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 <bass6@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #3542

cmd/zdb/zdb.c

@@ -1880,15 +1880,15 @@ dump_object(objset_t *os, uint64_t object, int verbosity, int *print_header)
 	dnode_t *dn;
 	void *bonus = NULL;
 	size_t bsize = 0;
-	char iblk[32], dblk[32], lsize[32], asize[32], fill[32];
+	char iblk[32], dblk[32], lsize[32], asize[32], fill[32], dnsize[32];
 	char bonus_size[32];
 	char aux[50];
 	int error;
 
 	if (*print_header) {
-		(void) printf("\n%10s  %3s  %5s  %5s  %5s  %5s  %6s  %s\n",
+		(void) printf("\n%10s  %3s  %5s  %5s  %5s  %6s  %5s  %6s  %s\n",
-		    "Object", "lvl", "iblk", "dblk", "dsize", "lsize",
+		    "Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
-		    "%full", "type");
+		    "lsize", "%full", "type");
 		*print_header = 0;
 	}
 
@@ -1910,6 +1910,7 @@ dump_object(objset_t *os, uint64_t object, int verbosity, int *print_header)
 	zdb_nicenum(doi.doi_max_offset, lsize);
 	zdb_nicenum(doi.doi_physical_blocks_512 << 9, asize);
 	zdb_nicenum(doi.doi_bonus_size, bonus_size);
+	zdb_nicenum(doi.doi_dnodesize, dnsize);
 	(void) sprintf(fill, "%6.2f", 100.0 * doi.doi_fill_count *
 	    doi.doi_data_block_size / (object == 0 ? DNODES_PER_BLOCK : 1) /
 	    doi.doi_max_offset);
@@ -1926,13 +1927,13 @@ dump_object(objset_t *os, uint64_t object, int verbosity, int *print_header)
 		    ZDB_COMPRESS_NAME(doi.doi_compress));
 	}
 
-	(void) printf("%10lld  %3u  %5s  %5s  %5s  %5s  %6s  %s%s\n",
+	(void) printf("%10lld  %3u  %5s  %5s  %5s  %6s  %5s  %6s  %s%s\n",
 	    (u_longlong_t)object, doi.doi_indirection, iblk, dblk,
-	    asize, lsize, fill, ZDB_OT_NAME(doi.doi_type), aux);
+	    asize, dnsize, lsize, fill, ZDB_OT_NAME(doi.doi_type), aux);
 
 	if (doi.doi_bonus_type != DMU_OT_NONE && verbosity > 3) {
-		(void) printf("%10s  %3s  %5s  %5s  %5s  %5s  %6s  %s\n",
+		(void) printf("%10s  %3s  %5s  %5s  %5s  %5s  %5s  %6s  %s\n",
-		    "", "", "", "", "", bonus_size, "bonus",
+		    "", "", "", "", "", "", bonus_size, "bonus",
 		    ZDB_OT_NAME(doi.doi_bonus_type));
 	}
 

cmd/zdb/zdb_il.c

@@ -80,8 +80,10 @@ zil_prt_rec_create(zilog_t *zilog, int txtype, lr_create_t *lr)
 	}
 
 	(void) printf("%s%s", prefix, ctime(&crtime));
-	(void) printf("%sdoid %llu, foid %llu, mode %llo\n", prefix,
+	(void) printf("%sdoid %llu, foid %llu, slots %llu, mode %llo\n", prefix,
-	    (u_longlong_t)lr->lr_doid, (u_longlong_t)lr->lr_foid,
+	    (u_longlong_t)lr->lr_doid,
+	    (u_longlong_t)LR_FOID_GET_OBJ(lr->lr_foid),
+	    (u_longlong_t)LR_FOID_GET_SLOTS(lr->lr_foid),
 	    (longlong_t)lr->lr_mode);
 	(void) printf("%suid %llu, gid %llu, gen %llu, rdev 0x%llx\n", prefix,
 	    (u_longlong_t)lr->lr_uid, (u_longlong_t)lr->lr_gid,

cmd/ztest/ztest.c

@@ -221,6 +221,7 @@ typedef struct ztest_block_tag {
 	uint64_t	bt_magic;
 	uint64_t	bt_objset;
 	uint64_t	bt_object;
+	uint64_t	bt_dnodesize;
 	uint64_t	bt_offset;
 	uint64_t	bt_gen;
 	uint64_t	bt_txg;
@@ -258,6 +259,7 @@ typedef struct ztest_od {
 	dmu_object_type_t od_crtype;
 	uint64_t	od_blocksize;
 	uint64_t	od_crblocksize;
+	uint64_t	od_crdnodesize;
 	uint64_t	od_gen;
 	uint64_t	od_crgen;
 	char		od_name[MAXNAMELEN];
@@ -329,6 +331,7 @@ ztest_func_t ztest_split_pool;
 ztest_func_t ztest_reguid;
 ztest_func_t ztest_spa_upgrade;
 ztest_func_t ztest_fletcher;
+ztest_func_t ztest_verify_dnode_bt;
 
 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
@@ -375,6 +378,7 @@ ztest_info_t ztest_info[] = {
 	ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
 	ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
 	ZTI_INIT(ztest_fletcher, 1, &zopt_rarely),
+	ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes),
 };
 
 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
@@ -1027,6 +1031,36 @@ ztest_random_blocksize(void)
 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
 }
 
+static int
+ztest_random_dnodesize(void)
+{
+	int slots;
+	int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
+
+	if (max_slots == DNODE_MIN_SLOTS)
+		return (DNODE_MIN_SIZE);
+
+	/*
+	 * Weight the random distribution more heavily toward smaller
+	 * dnode sizes since that is more likely to reflect real-world
+	 * usage.
+	 */
+	ASSERT3U(max_slots, >, 4);
+	switch (ztest_random(10)) {
+	case 0:
+		slots = 5 + ztest_random(max_slots - 4);
+		break;
+	case 1 ... 4:
+		slots = 2 + ztest_random(3);
+		break;
+	default:
+		slots = 1;
+		break;
+	}
+
+	return (slots << DNODE_SHIFT);
+}
+
 static int
 ztest_random_ibshift(void)
 {
@@ -1417,11 +1451,13 @@ ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
 
 static void
 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
-    uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
+    uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
+    uint64_t crtxg)
 {
 	bt->bt_magic = BT_MAGIC;
 	bt->bt_objset = dmu_objset_id(os);
 	bt->bt_object = object;
+	bt->bt_dnodesize = dnodesize;
 	bt->bt_offset = offset;
 	bt->bt_gen = gen;
 	bt->bt_txg = txg;
@@ -1430,11 +1466,13 @@ ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
 
 static void
 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
-    uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
+    uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
+    uint64_t crtxg)
 {
 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
 	ASSERT3U(bt->bt_object, ==, object);
+	ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
 	ASSERT3U(bt->bt_offset, ==, offset);
 	ASSERT3U(bt->bt_gen, <=, gen);
 	ASSERT3U(bt->bt_txg, <=, txg);
@@ -1455,6 +1493,52 @@ ztest_bt_bonus(dmu_buf_t *db)
 	return (bt);
 }
 
+/*
+ * Generate a token to fill up unused bonus buffer space.  Try to make
+ * it unique to the object, generation, and offset to verify that data
+ * is not getting overwritten by data from other dnodes.
+ */
+#define	ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
+	(((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
+
+/*
+ * Fill up the unused bonus buffer region before the block tag with a
+ * verifiable pattern. Filling the whole bonus area with non-zero data
+ * helps ensure that all dnode traversal code properly skips the
+ * interior regions of large dnodes.
+ */
+void
+ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
+    objset_t *os, uint64_t gen)
+{
+	uint64_t *bonusp;
+
+	ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
+
+	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
+		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
+		    gen, bonusp - (uint64_t *)db->db_data);
+		*bonusp = token;
+	}
+}
+
+/*
+ * Verify that the unused area of a bonus buffer is filled with the
+ * expected tokens.
+ */
+void
+ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
+    objset_t *os, uint64_t gen)
+{
+	uint64_t *bonusp;
+
+	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
+		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
+		    gen, bonusp - (uint64_t *)db->db_data);
+		VERIFY3U(*bonusp, ==, token);
+	}
+}
+
 /*
  * ZIL logging ops
  */
@@ -1463,7 +1547,7 @@ ztest_bt_bonus(dmu_buf_t *db)
 #define	lrz_blocksize	lr_uid
 #define	lrz_ibshift	lr_gid
 #define	lrz_bonustype	lr_rdev
-#define	lrz_bonuslen	lr_crtime[1]
+#define	lrz_dnodesize	lr_crtime[1]
 
 static void
 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
@@ -1578,6 +1662,7 @@ ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
 	dmu_tx_t *tx;
 	uint64_t txg;
 	int error = 0;
+	int bonuslen;
 
 	if (byteswap)
 		byteswap_uint64_array(lr, sizeof (*lr));
@@ -1600,26 +1685,27 @@ ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
 		return (ENOSPC);
 
