dn_num_slots;
+	DB_DNODE_EXIT(db);
+}
+
+void
+dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize)
+{
+	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
+	dnode_t *dn;
+
+	DB_DNODE_ENTER(db);
+	dn = DB_DNODE(db);
+	*dnsize = dn->dn_num_slots << DNODE_SHIFT;
 	DB_DNODE_EXIT(db);
 }
 
@@ -2020,6 +2033,7 @@ EXPORT_SYMBOL(dmu_object_info);
 EXPORT_SYMBOL(dmu_object_info_from_dnode);
 EXPORT_SYMBOL(dmu_object_info_from_db);
 EXPORT_SYMBOL(dmu_object_size_from_db);
+EXPORT_SYMBOL(dmu_object_dnsize_from_db);
 EXPORT_SYMBOL(dmu_object_set_blocksize);
 EXPORT_SYMBOL(dmu_object_set_checksum);
 EXPORT_SYMBOL(dmu_object_set_compress);
diff --git a/module/zfs/dmu_object.c b/module/zfs/dmu_object.c
index a5a53418b..e54043fc3 100644
--- a/module/zfs/dmu_object.c
+++ b/module/zfs/dmu_object.c
@@ -30,28 +30,55 @@
 #include <sys/dnode.h>
 #include <sys/zap.h>
 #include <sys/zfeature.h>
+#include <sys/dsl_dataset.h>
 
 uint64_t
 dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize,
     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 {
+	return dmu_object_alloc_dnsize(os, ot, blocksize, bonustype, bonuslen,
+	    0, tx);
+}
+
+uint64_t
+dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot, int blocksize,
+    dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
+{
 	uint64_t object;
 	uint64_t L1_dnode_count = DNODES_PER_BLOCK <<
 	    (DMU_META_DNODE(os)->dn_indblkshift - SPA_BLKPTRSHIFT);
 	dnode_t *dn = NULL;
+	int dn_slots = dnodesize >> DNODE_SHIFT;
+	boolean_t restarted = B_FALSE;
+
+	if (dn_slots == 0) {
+		dn_slots = DNODE_MIN_SLOTS;
+	} else {
+		ASSERT3S(dn_slots, >=, DNODE_MIN_SLOTS);
+		ASSERT3S(dn_slots, <=, DNODE_MAX_SLOTS);
+	}
 
 	mutex_enter(&os->os_obj_lock);
 	for (;;) {
 		object = os->os_obj_next;
 		/*
 		 * Each time we polish off a L1 bp worth of dnodes (2^12
-		 * objects), move to another L1 bp that's still reasonably
-		 * sparse (at most 1/4 full). Look from the beginning at most
-		 * once per txg, but after that keep looking from here.
+		 * objects), move to another L1 bp that's still
+		 * reasonably sparse (at most 1/4 full). Look from the
+		 * beginning at most once per txg. If we still can't
+		 * allocate from that L1 block, search for an empty L0
+		 * block, which will quickly skip to the end of the
+		 * metadnode if the no nearby L0 blocks are empty. This
+		 * fallback avoids a pathology where full dnode blocks
+		 * containing large dnodes appear sparse because they
+		 * have a low blk_fill, leading to many failed
+		 * allocation attempts. In the long term a better
+		 * mechanism to search for sparse metadnode regions,
+		 * such as spacemaps, could be implemented.
+		 *
 		 * os_scan_dnodes is set during txg sync if enough objects
 		 * have been freed since the previous rescan to justify
-		 * backfilling again. If we can't find a suitable block, just
-		 * keep going from here.
+		 * backfilling again.
 		 *
 		 * Note that dmu_traverse depends on the behavior that we use
 		 * multiple blocks of the dnode object before going back to
@@ -59,9 +86,10 @@ dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize,
 		 * that property or find another solution to the issues
 		 * described in traverse_visitbp.
 		 */
-
 		if (P2PHASE(object, L1_dnode_count) == 0) {
 			uint64_t offset;
+			uint64_t blkfill;
+			int minlvl;
 			int error;
 			if (os->os_rescan_dnodes) {
 				offset = 0;
@@ -69,13 +97,15 @@ dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize,
 			} else {
 				offset = object << DNODE_SHIFT;
 			}
+			blkfill = restarted ? 1 : DNODES_PER_BLOCK >> 2;
+			minlvl = restarted ? 1 : 2;
+			restarted = B_TRUE;
 			error = dnode_next_offset(DMU_META_DNODE(os),
-			    DNODE_FIND_HOLE,
-			    &offset, 2, DNODES_PER_BLOCK >> 2, 0);
+			    DNODE_FIND_HOLE, &offset, minlvl, blkfill, 0);
 			if (error == 0)
 				object = offset >> DNODE_SHIFT;
 		}
-		os->os_obj_next = ++object;
+		os->os_obj_next = object + dn_slots;
 
 		/*
 		 * XXX We should check for an i/o error here and return
@@ -83,16 +113,22 @@ dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize,
 		 * dmu_tx_assign(), but there is currently no mechanism
 		 * to do so.
 		 */
-		(void) dnode_hold_impl(os, object, DNODE_MUST_BE_FREE,
+		(void) dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, dn_slots,
 		    FTAG, &dn);
 		if (dn)
 			break;
 
 		if (dmu_object_next(os, &object, B_TRUE, 0) == 0)
-			os->os_obj_next = object - 1;
+			os->os_obj_next = object;
+		else
+			/*
+			 * Skip to next known valid starting point for a dnode.
+			 */
+			os->os_obj_next = P2ROUNDUP(object + 1,
+			    DNODES_PER_BLOCK);
 	}
 
-	dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, tx);
+	dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, dn_slots, tx);
 	dnode_rele(dn, FTAG);
 
 	mutex_exit(&os->os_obj_lock);
@@ -105,16 +141,33 @@ int
 dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 {
+	return (dmu_object_claim_dnsize(os, object, ot, blocksize, bonustype,
+	    bonuslen, 0, tx));
+}
+
+int
+dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
+    int blocksize, dmu_object_type_t bonustype, int bonuslen,
+    int dnodesize, dmu_tx_t *tx)
+{
 	dnode_t *dn;
+	int dn_slots = dnodesize >> DNODE_SHIFT;
 	int err;
 
+	if (dn_slots == 0)
+		dn_slots = DNODE_MIN_SLOTS;
+	ASSERT3S(dn_slots, >=, DNODE_MIN_SLOTS);
+	ASSERT3S(dn_slots, <=, DNODE_MAX_SLOTS);
+
 	if (object == DMU_META_DNODE_OBJECT && !dmu_tx_private_ok(tx))
 		return (SET_ERROR(EBADF));
 
-	err = dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, FTAG, &dn);
+	err = dnode_hold_impl(os, object, DNODE_MUST_BE_FREE, dn_slots,
+	    FTAG, &dn);
 	if (err)
 		return (err);
-	dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, tx);
+
+	dnode_allocate(dn, ot, blocksize, 0, bonustype, bonuslen, dn_slots, tx);
 	dnode_rele(dn, FTAG);
 
 	dmu_tx_add_new_object(tx, os, object);
@@ -125,23 +178,34 @@ int
 dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
     int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 {
+	return (dmu_object_reclaim_dnsize(os, object, ot, blocksize, bonustype,
+	    bonuslen, 0, tx));
+}
+
+int
+dmu_object_reclaim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
+    int blocksize, dmu_object_type_t bonustype, int bonuslen, int dnodesize,
+    dmu_tx_t *tx)
+{
 	dnode_t *dn;
+	int dn_slots = dnodesize >> DNODE_SHIFT;
 	int err;
 
 	if (object == DMU_META_DNODE_OBJECT)
 		return (SET_ERROR(EBADF));
 
-	err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED,
+	err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 0,
 	    FTAG, &dn);
 	if (err)
 		return (err);
 
-	dnode_reallocate(dn, ot, blocksize, bonustype, bonuslen, tx);
+	dnode_reallocate(dn, ot, blocksize, bonustype, bonuslen, dn_slots, tx);
 
 	dnode_rele(dn, FTAG);
 	return (err);
 }
 
+
 int
 dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx)
 {
@@ -150,7 +214,7 @@ dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx)
 
 	ASSERT(object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx));
 
-	err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED,
+	err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 0,
 	    FTAG, &dn);
 	if (err)
 		return (err);
@@ -171,9 +235,30 @@ dmu_object_free(objset_t *os, uint64_t object, dmu_tx_t *tx)
 int
 dmu_object_next(objset_t *os, uint64_t *objectp, boolean_t hole, uint64_t txg)
 {
-	uint64_t offset = (*objectp + 1) << DNODE_SHIFT;
+	uint64_t offset;
+	dmu_object_info_t doi;
+	struct dsl_dataset *ds = os->os_dsl_dataset;
+	int dnodesize;
 	int error;
 
+	/*
+	 * Avoid expensive dnode hold if this dataset doesn't use large dnodes.
+	 */
+	if (ds && ds->ds_feature_inuse[SPA_FEATURE_LARGE_DNODE]) {
+		error = dmu_object_info(os, *objectp, &doi);
+		if (error && !(error == EINVAL && *objectp == 0))
+			return (SET_ERROR(error));
+		else
+			dnodesize = doi.doi_dnodesize;
+	} else {
+		dnodesize = DNODE_MIN_SIZE;
+	}
+
+	if (*objectp == 0)
+		offset = 1 << DNODE_SHIFT;
+	else
+		offset = (*objectp << DNODE_SHIFT) + dnodesize;
+
 	error = dnode_next_offset(DMU_META_DNODE(os),
 	    (hole ? DNODE_FIND_HOLE : 0), &offset, 0, DNODES_PER_BLOCK, txg);
 
