4757 ZFS embedded-data block pointers ("zero block compression")

4913 zfs release should not be subject to space checks

Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Max Grossman <max.grossman@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Dan McDonald <danmcd@omniti.com>
Approved by: Dan McDonald <danmcd@omniti.com>

illumos/illumos-dist@5d7b4d438c

cmd/zdb/zdb.c

@@ -1032,8 +1032,17 @@ snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp)
 		return;
 	}
 
-	blkbuf[0] = '\0';
+	if (BP_IS_EMBEDDED(bp)) {
+		(void) sprintf(blkbuf,
+		    "EMBEDDED et=%u %llxL/%llxP B=%llu",
+		    (int)BPE_GET_ETYPE(bp),
+		    (u_longlong_t)BPE_GET_LSIZE(bp),
+		    (u_longlong_t)BPE_GET_PSIZE(bp),
+		    (u_longlong_t)bp->blk_birth);
+		return;
+	}
 
+	blkbuf[0] = '\0';
 	for (int i = 0; i < ndvas; i++)
 		(void) snprintf(blkbuf + strlen(blkbuf),
 		    buflen - strlen(blkbuf), "%llu:%llx:%llx ",
@@ -1051,7 +1060,7 @@ snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp)
 		    "%llxL/%llxP F=%llu B=%llu/%llu",
 		    (u_longlong_t)BP_GET_LSIZE(bp),
 		    (u_longlong_t)BP_GET_PSIZE(bp),
-		    (u_longlong_t)bp->blk_fill,
+		    (u_longlong_t)BP_GET_FILL(bp),
 		    (u_longlong_t)bp->blk_birth,
 		    (u_longlong_t)BP_PHYSICAL_BIRTH(bp));
 	}
@@ -1064,8 +1073,10 @@ print_indirect(blkptr_t *bp, const zbookmark_t *zb,
 	char blkbuf[BP_SPRINTF_LEN];
 	int l;
 
-	ASSERT3U(BP_GET_TYPE(bp), ==, dnp->dn_type);
-	ASSERT3U(BP_GET_LEVEL(bp), ==, zb->zb_level);
+	if (!BP_IS_EMBEDDED(bp)) {
+		ASSERT3U(BP_GET_TYPE(bp), ==, dnp->dn_type);
+		ASSERT3U(BP_GET_LEVEL(bp), ==, zb->zb_level);
+	}
 
 	(void) printf("%16llx ", (u_longlong_t)blkid2offset(dnp, bp, zb));
 
@@ -1119,10 +1130,10 @@ visit_indirect(spa_t *spa, const dnode_phys_t *dnp,
 			err = visit_indirect(spa, dnp, cbp, &czb);
 			if (err)
 				break;
-			fill += cbp->blk_fill;
+			fill += BP_GET_FILL(cbp);
 		}
 		if (!err)
-			ASSERT3U(fill, ==, bp->blk_fill);
+			ASSERT3U(fill, ==, BP_GET_FILL(bp));
 		(void) arc_buf_remove_ref(buf, &buf);
 	}
 
@@ -1789,14 +1800,14 @@ dump_dir(objset_t *os)
 
 	if (dds.dds_type == DMU_OST_META) {
 		dds.dds_creation_txg = TXG_INITIAL;
-		usedobjs = os->os_rootbp->blk_fill;
+		usedobjs = BP_GET_FILL(os->os_rootbp);
 		refdbytes = os->os_spa->spa_dsl_pool->
 		    dp_mos_dir->dd_phys->dd_used_bytes;
 	} else {
 		dmu_objset_space(os, &refdbytes, &scratch, &usedobjs, &scratch);
 	}
 
-	ASSERT3U(usedobjs, ==, os->os_rootbp->blk_fill);
+	ASSERT3U(usedobjs, ==, BP_GET_FILL(os->os_rootbp));
 
 	zdb_nicenum(refdbytes, numbuf);
 
@@ -2107,6 +2118,9 @@ typedef struct zdb_cb {
 	zdb_blkstats_t	zcb_type[ZB_TOTAL + 1][ZDB_OT_TOTAL + 1];
 	uint64_t	zcb_dedup_asize;
 	uint64_t	zcb_dedup_blocks;
+	uint64_t	zcb_embedded_blocks[NUM_BP_EMBEDDED_TYPES];
+	uint64_t	zcb_embedded_histogram[NUM_BP_EMBEDDED_TYPES]
+	    [BPE_PAYLOAD_SIZE];
 	uint64_t	zcb_start;
 	uint64_t	zcb_lastprint;
 	uint64_t	zcb_totalasize;
@@ -2161,6 +2175,13 @@ zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
 
 	}
 
+	if (BP_IS_EMBEDDED(bp)) {
+		zcb->zcb_embedded_blocks[BPE_GET_ETYPE(bp)]++;
+		zcb->zcb_embedded_histogram[BPE_GET_ETYPE(bp)]
+		    [BPE_GET_PSIZE(bp)]++;
+		return;
+	}
+
 	if (dump_opt['L'])
 		return;
 
@@ -2258,7 +2279,8 @@ zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 
 	is_metadata = (BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type));
 
-	if (dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata)) {
+	if (!BP_IS_EMBEDDED(bp) &&
+	    (dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata))) {
 		size_t size = BP_GET_PSIZE(bp);
 		void *data = zio_data_buf_alloc(size);
 		int flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW;
@@ -2450,7 +2472,7 @@ dump_block_stats(spa_t *spa)
 	zdb_blkstats_t *zb, *tzb;
 	uint64_t norm_alloc, norm_space, total_alloc, total_found;
 	int flags = TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA | TRAVERSE_HARD;
-	int leaks = 0;
+	boolean_t leaks = B_FALSE;
 
 	(void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
 	    (dump_opt['c'] || !dump_opt['L']) ? "to verify " : "",
@@ -2538,7 +2560,7 @@ dump_block_stats(spa_t *spa)
 		    (u_longlong_t)total_alloc,
 		    (dump_opt['L']) ? "unreachable" : "leaked",
 		    (longlong_t)(total_alloc - total_found));
-		leaks = 1;
+		leaks = B_TRUE;
 	}
 
 	if (tzb->zb_count == 0)
@@ -2570,6 +2592,23 @@ dump_block_stats(spa_t *spa)
 	(void) printf("\tSPA allocated: %10llu     used: %5.2f%%\n",
 	    (u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space);
 
+	for (bp_embedded_type_t i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) {
+		if (zcb.zcb_embedded_blocks[i] == 0)
+			continue;
+		(void) printf("\n");
+		(void) printf("\tadditional, non-pointer bps of type %u: "
+		    "%10llu\n",
+		    i, (u_longlong_t)zcb.zcb_embedded_blocks[i]);
+
+		if (dump_opt['b'] >= 3) {
+			(void) printf("\t number of (compressed) bytes:  "
+			    "number of bps\n");
+			dump_histogram(zcb.zcb_embedded_histogram[i],
+			    sizeof (zcb.zcb_embedded_histogram[i]) /
+			    sizeof (zcb.zcb_embedded_histogram[i][0]), 0);
+		}
+	}
+
 	if (tzb->zb_ditto_samevdev != 0) {
 		(void) printf("\tDittoed blocks on same vdev: %llu\n",
 		    (longlong_t)tzb->zb_ditto_samevdev);
@@ -2682,14 +2721,14 @@ zdb_ddt_add_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 	avl_index_t where;
 	zdb_ddt_entry_t *zdde, zdde_search;
 
-	if (BP_IS_HOLE(bp))
+	if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
 		return (0);
 
 	if (dump_opt['S'] > 1 && zb->zb_level == ZB_ROOT_LEVEL) {
 		(void) printf("traversing objset %llu, %llu objects, "
 		    "%lu blocks so far\n",
 		    (u_longlong_t)zb->zb_objset,
-		    (u_longlong_t)bp->blk_fill,
+		    (u_longlong_t)BP_GET_FILL(bp),
 		    avl_numnodes(t));
 	}
 

cmd/zfs/zfs_main.c

@@ -256,9 +256,9 @@ get_usage(zfs_help_t idx)
 	case HELP_ROLLBACK:
 		return (gettext("\trollback [-rRf] <snapshot>\n"));
 	case HELP_SEND:
-		return (gettext("\tsend [-DnPpRv] [-[iI] snapshot] "
+		return (gettext("\tsend [-DnPpRve] [-[iI] snapshot] "
 		    "<snapshot>\n"
-		    "\tsend [-i snapshot|bookmark] "
+		    "\tsend [-e] [-i snapshot|bookmark] "
 		    "<filesystem|volume|snapshot>\n"));
 	case HELP_SET:
 		return (gettext("\tset <property=value> "
@@ -573,6 +573,7 @@ finish_progress(char *done)
 	free(pt_header);
 	pt_header = NULL;
 }
+
 /*
  * zfs clone [-p] [-o prop=value] ... <snap> <fs | vol>
  *
@@ -3299,6 +3300,7 @@ rollback_check_dependent(zfs_handle_t *zhp, void *data)
 	zfs_close(zhp);
 	return (0);
 }
+
 /*
  * Report any snapshots more recent than the one specified.  Used when '-r' is
  * not specified.  We reuse this same callback for the snapshot dependents - if
@@ -3638,7 +3640,7 @@ zfs_do_send(int argc, char **argv)
 	boolean_t extraverbose = B_FALSE;
 
 	/* check options */
-	while ((c = getopt(argc, argv, ":i:I:RDpvnP")) != -1) {
+	while ((c = getopt(argc, argv, ":i:I:RDpvnPe")) != -1) {
 		switch (c) {
 		case 'i':
 			if (fromname)
@@ -3673,6 +3675,9 @@ zfs_do_send(int argc, char **argv)
 		case 'n':
 			flags.dryrun = B_TRUE;
 			break;
+		case 'e':
+			flags.embed_data = B_TRUE;
+			break;
 		case ':':
 			(void) fprintf(stderr, gettext("missing argument for "
 			    "'%c' option\n"), optopt);
@@ -3711,6 +3716,7 @@ zfs_do_send(int argc, char **argv)
 	if (strchr(argv[0], '@') == NULL ||
 	    (fromname && strchr(fromname, '#') != NULL)) {
 		char frombuf[ZFS_MAXNAMELEN];
+		enum lzc_send_flags lzc_flags = 0;
 
 		if (flags.replicate || flags.doall || flags.props ||
 		    flags.dedup || flags.dryrun || flags.verbose ||
@@ -3725,6 +3731,9 @@ zfs_do_send(int argc, char **argv)
 		if (zhp == NULL)
 			return (1);
 