 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
+	bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
 
 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
 		if (lr->lr_foid == 0) {
-			lr->lr_foid = zap_create(os,
+			lr->lr_foid = zap_create_dnsize(os,
 			    lr->lrz_type, lr->lrz_bonustype,
-			    lr->lrz_bonuslen, tx);
+			    bonuslen, lr->lrz_dnodesize, tx);
 		} else {
-			error = zap_create_claim(os, lr->lr_foid,
+			error = zap_create_claim_dnsize(os, lr->lr_foid,
 			    lr->lrz_type, lr->lrz_bonustype,
-			    lr->lrz_bonuslen, tx);
+			    bonuslen, lr->lrz_dnodesize, tx);
 		}
 	} else {
 		if (lr->lr_foid == 0) {
-			lr->lr_foid = dmu_object_alloc(os,
+			lr->lr_foid = dmu_object_alloc_dnsize(os,
 			    lr->lrz_type, 0, lr->lrz_bonustype,
-			    lr->lrz_bonuslen, tx);
+			    bonuslen, lr->lrz_dnodesize, tx);
 		} else {
-			error = dmu_object_claim(os, lr->lr_foid,
+			error = dmu_object_claim_dnsize(os, lr->lr_foid,
 			    lr->lrz_type, 0, lr->lrz_bonustype,
-			    lr->lrz_bonuslen, tx);
+			    bonuslen, lr->lrz_dnodesize, tx);
 		}
 	}
 
@@ -1639,7 +1725,9 @@ ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
 	bbt = ztest_bt_bonus(db);
 	dmu_buf_will_dirty(db, tx);
-	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
+	ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
+	    lr->lr_gen, txg, txg);
+	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
 	dmu_buf_rele(db, FTAG);
 
 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
@@ -1785,7 +1873,7 @@ ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
 			VERIFY(dmu_read(os, lr->lr_foid, offset,
 			    sizeof (rbt), &rbt, prefetch) == 0);
 			if (rbt.bt_magic == BT_MAGIC) {
-				ztest_bt_verify(&rbt, os, lr->lr_foid,
+				ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
 				    offset, gen, txg, crtxg);
 			}
 		}
@@ -1797,7 +1885,7 @@ ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
 		 * as it was when the write was generated.
 		 */
 		if (zd->zd_zilog->zl_replay) {
-			ztest_bt_verify(bt, os, lr->lr_foid, offset,
+			ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
 			    bt->bt_crtxg);
 		}
@@ -1806,7 +1894,8 @@ ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
 		 * so that all of the usual ASSERTs will work.
 		 */
-		ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
+		ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
+		    crtxg);
 	}
 
 	if (abuf == NULL) {
@@ -1874,7 +1963,7 @@ ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
 	dmu_tx_t *tx;
 	dmu_buf_t *db;
 	ztest_block_tag_t *bbt;
-	uint64_t txg, lrtxg, crtxg;
+	uint64_t txg, lrtxg, crtxg, dnodesize;
 
 	if (byteswap)
 		byteswap_uint64_array(lr, sizeof (*lr));
@@ -1897,6 +1986,7 @@ ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
 	crtxg = bbt->bt_crtxg;
 	lrtxg = lr->lr_common.lrc_txg;
+	dnodesize = bbt->bt_dnodesize;
 
 	if (zd->zd_zilog->zl_replay) {
 		ASSERT(lr->lr_size != 0);
@@ -1915,7 +2005,7 @@ ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
 	/*
 	 * Verify that the current bonus buffer is not newer than our txg.
 	 */
-	ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
+	ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
 	    MAX(txg, lrtxg), crtxg);
 
 	dmu_buf_will_dirty(db, tx);
@@ -1925,8 +2015,9 @@ ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
 	bbt = ztest_bt_bonus(db);
 
-	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
+	ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
-
+	    txg, crtxg);
+	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
 	dmu_buf_rele(db, FTAG);
 
 	(void) ztest_log_setattr(zd, tx, lr);
@@ -2171,7 +2262,7 @@ ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
 		lr->lrz_blocksize = od->od_crblocksize;
 		lr->lrz_ibshift = ztest_random_ibshift();
 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
-		lr->lrz_bonuslen = dmu_bonus_max();
+		lr->lrz_dnodesize = od->od_crdnodesize;
 		lr->lr_gen = od->od_crgen;
 		lr->lr_crtime[0] = time(NULL);
 
@@ -2351,7 +2442,8 @@ ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
 	switch (io_type) {
 
 	case ZTEST_IO_WRITE_TAG:
-		ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
+		ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
+		    offset, 0, 0, 0);
 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
 		break;
 
@@ -2414,13 +2506,15 @@ ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
  */
 static void
 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
-    dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
+    dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
+    uint64_t gen)
 {
 	od->od_dir = ZTEST_DIROBJ;
 	od->od_object = 0;
 
 	od->od_crtype = type;
 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
+	od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
 	od->od_crgen = gen;
 
 	od->od_type = DMU_OT_NONE;
@@ -3740,7 +3834,8 @@ ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
 	batchsize = OD_ARRAY_SIZE;
 
 	for (b = 0; b < batchsize; b++)
-		ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
+		ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER,
+		    0, 0, 0);
 
 	/*
 	 * Destroy the previous batch of objects, create a new batch,
@@ -3809,8 +3904,9 @@ ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
 	/*
 	 * Read the directory info.  If it's the first time, set things up.
 	 */
-	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
+	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize);
-	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
+	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
+	    chunksize);
 
 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
 		umem_free(od, size);
@@ -4090,8 +4186,9 @@ ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
 	/*
 	 * Read the directory info.  If it's the first time, set things up.
 	 */
-	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
+	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
-	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
+	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
+	    chunksize);
 
 
 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
@@ -4299,7 +4396,7 @@ ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
 	 * to verify that parallel writes to an object -- even to the
 	 * same blocks within the object -- doesn't cause any trouble.
 	 */
-	ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
+	ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
 
 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
 		return;
@@ -4322,7 +4419,7 @@ ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
 
 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
 
-	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
+	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
 
 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
 	    !ztest_random(2)) != 0) {
@@ -4375,7 +4472,7 @@ ztest_zap(ztest_ds_t *zd, uint64_t id)
 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
 
 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
-	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
+	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
 
 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
 			!ztest_random(2)) != 0)
@@ -4512,7 +4609,7 @@ ztest_fzap(ztest_ds_t *zd, uint64_t id)
 	int i;
 
 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
-	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
+	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
 
 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
 				!ztest_random(2)) != 0)
@@ -4561,7 +4658,7 @@ ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
 	void *data;
 
 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
-	ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
+	ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0);
 
 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
 		umem_free(od, sizeof (ztest_od_t));
@@ -4750,7 +4847,7 @@ ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
 	int i, error = 0;
 
 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
-	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
+	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
 
 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
 		umem_free(od, sizeof (ztest_od_t));
@@ -4871,6 +4968,41 @@ ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
 	umem_free(od, sizeof (ztest_od_t));
 }
 
+/*
+ * Visit each object in the dataset. Verify that its properties
+ * are consistent what was stored in the block tag when it was created,
+ * and that its unused bonus buffer space has not been overwritten.
+ */
+void
+ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
+{
+	objset_t *os = zd->zd_os;
+	uint64_t obj;
+	int err = 0;
+
+	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
+		ztest_block_tag_t *bt = NULL;
+		dmu_object_info_t doi;
+		dmu_buf_t *db;
+
+		if (dmu_bonus_hold(os, obj, FTAG, &db) != 0)
+			continue;
+
+		dmu_object_info_from_db(db, &doi);
+		if (doi.doi_bonus_size >= sizeof (*bt))
+			bt = ztest_bt_bonus(db);
+
+		if (bt && bt->bt_magic == BT_MAGIC) {
+			ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
+			    bt->bt_offset, bt->bt_gen, bt->bt_txg,
+			    bt->bt_crtxg);
+			ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
+		}
+
+		dmu_buf_rele(db, FTAG);
+	}
+}
+
 /* ARGSUSED */
 void
 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
@@ -5317,7 +5449,7 @@ ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
 