@@ -235,8 +320,11 @@ dmu_object_free_zapified(objset_t *mos, uint64_t object, dmu_tx_t *tx)
 
 #if defined(_KERNEL) && defined(HAVE_SPL)
 EXPORT_SYMBOL(dmu_object_alloc);
+EXPORT_SYMBOL(dmu_object_alloc_dnsize);
 EXPORT_SYMBOL(dmu_object_claim);
+EXPORT_SYMBOL(dmu_object_claim_dnsize);
 EXPORT_SYMBOL(dmu_object_reclaim);
+EXPORT_SYMBOL(dmu_object_reclaim_dnsize);
 EXPORT_SYMBOL(dmu_object_free);
 EXPORT_SYMBOL(dmu_object_next);
 EXPORT_SYMBOL(dmu_object_zapify);
diff --git a/module/zfs/dmu_objset.c b/module/zfs/dmu_objset.c
index 03b30dd3b..cdc897726 100644
--- a/module/zfs/dmu_objset.c
+++ b/module/zfs/dmu_objset.c
@@ -138,6 +138,12 @@ dmu_objset_id(objset_t *os)
 	return (ds ? ds->ds_object : 0);
 }
 
+uint64_t
+dmu_objset_dnodesize(objset_t *os)
+{
+	return (os->os_dnodesize);
+}
+
 zfs_sync_type_t
 dmu_objset_syncprop(objset_t *os)
 {
@@ -268,6 +274,34 @@ redundant_metadata_changed_cb(void *arg, uint64_t newval)
 }
 
 static void
+dnodesize_changed_cb(void *arg, uint64_t newval)
+{
+	objset_t *os = arg;
+
+	switch (newval) {
+	case ZFS_DNSIZE_LEGACY:
+		os->os_dnodesize = DNODE_MIN_SIZE;
+		break;
+	case ZFS_DNSIZE_AUTO:
+		/*
+		 * Choose a dnode size that will work well for most
+		 * workloads if the user specified "auto". Future code
+		 * improvements could dynamically select a dnode size
+		 * based on observed workload patterns.
+		 */
+		os->os_dnodesize = DNODE_MIN_SIZE * 2;
+		break;
+	case ZFS_DNSIZE_1K:
+	case ZFS_DNSIZE_2K:
+	case ZFS_DNSIZE_4K:
+	case ZFS_DNSIZE_8K:
+	case ZFS_DNSIZE_16K:
+		os->os_dnodesize = newval;
+		break;
+	}
+}
+
+static void
 logbias_changed_cb(void *arg, uint64_t newval)
 {
 	objset_t *os = arg;
@@ -421,6 +455,11 @@ dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 				    zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
 				    recordsize_changed_cb, os);
 			}
+			if (err == 0) {
+				err = dsl_prop_register(ds,
+				    zfs_prop_to_name(ZFS_PROP_DNODESIZE),
+				    dnodesize_changed_cb, os);
+			}
 		}
 		if (err != 0) {
 			VERIFY(arc_buf_remove_ref(os->os_phys_buf,
@@ -439,6 +478,7 @@ dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 		os->os_sync = ZFS_SYNC_STANDARD;
 		os->os_primary_cache = ZFS_CACHE_ALL;
 		os->os_secondary_cache = ZFS_CACHE_ALL;
+		os->os_dnodesize = DNODE_MIN_SIZE;
 	}
 
 	if (ds == NULL || !ds->ds_is_snapshot)
@@ -768,8 +808,8 @@ dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 
 	mdn = DMU_META_DNODE(os);
 
-	dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
-	    DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
+	dnode_allocate(mdn, DMU_OT_DNODE, DNODE_BLOCK_SIZE, DN_MAX_INDBLKSHIFT,
+	    DMU_OT_NONE, 0, DNODE_MIN_SLOTS, tx);
 
 	/*
 	 * We don't want to have to increase the meta-dnode's nlevels
@@ -1202,7 +1242,7 @@ do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
     uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
 {
 	if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
-		int64_t delta = DNODE_SIZE + used;
+		int64_t delta = DNODE_MIN_SIZE + used;
 		if (subtract)
 			delta = -delta;
 		VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
@@ -2023,6 +2063,7 @@ EXPORT_SYMBOL(dmu_objset_find);
 EXPORT_SYMBOL(dmu_objset_byteswap);
 EXPORT_SYMBOL(dmu_objset_evict_dbufs);
 EXPORT_SYMBOL(dmu_objset_snap_cmtime);
+EXPORT_SYMBOL(dmu_objset_dnodesize);
 
 EXPORT_SYMBOL(dmu_objset_sync);
 EXPORT_SYMBOL(dmu_objset_is_dirty);
diff --git a/module/zfs/dmu_send.c b/module/zfs/dmu_send.c
index 896a84b50..901386a5a 100644
--- a/module/zfs/dmu_send.c
+++ b/module/zfs/dmu_send.c
@@ -445,6 +445,7 @@ dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
 	drro->drr_bonustype = dnp->dn_bonustype;
 	drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
 	drro->drr_bonuslen = dnp->dn_bonuslen;
+	drro->drr_dn_slots = dnp->dn_extra_slots + 1;
 	drro->drr_checksumtype = dnp->dn_checksum;
 	drro->drr_compress = dnp->dn_compress;
 	drro->drr_toguid = dsp->dsa_toguid;
@@ -570,7 +571,6 @@ do_dump(dmu_sendarg_t *dsa, struct send_block_record *data)
 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
 	dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
 	int err = 0;
-	dnode_phys_t *blk;
 	uint64_t dnobj;
 
 	ASSERT3U(zb->zb_level, >=, 0);
@@ -590,7 +590,8 @@ do_dump(dmu_sendarg_t *dsa, struct send_block_record *data)
 	} else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
 		return (0);
 	} else if (type == DMU_OT_DNODE) {
-		int blksz = BP_GET_LSIZE(bp);
+		dnode_phys_t *blk;
+		int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
 		arc_flags_t aflags = ARC_FLAG_WAIT;
 		arc_buf_t *abuf;
 		int i;
@@ -603,8 +604,8 @@ do_dump(dmu_sendarg_t *dsa, struct send_block_record *data)
 			return (SET_ERROR(EIO));
 
 		blk = abuf->b_data;
-		dnobj = zb->zb_blkid * (blksz >> DNODE_SHIFT);
-		for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
+		dnobj = zb->zb_blkid * epb;
+		for (i = 0; i < epb; i += blk[i].dn_extra_slots + 1) {
 			err = dump_dnode(dsa, dnobj + i, blk + i);
 			if (err != 0)
 				break;
@@ -736,6 +737,8 @@ dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *to_ds,
 
 	if (large_block_ok && to_ds->ds_feature_inuse[SPA_FEATURE_LARGE_BLOCKS])
 		featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS;
+	if (to_ds->ds_feature_inuse[SPA_FEATURE_LARGE_DNODE])
+		featureflags |= DMU_BACKUP_FEATURE_LARGE_DNODE;
 	if (embedok &&
 	    spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) {
 		featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA;
@@ -1252,6 +1255,15 @@ dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
 	    !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
 		return (SET_ERROR(ENOTSUP));
 
+	/*
+	 * The receiving code doesn't know how to translate large dnodes
+	 * to smaller ones, so the pool must have the LARGE_DNODE
+	 * feature enabled if the stream has LARGE_DNODE.
+	 */
+	if ((featureflags & DMU_BACKUP_FEATURE_LARGE_DNODE) &&
+	    !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_DNODE))
+		return (SET_ERROR(ENOTSUP));
+
 	error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
 	if (error == 0) {
 		/* target fs already exists; recv into temp clone */
@@ -1658,7 +1670,8 @@ deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size)
 		return (1);
 	} else {
 		return (1 +
-		    ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT));
+		    ((DN_OLD_MAX_BONUSLEN -
+		    MIN(DN_OLD_MAX_BONUSLEN, bonus_size)) >> SPA_BLKPTRSHIFT));
 	}
 }
 
@@ -1679,7 +1692,8 @@ receive_object(struct receive_writer_arg *rwa, struct drr_object *drro,
 	    P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
 	    drro->drr_blksz < SPA_MINBLOCKSIZE ||
 	    drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(rwa->os)) ||
-	    drro->drr_bonuslen > DN_MAX_BONUSLEN) {
+	    drro->drr_bonuslen >
+	    DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(rwa->os)))) {
 		return (SET_ERROR(EINVAL));
 	}
 
@@ -1719,9 +1733,10 @@ receive_object(struct receive_writer_arg *rwa, struct drr_object *drro,
 
 	if (object == DMU_NEW_OBJECT) {
 		/* currently free, want to be allocated */
-		err = dmu_object_claim(rwa->os, drro->drr_object,
+		err = dmu_object_claim_dnsize(rwa->os, drro->drr_object,
 		    drro->drr_type, drro->drr_blksz,
-		    drro->drr_bonustype, drro->drr_bonuslen, tx);
+		    drro->drr_bonustype, drro->drr_bonuslen,
+		    drro->drr_dn_slots << DNODE_SHIFT, tx);
 	} else if (drro->drr_type != doi.doi_type ||
 	    drro->drr_blksz != doi.doi_data_block_size ||
 	    drro->drr_bonustype != doi.doi_bonus_type ||
@@ -1771,18 +1786,25 @@ receive_freeobjects(struct receive_writer_arg *rwa,
 	if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
 		return (SET_ERROR(EINVAL));
 