+		if (flags.embed_data)
+			lzc_flags |= LZC_SEND_FLAG_EMBED_DATA;
+
 		if (fromname != NULL &&
 		    (fromname[0] == '#' || fromname[0] == '@')) {
 			/*
@@ -3738,7 +3747,7 @@ zfs_do_send(int argc, char **argv)
 			(void) strlcat(frombuf, fromname, sizeof (frombuf));
 			fromname = frombuf;
 		}
-		err = zfs_send_one(zhp, fromname, STDOUT_FILENO);
+		err = zfs_send_one(zhp, fromname, STDOUT_FILENO, lzc_flags);
 		zfs_close(zhp);
 		return (err != 0);
 	}

cmd/zstreamdump/zstreamdump.c

@@ -49,7 +49,6 @@
  */
 #define	DUMP_GROUPING	4
 
-uint64_t drr_record_count[DRR_NUMTYPES];
 uint64_t total_write_size = 0;
 uint64_t total_stream_len = 0;
 FILE *send_stream = 0;
@@ -123,7 +122,7 @@ print_block(char *buf, int length)
 	 * Start printing ASCII characters at a constant offset, after
 	 * the hex prints. Leave 3 characters per byte on a line (2 digit
 	 * hex number plus 1 space) plus spaces between characters and
-	 * groupings
+	 * groupings.
 	 */
 	int ascii_start = BYTES_PER_LINE * 3 +
 	    BYTES_PER_LINE / DUMP_GROUPING + 2;
@@ -160,6 +159,8 @@ int
 main(int argc, char *argv[])
 {
 	char *buf = malloc(INITIAL_BUFLEN);
+	uint64_t drr_record_count[DRR_NUMTYPES] = { 0 };
+	uint64_t total_records = 0;
 	dmu_replay_record_t thedrr;
 	dmu_replay_record_t *drr = &thedrr;
 	struct drr_begin *drrb = &thedrr.drr_u.drr_begin;
@@ -170,6 +171,7 @@ main(int argc, char *argv[])
 	struct drr_write_byref *drrwbr = &thedrr.drr_u.drr_write_byref;
 	struct drr_free *drrf = &thedrr.drr_u.drr_free;
 	struct drr_spill *drrs = &thedrr.drr_u.drr_spill;
+	struct drr_write_embedded *drrwe = &thedrr.drr_u.drr_write_embedded;
 	char c;
 	boolean_t verbose = B_FALSE;
 	boolean_t first = B_TRUE;
@@ -264,6 +266,7 @@ main(int argc, char *argv[])
 		}
 
 		drr_record_count[drr->drr_type]++;
+		total_records++;
 
 		switch (drr->drr_type) {
 		case DRR_BEGIN:
@@ -376,8 +379,8 @@ main(int argc, char *argv[])
 				    drro->drr_bonuslen);
 			}
 			if (drro->drr_bonuslen > 0) {
-				(void) ssread(buf, P2ROUNDUP(drro->drr_bonuslen,
-				    8), &zc);
+				(void) ssread(buf,
+				    P2ROUNDUP(drro->drr_bonuslen, 8), &zc);
 				if (dump) {
 					print_block(buf,
 					    P2ROUNDUP(drro->drr_bonuslen, 8));
@@ -506,6 +509,38 @@ main(int argc, char *argv[])
 				print_block(buf, drrs->drr_length);
 			}
 			break;
+		case DRR_WRITE_EMBEDDED:
+			if (do_byteswap) {
+				drrwe->drr_object =
+				    BSWAP_64(drrwe->drr_object);
+				drrwe->drr_offset =
+				    BSWAP_64(drrwe->drr_offset);
+				drrwe->drr_length =
+				    BSWAP_64(drrwe->drr_length);
+				drrwe->drr_toguid =
+				    BSWAP_64(drrwe->drr_toguid);
+				drrwe->drr_lsize =
+				    BSWAP_32(drrwe->drr_lsize);
+				drrwe->drr_psize =
+				    BSWAP_32(drrwe->drr_psize);
+			}
+			if (verbose) {
+				(void) printf("WRITE_EMBEDDED object = %llu "
+				    "offset = %llu length = %llu\n"
+				    "toguid = %llx comp = %u etype = %u "
+				    "lsize = %u psize = %u\n",
+				    (u_longlong_t)drrwe->drr_object,
+				    (u_longlong_t)drrwe->drr_offset,
+				    (u_longlong_t)drrwe->drr_length,
+				    (u_longlong_t)drrwe->drr_toguid,
+				    drrwe->drr_compression,
+				    drrwe->drr_etype,
+				    drrwe->drr_lsize,
+				    drrwe->drr_psize);
+			}
+			(void) ssread(buf,
+			    P2ROUNDUP(drrwe->drr_psize, 8), &zc);
+			break;
 		}
 		pcksum = zc;
 	}
@@ -524,18 +559,16 @@ main(int argc, char *argv[])
 	    (u_longlong_t)drr_record_count[DRR_FREEOBJECTS]);
 	(void) printf("\tTotal DRR_WRITE records = %lld\n",
 	    (u_longlong_t)drr_record_count[DRR_WRITE]);
+	(void) printf("\tTotal DRR_WRITE_BYREF records = %lld\n",
+	    (u_longlong_t)drr_record_count[DRR_WRITE_BYREF]);
+	(void) printf("\tTotal DRR_WRITE_EMBEDDED records = %lld\n",
+	    (u_longlong_t)drr_record_count[DRR_WRITE_EMBEDDED]);
 	(void) printf("\tTotal DRR_FREE records = %lld\n",
 	    (u_longlong_t)drr_record_count[DRR_FREE]);
 	(void) printf("\tTotal DRR_SPILL records = %lld\n",
 	    (u_longlong_t)drr_record_count[DRR_SPILL]);
 	(void) printf("\tTotal records = %lld\n",
-	    (u_longlong_t)(drr_record_count[DRR_BEGIN] +
-	    drr_record_count[DRR_OBJECT] +
-	    drr_record_count[DRR_FREEOBJECTS] +
-	    drr_record_count[DRR_WRITE] +
-	    drr_record_count[DRR_FREE] +
-	    drr_record_count[DRR_SPILL] +
-	    drr_record_count[DRR_END]));
+	    (u_longlong_t)total_records);
 	(void) printf("\tTotal write size = %lld (0x%llx)\n",
 	    (u_longlong_t)total_write_size, (u_longlong_t)total_write_size);
 	(void) printf("\tTotal stream length = %lld (0x%llx)\n",

cmd/ztest/ztest.c

@@ -52,7 +52,7 @@
  *     At random times, the child self-immolates with a SIGKILL.
  *     This is the software equivalent of pulling the power cord.
  *     The parent then runs the test again, using the existing
- *     storage pool, as many times as desired. If backwards compatability
+ *     storage pool, as many times as desired. If backwards compatibility
  *     testing is enabled ztest will sometimes run the "older" version
  *     of ztest after a SIGKILL.
  *
@@ -1265,13 +1265,13 @@ 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)
 {
-	ASSERT(bt->bt_magic == BT_MAGIC);
-	ASSERT(bt->bt_objset == dmu_objset_id(os));
-	ASSERT(bt->bt_object == object);
-	ASSERT(bt->bt_offset == offset);
-	ASSERT(bt->bt_gen <= gen);
-	ASSERT(bt->bt_txg <= txg);
-	ASSERT(bt->bt_crtxg == crtxg);
+	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
+	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
+	ASSERT3U(bt->bt_object, ==, object);
+	ASSERT3U(bt->bt_offset, ==, offset);
+	ASSERT3U(bt->bt_gen, <=, gen);
+	ASSERT3U(bt->bt_txg, <=, txg);
+	ASSERT3U(bt->bt_crtxg, ==, crtxg);
 }
 
 static ztest_block_tag_t *
@@ -3470,6 +3470,11 @@ ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
 	if (error)
 		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
 	error = dsl_dataset_promote(clone2name, NULL);
+	if (error == ENOSPC) {
+		dmu_objset_disown(os, FTAG);
+		ztest_record_enospc(FTAG);
+		goto out;
+	}
 	if (error != EBUSY)
 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
 		    error);
@@ -3625,11 +3630,19 @@ ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
 		return;
 	}
 
-	dmu_object_set_checksum(os, bigobj,
-	    (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
+	enum zio_checksum cksum;
+	do {
+		cksum = (enum zio_checksum)
+		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
+	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
+	dmu_object_set_checksum(os, bigobj, cksum, tx);
 
-	dmu_object_set_compress(os, bigobj,
-	    (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
+	enum zio_compress comp;
+	do {
+		comp = (enum zio_compress)
+		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
+	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
+	dmu_object_set_compress(os, bigobj, comp, tx);
 
 	/*
 	 * For each index from n to n + s, verify that the existing bufwad
@@ -4709,8 +4722,13 @@ ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
 	error = dsl_dataset_user_hold(holds, 0, NULL);
 	fnvlist_free(holds);
 
-	if (error)
-		fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
+	if (error == ENOSPC) {
+		ztest_record_enospc("dsl_dataset_user_hold");
+		goto out;
+	} else if (error) {
+		fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
+		    fullname, tag, error);
+	}
 
 	error = dsl_destroy_snapshot(fullname, B_FALSE);
 	if (error != EBUSY) {
@@ -5163,7 +5181,7 @@ ztest_run_zdb(char *pool)
 	isa = strdup(isa);
 	/* LINTED */
 	(void) sprintf(bin,
-	    "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
+	    "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s",
 	    isalen,
 	    isa,
 	    ztest_opts.zo_verbose >= 3 ? "s" : "",

common/zfs/zfeature_common.c

@@ -215,4 +215,9 @@ zpool_feature_init(void)
 	    "com.joyent:filesystem_limits", "filesystem_limits",
 	    "Filesystem and snapshot limits.", B_TRUE, B_FALSE, B_FALSE,
 	    filesystem_limits_deps);
+
+	zfeature_register(SPA_FEATURE_EMBEDDED_DATA,
+	    "com.delphix:embedded_data", "embedded_data",
+	    "Blocks which compress very well use even less space.",
+	    B_FALSE, B_TRUE, B_TRUE, NULL);
 }

common/zfs/zfeature_common.h

@@ -48,6 +48,7 @@ typedef enum spa_feature {
 	SPA_FEATURE_ENABLED_TXG,
 	SPA_FEATURE_HOLE_BIRTH,
 	SPA_FEATURE_EXTENSIBLE_DATASET,
+	SPA_FEATURE_EMBEDDED_DATA,
 	SPA_FEATURE_BOOKMARKS,
 	SPA_FEATURE_FS_SS_LIMIT,
 	SPA_FEATURES
@@ -68,7 +69,7 @@ typedef struct zfeature_info {
 	const spa_feature_t *fi_depends;
 } zfeature_info_t;
 
-typedef int (zfeature_func_t)(zfeature_info_t *fi, void *arg);
+typedef int (zfeature_func_t)(zfeature_info_t *, void *);
 
 #define	ZFS_FEATURE_DEBUG
 
@@ -77,8 +78,8 @@ extern zfeature_info_t spa_feature_table[SPA_FEATURES];
 extern boolean_t zfeature_is_valid_guid(const char *);
 
 extern boolean_t zfeature_is_supported(const char *);
-extern int zfeature_lookup_name(const char *name, spa_feature_t *res);
-extern boolean_t zfeature_depends_on(spa_feature_t fid, spa_feature_t check);
+extern int zfeature_lookup_name(const char *, spa_feature_t *);
+extern boolean_t zfeature_depends_on(spa_feature_t, spa_feature_t);
 
 extern void zpool_feature_init(void);
 

lib/libzfs/common/libzfs.h

@@ -39,6 +39,7 @@
 #include <sys/fs/zfs.h>
 #include <sys/avl.h>
 #include <ucred.h>
+#include <libzfs_core.h>
 