 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
-	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
+	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
 
 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
 		umem_free(od, sizeof (ztest_od_t));
@@ -6194,7 +6326,7 @@ ztest_freeze(void)
 	    numloops++ < ztest_opts.zo_maxloops &&
 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
 		ztest_od_t od;
-		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
+		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
 		ztest_io(zd, od.od_object,
 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);

configure.ac

@@ -230,6 +230,7 @@ AC_CONFIG_FILES([
 	tests/zfs-tests/tests/functional/devices/Makefile
 	tests/zfs-tests/tests/functional/exec/Makefile
 	tests/zfs-tests/tests/functional/features/async_destroy/Makefile
+	tests/zfs-tests/tests/functional/features/large_dnode/Makefile
 	tests/zfs-tests/tests/functional/features/Makefile
 	tests/zfs-tests/tests/functional/grow_pool/Makefile
 	tests/zfs-tests/tests/functional/grow_replicas/Makefile

include/sys/dmu.h

@@ -334,10 +334,19 @@ typedef struct dmu_buf {
  */
 uint64_t dmu_object_alloc(objset_t *os, dmu_object_type_t ot,
     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
+uint64_t dmu_object_alloc_dnsize(objset_t *os, dmu_object_type_t ot,
+    int blocksize, dmu_object_type_t bonus_type, int bonus_len,
+    int dnodesize, dmu_tx_t *tx);
 int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
     int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
+int dmu_object_claim_dnsize(objset_t *os, uint64_t object, dmu_object_type_t ot,
+    int blocksize, dmu_object_type_t bonus_type, int bonus_len,
+    int dnodesize, dmu_tx_t *tx);
 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 *txp);
+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 *txp);
 
 /*
  * Free an object from this objset.
@@ -756,6 +765,7 @@ typedef struct dmu_object_info {
 	uint8_t doi_compress;
 	uint8_t doi_nblkptr;
 	uint8_t doi_pad[4];
+	uint64_t doi_dnodesize;
 	uint64_t doi_physical_blocks_512;	/* data + metadata, 512b blks */
 	uint64_t doi_max_offset;
 	uint64_t doi_fill_count;		/* number of non-empty blocks */
@@ -797,6 +807,8 @@ void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
 void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
     u_longlong_t *nblk512);
 
+void dmu_object_dnsize_from_db(dmu_buf_t *db, int *dnsize);
+
 typedef struct dmu_objset_stats {
 	uint64_t dds_num_clones; /* number of clones of this */
 	uint64_t dds_creation_txg;
@@ -854,6 +866,7 @@ extern struct dsl_dataset *dmu_objset_ds(objset_t *os);
 extern void dmu_objset_name(objset_t *os, char *buf);
 extern dmu_objset_type_t dmu_objset_type(objset_t *os);
 extern uint64_t dmu_objset_id(objset_t *os);
+extern uint64_t dmu_objset_dnodesize(objset_t *os);
 extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
 extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
 extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,

include/sys/dmu_objset.h

@@ -88,6 +88,7 @@ struct objset {
 	list_node_t os_evicting_node;
 
 	/* can change, under dsl_dir's locks: */
+	uint64_t os_dnodesize; /* default dnode size for new objects */
 	enum zio_checksum os_checksum;
 	enum zio_compress os_compress;
 	uint8_t os_copies;

include/sys/dnode.h

@@ -79,11 +79,18 @@ extern "C" {
 /*
  * Derived constants.
  */
-#define	DNODE_SIZE	(1 << DNODE_SHIFT)
+#define	DNODE_MIN_SIZE		(1 << DNODE_SHIFT)
-#define	DN_MAX_NBLKPTR	((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
+#define	DNODE_MAX_SIZE		(1 << DNODE_BLOCK_SHIFT)
-#define	DN_MAX_BONUSLEN	(DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT))
+#define	DNODE_BLOCK_SIZE	(1 << DNODE_BLOCK_SHIFT)
+#define	DNODE_MIN_SLOTS		(DNODE_MIN_SIZE >> DNODE_SHIFT)
+#define	DNODE_MAX_SLOTS		(DNODE_MAX_SIZE >> DNODE_SHIFT)
+#define	DN_BONUS_SIZE(dnsize)	((dnsize) - DNODE_CORE_SIZE - \
+	(1 << SPA_BLKPTRSHIFT))
+#define	DN_SLOTS_TO_BONUSLEN(slots)	DN_BONUS_SIZE((slots) << DNODE_SHIFT)
+#define	DN_OLD_MAX_BONUSLEN	(DN_BONUS_SIZE(DNODE_MIN_SIZE))
+#define	DN_MAX_NBLKPTR	((DNODE_MIN_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT)
 #define	DN_MAX_OBJECT	(1ULL << DN_MAX_OBJECT_SHIFT)
-#define	DN_ZERO_BONUSLEN	(DN_MAX_BONUSLEN + 1)
+#define	DN_ZERO_BONUSLEN	(DN_BONUS_SIZE(DNODE_MAX_SIZE) + 1)
 #define	DN_KILL_SPILLBLK (1)
 
 #define	DNODES_PER_BLOCK_SHIFT	(DNODE_BLOCK_SHIFT - DNODE_SHIFT)
@@ -131,7 +138,8 @@ typedef struct dnode_phys {
 	uint8_t dn_flags;		/* DNODE_FLAG_* */
 	uint16_t dn_datablkszsec;	/* data block size in 512b sectors */
 	uint16_t dn_bonuslen;		/* length of dn_bonus */
-	uint8_t dn_pad2[4];
+	uint8_t dn_extra_slots;		/* # of subsequent slots consumed */
+	uint8_t dn_pad2[3];
 
 	/* accounting is protected by dn_dirty_mtx */
 	uint64_t dn_maxblkid;		/* largest allocated block ID */
@@ -140,8 +148,11 @@ typedef struct dnode_phys {
 	uint64_t dn_pad3[4];
 
 	/*
-	 * The tail region is 448 bytes, and there are three ways to
+	 * The tail region is 448 bytes for a 512 byte dnode, and
-	 * look at it.
+	 * correspondingly larger for larger dnode sizes. The spill
+	 * block pointer, when present, is always at the end of the tail
+	 * region. There are three ways this space may be used, using
+	 * a 512 byte dnode for this diagram:
 	 *
 	 * 0       64      128     192     256     320     384     448 (offset)
 	 * +---------------+---------------+---------------+-------+
@@ -149,23 +160,27 @@ typedef struct dnode_phys {
 	 * +---------------+---------------+---------------+-------+
 	 * | dn_blkptr[0]  | dn_bonus[0..319]                      |
 	 * +---------------+-----------------------+---------------+
-	 * | dn_blkptr[0]  | /                     | dn_spill      |
+	 * | dn_blkptr[0]  | dn_bonus[0..191]      | dn_spill      |
 	 * +---------------+-----------------------+---------------+
 	 */
 	union {
-		blkptr_t dn_blkptr[1+DN_MAX_BONUSLEN/sizeof (blkptr_t)];
+		blkptr_t dn_blkptr[1+DN_OLD_MAX_BONUSLEN/sizeof (blkptr_t)];
 		struct {
 			blkptr_t __dn_ignore1;
-			uint8_t dn_bonus[DN_MAX_BONUSLEN];
+			uint8_t dn_bonus[DN_OLD_MAX_BONUSLEN];
 		};
 		struct {
 			blkptr_t __dn_ignore2;
-			uint8_t __dn_ignore3[DN_MAX_BONUSLEN-sizeof (blkptr_t)];
+			uint8_t __dn_ignore3[DN_OLD_MAX_BONUSLEN -
+			    sizeof (blkptr_t)];
 			blkptr_t dn_spill;
 		};
 	};
 } dnode_phys_t;
 