-	for (obj = drrfo->drr_firstobj;
+	for (obj = drrfo->drr_firstobj == 0 ? 1 : drrfo->drr_firstobj;
 	    obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
 	    (void) dmu_object_next(rwa->os, &obj, FALSE, 0)) {
+		dmu_object_info_t doi;
 		int err;
 
-		if (dmu_object_info(rwa->os, obj, NULL) != 0)
+		err = dmu_object_info(rwa->os, obj, &doi);
+		if (err == ENOENT) {
+			obj++;
 			continue;
+		} else if (err != 0) {
+			return (err);
+		}
 
 		err = dmu_free_long_object(rwa->os, obj);
 		if (err != 0)
 			return (err);
 	}
+
 	return (0);
 }
 
diff --git a/module/zfs/dmu_traverse.c b/module/zfs/dmu_traverse.c
index bba9efe14..44ba74181 100644
--- a/module/zfs/dmu_traverse.c
+++ b/module/zfs/dmu_traverse.c
@@ -331,13 +331,13 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
 			goto post;
 		cdnp = buf->b_data;
 
-		for (i = 0; i < epb; i++) {
+		for (i = 0; i < epb; i += cdnp[i].dn_extra_slots + 1) {
 			prefetch_dnode_metadata(td, &cdnp[i], zb->zb_objset,
 			    zb->zb_blkid * epb + i);
 		}
 
 		/* recursively visitbp() blocks below this */
-		for (i = 0; i < epb; i++) {
+		for (i = 0; i < epb; i += cdnp[i].dn_extra_slots + 1) {
 			err = traverse_dnode(td, &cdnp[i], zb->zb_objset,
 			    zb->zb_blkid * epb + i);
 			if (err != 0)
@@ -439,7 +439,7 @@ prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *dnp,
 
 	if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
 		SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
-		traverse_prefetch_metadata(td, &dnp->dn_spill, &czb);
+		traverse_prefetch_metadata(td, DN_SPILL_BLKPTR(dnp), &czb);
 	}
 }
 
@@ -470,7 +470,7 @@ traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
 
 	if (err == 0 && (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) {
 		SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
-		err = traverse_visitbp(td, dnp, &dnp->dn_spill, &czb);
+		err = traverse_visitbp(td, dnp, DN_SPILL_BLKPTR(dnp), &czb);
 	}
 
 	if (err == 0 && (td->td_flags & TRAVERSE_POST)) {
diff --git a/module/zfs/dmu_tx.c b/module/zfs/dmu_tx.c
index 74e323dbd..ed29bfbc6 100644
--- a/module/zfs/dmu_tx.c
+++ b/module/zfs/dmu_tx.c
@@ -1586,7 +1586,7 @@ dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object)
 	} else {
 		blkptr_t *bp;
 
-		bp = &dn->dn_phys->dn_spill;
+		bp = DN_SPILL_BLKPTR(dn->dn_phys);
 		if (dsl_dataset_block_freeable(dn->dn_objset->os_dsl_dataset,
 		    bp, bp->blk_birth))
 			txh->txh_space_tooverwrite += SPA_OLD_MAXBLOCKSIZE;
@@ -1618,7 +1618,7 @@ dmu_tx_hold_sa_create(dmu_tx_t *tx, int attrsize)
 
 	dmu_tx_sa_registration_hold(sa, tx);
 
-	if (attrsize <= DN_MAX_BONUSLEN && !sa->sa_force_spill)
+	if (attrsize <= DN_OLD_MAX_BONUSLEN && !sa->sa_force_spill)
 		return;
 
 	(void) dmu_tx_hold_object_impl(tx, tx->tx_objset, DMU_NEW_OBJECT,
diff --git a/module/zfs/dnode.c b/module/zfs/dnode.c
index 38bcecd46..975bd5fb8 100644
--- a/module/zfs/dnode.c
+++ b/module/zfs/dnode.c
@@ -248,6 +248,7 @@ dnode_verify(dnode_t *dn)
 	}
 	if (dn->dn_phys->dn_type != DMU_OT_NONE || dn->dn_allocated_txg != 0) {
 		int i;
+		int max_bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots);
 		ASSERT3U(dn->dn_indblkshift, <=, SPA_MAXBLOCKSHIFT);
 		if (dn->dn_datablkshift) {
 			ASSERT3U(dn->dn_datablkshift, >=, SPA_MINBLOCKSHIFT);
@@ -258,12 +259,12 @@ dnode_verify(dnode_t *dn)
 		ASSERT(DMU_OT_IS_VALID(dn->dn_type));
 		ASSERT3U(dn->dn_nblkptr, >=, 1);
 		ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR);
-		ASSERT3U(dn->dn_bonuslen, <=, DN_MAX_BONUSLEN);
+		ASSERT3U(dn->dn_bonuslen, <=, max_bonuslen);
 		ASSERT3U(dn->dn_datablksz, ==,
 		    dn->dn_datablkszsec << SPA_MINBLOCKSHIFT);
 		ASSERT3U(ISP2(dn->dn_datablksz), ==, dn->dn_datablkshift != 0);
 		ASSERT3U((dn->dn_nblkptr - 1) * sizeof (blkptr_t) +
-		    dn->dn_bonuslen, <=, DN_MAX_BONUSLEN);
+		    dn->dn_bonuslen, <=, max_bonuslen);
 		for (i = 0; i < TXG_SIZE; i++) {
 			ASSERT3U(dn->dn_next_nlevels[i], <=, dn->dn_nlevels);
 		}
@@ -294,6 +295,7 @@ dnode_byteswap(dnode_phys_t *dnp)
 
 	dnp->dn_datablkszsec = BSWAP_16(dnp->dn_datablkszsec);
 	dnp->dn_bonuslen = BSWAP_16(dnp->dn_bonuslen);
+	dnp->dn_extra_slots = BSWAP_8(dnp->dn_extra_slots);
 	dnp->dn_maxblkid = BSWAP_64(dnp->dn_maxblkid);
 	dnp->dn_used = BSWAP_64(dnp->dn_used);
 
@@ -320,7 +322,8 @@ dnode_byteswap(dnode_phys_t *dnp)
 		 * dnode buffer).
 		 */
 		int off = (dnp->dn_nblkptr-1) * sizeof (blkptr_t);
-		size_t len = DN_MAX_BONUSLEN - off;
+		int slots = dnp->dn_extra_slots + 1;
+		size_t len = DN_SLOTS_TO_BONUSLEN(slots) - off;
 		dmu_object_byteswap_t byteswap;
 		ASSERT(DMU_OT_IS_VALID(dnp->dn_bonustype));
 		byteswap = DMU_OT_BYTESWAP(dnp->dn_bonustype);
@@ -329,23 +332,24 @@ dnode_byteswap(dnode_phys_t *dnp)
 
 	/* Swap SPILL block if we have one */
 	if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)
-		byteswap_uint64_array(&dnp->dn_spill, sizeof (blkptr_t));
-
+		byteswap_uint64_array(DN_SPILL_BLKPTR(dnp), sizeof (blkptr_t));
 }
 
 void
 dnode_buf_byteswap(void *vbuf, size_t size)
 {
-	dnode_phys_t *buf = vbuf;
-	int i;
+	int i = 0;
 
 	ASSERT3U(sizeof (dnode_phys_t), ==, (1<<DNODE_SHIFT));
 	ASSERT((size & (sizeof (dnode_phys_t)-1)) == 0);
 
-	size >>= DNODE_SHIFT;
-	for (i = 0; i < size; i++) {
-		dnode_byteswap(buf);
-		buf++;
+	while (i < size) {
+		dnode_phys_t *dnp = vbuf + i;
+		dnode_byteswap(dnp);
+
+		i += DNODE_MIN_SIZE;
+		if (dnp->dn_type != DMU_OT_NONE)
+			i += dnp->dn_extra_slots * DNODE_MIN_SIZE;
 	}
 }
 
@@ -356,7 +360,7 @@ dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx)
 
 	dnode_setdirty(dn, tx);
 	rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
-	ASSERT3U(newsize, <=, DN_MAX_BONUSLEN -
+	ASSERT3U(newsize, <=, DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots) -
 	    (dn->dn_nblkptr-1) * sizeof (blkptr_t));
 	dn->dn_bonuslen = newsize;
 	if (newsize == 0)
@@ -434,6 +438,7 @@ dnode_create(objset_t *os, dnode_phys_t *dnp, dmu_buf_impl_t *db,
 	dn->dn_compress = dnp->dn_compress;
 	dn->dn_bonustype = dnp->dn_bonustype;
 	dn->dn_bonuslen = dnp->dn_bonuslen;
+	dn->dn_num_slots = dnp->dn_extra_slots + 1;
 	dn->dn_maxblkid = dnp->dn_maxblkid;
 	dn->dn_have_spill = ((dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0);
 	dn->dn_id_flags = 0;
@@ -534,10 +539,13 @@ dnode_destroy(dnode_t *dn)
 
 void
 dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
-    dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
+    dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx)
 {
 	int i;
 
+	ASSERT3U(dn_slots, >, 0);
+	ASSERT3U(dn_slots << DNODE_SHIFT, <=,
+	    spa_maxdnodesize(dmu_objset_spa(dn->dn_objset)));
 	ASSERT3U(blocksize, <=,
 	    spa_maxblocksize(dmu_objset_spa(dn->dn_objset)));
 	if (blocksize == 0)
@@ -550,8 +558,8 @@ dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
 
 	ibs = MIN(MAX(ibs, DN_MIN_INDBLKSHIFT), DN_MAX_INDBLKSHIFT);
 