 #ifdef	__cplusplus
 extern "C" {
@@ -589,13 +590,16 @@ typedef struct sendflags {
 
 	/* show progress (ie. -v) */
 	boolean_t progress;
+
+	/* WRITE_EMBEDDED records of type DATA are permitted */
+	boolean_t embed_data;
 } sendflags_t;
 
 typedef boolean_t (snapfilter_cb_t)(zfs_handle_t *, void *);
 
 extern int zfs_send(zfs_handle_t *, const char *, const char *,
     sendflags_t *, int, snapfilter_cb_t, void *, nvlist_t **);
-extern int zfs_send_one(zfs_handle_t *, const char *, int);
+extern int zfs_send_one(zfs_handle_t *, const char *, int, enum lzc_send_flags);
 
 extern int zfs_promote(zfs_handle_t *);
 extern int zfs_hold(zfs_handle_t *, const char *, const char *,

lib/libzfs/common/libzfs_sendrecv.c

@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  */
@@ -42,6 +42,7 @@
 #include <time.h>
 
 #include <libzfs.h>
+#include <libzfs_core.h>
 
 #include "zfs_namecheck.h"
 #include "zfs_prop.h"
@@ -213,6 +214,7 @@ cksummer(void *arg)
 	struct drr_object *drro = &thedrr.drr_u.drr_object;
 	struct drr_write *drrw = &thedrr.drr_u.drr_write;
 	struct drr_spill *drrs = &thedrr.drr_u.drr_spill;
+	struct drr_write_embedded *drrwe = &thedrr.drr_u.drr_write_embedded;
 	FILE *ofp;
 	int outfd;
 	dmu_replay_record_t wbr_drr = {0};
@@ -409,6 +411,20 @@ cksummer(void *arg)
 			break;
 		}
 
+		case DRR_WRITE_EMBEDDED:
+		{
+			if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
+			    &stream_cksum, outfd) == -1)
+				goto out;
+			(void) ssread(buf,
+			    P2ROUNDUP((uint64_t)drrwe->drr_psize, 8), ofp);
+			if (cksum_and_write(buf,
+			    P2ROUNDUP((uint64_t)drrwe->drr_psize, 8),
+			    &stream_cksum, outfd) == -1)
+				goto out;
+			break;
+		}
+
 		case DRR_FREE:
 		{
 			if (cksum_and_write(drr, sizeof (dmu_replay_record_t),
@@ -790,7 +806,7 @@ typedef struct send_dump_data {
 	char prevsnap[ZFS_MAXNAMELEN];
 	uint64_t prevsnap_obj;
 	boolean_t seenfrom, seento, replicate, doall, fromorigin;
-	boolean_t verbose, dryrun, parsable, progress;
+	boolean_t verbose, dryrun, parsable, progress, embed_data;
 	int outfd;
 	boolean_t err;
 	nvlist_t *fss;
@@ -870,7 +886,8 @@ estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
  */
 static int
 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
-    boolean_t fromorigin, int outfd, nvlist_t *debugnv)
+    boolean_t fromorigin, int outfd, enum lzc_send_flags flags,
+    nvlist_t *debugnv)
 {
 	zfs_cmd_t zc = { 0 };
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
@@ -884,6 +901,7 @@ dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
 	zc.zc_obj = fromorigin;
 	zc.zc_sendobj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID);
 	zc.zc_fromobj = fromsnap_obj;
+	zc.zc_flags = flags;
 
 	VERIFY(0 == nvlist_alloc(&thisdbg, NV_UNIQUE_NAME, 0));
 	if (fromsnap && fromsnap[0] != '\0') {
@@ -1134,8 +1152,12 @@ dump_snapshot(zfs_handle_t *zhp, void *arg)
 			}
 		}
 
+		enum lzc_send_flags flags = 0;
+		if (sdd->embed_data)
+			flags |= LZC_SEND_FLAG_EMBED_DATA;
+
 		err = dump_ioctl(zhp, sdd->prevsnap, sdd->prevsnap_obj,
-		    fromorigin, sdd->outfd, sdd->debugnv);
+		    fromorigin, sdd->outfd, flags, sdd->debugnv);
 
 		if (sdd->progress) {
 			(void) pthread_cancel(tid);
@@ -1479,6 +1501,7 @@ zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
 	sdd.parsable = flags->parsable;
 	sdd.progress = flags->progress;
 	sdd.dryrun = flags->dryrun;
+	sdd.embed_data = flags->embed_data;
 	sdd.filter_cb = filter_func;
 	sdd.filter_cb_arg = cb_arg;
 	if (debugnvp)
@@ -1610,7 +1633,8 @@ zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
 }
 
 int
-zfs_send_one(zfs_handle_t *zhp, const char *from, int fd)
+zfs_send_one(zfs_handle_t *zhp, const char *from, int fd,
+    enum lzc_send_flags flags)
 {
 	int err;
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
@@ -1619,7 +1643,7 @@ zfs_send_one(zfs_handle_t *zhp, const char *from, int fd)
 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
 	    "warning: cannot send '%s'"), zhp->zfs_name);
 
-	err = lzc_send(zhp->zfs_name, from, fd);
+	err = lzc_send(zhp->zfs_name, from, fd, flags);
 	if (err != 0) {
 		switch (errno) {
 		case EXDEV:
@@ -2537,6 +2561,16 @@ recv_skip(libzfs_handle_t *hdl, int fd, boolean_t byteswap)
 			(void) recv_read(hdl, fd, buf,
 			    drr->drr_u.drr_spill.drr_length, B_FALSE, NULL);
 			break;
+		case DRR_WRITE_EMBEDDED:
+			if (byteswap) {
+				drr->drr_u.drr_write_embedded.drr_psize =
+				    BSWAP_32(drr->drr_u.drr_write_embedded.
+				    drr_psize);
+			}
+			(void) recv_read(hdl, fd, buf,
+			    P2ROUNDUP(drr->drr_u.drr_write_embedded.drr_psize,
+			    8), B_FALSE, NULL);
+			break;
 		case DRR_WRITE_BYREF:
 		case DRR_FREEOBJECTS:
 		case DRR_FREE:

lib/libzfs_core/common/libzfs_core.c

@@ -439,6 +439,8 @@ lzc_get_holds(const char *snapname, nvlist_t **holdsp)
 }
 
 /*
+ * Generate a zfs send stream for the specified snapshot and write it to
+ * the specified file descriptor.
  *
  * "snapname" is the full name of the snapshot to send (e.g. "pool/fs@snap")
  *
@@ -452,9 +454,15 @@ lzc_get_holds(const char *snapname, nvlist_t **holdsp)
  * snapshot in the origin, etc.
  *
  * "fd" is the file descriptor to write the send stream to.
+ *
+ * If "flags" contains LZC_SEND_FLAG_EMBED_DATA, the stream is permitted
+ * to contain DRR_WRITE_EMBEDDED records with drr_etype==BP_EMBEDDED_TYPE_DATA,
+ * which the receiving system must support (as indicated by support
+ * for the "embedded_data" feature).
  */
 int
-lzc_send(const char *snapname, const char *from, int fd)
+lzc_send(const char *snapname, const char *from, int fd,
+    enum lzc_send_flags flags)
 {
 	nvlist_t *args;
 	int err;
@@ -463,6 +471,8 @@ lzc_send(const char *snapname, const char *from, int fd)
 	fnvlist_add_int32(args, "fd", fd);
 	if (from != NULL)
 		fnvlist_add_string(args, "fromsnap", from);
+	if (flags & LZC_SEND_FLAG_EMBED_DATA)
+		fnvlist_add_boolean(args, "embedok");
 	err = lzc_ioctl(ZFS_IOC_SEND_NEW, snapname, args, NULL);
 	nvlist_free(args);
 	return (err);

lib/libzfs_core/common/libzfs_core.h

@@ -52,7 +52,11 @@ int lzc_hold(nvlist_t *, int, nvlist_t **);
 int lzc_release(nvlist_t *, nvlist_t **);
 int lzc_get_holds(const char *, nvlist_t **);
 
-int lzc_send(const char *, const char *, int);
+enum lzc_send_flags {
+	LZC_SEND_FLAG_EMBED_DATA = 1 << 0
+};
+
+int lzc_send(const char *, const char *, int, enum lzc_send_flags);
 int lzc_receive(const char *, nvlist_t *, const char *, boolean_t, int);
 int lzc_send_space(const char *, const char *, uint64_t *);
 

man/man1m/zfs.1m

@@ -176,12 +176,12 @@ zfs \- configures ZFS file systems
 
 .LP
 .nf
-\fBzfs\fR \fBsend\fR [\fB-DnPpRv\fR] [\fB-\fR[\fBiI\fR] \fIsnapshot\fR] \fIsnapshot\fR
+\fBzfs\fR \fBsend\fR [\fB-DnPpRve\fR] [\fB-\fR[\fBiI\fR] \fIsnapshot\fR] \fIsnapshot\fR
 .fi
 
 .LP
 .nf
-\fBzfs\fR \fBsend\fR [\fB-i \fIsnapshot\fR|\fIbookmark\fR]\fR \fIfilesystem\fR|\fIvolume\fR|\fIsnapshot\fR
+\fBzfs\fR \fBsend\fR [\fB-e\fR] [\fB-i \fIsnapshot\fR|\fIbookmark\fR]\fR \fIfilesystem\fR|\fIvolume\fR|\fIsnapshot\fR
 .fi
 
 .LP
@@ -2923,7 +2923,7 @@ See \fBzpool-features\fR(5) for details on ZFS feature flags and the
 .sp
 .ne 2
 .na
-\fBzfs send\fR [\fB-DnPpRv\fR] [\fB-\fR[\fBiI\fR] \fIsnapshot\fR] \fIsnapshot\fR
+\fBzfs send\fR [\fB-DnPpRve\fR] [\fB-\fR[\fBiI\fR] \fIsnapshot\fR] \fIsnapshot\fR
 .ad
 .sp .6
 .RS 4n
@@ -2996,6 +2996,23 @@ will be much better if the filesystem uses a dedup-capable checksum (eg.
 \fBsha256\fR).
 .RE
 