+#define	DN_SPILL_BLKPTR(dnp)	(blkptr_t *)((char *)(dnp) + \
+	(((dnp)->dn_extra_slots + 1) << DNODE_SHIFT) - (1 << SPA_BLKPTRSHIFT))
+
 typedef struct dnode {
 	/*
 	 * Protects the structure of the dnode, including the number of levels
@@ -202,6 +217,7 @@ typedef struct dnode {
 	uint32_t dn_datablksz;		/* in bytes */
 	uint64_t dn_maxblkid;
 	uint8_t dn_next_type[TXG_SIZE];
+	uint8_t dn_num_slots;		/* metadnode slots consumed on disk */
 	uint8_t dn_next_nblkptr[TXG_SIZE];
 	uint8_t dn_next_nlevels[TXG_SIZE];
 	uint8_t dn_next_indblkshift[TXG_SIZE];
@@ -299,7 +315,7 @@ void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx);
 
 int dnode_hold(struct objset *dd, uint64_t object,
     void *ref, dnode_t **dnp);
-int dnode_hold_impl(struct objset *dd, uint64_t object, int flag,
+int dnode_hold_impl(struct objset *dd, uint64_t object, int flag, int dn_slots,
     void *ref, dnode_t **dnp);
 boolean_t dnode_add_ref(dnode_t *dn, void *ref);
 void dnode_rele(dnode_t *dn, void *ref);
@@ -307,9 +323,9 @@ void dnode_rele_and_unlock(dnode_t *dn, void *tag);
 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx);
 void dnode_sync(dnode_t *dn, dmu_tx_t *tx);
 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);
 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);
 void dnode_free(dnode_t *dn, dmu_tx_t *tx);
 void dnode_byteswap(dnode_phys_t *dnp);
 void dnode_buf_byteswap(void *buf, size_t size);

include/sys/dsl_dataset.h

@@ -91,6 +91,13 @@ struct dsl_pool;
  */
 #define	DS_FIELD_LARGE_BLOCKS "org.open-zfs:large_blocks"
 
+/*
+ * This field is present (with value=0) if this dataset may contain large
+ * dnodes (>512B).  If it is present, then this dataset is counted in the
+ * refcount of the SPA_FEATURE_LARGE_DNODE feature.
+ */
+#define	DS_FIELD_LARGE_DNODE "org.zfsonlinux:large_dnode"
+
 /*
  * DS_FLAG_CI_DATASET is set if the dataset contains a file system whose
  * name lookups should be performed case-insensitively.

include/sys/fs/zfs.h

@@ -137,6 +137,7 @@ typedef enum {
 	ZFS_PROP_DEDUP,
 	ZFS_PROP_MLSLABEL,
 	ZFS_PROP_SYNC,
+	ZFS_PROP_DNODESIZE,
 	ZFS_PROP_REFRATIO,
 	ZFS_PROP_WRITTEN,
 	ZFS_PROP_CLONES,
@@ -204,6 +205,7 @@ typedef enum {
 	ZPOOL_PROP_LEAKED,
 	ZPOOL_PROP_MAXBLOCKSIZE,
 	ZPOOL_PROP_TNAME,
+	ZPOOL_PROP_MAXDNODESIZE,
 	ZPOOL_NUM_PROPS
 } zpool_prop_t;
 
@@ -361,6 +363,16 @@ typedef enum {
 	ZFS_XATTR_SA = 2
 } zfs_xattr_type_t;
 
+typedef enum {
+	ZFS_DNSIZE_LEGACY = 0,
+	ZFS_DNSIZE_AUTO = 1,
+	ZFS_DNSIZE_1K = 1024,
+	ZFS_DNSIZE_2K = 2048,
+	ZFS_DNSIZE_4K = 4096,
+	ZFS_DNSIZE_8K = 8192,
+	ZFS_DNSIZE_16K = 16384
+} zfs_dnsize_type_t;
+
 typedef enum {
 	ZFS_REDUNDANT_METADATA_ALL,
 	ZFS_REDUNDANT_METADATA_MOST

include/sys/sa_impl.h

@@ -235,7 +235,7 @@ struct sa_handle {
 #define	SA_BONUSTYPE_FROM_DB(db) \
 	(dmu_get_bonustype((dmu_buf_t *)db))
 
-#define	SA_BLKPTR_SPACE	(DN_MAX_BONUSLEN - sizeof (blkptr_t))
+#define	SA_BLKPTR_SPACE	(DN_OLD_MAX_BONUSLEN - sizeof (blkptr_t))
 
 #define	SA_LAYOUT_NUM(x, type) \
 	((!IS_SA_BONUSTYPE(type) ? 0 : (((IS_SA_BONUSTYPE(type)) && \

include/sys/spa.h

@@ -822,6 +822,7 @@ extern boolean_t spa_is_root(spa_t *spa);
 extern boolean_t spa_writeable(spa_t *spa);
 extern boolean_t spa_has_pending_synctask(spa_t *spa);
 extern int spa_maxblocksize(spa_t *spa);
+extern int spa_maxdnodesize(spa_t *spa);
 extern void zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp);
 
 extern int spa_mode(spa_t *spa);

include/sys/zap.h

@@ -129,16 +129,30 @@ typedef enum zap_flags {
  *     MT_FIRST/MT_BEST matching will find entries that match without
  *     regard to case (eg. looking for "foo" can find an entry "Foo").
  * Eventually, other flags will permit unicode normalization as well.
+ *
+ * dnodesize specifies the on-disk size of the dnode for the new zapobj.
+ * Valid values are multiples of 512 up to DNODE_MAX_SIZE.
  */
 uint64_t zap_create(objset_t *ds, dmu_object_type_t ot,
     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
+uint64_t zap_create_dnsize(objset_t *ds, dmu_object_type_t ot,
+    dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx);
 uint64_t zap_create_norm(objset_t *ds, int normflags, dmu_object_type_t ot,
     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
+uint64_t zap_create_norm_dnsize(objset_t *ds, int normflags,
+    dmu_object_type_t ot, dmu_object_type_t bonustype, int bonuslen,
+    int dnodesize, dmu_tx_t *tx);
 uint64_t 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 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 zap_create_link(objset_t *os, dmu_object_type_t ot,
     uint64_t parent_obj, const char *name, dmu_tx_t *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);
 
 /*
  * Initialize an already-allocated object.
@@ -152,9 +166,14 @@ void mzap_create_impl(objset_t *os, uint64_t obj, int normflags,
  */
 int zap_create_claim(objset_t *ds, uint64_t obj, dmu_object_type_t ot,
     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
+int zap_create_claim_dnsize(objset_t *ds, uint64_t obj, dmu_object_type_t ot,
+    dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx);
 int zap_create_claim_norm(objset_t *ds, uint64_t obj,
     int normflags, dmu_object_type_t ot,
     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx);
+int zap_create_claim_norm_dnsize(objset_t *ds, uint64_t obj,
+    int normflags, dmu_object_type_t ot,
+    dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx);
 
 /*
  * The zapobj passed in must be a valid ZAP object for all of the

include/sys/zfs_ioctl.h

@@ -98,6 +98,7 @@ typedef enum drr_headertype {
 #define	DMU_BACKUP_FEATURE_EMBED_DATA_LZ4	(1<<17)
 /* flag #18 is reserved for a Delphix feature */
 #define	DMU_BACKUP_FEATURE_LARGE_BLOCKS		(1<<19)
+#define	DMU_BACKUP_FEATURE_LARGE_DNODE		(1<<20)
 
 /*
  * Mask of all supported backup features
@@ -105,7 +106,7 @@ typedef enum drr_headertype {
 #define	DMU_BACKUP_FEATURE_MASK	(DMU_BACKUP_FEATURE_DEDUP | \
     DMU_BACKUP_FEATURE_DEDUPPROPS | DMU_BACKUP_FEATURE_SA_SPILL | \
     DMU_BACKUP_FEATURE_EMBED_DATA | DMU_BACKUP_FEATURE_EMBED_DATA_LZ4 | \
-    DMU_BACKUP_FEATURE_LARGE_BLOCKS)
+    DMU_BACKUP_FEATURE_LARGE_BLOCKS | DMU_BACKUP_FEATURE_LARGE_DNODE)
 