-	dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d\n", dn->dn_objset,
-	    dn->dn_object, tx->tx_txg, blocksize, ibs);
+	dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d dn_slots=%d\n",
+	    dn->dn_objset, dn->dn_object, tx->tx_txg, blocksize, ibs, dn_slots);
 
 	ASSERT(dn->dn_type == DMU_OT_NONE);
 	ASSERT(bcmp(dn->dn_phys, &dnode_phys_zero, sizeof (dnode_phys_t)) == 0);
@@ -562,7 +570,7 @@ dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
 	    (bonustype == DMU_OT_SA && bonuslen == 0) ||
 	    (bonustype != DMU_OT_NONE && bonuslen != 0));
 	ASSERT(DMU_OT_IS_VALID(bonustype));
-	ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN);
+	ASSERT3U(bonuslen, <=, DN_SLOTS_TO_BONUSLEN(dn_slots));
 	ASSERT(dn->dn_type == DMU_OT_NONE);
 	ASSERT0(dn->dn_maxblkid);
 	ASSERT0(dn->dn_allocated_txg);
@@ -588,11 +596,15 @@ dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
 	dnode_setdblksz(dn, blocksize);
 	dn->dn_indblkshift = ibs;
 	dn->dn_nlevels = 1;
+	dn->dn_num_slots = dn_slots;
 	if (bonustype == DMU_OT_SA) /* Maximize bonus space for SA */
 		dn->dn_nblkptr = 1;
-	else
-		dn->dn_nblkptr = 1 +
-		    ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT);
+	else {
+		dn->dn_nblkptr = MIN(DN_MAX_NBLKPTR,
+		    1 + ((DN_SLOTS_TO_BONUSLEN(dn_slots) - bonuslen) >>
+		    SPA_BLKPTRSHIFT));
+	}
+
 	dn->dn_bonustype = bonustype;
 	dn->dn_bonuslen = bonuslen;
 	dn->dn_checksum = ZIO_CHECKSUM_INHERIT;
@@ -617,7 +629,7 @@ dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
 
 void
 dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
-    dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
+    dmu_object_type_t bonustype, int bonuslen, int dn_slots, dmu_tx_t *tx)
 {
 	int nblkptr;
 
@@ -631,7 +643,10 @@ dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
 	    (bonustype != DMU_OT_NONE && bonuslen != 0) ||
 	    (bonustype == DMU_OT_SA && bonuslen == 0));
 	ASSERT(DMU_OT_IS_VALID(bonustype));
-	ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN);
+	ASSERT3U(bonuslen, <=,
+		DN_BONUS_SIZE(spa_maxdnodesize(dmu_objset_spa(dn->dn_objset))));
+
+	dn_slots = dn_slots > 0 ? dn_slots : DNODE_MIN_SLOTS;
 
 	/* clean up any unreferenced dbufs */
 	dnode_evict_dbufs(dn);
@@ -654,7 +669,9 @@ dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
 	if (bonustype == DMU_OT_SA) /* Maximize bonus space for SA */
 		nblkptr = 1;
 	else
-		nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT);
+		nblkptr = MIN(DN_MAX_NBLKPTR,
+		    1 + ((DN_SLOTS_TO_BONUSLEN(dn_slots) - bonuslen) >>
+		    SPA_BLKPTRSHIFT));
 	if (dn->dn_bonustype != bonustype)
 		dn->dn_next_bonustype[tx->tx_txg&TXG_MASK] = bonustype;
 	if (dn->dn_nblkptr != nblkptr)
@@ -672,6 +689,7 @@ dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
 	mutex_enter(&dn->dn_mtx);
 	dn->dn_bonustype = bonustype;
 	dn->dn_bonuslen = bonuslen;
+	dn->dn_num_slots = dn_slots;
 	dn->dn_nblkptr = nblkptr;
 	dn->dn_checksum = ZIO_CHECKSUM_INHERIT;
 	dn->dn_compress = ZIO_COMPRESS_INHERIT;
@@ -680,7 +698,8 @@ dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
 	/* fix up the bonus db_size */
 	if (dn->dn_bonus) {
 		dn->dn_bonus->db.db_size =
-		    DN_MAX_BONUSLEN - (dn->dn_nblkptr-1) * sizeof (blkptr_t);
+		    DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots) -
+		    (dn->dn_nblkptr-1) * sizeof (blkptr_t);
 		ASSERT(dn->dn_bonuslen <= dn->dn_bonus->db.db_size);
 	}
 
@@ -1053,24 +1072,150 @@ dnode_buf_pageout(void *dbu)
 }
 
 /*
+ * Return true if the given index is interior to a dnode already
+ * allocated in the block. That is, the index is neither free nor
+ * allocated, but is consumed by a large dnode.
+ *
+ * The dnode_phys_t buffer may not be in sync with the in-core dnode
+ * structure, so we try to check the dnode structure first and fall back
+ * to the dnode_phys_t buffer it doesn't exist.
+ */
+static boolean_t
+dnode_is_consumed(dmu_buf_impl_t *db, int idx)
+{
+	dnode_handle_t *dnh;
+	dmu_object_type_t ot;
+	dnode_children_t *children_dnodes;
+	dnode_phys_t *dn_block;
+	int skip;
+	int i;
+
+	children_dnodes = dmu_buf_get_user(&db->db);
+	dn_block = (dnode_phys_t *)db->db.db_data;
+
+	for (i = 0; i < idx; i += skip) {
+		dnh = &children_dnodes->dnc_children[i];
+
+		zrl_add(&dnh->dnh_zrlock);
+		if (dnh->dnh_dnode != NULL) {
+			ot = dnh->dnh_dnode->dn_type;
+			skip = dnh->dnh_dnode->dn_num_slots;
+		} else {
+			ot = dn_block[i].dn_type;
+			skip = dn_block[i].dn_extra_slots + 1;
+		}
+		zrl_remove(&dnh->dnh_zrlock);
+
+		if (ot == DMU_OT_NONE)
+			skip = 1;
+	}
+
+	return (i > idx);
+}
+
+/*
+ * Return true if the given index in the dnode block is a valid
+ * allocated dnode. That is, the index is not consumed by a large
+ * dnode and is not free.
+ *
+ * The dnode_phys_t buffer may not be in sync with the in-core dnode
+ * structure, so we try to check the dnode structure first and fall back
+ * to the dnode_phys_t buffer it doesn't exist.
+ */
+static boolean_t
+dnode_is_allocated(dmu_buf_impl_t *db, int idx)
+{
+	dnode_handle_t *dnh;
+	dmu_object_type_t ot;
+	dnode_children_t *children_dnodes;
+	dnode_phys_t *dn_block;
+
+	if (dnode_is_consumed(db, idx))
+		return (B_FALSE);
+
+	children_dnodes = dmu_buf_get_user(&db->db);
+	dn_block = (dnode_phys_t *)db->db.db_data;
+
+	dnh = &children_dnodes->dnc_children[idx];
+
+	zrl_add(&dnh->dnh_zrlock);
+	if (dnh->dnh_dnode != NULL)
+		ot = dnh->dnh_dnode->dn_type;
+	else
+		ot = dn_block[idx].dn_type;
+	zrl_remove(&dnh->dnh_zrlock);
+
+	return (ot != DMU_OT_NONE);
+}
+
+/*
+ * Return true if the given range of indices in the dnode block are
+ * free. That is, the starting index is not consumed by a large dnode
+ * and none of the indices are allocated.
+ *
+ * The dnode_phys_t buffer may not be in sync with the in-core dnode
+ * structure, so we try to check the dnode structure first and fall back
+ * to the dnode_phys_t buffer it doesn't exist.
+ */
+static boolean_t
+dnode_is_free(dmu_buf_impl_t *db, int idx, int slots)
+{
+	dnode_handle_t *dnh;
+	dmu_object_type_t ot;
+	dnode_children_t *children_dnodes;
+	dnode_phys_t *dn_block;
+	int i;
+
+	if (idx + slots > DNODES_PER_BLOCK)
+		return (B_FALSE);
+
+	children_dnodes = dmu_buf_get_user(&db->db);
+	dn_block = (dnode_phys_t *)db->db.db_data;
+
+	if (dnode_is_consumed(db, idx))
+		return (B_FALSE);
+
+	for (i = idx; i < idx + slots; i++) {
+		dnh = &children_dnodes->dnc_children[i];
+
+		zrl_add(&dnh->dnh_zrlock);
+		if (dnh->dnh_dnode != NULL)
+			ot = dnh->dnh_dnode->dn_type;
+		else
+			ot = dn_block[i].dn_type;
+		zrl_remove(&dnh->dnh_zrlock);
+
+		if (ot != DMU_OT_NONE)
+			return (B_FALSE);
+	}
+
+	return (B_TRUE);
+}
+
+/*
  * errors:
  * EINVAL - invalid object number.
+ * ENOSPC - hole too small to fulfill "slots" request
  * EIO - i/o error.
  * succeeds even for free dnodes.
  */
 int
-dnode_hold_impl(objset_t *os, uint64_t object, int flag,
+dnode_hold_impl(objset_t *os, uint64_t object, int flag, int slots,
     void *tag, dnode_t **dnp)
 {
-	int epb, idx, err;
+	int epb, idx, err, i;
 	int drop_struct_lock = FALSE;
 	int type;
 	uint64_t blk;
 	dnode_t *mdn, *dn;
 	dmu_buf_impl_t *db;
 	dnode_children_t *children_dnodes;
+	dnode_phys_t *dn_block_begin;
 	dnode_handle_t *dnh;
 