+.sp
+.ne 2
+.na
+\fB\fB-e\fR\fR
+.ad
+.sp .6
+.RS 4n
+Generate a more compact stream by using WRITE_EMBEDDED records for blocks
+which are stored more compactly on disk by the \fBembedded_data\fR pool
+feature.  This flag has no effect if the \fBembedded_data\fR feature is
+disabled.  The receiving system must have the \fBembedded_data\fR feature
+enabled.  If the \fBlz4_compress\fR feature is active on the sending system,
+then the receiving system must have that feature enabled as well. See
+\fBzpool-features\fR(5) for details on ZFS feature flags and the
+\fBembedded_data\fR feature.
+.RE
+
 .sp
 .ne 2
 .na
@@ -3047,7 +3064,7 @@ on future versions of \fBZFS\fR.
 .sp
 .ne 2
 .na
-\fBzfs send\fR [\fB-i\fR \fIsnapshot\fR|\fIbookmark\fR] \fIfilesystem\fR|\fIvolume\fR|\fIsnapshot\fR
+\fBzfs send\fR [\fB-e\fR] [\fB-i\fR \fIsnapshot\fR|\fIbookmark\fR] \fIfilesystem\fR|\fIvolume\fR|\fIsnapshot\fR
 .ad
 .sp .6
 .RS 4n
@@ -3075,6 +3092,22 @@ be the origin snapshot, or an earlier snapshot in the origin's filesystem,
 or the origin's origin, etc.
 .RE
 
+.sp
+.ne 2
+.na
+\fB\fB-e\fR\fR
+.ad
+.sp .6
+.RS 4n
+Generate a more compact stream by using WRITE_EMBEDDED records for blocks
+which are stored more compactly on disk by the \fBembedded_data\fR pool
+feature.  This flag has no effect if the \fBembedded_data\fR feature is
+disabled.  The receiving system must have the \fBembedded_data\fR feature
+enabled.  If the \fBlz4_compress\fR feature is active on the sending system,
+then the receiving system must have that feature enabled as well. See
+\fBzpool-features\fR(5) for details on ZFS feature flags and the
+\fBembedded_data\fR feature.
+.RE
 .RE
 
 .sp

man/man5/zpool-features.5

@@ -401,6 +401,33 @@ never return to being \fBenabled\fB.
 
 .RE
 
+.sp
+.ne 2
+.na
+\fB\fBembedded_data\fR\fR
+.ad
+.RS 4n
+.TS
+l l .
+GUID	com.delphix:embedded_data
+READ\-ONLY COMPATIBLE	no
+DEPENDENCIES	none
+.TE
+
+This feature improves the performance and compression ratio of
+highly-compressible blocks.  Blocks whose contents can compress to 112 bytes
+or smaller can take advantage of this feature.
+
+When this feature is enabled, the contents of highly-compressible blocks are
+stored in the block "pointer" itself (a misnomer in this case, as it contains
+the compresseed data, rather than a pointer to its location on disk).  Thus
+the space of the block (one sector, typically 512 bytes or 4KB) is saved,
+and no additional i/o is needed to read and write the data block.
+
+This feature becomes \fBactive\fR as soon as it is enabled and will
+never return to being \fBenabled\fR.
+
+.RE
 
 .SH "SEE ALSO"
 \fBzpool\fR(1M)

uts/common/fs/zfs/arc.c

@@ -744,8 +744,10 @@ buf_discard_identity(arc_buf_hdr_t *hdr)
 }
 
 static arc_buf_hdr_t *
-buf_hash_find(uint64_t spa, const dva_t *dva, uint64_t birth, kmutex_t **lockp)
+buf_hash_find(uint64_t spa, const blkptr_t *bp, kmutex_t **lockp)
 {
+	const dva_t *dva = BP_IDENTITY(bp);
+	uint64_t birth = BP_PHYSICAL_BIRTH(bp);
 	uint64_t idx = BUF_HASH_INDEX(spa, dva, birth);
 	kmutex_t *hash_lock = BUF_HASH_LOCK(idx);
 	arc_buf_hdr_t *buf;
@@ -777,6 +779,8 @@ buf_hash_insert(arc_buf_hdr_t *buf, kmutex_t **lockp)
 	arc_buf_hdr_t *fbuf;
 	uint32_t i;
 
+	ASSERT(!DVA_IS_EMPTY(&buf->b_dva));
+	ASSERT(buf->b_birth != 0);
 	ASSERT(!HDR_IN_HASH_TABLE(buf));
 	*lockp = hash_lock;
 	mutex_enter(hash_lock);
@@ -2738,10 +2742,10 @@ arc_getbuf_func(zio_t *zio, arc_buf_t *buf, void *arg)
 static void
 arc_read_done(zio_t *zio)
 {
-	arc_buf_hdr_t	*hdr, *found;
+	arc_buf_hdr_t	*hdr;
 	arc_buf_t	*buf;
 	arc_buf_t	*abuf;	/* buffer we're assigning to callback */
-	kmutex_t	*hash_lock;
+	kmutex_t	*hash_lock = NULL;
 	arc_callback_t	*callback_list, *acb;
 	int		freeable = FALSE;
 
@@ -2756,12 +2760,22 @@ arc_read_done(zio_t *zio)
 	 * reason for it not to be found is if we were freed during the
 	 * read.
 	 */
-	found = buf_hash_find(hdr->b_spa, &hdr->b_dva, hdr->b_birth,
-	    &hash_lock);
+	if (HDR_IN_HASH_TABLE(hdr)) {
+		ASSERT3U(hdr->b_birth, ==, BP_PHYSICAL_BIRTH(zio->io_bp));
+		ASSERT3U(hdr->b_dva.dva_word[0], ==,
+		    BP_IDENTITY(zio->io_bp)->dva_word[0]);
+		ASSERT3U(hdr->b_dva.dva_word[1], ==,
+		    BP_IDENTITY(zio->io_bp)->dva_word[1]);
 
-	ASSERT((found == NULL && HDR_FREED_IN_READ(hdr) && hash_lock == NULL) ||
-	    (found == hdr && DVA_EQUAL(&hdr->b_dva, BP_IDENTITY(zio->io_bp))) ||
-	    (found == hdr && HDR_L2_READING(hdr)));
+		arc_buf_hdr_t *found = buf_hash_find(hdr->b_spa, zio->io_bp,
+		    &hash_lock);
+
+		ASSERT((found == NULL && HDR_FREED_IN_READ(hdr) &&
+		    hash_lock == NULL) ||
+		    (found == hdr &&
+		    DVA_EQUAL(&hdr->b_dva, BP_IDENTITY(zio->io_bp))) ||
+		    (found == hdr && HDR_L2_READING(hdr)));
+	}
 
 	hdr->b_flags &= ~ARC_L2_EVICTED;
 	if (l2arc_noprefetch && (hdr->b_flags & ARC_PREFETCH))
@@ -2885,16 +2899,25 @@ arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_done_func_t *done,
     void *private, zio_priority_t priority, int zio_flags, uint32_t *arc_flags,
     const zbookmark_t *zb)
 {
-	arc_buf_hdr_t *hdr;
+	arc_buf_hdr_t *hdr = NULL;
 	arc_buf_t *buf = NULL;
-	kmutex_t *hash_lock;
+	kmutex_t *hash_lock = NULL;
 	zio_t *rzio;
 	uint64_t guid = spa_load_guid(spa);
 
+	ASSERT(!BP_IS_EMBEDDED(bp) ||
+	    BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
+
 top:
-	hdr = buf_hash_find(guid, BP_IDENTITY(bp), BP_PHYSICAL_BIRTH(bp),
-	    &hash_lock);
-	if (hdr && hdr->b_datacnt > 0) {
+	if (!BP_IS_EMBEDDED(bp)) {
+		/*
+		 * Embedded BP's have no DVA and require no I/O to "read".
+		 * Create an anonymous arc buf to back it.
+		 */
+		hdr = buf_hash_find(guid, bp, &hash_lock);
+	}
+
+	if (hdr != NULL && hdr->b_datacnt > 0) {
 
 		*arc_flags |= ARC_CACHED;
 
@@ -2968,7 +2991,7 @@ arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_done_func_t *done,
 			done(NULL, buf, private);
 	} else {
 		uint64_t size = BP_GET_LSIZE(bp);
-		arc_callback_t	*acb;
+		arc_callback_t *acb;
 		vdev_t *vd = NULL;
 		uint64_t addr = 0;
 		boolean_t devw = B_FALSE;
@@ -2977,15 +3000,17 @@ arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_done_func_t *done,
 
 		if (hdr == NULL) {
 			/* this block is not in the cache */
-			arc_buf_hdr_t	*exists;
+			arc_buf_hdr_t *exists = NULL;
 			arc_buf_contents_t type = BP_GET_BUFC_TYPE(bp);
 			buf = arc_buf_alloc(spa, size, private, type);
 			hdr = buf->b_hdr;
-			hdr->b_dva = *BP_IDENTITY(bp);
-			hdr->b_birth = BP_PHYSICAL_BIRTH(bp);
-			hdr->b_cksum0 = bp->blk_cksum.zc_word[0];
-			exists = buf_hash_insert(hdr, &hash_lock);
-			if (exists) {
+			if (!BP_IS_EMBEDDED(bp)) {
+				hdr->b_dva = *BP_IDENTITY(bp);
+				hdr->b_birth = BP_PHYSICAL_BIRTH(bp);
+				hdr->b_cksum0 = bp->blk_cksum.zc_word[0];
+				exists = buf_hash_insert(hdr, &hash_lock);
+			}
+			if (exists != NULL) {
 				/* somebody beat us to the hash insert */
 				mutex_exit(hash_lock);
 				buf_discard_identity(hdr);
@@ -3057,7 +3082,8 @@ arc_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, arc_done_func_t *done,
 				vd = NULL;
 		}
 
-		mutex_exit(hash_lock);
+		if (hash_lock != NULL)
+			mutex_exit(hash_lock);
 
 		/*
 		 * At this point, we have a level 1 cache miss.  Try again in
@@ -3192,8 +3218,9 @@ arc_freed(spa_t *spa, const blkptr_t *bp)
 	kmutex_t *hash_lock;
 	uint64_t guid = spa_load_guid(spa);
 
-	hdr = buf_hash_find(guid, BP_IDENTITY(bp), BP_PHYSICAL_BIRTH(bp),
-	    &hash_lock);
+	ASSERT(!BP_IS_EMBEDDED(bp));
+
+	hdr = buf_hash_find(guid, bp, &hash_lock);
 	if (hdr == NULL)
 		return;
 	if (HDR_BUF_AVAILABLE(hdr)) {
@@ -3509,7 +3536,7 @@ arc_write_done(zio_t *zio)
 	ASSERT(hdr->b_acb == NULL);
 
 	if (zio->io_error == 0) {
-		if (BP_IS_HOLE(zio->io_bp)) {
+		if (BP_IS_HOLE(zio->io_bp) || BP_IS_EMBEDDED(zio->io_bp)) {
 			buf_discard_identity(hdr);
 		} else {
 			hdr->b_dva = *BP_IDENTITY(zio->io_bp);
@@ -3521,10 +3548,10 @@ arc_write_done(zio_t *zio)
 	}
 
 	/*
-	 * If the block to be written was all-zero, we may have
-	 * compressed it away.  In this case no write was performed
-	 * so there will be no dva/birth/checksum.  The buffer must
-	 * therefore remain anonymous (and uncached).
+	 * If the block to be written was all-zero or compressed enough to be
+	 * embedded in the BP, no write was performed so there will be no
+	 * dva/birth/checksum.  The buffer must therefore remain anonymous
+	 * (and uncached).
 	 */
 	if (!BUF_EMPTY(hdr)) {
 		arc_buf_hdr_t *exists;
@@ -4818,7 +4845,7 @@ static boolean_t
 l2arc_compress_buf(l2arc_buf_hdr_t *l2hdr)
 {
 	void *cdata;
-	size_t csize, len;
+	size_t csize, len, rounded;
 