 /* Are all features in the given flag word currently supported? */
 #define	DMU_STREAM_SUPPORTED(x)	(!((x) & ~DMU_BACKUP_FEATURE_MASK))
@@ -173,7 +174,8 @@ typedef struct dmu_replay_record {
 			uint32_t drr_bonuslen;
 			uint8_t drr_checksumtype;
 			uint8_t drr_compress;
-			uint8_t drr_pad[6];
+			uint8_t drr_dn_slots;
+			uint8_t drr_pad[5];
 			uint64_t drr_toguid;
 			/* bonus content follows */
 		} drr_object;

include/sys/zfs_znode.h

@@ -196,6 +196,7 @@ typedef struct znode {
 	uint_t		z_blksz;	/* block size in bytes */
 	uint_t		z_seq;		/* modification sequence number */
 	uint64_t	z_mapcnt;	/* number of pages mapped to file */
+	uint64_t	z_dnodesize;	/* dnode size */
 	uint64_t	z_size;		/* file size (cached) */
 	uint64_t	z_links;	/* file links (cached) */
 	uint64_t	z_pflags;	/* pflags (cached) */

include/sys/zil.h

@@ -172,6 +172,19 @@ typedef enum zil_create {
 	(txtype) == TX_ACL ||		\
 	(txtype) == TX_WRITE2)
 
+/*
+ * The number of dnode slots consumed by the object is stored in the 8
+ * unused upper bits of the object ID. We subtract 1 from the value
+ * stored on disk for compatibility with implementations that don't
+ * support large dnodes. The slot count for a single-slot dnode will
+ * contain 0 for those bits to preserve the log record format for
+ * "small" dnodes.
+ */
+#define	LR_FOID_GET_SLOTS(oid) (BF64_GET((oid), 56, 8) + 1)
+#define	LR_FOID_SET_SLOTS(oid, x) BF64_SET((oid), 56, 8, (x) - 1)
+#define	LR_FOID_GET_OBJ(oid) BF64_GET((oid), 0, DN_MAX_OBJECT_SHIFT)
+#define	LR_FOID_SET_OBJ(oid, x) BF64_SET((oid), 0, DN_MAX_OBJECT_SHIFT, (x))
+
 /*
  * Format of log records.
  * The fields are carefully defined to allow them to be aligned

include/zfeature_common.h

@@ -50,6 +50,7 @@ typedef enum spa_feature {
 	SPA_FEATURE_BOOKMARKS,
 	SPA_FEATURE_FS_SS_LIMIT,
 	SPA_FEATURE_LARGE_BLOCKS,
+	SPA_FEATURE_LARGE_DNODE,
 	SPA_FEATURES
 } spa_feature_t;
 

lib/libzfs/libzfs_dataset.c

@@ -1473,6 +1473,7 @@ zfs_setprop_error(libzfs_handle_t *hdl, zfs_prop_t prop, int err,
 
 	case ERANGE:
 		if (prop == ZFS_PROP_COMPRESSION ||
+		    prop == ZFS_PROP_DNODESIZE ||
 		    prop == ZFS_PROP_RECORDSIZE) {
 			(void) zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 			    "property setting is not allowed on "

man/man5/zpool-features.5

@@ -432,5 +432,30 @@ set larger than 128KB, and will return to being \fBenabled\fR once all
 filesystems that have ever had their recordsize larger than 128KB are destroyed.
 .RE
 
+.sp
+.ne 2
+.na
+\fB\fBlarge_dnode\fR\fR
+.ad
+.RS 4n
+.TS
+l l .
+GUID	org.zfsonlinux:large_dnode
+READ\-ONLY COMPATIBLE	no
+DEPENDENCIES	extensible_dataset
+.TE
+
+The \fBlarge_dnode\fR feature allows the size of dnodes in a dataset to be
+set larger than 512B.
+
+This feature becomes \fBactive\fR once a dataset contains an object with
+a dnode larger than 512B, which occurs as a result of setting the
+\fBdnodesize\fR dataset property to a value other than \fBlegacy\fR. The
+feature will return to being \fBenabled\fR once all filesystems that
+have ever contained a dnode larger than 512B are destroyed. Large dnodes
+allow more data to be stored in the bonus buffer, thus potentially
+improving performance by avoiding the use of spill blocks.
+.RE
+
 .SH "SEE ALSO"
 \fBzpool\fR(8)

man/man8/zfs.8

@@ -910,6 +910,35 @@ The values \fBon\fR and \fBoff\fR are equivalent to the \fBdev\fR and \fBnodev\f
 .sp
 .ne 2
 .na
+\fB\fBdnodesize\fR=\fBlegacy\fR | \fBauto\fR | \fB1k\fR | \fB2k\fR | \fB4k\fR | \fB8k\fR | \fB16k\fR\fR
+.ad
+.sp .6
+.RS 4n
+Specifies a compatibility mode or literal value for the size of dnodes
+in the file system. The default value is \fBlegacy\fR. Setting this
+property to a value other than \fBlegacy\fR requires the
+\fBlarge_dnode\fR pool feature to be enabled.
+.sp
+Consider setting \fBdnodesize\fR to \fBauto\fR if the dataset uses the
+\fBxattr=sa\fR property setting and the workload makes heavy use of
+extended attributes. This may be applicable to SELinux-enabled systems,
+Lustre servers, and Samba servers, for example. Literal values are
+supported for cases where the optimal size is known in advance and for
+performance testing.
+.sp
+Leave \fBdnodesize\fR set to \fBlegacy\fR if you need to receive
+a \fBzfs send\fR stream of this dataset on a pool that doesn't enable
+the \fBlarge_dnode\fR feature, or if you need to import this pool on a
+system that doesn't support the \fBlarge_dnode\fR feature.
+.sp
+This property can also be referred to by its shortened column name,
+\fBdnsize\fR.
+.RE
+
+.sp
+.ne 2
+.mk
+.na
 \fB\fBexec\fR=\fBon\fR | \fBoff\fR\fR
 .ad
 .sp .6

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,

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");
 }
 
 /*

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);
+		db->db.db_data = zio_buf_alloc(max_bonuslen);
-		arc_space_consume(DN_MAX_BONUSLEN, ARC_SPACE_OTHER);
+		arc_space_consume(max_bonuslen, ARC_SPACE_OTHER);
-		if (bonuslen < DN_MAX_BONUSLEN)
+		if (bonuslen < max_bonuslen)
-			bzero(db->db.db_data, DN_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);
+		dnode_t *dn = DB_DNODE(db);
-		arc_space_consume(DN_MAX_BONUSLEN, ARC_SPACE_OTHER);
+		int bonuslen = DN_SLOTS_TO_BONUSLEN(dn->dn_num_slots);
-		bcopy(db->db.db_data, dr->dt.dl.dr_data, DN_MAX_BONUSLEN);
+		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);
+			int slots = DB_DNODE(db)->dn_num_slots;
-			arc_space_return(DN_MAX_BONUSLEN, ARC_SPACE_OTHER);
+			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);
+			int slots = DB_DNODE(db)->dn_num_slots;
-			arc_space_return(DN_MAX_BONUSLEN, ARC_SPACE_OTHER);
+			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;
+			i = 0;
-			for (i = db->db.db_size >> DNODE_SHIFT; i > 0;
+			while (i < db->db.db_size) {
-			    i--, dnp++) {
+				dnode_phys_t *dnp = db->db.db_data + i;
-				if (dnp->dn_type != DMU_OT_NONE)
+
+				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

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);

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
+		 * objects), move to another L1 bp that's still
-		 * sparse (at most 1/4 full). Look from the beginning at most
+		 * reasonably sparse (at most 1/4 full). Look from the
-		 * once per txg, but after that keep looking from here.
+		 * 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
+		 * backfilling again.
-		 * keep going from here.
 		 *
 		 * 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,
+			    DNODE_FIND_HOLE, &offset, minlvl, blkfill, 0);
-			    &offset, 2, DNODES_PER_BLOCK >> 2, 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);
@@ -104,17 +140,34 @@ dmu_object_alloc(objset_t *os, dmu_object_type_t ot, int blocksize,
 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);
@@ -124,24 +177,35 @@ dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
 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);