+	ASSERT(!(flag & DNODE_MUST_BE_ALLOCATED) || (slots == 0));
+	ASSERT(!(flag & DNODE_MUST_BE_FREE) || (slots > 0));
+
 	/*
 	 * If you are holding the spa config lock as writer, you shouldn't
 	 * be asking the DMU to do *anything* unless it's the root pool
@@ -1126,12 +1271,9 @@ dnode_hold_impl(objset_t *os, uint64_t object, int flag,
 	ASSERT3U(db->db.db_size, >=, 1<<DNODE_SHIFT);
 	epb = db->db.db_size >> DNODE_SHIFT;
 
-	idx = object & (epb-1);
-
 	ASSERT(DB_DNODE(db)->dn_type == DMU_OT_DNODE);
 	children_dnodes = dmu_buf_get_user(&db->db);
 	if (children_dnodes == NULL) {
-		int i;
 		dnode_children_t *winner;
 		children_dnodes = kmem_zalloc(sizeof (dnode_children_t) +
 		    epb * sizeof (dnode_handle_t), KM_SLEEP);
@@ -1156,21 +1298,28 @@ dnode_hold_impl(objset_t *os, uint64_t object, int flag,
 	}
 	ASSERT(children_dnodes->dnc_count == epb);
 
+	idx = object & (epb - 1);
+	dn_block_begin = (dnode_phys_t *)db->db.db_data;
+
+	if ((flag & DNODE_MUST_BE_FREE) && !dnode_is_free(db, idx, slots)) {
+		dbuf_rele(db, FTAG);
+		return (ENOSPC);
+	} else if ((flag & DNODE_MUST_BE_ALLOCATED) &&
+	    !dnode_is_allocated(db, idx)) {
+		dbuf_rele(db, FTAG);
+		return (ENOENT);
+	}
+
 	dnh = &children_dnodes->dnc_children[idx];
 	zrl_add(&dnh->dnh_zrlock);
 	dn = dnh->dnh_dnode;
-	if (dn == NULL) {
-		dnode_phys_t *phys = (dnode_phys_t *)db->db.db_data+idx;
-
-		dn = dnode_create(os, phys, db, object, dnh);
-	}
+	if (dn == NULL)
+		dn = dnode_create(os, dn_block_begin + idx, db, object, dnh);
 
 	mutex_enter(&dn->dn_mtx);
 	type = dn->dn_type;
 	if (dn->dn_free_txg ||
-	    ((flag & DNODE_MUST_BE_ALLOCATED) && type == DMU_OT_NONE) ||
-	    ((flag & DNODE_MUST_BE_FREE) &&
-	    (type != DMU_OT_NONE || !refcount_is_zero(&dn->dn_holds)))) {
+	    ((flag & DNODE_MUST_BE_FREE) && !refcount_is_zero(&dn->dn_holds))) {
 		mutex_exit(&dn->dn_mtx);
 		zrl_remove(&dnh->dnh_zrlock);
 		dbuf_rele(db, FTAG);
@@ -1198,7 +1347,8 @@ dnode_hold_impl(objset_t *os, uint64_t object, int flag,
 int
 dnode_hold(objset_t *os, uint64_t object, void *tag, dnode_t **dnp)
 {
-	return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, tag, dnp));
+	return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 0, tag,
+	    dnp));
 }
 
 /*
@@ -1908,17 +2058,21 @@ dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset,
 		error = SET_ERROR(ESRCH);
 	} else if (lvl == 0) {
 		dnode_phys_t *dnp = data;
-		span = DNODE_SHIFT;
+
 		ASSERT(dn->dn_type == DMU_OT_DNODE);
+		ASSERT(!(flags & DNODE_FIND_BACKWARDS));
 
-		for (i = (*offset >> span) & (blkfill - 1);
-		    i >= 0 && i < blkfill; i += inc) {
+		for (i = (*offset >> DNODE_SHIFT) & (blkfill - 1);
+		    i < blkfill; i += dnp[i].dn_extra_slots + 1) {
 			if ((dnp[i].dn_type == DMU_OT_NONE) == hole)
 				break;
-			*offset += (1ULL << span) * inc;
 		}
-		if (i < 0 || i == blkfill)
+
+		if (i == blkfill)
 			error = SET_ERROR(ESRCH);
+
+		*offset = (*offset & ~(DNODE_BLOCK_SIZE - 1)) +
+		    (i << DNODE_SHIFT);
 	} else {
 		blkptr_t *bp = data;
 		uint64_t start = *offset;
diff --git a/module/zfs/dnode_sync.c b/module/zfs/dnode_sync.c
index bea7be186..54066e2e3 100644
--- a/module/zfs/dnode_sync.c
+++ b/module/zfs/dnode_sync.c
@@ -524,7 +524,7 @@ dnode_sync_free(dnode_t *dn, dmu_tx_t *tx)
 	ASSERT(dn->dn_free_txg > 0);
 	if (dn->dn_allocated_txg != dn->dn_free_txg)
 		dmu_buf_will_dirty(&dn->dn_dbuf->db, tx);
-	bzero(dn->dn_phys, sizeof (dnode_phys_t));
+	bzero(dn->dn_phys, sizeof (dnode_phys_t) * dn->dn_num_slots);
 
 	mutex_enter(&dn->dn_mtx);
 	dn->dn_type = DMU_OT_NONE;
@@ -559,7 +559,7 @@ dnode_sync(dnode_t *dn, dmu_tx_t *tx)
 	ASSERT(dmu_tx_is_syncing(tx));
 	ASSERT(dnp->dn_type != DMU_OT_NONE || dn->dn_allocated_txg);
 	ASSERT(dnp->dn_type != DMU_OT_NONE ||
-	    bcmp(dnp, &zerodn, DNODE_SIZE) == 0);
+	    bcmp(dnp, &zerodn, DNODE_MIN_SIZE) == 0);
 	DNODE_VERIFY(dn);
 
 	ASSERT(dn->dn_dbuf == NULL || arc_released(dn->dn_dbuf->db_buf));
@@ -591,6 +591,9 @@ dnode_sync(dnode_t *dn, dmu_tx_t *tx)
 		dnp->dn_bonustype = dn->dn_bonustype;
 		dnp->dn_bonuslen = dn->dn_bonuslen;
 	}
+
+	dnp->dn_extra_slots = dn->dn_num_slots - 1;
+
 	ASSERT(dnp->dn_nlevels > 1 ||
 	    BP_IS_HOLE(&dnp->dn_blkptr[0]) ||
 	    BP_IS_EMBEDDED(&dnp->dn_blkptr[0]) ||
@@ -623,7 +626,8 @@ dnode_sync(dnode_t *dn, dmu_tx_t *tx)
 			dnp->dn_bonuslen = 0;
 		else
 			dnp->dn_bonuslen = dn->dn_next_bonuslen[txgoff];
-		ASSERT(dnp->dn_bonuslen <= DN_MAX_BONUSLEN);
+		ASSERT(dnp->dn_bonuslen <=
+		    DN_SLOTS_TO_BONUSLEN(dnp->dn_extra_slots + 1));
 		dn->dn_next_bonuslen[txgoff] = 0;
 	}
 
@@ -662,7 +666,7 @@ dnode_sync(dnode_t *dn, dmu_tx_t *tx)
 	mutex_exit(&dn->dn_mtx);
 
 	if (kill_spill) {
-		free_blocks(dn, &dn->dn_phys->dn_spill, 1, tx);
+		free_blocks(dn, DN_SPILL_BLKPTR(dn->dn_phys), 1, tx);
 		mutex_enter(&dn->dn_mtx);
 		dnp->dn_flags &= ~DNODE_FLAG_SPILL_BLKPTR;
 		mutex_exit(&dn->dn_mtx);
@@ -687,6 +691,14 @@ dnode_sync(dnode_t *dn, dmu_tx_t *tx)
 		return;
 	}
 
+	if (dn->dn_num_slots > DNODE_MIN_SLOTS) {
+		dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset;
+		mutex_enter(&ds->ds_lock);
+		ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_DNODE] =
+		    B_TRUE;
+		mutex_exit(&ds->ds_lock);
+	}
+
 	if (dn->dn_next_nlevels[txgoff]) {
 		dnode_increase_indirection(dn, tx);
 		dn->dn_next_nlevels[txgoff] = 0;
diff --git a/module/zfs/dsl_scan.c b/module/zfs/dsl_scan.c
index b5e272fb9..72163521e 100644
--- a/module/zfs/dsl_scan.c
+++ b/module/zfs/dsl_scan.c
@@ -709,14 +709,18 @@ dsl_scan_recurse(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype,
 			scn->scn_phys.scn_errors++;
 			return (err);
 		}
-		for (i = 0, cdnp = buf->b_data; i < epb; i++, cdnp++) {
+		for (i = 0, cdnp = buf->b_data; i < epb;
+		    i += cdnp->dn_extra_slots + 1,
+		    cdnp += cdnp->dn_extra_slots + 1) {
 			for (j = 0; j < cdnp->dn_nblkptr; j++) {
 				blkptr_t *cbp = &cdnp->dn_blkptr[j];
 				dsl_scan_prefetch(scn, buf, cbp,
 				    zb->zb_objset, zb->zb_blkid * epb + i, j);
 			}
 		}
-		for (i = 0, cdnp = buf->b_data; i < epb; i++, cdnp++) {
+		for (i = 0, cdnp = buf->b_data; i < epb;
+		    i += cdnp->dn_extra_slots + 1,
+		    cdnp += cdnp->dn_extra_slots + 1) {
 			dsl_scan_visitdnode(scn, ds, ostype,
 			    cdnp, zb->zb_blkid * epb + i, tx);
 		}
@@ -779,7 +783,7 @@ dsl_scan_visitdnode(dsl_scan_t *scn, dsl_dataset_t *ds,
 		zbookmark_phys_t czb;
 		SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object,
 		    0, DMU_SPILL_BLKID);
-		dsl_scan_visitbp(&dnp->dn_spill,
+		dsl_scan_visitbp(DN_SPILL_BLKPTR(dnp),
 		    &czb, dnp, ds, scn, ostype, tx);
 	}
 }
diff --git a/module/zfs/sa.c b/module/zfs/sa.c
index d6ac5fcc7..adc301512 100644
--- a/module/zfs/sa.c
+++ b/module/zfs/sa.c
@@ -33,6 +33,7 @@
 #include <sys/dmu.h>
 #include <sys/dmu_impl.h>
 #include <sys/dmu_objset.h>
+#include <sys/dmu_tx.h>
 #include <sys/dbuf.h>
 #include <sys/dnode.h>
 #include <sys/zap.h>
@@ -553,12 +554,11 @@ sa_copy_data(sa_data_locator_t *func, void *datastart, void *target, int buflen)
  */
 static int
 sa_find_sizes(sa_os_t *sa, sa_bulk_attr_t *attr_desc, int attr_count,
-    dmu_buf_t *db, sa_buf_type_t buftype, int *index, int *total,
-    boolean_t *will_spill)
+    dmu_buf_t *db, sa_buf_type_t buftype, int full_space, int *index,
+    int *total, boolean_t *will_spill)
 {
 	int var_size_count = 0;
 	int i;
-	int full_space;
 	int hdrsize;
 	int extra_hdrsize;
 