 	ASSERT(l2hdr->b_compress == ZIO_COMPRESS_OFF);
 	ASSERT(l2hdr->b_tmp_cdata != NULL);
@@ -4828,6 +4855,12 @@ l2arc_compress_buf(l2arc_buf_hdr_t *l2hdr)
 	csize = zio_compress_data(ZIO_COMPRESS_LZ4, l2hdr->b_tmp_cdata,
 	    cdata, l2hdr->b_asize);
 
+	rounded = P2ROUNDUP(csize, (size_t)SPA_MINBLOCKSIZE);
+	if (rounded > csize) {
+		bzero((char *)cdata + csize, rounded - csize);
+		csize = rounded;
+	}
+
 	if (csize == 0) {
 		/* zero block, indicate that there's nothing to write */
 		zio_data_buf_free(cdata, len);

uts/common/fs/zfs/blkptr.c

@@ -0,0 +1,119 @@
+/*
+ * CDDL HEADER START
+ *
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright (c) 2013 by Delphix. All rights reserved.
+ */
+
+#include <sys/zfs_context.h>
+#include <sys/zio.h>
+#include <sys/zio_compress.h>
+
+/*
+ * Embedded-data Block Pointers
+ *
+ * Normally, block pointers point (via their DVAs) to a block which holds data.
+ * If the data that we need to store is very small, this is an inefficient
+ * use of space, because a block must be at minimum 1 sector (typically 512
+ * bytes or 4KB).  Additionally, reading these small blocks tends to generate
+ * more random reads.
+ *
+ * Embedded-data Block Pointers allow small pieces of data (the "payload",
+ * up to 112 bytes) to be stored in the block pointer itself, instead of
+ * being pointed to.  The "Pointer" part of this name is a bit of a
+ * misnomer, as nothing is pointed to.
+ *
+ * BP_EMBEDDED_TYPE_DATA block pointers allow highly-compressible data to
+ * be embedded in the block pointer.  The logic for this is handled in
+ * the SPA, by the zio pipeline.  Therefore most code outside the zio
+ * pipeline doesn't need special-cases to handle these block pointers.
+ *
+ * See spa.h for details on the exact layout of embedded block pointers.
+ */
+
+void
+encode_embedded_bp_compressed(blkptr_t *bp, void *data,
+    enum zio_compress comp, int uncompressed_size, int compressed_size)
+{
+	uint64_t *bp64 = (uint64_t *)bp;
+	uint64_t w = 0;
+	uint8_t *data8 = data;
+
+	ASSERT3U(compressed_size, <=, BPE_PAYLOAD_SIZE);
+	ASSERT(uncompressed_size == compressed_size ||
+	    comp != ZIO_COMPRESS_OFF);
+	ASSERT3U(comp, >=, ZIO_COMPRESS_OFF);
+	ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS);
+
+	bzero(bp, sizeof (*bp));
+	BP_SET_EMBEDDED(bp, B_TRUE);
+	BP_SET_COMPRESS(bp, comp);
+	BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
+	BPE_SET_LSIZE(bp, uncompressed_size);
+	BPE_SET_PSIZE(bp, compressed_size);
+
+	/*
+	 * Encode the byte array into the words of the block pointer.
+	 * First byte goes into low bits of first word (little endian).
+	 */
+	for (int i = 0; i < compressed_size; i++) {
+		BF64_SET(w, (i % sizeof (w)) * NBBY, NBBY, data8[i]);
+		if (i % sizeof (w) == sizeof (w) - 1) {
+			/* we've reached the end of a word */
+			ASSERT3P(bp64, <, bp + 1);
+			*bp64 = w;
+			bp64++;
+			if (!BPE_IS_PAYLOADWORD(bp, bp64))
+				bp64++;
+			w = 0;
+		}
+	}
+	/* write last partial word */
+	if (bp64 < (uint64_t *)(bp + 1))
+		*bp64 = w;
+}
+
+/*
+ * buf must be at least BPE_GET_PSIZE(bp) bytes long (which will never be
+ * more than BPE_PAYLOAD_SIZE bytes).
+ */
+void
+decode_embedded_bp_compressed(const blkptr_t *bp, void *buf)
+{
+	int psize;
+	uint8_t *buf8 = buf;
+	uint64_t w = 0;
+	const uint64_t *bp64 = (const uint64_t *)bp;
+
+	ASSERT(BP_IS_EMBEDDED(bp));
+
+	psize = BPE_GET_PSIZE(bp);
+
+	/*
+	 * Decode the words of the block pointer into the byte array.
+	 * Low bits of first word are the first byte (little endian).
+	 */
+	for (int i = 0; i < psize; i++) {
+		if (i % sizeof (w) == 0) {
+			/* beginning of a word */
+			ASSERT3P(bp64, <, bp + 1);
+			w = *bp64;
+			bp64++;
+			if (!BPE_IS_PAYLOADWORD(bp, bp64))
+				bp64++;
+		}
+		buf8[i] = BF64_GET(w, (i % sizeof (w)) * NBBY, NBBY);
+	}
+}

uts/common/fs/zfs/bpobj.c

@@ -192,6 +192,13 @@ bpobj_close(bpobj_t *bpo)
 	mutex_destroy(&bpo->bpo_lock);
 }
 
+static boolean_t
+bpobj_hasentries(bpobj_t *bpo)
+{
+	return (bpo->bpo_phys->bpo_num_blkptrs != 0 ||
+	    (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_num_subobjs != 0));
+}
+
 static int
 bpobj_iterate_impl(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx,
     boolean_t free)
@@ -332,9 +339,11 @@ bpobj_iterate_impl(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx,
 
 out:
 	/* If there are no entries, there should be no bytes. */
-	ASSERT(bpo->bpo_phys->bpo_num_blkptrs > 0 ||
-	    (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_num_subobjs > 0) ||
-	    bpo->bpo_phys->bpo_bytes == 0);
+	if (!bpobj_hasentries(bpo)) {
+		ASSERT0(bpo->bpo_phys->bpo_bytes);
+		ASSERT0(bpo->bpo_phys->bpo_comp);
+		ASSERT0(bpo->bpo_phys->bpo_uncomp);
+	}
 
 	mutex_exit(&bpo->bpo_lock);
 	return (err);
@@ -377,7 +386,7 @@ bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx)
 	VERIFY3U(0, ==, bpobj_open(&subbpo, bpo->bpo_os, subobj));
 	VERIFY3U(0, ==, bpobj_space(&subbpo, &used, &comp, &uncomp));
 
-	if (used == 0) {
+	if (!bpobj_hasentries(&subbpo)) {
 		/* No point in having an empty subobj. */
 		bpobj_close(&subbpo);
 		bpobj_free(bpo->bpo_os, subobj, tx);
@@ -453,13 +462,29 @@ bpobj_enqueue(bpobj_t *bpo, const blkptr_t *bp, dmu_tx_t *tx)
 	ASSERT(!BP_IS_HOLE(bp));
 	ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
 
+	if (BP_IS_EMBEDDED(bp)) {
+		/*
+		 * The bpobj will compress better without the payload.
+		 *
+		 * Note that we store EMBEDDED bp's because they have an
+		 * uncompressed size, which must be accounted for.  An
+		 * alternative would be to add their size to bpo_uncomp
+		 * without storing the bp, but that would create additional
+		 * complications: bpo_uncomp would be inconsistent with the
+		 * set of BP's stored, and bpobj_iterate() wouldn't visit
+		 * all the space accounted for in the bpobj.
+		 */
+		bzero(&stored_bp, sizeof (stored_bp));
+		stored_bp.blk_prop = bp->blk_prop;
+		stored_bp.blk_birth = bp->blk_birth;
+	} else if (!BP_GET_DEDUP(bp)) {
+		/* The bpobj will compress better without the checksum */
+		bzero(&stored_bp.blk_cksum, sizeof (stored_bp.blk_cksum));
+	}
+
 	/* We never need the fill count. */
 	stored_bp.blk_fill = 0;
 
-	/* The bpobj will compress better if we can leave off the checksum */
-	if (!BP_GET_DEDUP(bp))
-		bzero(&stored_bp.blk_cksum, sizeof (stored_bp.blk_cksum));
-
 	mutex_enter(&bpo->bpo_lock);
 
 	offset = bpo->bpo_phys->bpo_num_blkptrs * sizeof (stored_bp);

uts/common/fs/zfs/dbuf.c

@@ -40,6 +40,8 @@
 #include <sys/dmu_zfetch.h>
 #include <sys/sa.h>
 #include <sys/sa_impl.h>
+#include <sys/zfeature.h>
+#include <sys/blkptr.h>
 #include <sys/range_tree.h>
 
 /*
@@ -1435,6 +1437,38 @@ dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx)
 	mutex_exit(&db->db_mtx);
 }
 
+void
+dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
+    bp_embedded_type_t etype, enum zio_compress comp,
+    int uncompressed_size, int compressed_size, int byteorder,
+    dmu_tx_t *tx)
+{
+	dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf;
+	struct dirty_leaf *dl;
+	dmu_object_type_t type;
+
+	DB_DNODE_ENTER(db);
+	type = DB_DNODE(db)->dn_type;
+	DB_DNODE_EXIT(db);
+
+	ASSERT0(db->db_level);
+	ASSERT(db->db_blkid != DMU_BONUS_BLKID);
+
+	dmu_buf_will_not_fill(dbuf, tx);
+
+	ASSERT3U(db->db_last_dirty->dr_txg, ==, tx->tx_txg);
+	dl = &db->db_last_dirty->dt.dl;
+	encode_embedded_bp_compressed(&dl->dr_overridden_by,
+	    data, comp, uncompressed_size, compressed_size);
+	BPE_SET_ETYPE(&dl->dr_overridden_by, etype);
+	BP_SET_TYPE(&dl->dr_overridden_by, type);
+	BP_SET_LEVEL(&dl->dr_overridden_by, 0);
+	BP_SET_BYTEORDER(&dl->dr_overridden_by, byteorder);
+
+	dl->dr_override_state = DR_OVERRIDDEN;
+	dl->dr_overridden_by.blk_birth = db->db_last_dirty->dr_txg;
+}
+
 /*
  * Directly assign a provided arc buf to a given dbuf if it's not referenced
  * by anybody except our caller. Otherwise copy arcbuf's contents to dbuf.
@@ -1819,7 +1853,7 @@ dbuf_prefetch(dnode_t *dn, uint64_t blkid, zio_priority_t prio)
 	}
 
 	if (dbuf_findbp(dn, 0, blkid, TRUE, &db, &bp) == 0) {
-		if (bp && !BP_IS_HOLE(bp)) {
+		if (bp && !BP_IS_HOLE(bp) && !BP_IS_EMBEDDED(bp)) {
 			dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset;
 			uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
 			zbookmark_t zb;
@@ -2451,7 +2485,7 @@ dbuf_write_ready(zio_t *zio, arc_buf_t *buf, void *vdb)
 	uint64_t fill = 0;
 	int i;
 