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)
 {
@@ -267,6 +273,34 @@ redundant_metadata_changed_cb(void *arg, uint64_t newval)
 	os->os_redundant_metadata = 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)
 {
@@ -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,
+	dnode_allocate(mdn, DMU_OT_DNODE, DNODE_BLOCK_SIZE, DN_MAX_INDBLKSHIFT,
-	    DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
+	    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);

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);
+		dnobj = zb->zb_blkid * epb;
-		for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
+		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);
 }
 

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)) {

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,

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 = 0;
-	int i;
 
 	ASSERT3U(sizeof (dnode_phys_t), ==, (1<<DNODE_SHIFT));
 	ASSERT((size & (sizeof (dnode_phys_t)-1)) == 0);
 
-	size >>= DNODE_SHIFT;
+	while (i < size) {
-	for (i = 0; i < size; i++) {
+		dnode_phys_t *dnp = vbuf + i;
-		dnode_byteswap(buf);
+		dnode_byteswap(dnp);
-		buf++;
+
+		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,
+	dprintf("os=%p obj=%llu txg=%llu blocksize=%d ibs=%d dn_slots=%d\n",
-	    dn->dn_object, tx->tx_txg, blocksize, ibs);
+	    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
+	else {
-		dn->dn_nblkptr = 1 +
+		dn->dn_nblkptr = MIN(DN_MAX_NBLKPTR,
-		    ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT);
+		    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);
 	}
 
@@ -1052,25 +1071,151 @@ dnode_buf_pageout(void *dbu)
 	    children_dnodes->dnc_count * sizeof (dnode_handle_t));
 }
 
+/*
+ * 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) {
+	if (dn == NULL)
-		dnode_phys_t *phys = (dnode_phys_t *)db->db.db_data+idx;
+		dn = dnode_create(os, dn_block_begin + idx, db, object, dnh);
-
-		dn = dnode_create(os, phys, 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) && !refcount_is_zero(&dn->dn_holds))) {
-	    ((flag & DNODE_MUST_BE_FREE) &&
-	    (type != DMU_OT_NONE || !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);
 
-		for (i = (*offset >> span) & (blkfill - 1);
+		ASSERT(dn->dn_type == DMU_OT_DNODE);
-		    i >= 0 && i < blkfill; i += inc) {
+		ASSERT(!(flags & DNODE_FIND_BACKWARDS));
+
+		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;

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;

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);
 	}
 }

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,
+    dmu_buf_t *db, sa_buf_type_t buftype, int full_space, int *index,
-    boolean_t *will_spill)
+    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 ?
+	VERIFY0(dmu_set_bonus(hdl->sa_bonus, spilling ?
-	    MIN(DN_MAX_BONUSLEN - sizeof (blkptr_t), used + hdrsize) :
+	    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,
+		    attr_count - spill_idx, hdl->sa_spill, SA_SPILL,
-		    &spill_used, &dummy);
+		    hdl->sa_spill->db_size, &i, &spill_used, &dummy);
 
 		if (spill_used > SPA_OLD_MAXBLOCKSIZE)
 			return (SET_ERROR(EFBIG));

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);
 				}

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);

module/zfs/zap.c

@@ -967,10 +967,18 @@ fzap_prefetch(zap_name_t *zn)
 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));
 

module/zfs/zap_micro.c

@@ -630,18 +630,36 @@ 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,
+	return (zap_create_claim_dnsize(os, obj, ot, bonustype, bonuslen,
-	    0, ot, bonustype, bonuslen, tx));
+	    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
 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);
@@ -655,11 +673,28 @@ zap_create(objset_t *os, dmu_object_type_t ot,
 	return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
 }
 
+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);

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);
+	}
 }

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);

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));

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;

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()
+	 * creation time, generation number, and dnode size. The generic
-	 * doesn't have either concept, so we smuggle the values inside
+	 * zfs_create() has no concept of these attributes, so we smuggle
-	 * the vattr's otherwise unused va_ctime and va_nblocks fields.
+	 * 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()
+	 * creation time, generation number, and dnode slot count. The
-	 * doesn't have either concept, so we smuggle the values inside
+	 * generic zfs_create() has no concept of these attributes, so
-	 * the vattr's otherwise unused va_ctime and va_nblocks fields.
+	 * 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;
 

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,
+		VERIFY0(dmu_set_bonus(db, len + ZFS_OLD_ZNODE_PHYS_SIZE, tx));
-		    len + ZFS_OLD_ZNODE_PHYS_SIZE, tx) == 0);
 		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,
+		VERIFY0(dmu_buf_hold(ZTOZSB(zp)->z_os, zp->z_id, 0, FTAG, &dbp,
-		    zp->z_id, 0, FTAG, &dbp, DMU_READ_NO_PREFETCH));
+		    DMU_READ_NO_PREFETCH));
 
 		dmu_buf_will_dirty(dbp, tx);
 

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) {
@@ -1766,6 +1775,14 @@ zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
 	    DMU_OT_NONE, 0, 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.
 	 */

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;
 
@@ -1918,6 +1919,7 @@ zil_close(zilog_t *zilog)
 	mutex_exit(&zilog->zl_lock);
 	if (txg)
 		txg_wait_synced(zilog->zl_dmu_pool, txg);
+	if (txg < spa_freeze_txg(zilog->zl_spa))
 		ASSERT(!zilog_is_dirty(zilog));
 
 	taskq_destroy(zilog->zl_clean_taskq);
@@ -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);
 	}

tests/runfiles/linux.run

@@ -411,6 +411,11 @@ tests = ['exec_001_pos']
 [tests/functional/features/async_destroy]
 tests = ['async_destroy_001_pos']
 
+[tests/functional/features/large_dnode]
+tests = ['large_dnode_001_pos', 'large_dnode_002_pos', 'large_dnode_003_pos',
+         'large_dnode_004_neg', 'large_dnode_005_pos', 'large_dnode_006_pos',
+         'large_dnode_007_neg']
+
 # DISABLED: needs investigation
 #[tests/functional/grow_pool]
 #tests = ['grow_pool_001_pos']

tests/zfs-tests/include/default.cfg.in

@@ -80,7 +80,7 @@ export READMMAP=${READMMAP:-${helperdir}/readmmap}
 export RENAME_DIR=${RENAME_DIR:-${helperdir}/rename_dir}
 export RM_LNKCNT_ZERO_FILE=${RM_LNKCNT_ZERO_FILE:-${helperdir}/rm_lnkcnt_zero_file}
 export THREADSAPPEND=${THREADSAPPEND:-${helperdir}/threadsappend}
-export XATTRTEST=${XATTRTEST:-${helpdir}/xattrtest}
+export XATTRTEST=${XATTRTEST:-${helperdir}/xattrtest}
 
 # ensure we're running in the C locale, since
 # localised messages may result in test failures

tests/zfs-tests/tests/functional/cli_root/zpool_get/zpool_get.cfg

@@ -35,7 +35,7 @@ typeset -a properties=("size" "capacity" "altroot" "health" "guid" "version"
     "free" "allocated" "readonly" "comment" "expandsize" "freeing" "failmode"
     "listsnapshots" "autoexpand" "fragmentation" "leaked" "ashift"
     "feature@async_destroy" "feature@empty_bpobj" "feature@lz4_compress"
-    "feature@large_blocks" "feature@filesystem_limits"
+    "feature@large_blocks" "feature@large_dnode" "feature@filesystem_limits"
     "feature@spacemap_histogram" "feature@enabled_txg" "feature@hole_birth"
     "feature@extensible_dataset" "feature@bookmarks" "feature@embedded_data")
 else

tests/zfs-tests/tests/functional/delegate/delegate_common.kshlib

@@ -250,6 +250,10 @@ function check_fs_perm
 			verify_fs_canmount $user $perm $fs
 			ret=$?
 			;;
+		dnodesize)
+			verify_fs_dnodesize $user $perm $fs
+			ret=$?
+			;;
 		recordsize)
 			verify_fs_recordsize $user $perm $fs
 			ret=$?
@@ -1112,6 +1116,21 @@ function verify_fs_recordsize
 	return 0
 }
 