@@ -577,7 +577,6 @@ sa_find_sizes(sa_os_t *sa, sa_bulk_attr_t *attr_desc, int attr_count,
 	hdrsize = (SA_BONUSTYPE_FROM_DB(db) == DMU_OT_ZNODE) ? 0 :
 	    sizeof (sa_hdr_phys_t);
 
-	full_space = (buftype == SA_BONUS) ? DN_MAX_BONUSLEN : db->db_size;
 	ASSERT(IS_P2ALIGNED(full_space, 8));
 
 	for (i = 0; i != attr_count; i++) {
@@ -668,6 +667,7 @@ sa_build_layouts(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, int attr_count,
 	void *data_start;
 	sa_attr_type_t *attrs, *attrs_start;
 	int i, lot_count;
+	int dnodesize;
 	int spill_idx;
 	int hdrsize;
 	int spillhdrsize = 0;
@@ -676,20 +676,23 @@ sa_build_layouts(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, int attr_count,
 	sa_lot_t *lot;
 	int len_idx;
 	int spill_used;
+	int bonuslen;
 	boolean_t spilling;
 
 	dmu_buf_will_dirty(hdl->sa_bonus, tx);
 	bonustype = SA_BONUSTYPE_FROM_DB(hdl->sa_bonus);
+	dmu_object_dnsize_from_db(hdl->sa_bonus, &dnodesize);
+	bonuslen = DN_BONUS_SIZE(dnodesize);
 
 	/* first determine bonus header size and sum of all attributes */
 	hdrsize = sa_find_sizes(sa, attr_desc, attr_count, hdl->sa_bonus,
-	    SA_BONUS, &spill_idx, &used, &spilling);
+	    SA_BONUS, bonuslen, &spill_idx, &used, &spilling);
 
 	if (used > SPA_OLD_MAXBLOCKSIZE)
 		return (SET_ERROR(EFBIG));
 
-	VERIFY(0 == dmu_set_bonus(hdl->sa_bonus, spilling ?
-	    MIN(DN_MAX_BONUSLEN - sizeof (blkptr_t), used + hdrsize) :
+	VERIFY0(dmu_set_bonus(hdl->sa_bonus, spilling ?
+	    MIN(bonuslen - sizeof (blkptr_t), used + hdrsize) :
 	    used + hdrsize, tx));
 
 	ASSERT((bonustype == DMU_OT_ZNODE && spilling == 0) ||
@@ -706,8 +709,8 @@ sa_build_layouts(sa_handle_t *hdl, sa_bulk_attr_t *attr_desc, int attr_count,
 		dmu_buf_will_dirty(hdl->sa_spill, tx);
 
 		spillhdrsize = sa_find_sizes(sa, &attr_desc[spill_idx],
-		    attr_count - spill_idx, hdl->sa_spill, SA_SPILL, &i,
-		    &spill_used, &dummy);
+		    attr_count - spill_idx, hdl->sa_spill, SA_SPILL,
+		    hdl->sa_spill->db_size, &i, &spill_used, &dummy);
 
 		if (spill_used > SPA_OLD_MAXBLOCKSIZE)
 			return (SET_ERROR(EFBIG));
diff --git a/module/zfs/spa.c b/module/zfs/spa.c
index c23fd7a3a..d1aefe585 100644
--- a/module/zfs/spa.c
+++ b/module/zfs/spa.c
@@ -281,6 +281,14 @@ spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
 		    SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
 	}
 
+	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) {
+		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
+		    DNODE_MAX_SIZE, ZPROP_SRC_NONE);
+	} else {
+		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
+		    DNODE_MIN_SIZE, ZPROP_SRC_NONE);
+	}
+
 	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
 		if (dp->scd_path == NULL) {
 			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
@@ -512,7 +520,8 @@ spa_prop_validate(spa_t *spa, nvlist_t *props)
 				/*
 				 * Must be ZPL, and its property settings
 				 * must be supported by GRUB (compression
-				 * is not gzip, and large blocks are not used).
+				 * is not gzip, and large blocks or large
+				 * dnodes are not used).
 				 */
 
 				if (dmu_objset_type(os) != DMU_OST_ZFS) {
@@ -529,6 +538,12 @@ spa_prop_validate(spa_t *spa, nvlist_t *props)
 				    &propval)) == 0 &&
 				    propval > SPA_OLD_MAXBLOCKSIZE) {
 					error = SET_ERROR(ENOTSUP);
+				} else if ((error =
+				    dsl_prop_get_int_ds(dmu_objset_ds(os),
+				    zfs_prop_to_name(ZFS_PROP_DNODESIZE),
+				    &propval)) == 0 &&
+				    propval != ZFS_DNSIZE_LEGACY) {
+					error = SET_ERROR(ENOTSUP);
 				} else {
 					objnum = dmu_objset_id(os);
 				}
diff --git a/module/zfs/spa_misc.c b/module/zfs/spa_misc.c
index e3e7e36fe..d1303b5c2 100644
--- a/module/zfs/spa_misc.c
+++ b/module/zfs/spa_misc.c
@@ -2000,6 +2000,15 @@ spa_maxblocksize(spa_t *spa)
 		return (SPA_OLD_MAXBLOCKSIZE);
 }
 
+int
+spa_maxdnodesize(spa_t *spa)
+{
+	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE))
+		return (DNODE_MAX_SIZE);
+	else
+		return (DNODE_MIN_SIZE);
+}
+
 #if defined(_KERNEL) && defined(HAVE_SPL)
 /* Namespace manipulation */
 EXPORT_SYMBOL(spa_lookup);
@@ -2056,6 +2065,7 @@ EXPORT_SYMBOL(spa_bootfs);
 EXPORT_SYMBOL(spa_delegation);
 EXPORT_SYMBOL(spa_meta_objset);
 EXPORT_SYMBOL(spa_maxblocksize);
+EXPORT_SYMBOL(spa_maxdnodesize);
 
 /* Miscellaneous support routines */
 EXPORT_SYMBOL(spa_rename);
diff --git a/module/zfs/zap.c b/module/zfs/zap.c
index 454b4be62..9e4f05049 100644
--- a/module/zfs/zap.c
+++ b/module/zfs/zap.c
@@ -968,9 +968,17 @@ uint64_t
 zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
     const char *name, dmu_tx_t *tx)
 {
+	return (zap_create_link_dnsize(os, ot, parent_obj, name, 0, tx));
+}
+
+uint64_t
+zap_create_link_dnsize(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
+    const char *name, int dnodesize, dmu_tx_t *tx)
+{
 	uint64_t new_obj;
 
-	VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0);
+	VERIFY((new_obj = zap_create_dnsize(os, ot, DMU_OT_NONE, 0,
+	    dnodesize, tx)) > 0);
 	VERIFY0(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
 	    tx));
 
diff --git a/module/zfs/zap_micro.c b/module/zfs/zap_micro.c
index 3faf27ce3..f3153cc18 100644
--- a/module/zfs/zap_micro.c
+++ b/module/zfs/zap_micro.c
@@ -630,8 +630,16 @@ int
 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 {
-	return (zap_create_claim_norm(os, obj,
-	    0, ot, bonustype, bonuslen, tx));
+	return (zap_create_claim_dnsize(os, obj, ot, bonustype, bonuslen,
+	    0, tx));
+}
+
+int
+zap_create_claim_dnsize(objset_t *os, uint64_t obj, dmu_object_type_t ot,
+    dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
+{
+	return (zap_create_claim_norm_dnsize(os, obj,
+	    0, ot, bonustype, bonuslen, dnodesize, tx));
 }
 
 int
@@ -639,9 +647,19 @@ zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
     dmu_object_type_t ot,
     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 {
+	return (zap_create_claim_norm_dnsize(os, obj, normflags, ot, bonustype,
+	    bonuslen, 0, tx));
+}
+
+int
+zap_create_claim_norm_dnsize(objset_t *os, uint64_t obj, int normflags,
+    dmu_object_type_t ot, dmu_object_type_t bonustype, int bonuslen,
+    int dnodesize, dmu_tx_t *tx)
+{
 	int err;
 