-	ASSERT(db->db_blkptr == bp);
+	ASSERT3P(db->db_blkptr, ==, bp);
 
 	DB_DNODE_ENTER(db);
 	dn = DB_DNODE(db);
@@ -2463,7 +2497,8 @@ dbuf_write_ready(zio_t *zio, arc_buf_t *buf, void *vdb)
 		ASSERT((db->db_blkid != DMU_SPILL_BLKID &&
 		    BP_GET_TYPE(bp) == dn->dn_type) ||
 		    (db->db_blkid == DMU_SPILL_BLKID &&
-		    BP_GET_TYPE(bp) == dn->dn_bonustype));
+		    BP_GET_TYPE(bp) == dn->dn_bonustype) ||
+		    BP_IS_EMBEDDED(bp));
 		ASSERT(BP_GET_LEVEL(bp) == db->db_level);
 	}
 
@@ -2504,12 +2539,13 @@ dbuf_write_ready(zio_t *zio, arc_buf_t *buf, void *vdb)
 		for (i = db->db.db_size >> SPA_BLKPTRSHIFT; i > 0; i--, ibp++) {
 			if (BP_IS_HOLE(ibp))
 				continue;
-			fill += ibp->blk_fill;
+			fill += BP_GET_FILL(ibp);
 		}
 	}
 	DB_DNODE_EXIT(db);
 
-	bp->blk_fill = fill;
+	if (!BP_IS_EMBEDDED(bp))
+		bp->blk_fill = fill;
 
 	mutex_exit(&db->db_mtx);
 }
@@ -2621,7 +2657,8 @@ dbuf_write_done(zio_t *zio, arc_buf_t *buf, void *vdb)
 			    dn->dn_phys->dn_maxblkid >> (db->db_level * epbs));
 			ASSERT3U(BP_GET_LSIZE(db->db_blkptr), ==,
 			    db->db.db_size);
-			arc_set_callback(db->db_buf, dbuf_do_evict, db);
+			if (!arc_released(db->db_buf))
+				arc_set_callback(db->db_buf, dbuf_do_evict, db);
 		}
 		DB_DNODE_EXIT(db);
 		mutex_destroy(&dr->dt.di.dr_mtx);
@@ -2747,10 +2784,16 @@ dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx)
 	dmu_write_policy(os, dn, db->db_level, wp_flag, &zp);
 	DB_DNODE_EXIT(db);
 
-	if (db->db_level == 0 && dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
-		ASSERT(db->db_state != DB_NOFILL);
+	if (db->db_level == 0 &&
+	    dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
+		/*
+		 * The BP for this block has been provided by open context
+		 * (by dmu_sync() or dmu_buf_write_embedded()).
+		 */
+		void *contents = (data != NULL) ? data->b_data : NULL;
+
 		dr->dr_zio = zio_write(zio, os->os_spa, txg,
-		    db->db_blkptr, data->b_data, arc_buf_size(data), &zp,
+		    db->db_blkptr, contents, db->db.db_size, &zp,
 		    dbuf_write_override_ready, NULL, dbuf_write_override_done,
 		    dr, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
 		mutex_enter(&db->db_mtx);

uts/common/fs/zfs/dmu.c

@@ -125,17 +125,13 @@ const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = {
 };
 
 int
-dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
-    void *tag, dmu_buf_t **dbp, int flags)
+dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
+    void *tag, dmu_buf_t **dbp)
 {
 	dnode_t *dn;
 	uint64_t blkid;
 	dmu_buf_impl_t *db;
 	int err;
-	int db_flags = DB_RF_CANFAIL;
-
-	if (flags & DMU_READ_NO_PREFETCH)
-		db_flags |= DB_RF_NOPREFETCH;
 
 	err = dnode_hold(os, object, FTAG, &dn);
 	if (err)
@@ -144,18 +140,37 @@ dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
 	rw_enter(&dn->dn_struct_rwlock, RW_READER);
 	db = dbuf_hold(dn, blkid, tag);
 	rw_exit(&dn->dn_struct_rwlock);
+	dnode_rele(dn, FTAG);
+
 	if (db == NULL) {
-		err = SET_ERROR(EIO);
-	} else {
+		*dbp = NULL;
+		return (SET_ERROR(EIO));
+	}
+
+	*dbp = &db->db;
+	return (err);
+}
+
+int
+dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
+    void *tag, dmu_buf_t **dbp, int flags)
+{
+	int err;
+	int db_flags = DB_RF_CANFAIL;
+
+	if (flags & DMU_READ_NO_PREFETCH)
+		db_flags |= DB_RF_NOPREFETCH;
+
+	err = dmu_buf_hold_noread(os, object, offset, tag, dbp);
+	if (err == 0) {
+		dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp);
 		err = dbuf_read(db, NULL, db_flags);
-		if (err) {
+		if (err != 0) {
 			dbuf_rele(db, tag);
-			db = NULL;
+			*dbp = NULL;
 		}
 	}
 
-	dnode_rele(dn, FTAG);
-	*dbp = &db->db; /* NULL db plus first field offset is NULL */
 	return (err);
 }
 
@@ -848,6 +863,25 @@ dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 	dmu_buf_rele_array(dbp, numbufs, FTAG);
 }
 
+void
+dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
+    void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
+    int compressed_size, int byteorder, dmu_tx_t *tx)
+{
+	dmu_buf_t *db;
+
+	ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES);
+	ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS);
+	VERIFY0(dmu_buf_hold_noread(os, object, offset,
+	    FTAG, &db));
+
+	dmu_buf_write_embedded(db,
+	    data, (bp_embedded_type_t)etype, (enum zio_compress)comp,
+	    uncompressed_size, compressed_size, byteorder, tx);
+
+	dmu_buf_rele(db, FTAG);
+}
+
 /*
  * DMU support for xuio
  */
@@ -1263,7 +1297,7 @@ dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
 			 * block size still needs to be known for replay.
 			 */
 			BP_SET_LSIZE(bp, db->db_size);
-		} else {
+		} else if (!BP_IS_EMBEDDED(bp)) {
 			ASSERT(BP_GET_LEVEL(bp) == 0);
 			bp->blk_fill = 1;
 		}
@@ -1534,9 +1568,15 @@ dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
 {
 	dnode_t *dn;
 
-	/* XXX assumes dnode_hold will not get an i/o error */
-	(void) dnode_hold(os, object, FTAG, &dn);
-	ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS);
+	/*
+	 * Send streams include each object's checksum function.  This
+	 * check ensures that the receiving system can understand the
+	 * checksum function transmitted.
+	 */
+	ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS);
+
+	VERIFY0(dnode_hold(os, object, FTAG, &dn));
+	ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS);
 	dn->dn_checksum = checksum;
 	dnode_setdirty(dn, tx);
 	dnode_rele(dn, FTAG);
@@ -1548,9 +1588,14 @@ dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
 {
 	dnode_t *dn;
 
-	/* XXX assumes dnode_hold will not get an i/o error */
-	(void) dnode_hold(os, object, FTAG, &dn);
-	ASSERT(compress < ZIO_COMPRESS_FUNCTIONS);
+	/*
+	 * Send streams include each object's compression function.  This
+	 * check ensures that the receiving system can understand the
+	 * compression function transmitted.
+	 */
+	ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS);
+
+	VERIFY0(dnode_hold(os, object, FTAG, &dn));
 	dn->dn_compress = compress;
 	dnode_setdirty(dn, tx);
 	dnode_rele(dn, FTAG);
@@ -1717,7 +1762,7 @@ dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
 	doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz;
 	doi->doi_fill_count = 0;
 	for (int i = 0; i < dnp->dn_nblkptr; i++)
-		doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill;
+		doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]);
 
 	mutex_exit(&dn->dn_mtx);
 	rw_exit(&dn->dn_struct_rwlock);

uts/common/fs/zfs/dmu_objset.c

@@ -338,7 +338,7 @@ dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 	 * default (fletcher2/off).  Snapshots don't need to know about
 	 * checksum/compression/copies.
 	 */
-	if (ds) {
+	if (ds != NULL) {
 		err = dsl_prop_register(ds,
 		    zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
 		    primary_cache_changed_cb, os);
@@ -391,7 +391,7 @@ dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 			kmem_free(os, sizeof (objset_t));
 			return (err);
 		}
-	} else if (ds == NULL) {
+	} else {
 		/* It's the meta-objset. */
 		os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
 		os->os_compress = ZIO_COMPRESS_LZJB;
@@ -435,17 +435,6 @@ dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 		    &os->os_groupused_dnode);
 	}
 
-	/*
-	 * We should be the only thread trying to do this because we
-	 * have ds_opening_lock
-	 */
-	if (ds) {
-		mutex_enter(&ds->ds_lock);
-		ASSERT(ds->ds_objset == NULL);
-		ds->ds_objset = os;
-		mutex_exit(&ds->ds_lock);
-	}
-
 	*osp = os;
 	return (0);
 }
@@ -456,11 +445,19 @@ dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
 	int err = 0;
 
 	mutex_enter(&ds->ds_opening_lock);
-	*osp = ds->ds_objset;
-	if (*osp == NULL) {
+	if (ds->ds_objset == NULL) {
+		objset_t *os;
 		err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
-		    ds, dsl_dataset_get_blkptr(ds), osp);
+		    ds, dsl_dataset_get_blkptr(ds), &os);
+
+		if (err == 0) {
+			mutex_enter(&ds->ds_lock);
+			ASSERT(ds->ds_objset == NULL);
+			ds->ds_objset = os;
+			mutex_exit(&ds->ds_lock);
+		}
 	}
+	*osp = ds->ds_objset;
 	mutex_exit(&ds->ds_opening_lock);
 	return (err);
 }
@@ -986,6 +983,7 @@ dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
 	objset_t *os = arg;
 	dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
 
+	ASSERT(!BP_IS_EMBEDDED(bp));
 	ASSERT3P(bp, ==, os->os_rootbp);
 	ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
 	ASSERT0(BP_GET_LEVEL(bp));
@@ -998,7 +996,7 @@ dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
 	 */
 	bp->blk_fill = 0;
 	for (int i = 0; i < dnp->dn_nblkptr; i++)
-		bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
+		bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
 }
 
 /* ARGSUSED */

uts/common/fs/zfs/dmu_send.c

@@ -48,7 +48,9 @@
 #include <sys/zfs_onexit.h>
 #include <sys/dmu_send.h>
 #include <sys/dsl_destroy.h>
+#include <sys/blkptr.h>
 #include <sys/dsl_bookmark.h>
+#include <sys/zfeature.h>
 