+function verify_fs_dnodesize
+{
+	typeset user=$1
+	typeset perm=$2
+	typeset fs=$3
+	value="2k"
+
+	user_run $user $ZFS set dnodesize=$value $fs
+	if [[ $value != $(get_prop dnodesize $fs) ]]; then
+		return 1
+	fi
+
+	return 0
+}
+
 function verify_fs_quota
 {
 	typeset user=$1

tests/zfs-tests/tests/functional/delegate/zfs_allow_010_pos.ksh

@@ -70,6 +70,7 @@ set -A perms	create		true		false	\
 		allow		true		true	\
 		quota		true		false	\
 		reservation	true		true	\
+		dnodesize	true		false	\
 		recordsize	true		false	\
 		checksum	true		true	\
 		compression	true		true	\
@@ -95,6 +96,7 @@ set -A perms	create		true		false	\
 		allow		true		true	\
 		quota		true		false	\
 		reservation	true		true	\
+		dnodesize	true		false	\
 		recordsize	true		false	\
 		mountpoint	true		false	\
 		checksum	true		true	\

tests/zfs-tests/tests/functional/delegate/zfs_allow_012_neg.ksh

@@ -59,13 +59,13 @@ if is_linux; then
 set -A perms	create snapshot mount send allow quota reservation \
 		recordsize mountpoint checksum compression canmount atime \
 		devices exec volsize setuid readonly snapdir userprop \
-		rollback clone rename promote \
+		rollback clone rename promote dnodesize \
 		zoned xattr receive destroy
 else
 set -A perms	create snapshot mount send allow quota reservation \
 		recordsize mountpoint checksum compression canmount atime \
 		devices exec volsize setuid readonly snapdir userprop \
-		aclmode aclinherit rollback clone rename promote \
+		aclmode aclinherit rollback clone rename promote dnodesize \
 		zoned xattr receive destroy sharenfs share
 fi
 

tests/zfs-tests/tests/functional/features/Makefile.am

@@ -1 +1,3 @@
-SUBDIRS = async_destroy
+SUBDIRS = \
+	async_destroy \
+	large_dnode

tests/zfs-tests/tests/functional/features/large_dnode/Makefile.am

@@ -0,0 +1,11 @@
+pkgdatadir = $(datadir)/@PACKAGE@/zfs-tests/tests/functional/features/large_dnode
+dist_pkgdata_SCRIPTS = \
+	cleanup.ksh \
+	setup.ksh \
+	large_dnode_001_pos.ksh \
+	large_dnode_002_pos.ksh \
+	large_dnode_003_pos.ksh \
+	large_dnode_004_neg.ksh \
+	large_dnode_005_pos.ksh \
+	large_dnode_006_pos.ksh \
+	large_dnode_007_neg.ksh

tests/zfs-tests/tests/functional/features/large_dnode/cleanup.ksh

@@ -0,0 +1,25 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+default_cleanup

tests/zfs-tests/tests/functional/features/large_dnode/large_dnode_001_pos.ksh

@@ -0,0 +1,77 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+#
+# DESCRIPTION:
+# Verify that the dnode sizes of newly created files are consistent
+# with the dnodesize dataset property.
+#
+# STRATEGY:
+# 1. Create a file system
+# 2. Set dnodesize to a legal literal value
+# 3. Create a file
+# 4. Repeat 2-3 for all legal literal values of dnodesize values
+# 5. Unmount the file system
+# 6. Use zdb to check expected dnode sizes
+#
+
+TEST_FS=$TESTPOOL/large_dnode
+
+verify_runnable "both"
+
+function cleanup
+{
+	datasetexists $TEST_FS && log_must $ZFS destroy $TEST_FS
+}
+
+log_onexit cleanup
+log_assert "dnode sizes are consistent with dnodesize dataset property"
+
+log_must $ZFS create $TEST_FS
+
+set -A dnsizes "512" "1k" "2k" "4k" "8k" "16k"
+set -A inodes
+
+for ((i=0; i < ${#dnsizes[*]}; i++)) ; do
+	size=${dnsizes[$i]}
+	if [[ $size == "512" ]] ; then
+		size="legacy"
+	fi
+	file=/$TEST_FS/file.$size
+	log_must $ZFS set dnsize=$size $TEST_FS
+	touch $file
+	inodes[$i]=$(ls -li $file | awk '{print $1}')
+done
+
+log_must $ZFS umount $TEST_FS
+
+for ((i=0; i < ${#dnsizes[*]}; i++)) ; do
+	dnsize=$($ZDB -dddd $TEST_FS ${inodes[$i]} |
+	    awk '/ZFS plain file/ {print $6}' | tr K k)
+	if [[ "$dnsize" != "${dnsizes[$i]}" ]]; then
+		log_fail "dnode size is $dnsize (expected ${dnsizes[$i]})"
+	fi
+done
+
+log_pass "dnode sizes are consistent with dnodesize dataset property"

tests/zfs-tests/tests/functional/features/large_dnode/large_dnode_002_pos.ksh

@@ -0,0 +1,78 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+#
+# DESCRIPTION:
+# Verify that extended attributes can use extra bonus space of a large
+# dnode without kicking in a spill block.
+#
+# STRATEGY:
+# 1. Create a file system with xattr=sa
+# 2. Set dnodesize to a legal literal value
+# 3. Create a file
+# 4  Store an xattr that fits within the dnode size
+# 4. Repeat 2-3 for all legal literal values of dnodesize values
+# 5. Unmount the file system
+# 6. Use zdb to check for missing SPILL_BLKPTR flag
+#
+
+TEST_FS=$TESTPOOL/large_dnode
+
+verify_runnable "both"
+
+function cleanup
+{
+	datasetexists $TEST_FS && log_must $ZFS destroy $TEST_FS
+}
+
+log_onexit cleanup
+log_assert "extended attributes use extra bonus space of a large dnode"
+
+log_must $ZFS create -o xattr=sa $TEST_FS
+
+# Store dnode size minus 512 in an xattr
+set -A xattr_sizes "512" "1536" "3584" "7680" "15872"
+set -A prop_values "1k"  "2k"   "4k"   "8k"   "16k"
+set -A inodes
+
+for ((i=0; i < ${#prop_values[*]}; i++)) ; do
+	prop_val=${prop_values[$i]}
+	file=/$TEST_FS/file.$prop_val
+	log_must $ZFS set dnsize=$prop_val $TEST_FS
+	touch $file
+	xattr_size=${xattr_sizes[$i]}
+	xattr_name=user.foo
+	xattr_val=$(dd if=/dev/urandom bs=1 count=$xattr_size |
+	    openssl enc -a -A)
+	log_must setfattr -n $xattr_name -v 0s$xattr_val $file
+	inodes[$i]=$(ls -li $file | awk '{print $1}')
+done
+
+log_must $ZFS umount $TEST_FS
+
+for ((i=0; i < ${#inodes[*]}; i++)) ; do
+	log_mustnot eval "$ZDB -dddd $TEST_FS ${inodes[$i]} | grep SPILL_BLKPTR"
+done
+
+log_pass "extended attributes use extra bonus space of a large dnode"

tests/zfs-tests/tests/functional/features/large_dnode/large_dnode_003_pos.ksh

@@ -0,0 +1,65 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+verify_runnable "both"
+
+function cleanup
+{
+        if datasetexists $LDNPOOL ; then
+                log_must $ZPOOL destroy -f $LDNPOOL
+        fi
+}
+
+log_onexit cleanup
+
+log_assert "feature correctly switches between enabled and active"
+
+LDNPOOL=ldnpool
+LDNFS=$LDNPOOL/large_dnode
+log_must $MKFILE 64M  $TESTDIR/$LDNPOOL
+log_must $ZPOOL create $LDNPOOL $TESTDIR/$LDNPOOL
+
+
+state=$($ZPOOL list -Ho feature@large_dnode $LDNPOOL)
+if [[ "$state" != "enabled" ]]; then
+        log_fail "large_dnode has state $state (expected enabled)"
+fi
+
+log_must $ZFS create -o dnodesize=1k $LDNFS
+log_must touch /$LDNFS/foo
+log_must $ZFS unmount $LDNFS
+
+state=$($ZPOOL list -Ho feature@large_dnode $LDNPOOL)
+if [[ "$state" != "active" ]]; then
+        log_fail "large_dnode has state $state (expected active)"
+fi
+
+log_must $ZFS destroy $LDNFS
+
+state=$($ZPOOL list -Ho feature@large_dnode $LDNPOOL)
+if [[ "$state" != "enabled" ]]; then
+        log_fail "large_dnode has state $state (expected enabled)"
+fi
+
+log_pass