-	err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
+	err = dmu_object_claim_dnsize(os, obj, ot, 0, bonustype, bonuslen,
+	    dnodesize, tx);
 	if (err != 0)
 		return (err);
 	mzap_create_impl(os, obj, normflags, 0, tx);
@@ -656,10 +674,27 @@ zap_create(objset_t *os, dmu_object_type_t ot,
 }
 
 uint64_t
+zap_create_dnsize(objset_t *os, dmu_object_type_t ot,
+    dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
+{
+	return (zap_create_norm_dnsize(os, 0, ot, bonustype, bonuslen,
+	    dnodesize, tx));
+}
+
+uint64_t
 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 {
-	uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
+	return (zap_create_norm_dnsize(os, normflags, ot, bonustype, bonuslen,
+	    0, tx));
+}
+
+uint64_t
+zap_create_norm_dnsize(objset_t *os, int normflags, dmu_object_type_t ot,
+    dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
+{
+	uint64_t obj = dmu_object_alloc_dnsize(os, ot, 0, bonustype, bonuslen,
+	    dnodesize, tx);
 
 	mzap_create_impl(os, obj, normflags, 0, tx);
 	return (obj);
@@ -670,7 +705,17 @@ zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
     dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
 {
-	uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
+	return (zap_create_flags_dnsize(os, normflags, flags, ot,
+	    leaf_blockshift, indirect_blockshift, bonustype, bonuslen, 0, tx));
+}
+
+uint64_t
+zap_create_flags_dnsize(objset_t *os, int normflags, zap_flags_t flags,
+    dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
+    dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
+{
+	uint64_t obj = dmu_object_alloc_dnsize(os, ot, 0, bonustype, bonuslen,
+	    dnodesize, tx);
 
 	ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
 	    leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
@@ -1458,10 +1503,14 @@ zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
 
 #if defined(_KERNEL) && defined(HAVE_SPL)
 EXPORT_SYMBOL(zap_create);
+EXPORT_SYMBOL(zap_create_dnsize);
 EXPORT_SYMBOL(zap_create_norm);
+EXPORT_SYMBOL(zap_create_norm_dnsize);
 EXPORT_SYMBOL(zap_create_flags);
+EXPORT_SYMBOL(zap_create_flags_dnsize);
 EXPORT_SYMBOL(zap_create_claim);
 EXPORT_SYMBOL(zap_create_claim_norm);
+EXPORT_SYMBOL(zap_create_claim_norm_dnsize);
 EXPORT_SYMBOL(zap_destroy);
 EXPORT_SYMBOL(zap_lookup);
 EXPORT_SYMBOL(zap_lookup_norm);
diff --git a/module/zfs/zfeature_common.c b/module/zfs/zfeature_common.c
index f57e5489c..3264f6235 100644
--- a/module/zfs/zfeature_common.c
+++ b/module/zfs/zfeature_common.c
@@ -242,4 +242,15 @@ zpool_feature_init(void)
 	    "Support for blocks larger than 128KB.",
 	    ZFEATURE_FLAG_PER_DATASET, large_blocks_deps);
 	}
+
+	{
+	static const spa_feature_t large_dnode_deps[] = {
+		SPA_FEATURE_EXTENSIBLE_DATASET,
+		SPA_FEATURE_NONE
+	};
+	zfeature_register(SPA_FEATURE_LARGE_DNODE,
+	    "org.zfsonlinux:large_dnode", "large_dnode",
+	    "Variable on-disk size of dnodes.",
+	    ZFEATURE_FLAG_PER_DATASET, large_dnode_deps);
+	}
 }
diff --git a/module/zfs/zfs_acl.c b/module/zfs/zfs_acl.c
index 961083d4a..f820cdfd6 100644
--- a/module/zfs/zfs_acl.c
+++ b/module/zfs/zfs_acl.c
@@ -1394,7 +1394,7 @@ zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
 				    otype == DMU_OT_ACL ?
 				    DMU_OT_SYSACL : DMU_OT_NONE,
 				    otype == DMU_OT_ACL ?
-				    DN_MAX_BONUSLEN : 0, tx);
+				    DN_OLD_MAX_BONUSLEN : 0, tx);
 			} else {
 				(void) dmu_object_set_blocksize(zsb->z_os,
 				    aoid, aclp->z_acl_bytes, 0, tx);
diff --git a/module/zfs/zfs_ioctl.c b/module/zfs/zfs_ioctl.c
index c63af167a..30338ac14 100644
--- a/module/zfs/zfs_ioctl.c
+++ b/module/zfs/zfs_ioctl.c
@@ -3785,7 +3785,7 @@ zfs_check_settable(const char *dsname, nvpair_t *pair, cred_t *cr)
 
 			/*
 			 * If this is a bootable dataset then
-			 * the we don't allow large (>128K) blocks,
+			 * we don't allow large (>128K) blocks,
 			 * because GRUB doesn't support them.
 			 */
 			if (zfs_is_bootfs(dsname) &&
@@ -3813,6 +3813,34 @@ zfs_check_settable(const char *dsname, nvpair_t *pair, cred_t *cr)
 		}
 		break;
 
+	case ZFS_PROP_DNODESIZE:
+		/* Dnode sizes above 512 need the feature to be enabled */
+		if (nvpair_value_uint64(pair, &intval) == 0 &&
+		    intval != ZFS_DNSIZE_LEGACY) {
+			spa_t *spa;
+
+			/*
+			 * If this is a bootable dataset then
+			 * we don't allow large (>512B) dnodes,
+			 * because GRUB doesn't support them.
+			 */
+			if (zfs_is_bootfs(dsname) &&
+				intval != ZFS_DNSIZE_LEGACY) {
+				return (SET_ERROR(EDOM));
+			}
+
+			if ((err = spa_open(dsname, &spa, FTAG)) != 0)
+				return (err);
+
+			if (!spa_feature_is_enabled(spa,
+			    SPA_FEATURE_LARGE_DNODE)) {
+				spa_close(spa, FTAG);
+				return (SET_ERROR(ENOTSUP));
+			}
+			spa_close(spa, FTAG);
+		}
+		break;
+
 	case ZFS_PROP_SHARESMB:
 		if (zpl_earlier_version(dsname, ZPL_VERSION_FUID))
 			return (SET_ERROR(ENOTSUP));
diff --git a/module/zfs/zfs_log.c b/module/zfs/zfs_log.c
index 38d8de0eb..4d89cb04b 100644
--- a/module/zfs/zfs_log.c
+++ b/module/zfs/zfs_log.c
@@ -279,6 +279,8 @@ zfs_log_create(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,
 	lr = (lr_create_t *)&itx->itx_lr;
 	lr->lr_doid = dzp->z_id;
 	lr->lr_foid = zp->z_id;
+	/* Store dnode slot count in 8 bits above object id. */
+	LR_FOID_SET_SLOTS(lr->lr_foid, zp->z_dnodesize >> DNODE_SHIFT);
 	lr->lr_mode = zp->z_mode;
 	if (!IS_EPHEMERAL(zp->z_uid)) {
 		lr->lr_uid = (uint64_t)zp->z_uid;
diff --git a/module/zfs/zfs_replay.c b/module/zfs/zfs_replay.c
index b97a60ed8..54c175437 100644
--- a/module/zfs/zfs_replay.c
+++ b/module/zfs/zfs_replay.c
@@ -279,6 +279,8 @@ zfs_replay_create_acl(zfs_sb_t *zsb, lr_acl_create_t *lracl, boolean_t byteswap)
 	void *fuidstart;
 	size_t xvatlen = 0;
 	uint64_t txtype;
+	uint64_t objid;
+	uint64_t dnodesize;
 	int error;
 
 	txtype = (lr->lr_common.lrc_txtype & ~TX_CI);
@@ -304,19 +306,24 @@ zfs_replay_create_acl(zfs_sb_t *zsb, lr_acl_create_t *lracl, boolean_t byteswap)
 	if ((error = zfs_zget(zsb, lr->lr_doid, &dzp)) != 0)
 		return (error);
 
+	objid = LR_FOID_GET_OBJ(lr->lr_foid);
+	dnodesize = LR_FOID_GET_SLOTS(lr->lr_foid) << DNODE_SHIFT;
+
 	xva_init(&xva);
 	zfs_init_vattr(&xva.xva_vattr, ATTR_MODE | ATTR_UID | ATTR_GID,
-	    lr->lr_mode, lr->lr_uid, lr->lr_gid, lr->lr_rdev, lr->lr_foid);
+	    lr->lr_mode, lr->lr_uid, lr->lr_gid, lr->lr_rdev, objid);
 
 	/*
 	 * All forms of zfs create (create, mkdir, mkxattrdir, symlink)
 	 * eventually end up in zfs_mknode(), which assigns the object's
-	 * creation time and generation number.  The generic zfs_create()
-	 * doesn't have either concept, so we smuggle the values inside
-	 * the vattr's otherwise unused va_ctime and va_nblocks fields.
+	 * creation time, generation number, and dnode size. The generic
+	 * zfs_create() has no concept of these attributes, so we smuggle
+	 * the values inside the vattr's otherwise unused va_ctime,
+	 * va_nblocks, and va_fsid fields.
 	 */
 	ZFS_TIME_DECODE(&xva.xva_vattr.va_ctime, lr->lr_crtime);
 	xva.xva_vattr.va_nblocks = lr->lr_gen;
+	xva.xva_vattr.va_fsid = dnodesize;
 
 	error = dmu_object_info(zsb->z_os, lr->lr_foid, NULL);
 	if (error != ENOENT)
@@ -418,6 +425,8 @@ zfs_replay_create(zfs_sb_t *zsb, lr_create_t *lr, boolean_t byteswap)
 	void *start;
 	size_t xvatlen;
 	uint64_t txtype;
+	uint64_t objid;
+	uint64_t dnodesize;
 	int error;
 
 	txtype = (lr->lr_common.lrc_txtype & ~TX_CI);
@@ -431,21 +440,26 @@ zfs_replay_create(zfs_sb_t *zsb, lr_create_t *lr, boolean_t byteswap)
 	if ((error = zfs_zget(zsb, lr->lr_doid, &dzp)) != 0)
 		return (error);
 