 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
 int zfs_send_corrupt_data = B_FALSE;
@@ -168,7 +170,7 @@ dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
 }
 
 static int
-dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
+dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type,
     uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
 {
 	struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
@@ -203,13 +205,22 @@ dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
 	drrw->drr_offset = offset;
 	drrw->drr_length = blksz;
 	drrw->drr_toguid = dsp->dsa_toguid;
-	drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
-	if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
-		drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
-	DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
-	DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
-	DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
-	drrw->drr_key.ddk_cksum = bp->blk_cksum;
+	if (BP_IS_EMBEDDED(bp)) {
+		/*
+		 * There's no pre-computed checksum of embedded BP's, so
+		 * (like fletcher4-checkummed blocks) userland will have
+		 * to compute a dedup-capable checksum itself.
+		 */
+		drrw->drr_checksumtype = ZIO_CHECKSUM_OFF;
+	} else {
+		drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
+		if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
+			drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
+		DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
+		DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
+		DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
+		drrw->drr_key.ddk_cksum = bp->blk_cksum;
+	}
 
 	if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
 		return (SET_ERROR(EINTR));
@@ -218,6 +229,43 @@ dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
 	return (0);
 }
 
+static int
+dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
+    int blksz, const blkptr_t *bp)
+{
+	char buf[BPE_PAYLOAD_SIZE];
+	struct drr_write_embedded *drrw =
+	    &(dsp->dsa_drr->drr_u.drr_write_embedded);
+
+	if (dsp->dsa_pending_op != PENDING_NONE) {
+		if (dump_bytes(dsp, dsp->dsa_drr,
+		    sizeof (dmu_replay_record_t)) != 0)
+			return (EINTR);
+		dsp->dsa_pending_op = PENDING_NONE;
+	}
+
+	ASSERT(BP_IS_EMBEDDED(bp));
+
+	bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
+	dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED;
+	drrw->drr_object = object;
+	drrw->drr_offset = offset;
+	drrw->drr_length = blksz;
+	drrw->drr_toguid = dsp->dsa_toguid;
+	drrw->drr_compression = BP_GET_COMPRESS(bp);
+	drrw->drr_etype = BPE_GET_ETYPE(bp);
+	drrw->drr_lsize = BPE_GET_LSIZE(bp);
+	drrw->drr_psize = BPE_GET_PSIZE(bp);
+
+	decode_embedded_bp_compressed(bp, buf);
+
+	if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
+		return (EINTR);
+	if (dump_bytes(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0)
+		return (EINTR);
+	return (0);
+}
+
 static int
 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
 {
@@ -338,6 +386,33 @@ dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
 	return (0);
 }
 
+static boolean_t
+backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp)
+{
+	if (!BP_IS_EMBEDDED(bp))
+		return (B_FALSE);
+
+	/*
+	 * Compression function must be legacy, or explicitly enabled.
+	 */
+	if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS &&
+	    !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4)))
+		return (B_FALSE);
+
+	/*
+	 * Embed type must be explicitly enabled.
+	 */
+	switch (BPE_GET_ETYPE(bp)) {
+	case BP_EMBEDDED_TYPE_DATA:
+		if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)
+			return (B_TRUE);
+		break;
+	default:
+		return (B_FALSE);
+	}
+	return (B_FALSE);
+}
+
 #define	BP_SPAN(dnp, level) \
 	(((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
 	(level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
@@ -406,11 +481,17 @@ backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 
 		err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
 		(void) arc_buf_remove_ref(abuf, &abuf);
+	} else if (backup_do_embed(dsp, bp)) {
+		/* it's an embedded level-0 block of a regular object */
+		int blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
+		err = dump_write_embedded(dsp, zb->zb_object,
+		    zb->zb_blkid * blksz, blksz, bp);
 	} else { /* it's a level-0 block of a regular object */
 		uint32_t aflags = ARC_WAIT;
 		arc_buf_t *abuf;
 		int blksz = BP_GET_LSIZE(bp);
 
+		ASSERT3U(blksz, ==, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
 		ASSERT0(zb->zb_level);
 		if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
@@ -429,7 +510,7 @@ backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 			}
 		}
 
-		err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
+		err = dump_write(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
 		    blksz, bp, abuf->b_data);
 		(void) arc_buf_remove_ref(abuf, &abuf);
 	}
@@ -443,14 +524,15 @@ backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
  */
 static int
 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
-    zfs_bookmark_phys_t *fromzb, boolean_t is_clone, int outfd,
-    vnode_t *vp, offset_t *off)
+    zfs_bookmark_phys_t *fromzb, boolean_t is_clone, boolean_t embedok,
+    int outfd, vnode_t *vp, offset_t *off)
 {
 	objset_t *os;
 	dmu_replay_record_t *drr;
 	dmu_sendarg_t *dsp;
 	int err;
 	uint64_t fromtxg = 0;
+	uint64_t featureflags = 0;
 
 	err = dmu_objset_from_ds(ds, &os);
 	if (err != 0) {
@@ -473,13 +555,23 @@ dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 			return (SET_ERROR(EINVAL));
 		}
 		if (version >= ZPL_VERSION_SA) {
-			DMU_SET_FEATUREFLAGS(
-			    drr->drr_u.drr_begin.drr_versioninfo,
-			    DMU_BACKUP_FEATURE_SA_SPILL);
+			featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
 		}
 	}
 #endif
 
+	if (embedok &&
+	    spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) {
+		featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA;
+		if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
+			featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA_LZ4;
+	} else {
+		embedok = B_FALSE;
+	}
+
+	DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo,
+	    featureflags);
+
 	drr->drr_u.drr_begin.drr_creation_time =
 	    ds->ds_phys->ds_creation_time;
 	drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
@@ -511,6 +603,7 @@ dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 	ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
 	dsp->dsa_pending_op = PENDING_NONE;
 	dsp->dsa_incremental = (fromzb != NULL);
+	dsp->dsa_featureflags = featureflags;
 
 	mutex_enter(&ds->ds_sendstream_lock);
 	list_insert_head(&ds->ds_sendstreams, dsp);
@@ -562,7 +655,7 @@ dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 
 int
 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
-    int outfd, vnode_t *vp, offset_t *off)
+    boolean_t embedok, int outfd, vnode_t *vp, offset_t *off)
 {
 	dsl_pool_t *dp;
 	dsl_dataset_t *ds;
@@ -596,10 +689,10 @@ dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
 		zb.zbm_guid = fromds->ds_phys->ds_guid;
 		is_clone = (fromds->ds_dir != ds->ds_dir);
 		dsl_dataset_rele(fromds, FTAG);
-		err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
+		err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, embedok,
 		    outfd, vp, off);
 	} else {
-		err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
+		err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, embedok,
 		    outfd, vp, off);
 	}
 	dsl_dataset_rele(ds, FTAG);
@@ -607,7 +700,7 @@ dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
 }
 
 int
-dmu_send(const char *tosnap, const char *fromsnap,
+dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok,
     int outfd, vnode_t *vp, offset_t *off)
 {
 	dsl_pool_t *dp;
@@ -674,10 +767,10 @@ dmu_send(const char *tosnap, const char *fromsnap,
 			dsl_pool_rele(dp, FTAG);
 			return (err);
 		}
-		err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
+		err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone, embedok,
 		    outfd, vp, off);
 	} else {
-		err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
+		err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE, embedok,
 		    outfd, vp, off);
 	}
 	if (owned)
@@ -847,6 +940,7 @@ dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
 	uint64_t fromguid = drrb->drr_fromguid;
 	int flags = drrb->drr_flags;
 	int error;
+	uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
 	dsl_dataset_t *ds;
 	const char *tofs = drba->drba_cookie->drc_tofs;
 
@@ -860,11 +954,22 @@ dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
 		return (SET_ERROR(EINVAL));
 
 	/* Verify pool version supports SA if SA_SPILL feature set */
-	if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
-	    DMU_BACKUP_FEATURE_SA_SPILL) &&
-	    spa_version(dp->dp_spa) < SPA_VERSION_SA) {
+	if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
+	    spa_version(dp->dp_spa) < SPA_VERSION_SA)
+		return (SET_ERROR(ENOTSUP));
+
+	/*
+	 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
+	 * record to a plan WRITE record, so the pool must have the
+	 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
+	 * records.  Same with WRITE_EMBEDDED records that use LZ4 compression.
+	 */
+	if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) &&
+	    !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA))
+		return (SET_ERROR(ENOTSUP));
+	if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) &&
+	    !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
 		return (SET_ERROR(ENOTSUP));
-	}
 
 	error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
 	if (error == 0) {
@@ -1159,7 +1264,6 @@ backup_byteswap(dmu_replay_record_t *drr)
 		break;
 	case DRR_OBJECT:
 		DO64(drr_object.drr_object);
-		/* DO64(drr_object.drr_allocation_txg); */
 		DO32(drr_object.drr_type);
 		DO32(drr_object.drr_bonustype);
 		DO32(drr_object.drr_blksz);
@@ -1197,6 +1301,14 @@ backup_byteswap(dmu_replay_record_t *drr)
 		DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
 		DO64(drr_write_byref.drr_key.ddk_prop);
 		break;
+	case DRR_WRITE_EMBEDDED:
+		DO64(drr_write_embedded.drr_object);
+		DO64(drr_write_embedded.drr_offset);
+		DO64(drr_write_embedded.drr_length);
+		DO64(drr_write_embedded.drr_toguid);
+		DO32(drr_write_embedded.drr_lsize);
+		DO32(drr_write_embedded.drr_psize);
+		break;
 	case DRR_FREE:
 		DO64(drr_free.drr_object);
 		DO64(drr_free.drr_offset);
@@ -1384,7 +1496,7 @@ restore_write_byref(struct restorearg *ra, objset_t *os,
 	int err;
 	guid_map_entry_t gmesrch;
 	guid_map_entry_t *gmep;
-	avl_index_t	where;
+	avl_index_t where;
 	objset_t *ref_os = NULL;
 	dmu_buf_t *dbp;
 
@@ -1407,8 +1519,9 @@ restore_write_byref(struct restorearg *ra, objset_t *os,
 		ref_os = os;
 	}
 
-	if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
-	    drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
+	err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
+	    drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
+	if (err != 0)
 		return (err);
 
 	tx = dmu_tx_create(os);
@@ -1427,6 +1540,48 @@ restore_write_byref(struct restorearg *ra, objset_t *os,
 	return (0);
 }
 