tests/zfs-tests/tests/functional/features/large_dnode/large_dnode_004_neg.ksh

@@ -0,0 +1,59 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+verify_runnable "both"
+
+TEST_FS=$TESTPOOL/large_dnode
+TEST_SNAP=$TESTPOOL/large_dnode@ldnsnap
+TEST_STREAM=$TESTDIR/ldnsnap
+
+function cleanup
+{
+	if datasetexists $TEST_FS ; then
+		log_must $ZFS destroy -r $TEST_FS
+	fi
+
+	if datasetexists $LGCYPOOL ; then
+		log_must $ZPOOL destroy -f $LGCYPOOL
+	fi
+
+	rm -f $TEST_STREAM
+}
+
+log_onexit cleanup
+log_assert "zfs send stream with large dnodes not accepted by legacy pool"
+
+log_must $ZFS create -o dnodesize=1k $TEST_FS
+log_must touch /$TEST_FS/foo
+log_must $ZFS umount $TEST_FS
+log_must $ZFS snap $TEST_SNAP
+log_must eval "$ZFS send $TEST_SNAP > $TEST_STREAM"
+
+LGCYPOOL=ldnpool
+LGCYFS=$LGCYPOOL/legacy
+log_must $MKFILE 64M  $TESTDIR/$LGCYPOOL
+log_must $ZPOOL create -d $LGCYPOOL $TESTDIR/$LGCYPOOL
+log_mustnot eval "$ZFS recv $LGCYFS < $TEST_STREAM"
+
+log_pass

tests/zfs-tests/tests/functional/features/large_dnode/large_dnode_005_pos.ksh

@@ -0,0 +1,64 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+verify_runnable "both"
+
+TEST_SEND_FS=$TESTPOOL/send_large_dnode
+TEST_RECV_FS=$TESTPOOL/recv_large_dnode
+TEST_SNAP=$TEST_SEND_FS@ldnsnap
+TEST_STREAM=$TESTDIR/ldnsnap
+TEST_FILE=foo
+
+
+function cleanup
+{
+	if datasetexists $TEST_SEND_FS ; then
+		log_must $ZFS destroy -r $TEST_SEND_FS
+	fi
+
+	if datasetexists $TEST_RECV_FS ; then
+		log_must $ZFS destroy -r $TEST_RECV_FS
+	fi
+
+	rm -f $TEST_STREAM
+}
+
+log_onexit cleanup
+
+log_assert "zfs send stream with large dnodes accepted by new pool"
+
+log_must $ZFS create -o dnodesize=1k $TEST_SEND_FS
+log_must touch /$TEST_SEND_FS/$TEST_FILE
+log_must $ZFS umount $TEST_SEND_FS
+log_must $ZFS snap $TEST_SNAP
+log_must $ZFS send $TEST_SNAP > $TEST_STREAM
+
+log_must eval "$ZFS recv $TEST_RECV_FS < $TEST_STREAM"
+inode=$(ls -li /$TEST_RECV_FS/$TEST_FILE | awk '{print $1}')
+dnsize=$($ZDB -dddd $TEST_RECV_FS $inode | awk '/ZFS plain file/ {print $6}')
+if [[ "$dnsize" != "1K" ]]; then
+	log_fail "dnode size is $dnsize (expected 1K)"
+fi
+
+log_pass

tests/zfs-tests/tests/functional/features/large_dnode/large_dnode_006_pos.ksh

@@ -0,0 +1,58 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+#
+# DESCRIPTION:
+# Run xattrtest on a dataset with large dnodes and xattr=sa
+# to stress xattr usage of the extra bonus space and verify
+# contents
+#
+
+TEST_FS=$TESTPOOL/large_dnode
+
+verify_runnable "both"
+
+function cleanup
+{
+	datasetexists $TEST_FS && log_must $ZFS destroy $TEST_FS
+}
+
+log_onexit cleanup
+log_assert "xattrtest runs cleanly on dataset with large dnodes"
+
+log_must $ZFS create $TEST_FS
+
+set -A xattr_sizes "512" "1536" "3584" "7680" "15872"
+set -A prop_values "1k"  "2k"   "4k"   "8k"   "16k"
+
+for ((i=0; i < ${#prop_values[*]}; i++)) ; do
+	prop_val=${prop_values[$i]}
+	dir=/$TEST_FS/$prop_val
+	xattr_size=${xattr_sizes[$i]}
+	log_must $ZFS set dnsize=$prop_val $TEST_FS
+	log_must mkdir $dir
+	log_must $XATTRTEST -R -y -s $xattr_size -f 1024 -p $dir
+done
+
+log_pass

tests/zfs-tests/tests/functional/features/large_dnode/large_dnode_007_neg.ksh

@@ -0,0 +1,56 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+#
+# DESCRIPTION:
+# Verify that the dnodesize dataset property won't accept a value
+# other than "legacy" if the large_dnode feature is not enabled.
+#
+
+verify_runnable "both"
+
+function cleanup
+{
+        if datasetexists $LGCYPOOL ; then
+                log_must $ZPOOL destroy -f $LGCYPOOL
+        fi
+}
+
+log_onexit cleanup
+
+log_assert "values other than dnodesize=legacy rejected by legacy pool"
+
+set -A prop_vals "auto" "1k" "2k" "4k" "8k" "16k"
+
+LGCYPOOL=lgcypool
+LGCYFS=$LGCYPOOL/legacy
+log_must $MKFILE 64M  $TESTDIR/$LGCYPOOL
+log_must $ZPOOL create -d $LGCYPOOL $TESTDIR/$LGCYPOOL
+log_must $ZFS create $LGCYFS
+
+for val in ${prop_vals[@]} ; do
+	log_mustnot $ZFS set dnodesize=$val $LGCYFS
+done
+
+log_pass

tests/zfs-tests/tests/functional/features/large_dnode/setup.ksh

@@ -0,0 +1,27 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+DISK=${DISKS%% *}
+
+default_setup $DISK

tests/zfs-tests/tests/functional/rsend/rsend.kshlib

@@ -201,8 +201,8 @@ function cmp_ds_prop
 
 	for item in "type" "origin" "volblocksize" "aclinherit" "aclmode" \
 	    "atime" "canmount" "checksum" "compression" "copies" "devices" \
-	    "exec" "quota" "readonly" "recordsize" "reservation" "setuid" \
+	    "dnodesize" "exec" "quota" "readonly" "recordsize" "reservation" \
-	    "sharenfs" "snapdir" "version" "volsize" "xattr" "zoned" \
+	    "setuid" "sharenfs" "snapdir" "version" "volsize" "xattr" "zoned" \
 	    "mountpoint";
 	do
 		$ZFS get -H -o property,value,source $item $dtst1 >> \

tests/zfs-tests/tests/functional/rsend/rsend_012_pos.ksh

@@ -142,6 +142,7 @@ for fs in "$POOL" "$POOL/pclone" "$POOL/$FS" "$POOL/$FS/fs1" \
 	rand_set_prop $fs exec "on" "off"
 	rand_set_prop $fs quota "512M" "1024M"
 	rand_set_prop $fs recordsize "512" "2K" "8K" "32K" "128K"
+	rand_set_prop $fs dnodesize "legacy" "auto" "1k" "2k" "4k" "8k" "16k"
 	rand_set_prop $fs setuid "on" "off"
 	rand_set_prop $fs snapdir "hidden" "visible"
 	rand_set_prop $fs xattr "on" "off"

zfs-script-config.sh.in

@@ -59,7 +59,7 @@ export READMMAP=${TESTSDIR}/zfs-tests/cmd/readmmap/readmmap
 export RENAME_DIR=${TESTSDIR}/zfs-tests/cmd/rename_dir/rename_dir
 export RM_LNKCNT_ZERO_FILE=${TESTSDIR}/zfs-tests/cmd/rm_lnkcnt_zero_file/rm_lnkcnt_zero_file
 export THREADSAPPEND=${TESTSDIR}/zfs-tests/cmd/threadsappend/threadsappend
-export XATTRTEST=${TESTDIR}/zfs-tests/cmd/xattrtest
+export XATTRTEST=${TESTSDIR}/zfs-tests/cmd/xattrtest/xattrtest
 
 export INTREE=1
 export LDMOD=/sbin/insmod