+	objid = LR_FOID_GET_OBJ(lr->lr_foid);
+	dnodesize = LR_FOID_GET_SLOTS(lr->lr_foid) << DNODE_SHIFT;
+
 	xva_init(&xva);
 	zfs_init_vattr(&xva.xva_vattr, ATTR_MODE | ATTR_UID | ATTR_GID,
-	    lr->lr_mode, lr->lr_uid, lr->lr_gid, lr->lr_rdev, lr->lr_foid);
+	    lr->lr_mode, lr->lr_uid, lr->lr_gid, lr->lr_rdev, objid);
 
 	/*
 	 * All forms of zfs create (create, mkdir, mkxattrdir, symlink)
 	 * eventually end up in zfs_mknode(), which assigns the object's
-	 * creation time and generation number.  The generic zfs_create()
-	 * doesn't have either concept, so we smuggle the values inside
-	 * the vattr's otherwise unused va_ctime and va_nblocks fields.
+	 * creation time, generation number, and dnode slot count. The
+	 * generic zfs_create() has no concept of these attributes, so
+	 * we smuggle the values inside * the vattr's otherwise unused
+	 * va_ctime, va_nblocks, and va_nlink fields.
 	 */
 	ZFS_TIME_DECODE(&xva.xva_vattr.va_ctime, lr->lr_crtime);
 	xva.xva_vattr.va_nblocks = lr->lr_gen;
+	xva.xva_vattr.va_fsid = dnodesize;
 
-	error = dmu_object_info(zsb->z_os, lr->lr_foid, NULL);
+	error = dmu_object_info(zsb->z_os, objid, NULL);
 	if (error != ENOENT)
 		goto out;
 
diff --git a/module/zfs/zfs_sa.c b/module/zfs/zfs_sa.c
index f4841435b..f3eac51f8 100644
--- a/module/zfs/zfs_sa.c
+++ b/module/zfs/zfs_sa.c
@@ -97,8 +97,7 @@ zfs_sa_symlink(znode_t *zp, char *link, int len, dmu_tx_t *tx)
 	dmu_buf_t *db = sa_get_db(zp->z_sa_hdl);
 
 	if (ZFS_OLD_ZNODE_PHYS_SIZE + len <= dmu_bonus_max()) {
-		VERIFY(dmu_set_bonus(db,
-		    len + ZFS_OLD_ZNODE_PHYS_SIZE, tx) == 0);
+		VERIFY0(dmu_set_bonus(db, len + ZFS_OLD_ZNODE_PHYS_SIZE, tx));
 		if (len) {
 			bcopy(link, (caddr_t)db->db_data +
 			    ZFS_OLD_ZNODE_PHYS_SIZE, len);
@@ -107,8 +106,8 @@ zfs_sa_symlink(znode_t *zp, char *link, int len, dmu_tx_t *tx)
 		dmu_buf_t *dbp;
 
 		zfs_grow_blocksize(zp, len, tx);
-		VERIFY(0 == dmu_buf_hold(ZTOZSB(zp)->z_os,
-		    zp->z_id, 0, FTAG, &dbp, DMU_READ_NO_PREFETCH));
+		VERIFY0(dmu_buf_hold(ZTOZSB(zp)->z_os, zp->z_id, 0, FTAG, &dbp,
+		    DMU_READ_NO_PREFETCH));
 
 		dmu_buf_will_dirty(dbp, tx);
 
diff --git a/module/zfs/zfs_znode.c b/module/zfs/zfs_znode.c
index 19cb414a9..310d4827b 100644
--- a/module/zfs/zfs_znode.c
+++ b/module/zfs/zfs_znode.c
@@ -62,6 +62,7 @@
 
 #include <sys/dmu.h>
 #include <sys/dmu_objset.h>
+#include <sys/dmu_tx.h>
 #include <sys/refcount.h>
 #include <sys/stat.h>
 #include <sys/zap.h>
@@ -728,6 +729,7 @@ zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
 	timestruc_t	now;
 	uint64_t	gen, obj;
 	int		bonuslen;
+	int		dnodesize;
 	sa_handle_t	*sa_hdl;
 	dmu_object_type_t obj_type;
 	sa_bulk_attr_t	*sa_attrs;
@@ -739,15 +741,21 @@ zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
 		obj = vap->va_nodeid;
 		now = vap->va_ctime;		/* see zfs_replay_create() */
 		gen = vap->va_nblocks;		/* ditto */
+		dnodesize = vap->va_fsid;	/* ditto */
 	} else {
 		obj = 0;
 		gethrestime(&now);
 		gen = dmu_tx_get_txg(tx);
+		dnodesize = dmu_objset_dnodesize(zsb->z_os);
 	}
 
+	if (dnodesize == 0)
+		dnodesize = DNODE_MIN_SIZE;
+
 	obj_type = zsb->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
+
 	bonuslen = (obj_type == DMU_OT_SA) ?
-	    DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
+	    DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
 
 	/*
 	 * Create a new DMU object.
@@ -760,23 +768,23 @@ zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
 	 */
 	if (S_ISDIR(vap->va_mode)) {
 		if (zsb->z_replay) {
-			VERIFY0(zap_create_claim_norm(zsb->z_os, obj,
+			VERIFY0(zap_create_claim_norm_dnsize(zsb->z_os, obj,
 			    zsb->z_norm, DMU_OT_DIRECTORY_CONTENTS,
-			    obj_type, bonuslen, tx));
+			    obj_type, bonuslen, dnodesize, tx));
 		} else {
-			obj = zap_create_norm(zsb->z_os,
+			obj = zap_create_norm_dnsize(zsb->z_os,
 			    zsb->z_norm, DMU_OT_DIRECTORY_CONTENTS,
-			    obj_type, bonuslen, tx);
+			    obj_type, bonuslen, dnodesize, tx);
 		}
 	} else {
 		if (zsb->z_replay) {
-			VERIFY0(dmu_object_claim(zsb->z_os, obj,
+			VERIFY0(dmu_object_claim_dnsize(zsb->z_os, obj,
 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
-			    obj_type, bonuslen, tx));
+			    obj_type, bonuslen, dnodesize, tx));
 		} else {
-			obj = dmu_object_alloc(zsb->z_os,
+			obj = dmu_object_alloc_dnsize(zsb->z_os,
 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
-			    obj_type, bonuslen, tx);
+			    obj_type, bonuslen, dnodesize, tx);
 		}
 	}
 
@@ -948,6 +956,7 @@ zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
 
 	(*zpp)->z_pflags = pflags;
 	(*zpp)->z_mode = mode;
+	(*zpp)->z_dnodesize = dnodesize;
 
 	if (obj_type == DMU_OT_ZNODE ||
 	    acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
@@ -1767,6 +1776,14 @@ zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
 	ASSERT(error == 0);
 
 	/*
+	 * Give dmu_object_alloc() a hint about where to start
+	 * allocating new objects. Otherwise, since the metadnode's
+	 * dnode_phys_t structure isn't initialized yet, dmu_object_next()
+	 * would fail and we'd have to skip to the next dnode block.
+	 */
+	os->os_obj_next = moid + 1;
+
+	/*
 	 * Set starting attributes.
 	 */
 	version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
diff --git a/module/zfs/zil.c b/module/zfs/zil.c
index 289b23c7f..988ffec29 100644
--- a/module/zfs/zil.c
+++ b/module/zfs/zil.c
@@ -1372,7 +1372,8 @@ zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx)
 		itxg->itxg_sod += itx->itx_sod;
 	} else {
 		avl_tree_t *t = &itxs->i_async_tree;
-		uint64_t foid = ((lr_ooo_t *)&itx->itx_lr)->lr_foid;
+		uint64_t foid =
+		    LR_FOID_GET_OBJ(((lr_ooo_t *)&itx->itx_lr)->lr_foid);
 		itx_async_node_t *ian;
 		avl_index_t where;
 
@@ -1918,7 +1919,8 @@ zil_close(zilog_t *zilog)
 	mutex_exit(&zilog->zl_lock);
 	if (txg)
 		txg_wait_synced(zilog->zl_dmu_pool, txg);
-	ASSERT(!zilog_is_dirty(zilog));
+	if (txg < spa_freeze_txg(zilog->zl_spa))
+		ASSERT(!zilog_is_dirty(zilog));
 
 	taskq_destroy(zilog->zl_clean_taskq);
 	zilog->zl_clean_taskq = NULL;
@@ -2122,7 +2124,7 @@ zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg)
 	 */
 	if (TX_OOO(txtype)) {
 		error = dmu_object_info(zilog->zl_os,
-		    ((lr_ooo_t *)lr)->lr_foid, NULL);
+		    LR_FOID_GET_OBJ(((lr_ooo_t *)lr)->lr_foid), NULL);
 		if (error == ENOENT || error == EEXIST)
 			return (0);
 	}