+static int
+restore_write_embedded(struct restorearg *ra, objset_t *os,
+    struct drr_write_embedded *drrwnp)
+{
+	dmu_tx_t *tx;
+	int err;
+	void *data;
+
+	if (drrwnp->drr_offset + drrwnp->drr_length < drrwnp->drr_offset)
+		return (EINVAL);
+
+	if (drrwnp->drr_psize > BPE_PAYLOAD_SIZE)
+		return (EINVAL);
+
+	if (drrwnp->drr_etype >= NUM_BP_EMBEDDED_TYPES)
+		return (EINVAL);
+	if (drrwnp->drr_compression >= ZIO_COMPRESS_FUNCTIONS)
+		return (EINVAL);
+
+	data = restore_read(ra, P2ROUNDUP(drrwnp->drr_psize, 8));
+	if (data == NULL)
+		return (ra->err);
+
+	tx = dmu_tx_create(os);
+
+	dmu_tx_hold_write(tx, drrwnp->drr_object,
+	    drrwnp->drr_offset, drrwnp->drr_length);
+	err = dmu_tx_assign(tx, TXG_WAIT);
+	if (err != 0) {
+		dmu_tx_abort(tx);
+		return (err);
+	}
+
+	dmu_write_embedded(os, drrwnp->drr_object,
+	    drrwnp->drr_offset, data, drrwnp->drr_etype,
+	    drrwnp->drr_compression, drrwnp->drr_lsize, drrwnp->drr_psize,
+	    ra->byteswap ^ ZFS_HOST_BYTEORDER, tx);
+
+	dmu_tx_commit(tx);
+	return (0);
+}
+
 static int
 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
 {
@@ -1621,6 +1776,13 @@ dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
 			ra.err = restore_write_byref(&ra, os, &drrwbr);
 			break;
 		}
+		case DRR_WRITE_EMBEDDED:
+		{
+			struct drr_write_embedded drrwe =
+			    drr->drr_u.drr_write_embedded;
+			ra.err = restore_write_embedded(&ra, os, &drrwe);
+			break;
+		}
 		case DRR_FREE:
 		{
 			struct drr_free drrf = drr->drr_u.drr_free;

uts/common/fs/zfs/dmu_traverse.c

@@ -458,7 +458,7 @@ traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 	if (pfd->pd_cancel)
 		return (SET_ERROR(EINTR));
 
-	if (BP_IS_HOLE(bp) ||
+	if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) ||
 	    !((pfd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
 	    BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0) ||
 	    BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG)

uts/common/fs/zfs/dnode.c

@@ -1811,8 +1811,8 @@ dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset,
 		*offset = *offset >> span;
 		for (i = BF64_GET(*offset, 0, epbs);
 		    i >= 0 && i < epb; i += inc) {
-			if (bp[i].blk_fill >= minfill &&
-			    bp[i].blk_fill <= maxfill &&
+			if (BP_GET_FILL(&bp[i]) >= minfill &&
+			    BP_GET_FILL(&bp[i]) <= maxfill &&
 			    (hole || bp[i].blk_birth > txg))
 				break;
 			if (inc > 0 || *offset > 0)

uts/common/fs/zfs/dnode_sync.c

@@ -233,8 +233,6 @@ free_verify(dmu_buf_impl_t *db, uint64_t start, uint64_t end, dmu_tx_t *tx)
 }
 #endif
 
-#define	ALL -1
-
 static void
 free_children(dmu_buf_impl_t *db, uint64_t blkid, uint64_t nblks,
     dmu_tx_t *tx)
@@ -362,7 +360,6 @@ dnode_sync_free_range_impl(dnode_t *dn, uint64_t blkid, uint64_t nblks,
 
 			free_children(db, blkid, nblks, tx);
 			dbuf_rele(db, FTAG);
-
 		}
 	}
 
@@ -591,11 +588,14 @@ dnode_sync(dnode_t *dn, dmu_tx_t *tx)
 		dnp->dn_bonustype = dn->dn_bonustype;
 		dnp->dn_bonuslen = dn->dn_bonuslen;
 	}
-
 	ASSERT(dnp->dn_nlevels > 1 ||
 	    BP_IS_HOLE(&dnp->dn_blkptr[0]) ||
+	    BP_IS_EMBEDDED(&dnp->dn_blkptr[0]) ||
 	    BP_GET_LSIZE(&dnp->dn_blkptr[0]) ==
 	    dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
+	ASSERT(dnp->dn_nlevels < 2 ||
+	    BP_IS_HOLE(&dnp->dn_blkptr[0]) ||
+	    BP_GET_LSIZE(&dnp->dn_blkptr[0]) == 1 << dnp->dn_indblkshift);
 
 	if (dn->dn_next_type[txgoff] != 0) {
 		dnp->dn_type = dn->dn_type;

uts/common/fs/zfs/dsl_dataset.c

@@ -1641,7 +1641,7 @@ dsl_dataset_space(dsl_dataset_t *ds,
 		else
 			*availbytesp = 0;
 	}
-	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
+	*usedobjsp = BP_GET_FILL(&ds->ds_phys->ds_bp);
 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
 }
 

uts/common/fs/zfs/dsl_destroy.c

@@ -539,7 +539,7 @@ kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 	struct killarg *ka = arg;
 	dmu_tx_t *tx = ka->tx;
 
-	if (BP_IS_HOLE(bp))
+	if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
 		return (0);
 
 	if (zb->zb_level == ZB_ZIL_LEVEL) {
@@ -589,6 +589,7 @@ dsl_destroy_head_check_impl(dsl_dataset_t *ds, int expected_holds)
 	uint64_t count;
 	objset_t *mos;
 
+	ASSERT(!dsl_dataset_is_snapshot(ds));
 	if (dsl_dataset_is_snapshot(ds))
 		return (SET_ERROR(EINVAL));
 
@@ -711,7 +712,7 @@ dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx)
 	    ds->ds_prev->ds_phys->ds_num_children == 2 &&
 	    ds->ds_prev->ds_userrefs == 0);
 
-	/* Remove our reservation */
+	/* Remove our reservation. */
 	if (ds->ds_reserved != 0) {
 		dsl_dataset_set_refreservation_sync_impl(ds,
 		    (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),

uts/common/fs/zfs/dsl_scan.c

@@ -1456,6 +1456,10 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx)
 		}
 		if (err == ERESTART)
 			return;
+		/* finished; verify that space accounting went to zero */
+		ASSERT0(dp->dp_free_dir->dd_phys->dd_used_bytes);
+		ASSERT0(dp->dp_free_dir->dd_phys->dd_compressed_bytes);
+		ASSERT0(dp->dp_free_dir->dd_phys->dd_uncompressed_bytes);
 	}
 
 	if (scn->scn_phys.scn_state != DSS_SCANNING)
@@ -1638,6 +1642,9 @@ dsl_scan_scrub_cb(dsl_pool_t *dp,
 
 	count_block(dp->dp_blkstats, bp);
 
+	if (BP_IS_EMBEDDED(bp))
+		return (0);
+
 	ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn));
 	if (scn->scn_phys.scn_func == POOL_SCAN_SCRUB) {
 		zio_flags |= ZIO_FLAG_SCRUB;

uts/common/fs/zfs/dsl_userhold.c

@@ -600,8 +600,7 @@ dsl_dataset_user_release_impl(nvlist_t *holds, nvlist_t *errlist,
 	ddura.ddura_chkholds = fnvlist_alloc();
 
 	error = dsl_sync_task(pool, dsl_dataset_user_release_check,
-	    dsl_dataset_user_release_sync, &ddura,
-	    fnvlist_num_pairs(holds));
+	    dsl_dataset_user_release_sync, &ddura, 0);
 	fnvlist_free(ddura.ddura_todelete);
 	fnvlist_free(ddura.ddura_chkholds);
 

uts/common/fs/zfs/spa.c

@@ -1827,7 +1827,7 @@ static int
 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
 {
-	if (!BP_IS_HOLE(bp)) {
+	if (!BP_IS_HOLE(bp) && !BP_IS_EMBEDDED(bp)) {
 		zio_t *rio = arg;
 		size_t size = BP_GET_PSIZE(bp);
 		void *data = zio_data_buf_alloc(size);
@@ -2375,9 +2375,8 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
 
 	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
 		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
-		    &spa->spa_feat_enabled_txg_obj) != 0) {
+		    &spa->spa_feat_enabled_txg_obj) != 0)
 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-		}
 	}
 
 	spa->spa_is_initializing = B_TRUE;
@@ -5260,11 +5259,6 @@ spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
 		ASSERT(!locked);
 		ASSERT(vd == vd->vdev_top);
 
-		/*
-		 * XXX - Once we have bp-rewrite this should
-		 * become the common case.
-		 */
-
 		mg = vd->vdev_mg;
 
 		/*
@@ -6438,7 +6432,7 @@ spa_upgrade(spa_t *spa, uint64_t version)
 	 * possible.
 	 */
 	ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
-	ASSERT(version >= spa->spa_uberblock.ub_version);
+	ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
 
 	spa->spa_uberblock.ub_version = version;
 	vdev_config_dirty(spa->spa_root_vdev);

uts/common/fs/zfs/spa_misc.c

@@ -1341,7 +1341,10 @@ snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
 			(void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
 			    sizeof (type));
 		}
-		checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
+		if (!BP_IS_EMBEDDED(bp)) {
+			checksum =
+			    zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
+		}
 		compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
 	}
 
@@ -1643,7 +1646,7 @@ bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
 {
 	uint64_t dsize = 0;
 
-	for (int d = 0; d < SPA_DVAS_PER_BP; d++)
+	for (int d = 0; d < BP_GET_NDVAS(bp); d++)
 		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
 
 	return (dsize);
@@ -1656,7 +1659,7 @@ bp_get_dsize(spa_t *spa, const blkptr_t *bp)
 
 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
 
-	for (int d = 0; d < SPA_DVAS_PER_BP; d++)
+	for (int d = 0; d < BP_GET_NDVAS(bp); d++)
 		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
 
 	spa_config_exit(spa, SCL_VDEV, FTAG);

uts/common/fs/zfs/sys/blkptr.h

@@ -0,0 +1,38 @@
+/*
+ * CDDL HEADER START
+ *
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright (c) 2013 by Delphix. All rights reserved.
+ */
+
+#ifndef _SYS_BLKPTR_H
+#define	_SYS_BLKPTR_H
+
+#include <sys/spa.h>
+#include <sys/zio.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+void encode_embedded_bp_compressed(blkptr_t *, void *,
+    enum zio_compress, int, int);
+void decode_embedded_bp_compressed(const blkptr_t *, void *);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif	/* _SYS_BLKPTR_H */

uts/common/fs/zfs/sys/dbuf.h

@@ -274,6 +274,9 @@ void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
+void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
+    bp_embedded_type_t etype, enum zio_compress comp,
+    int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
 
 void dbuf_clear(dmu_buf_impl_t *db);
 void dbuf_evict(dmu_buf_impl_t *db);

uts/common/fs/zfs/sys/dmu.h

@@ -119,6 +119,14 @@ typedef enum dmu_object_byteswap {
 	((ot) & DMU_OT_METADATA) : \
 	dmu_ot[(ot)].ot_metadata)
 
+/*
+ * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
+ * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
+ * is repurposed for embedded BPs.
+ */
+#define	DMU_OT_HAS_FILL(ot) \
+	((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
+
 #define	DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
 	((ot) & DMU_OT_BYTESWAP_MASK) : \
 	dmu_ot[(ot)].ot_byteswap)
@@ -394,6 +402,11 @@ void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
 void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
     dmu_tx_t *tx);
 
+void
+dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
+    void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
+    int compressed_size, int byteorder, dmu_tx_t *tx);
+
 /*
  * Decide how to write a block: checksum, compression, number of copies, etc.
  */

uts/common/fs/zfs/sys/dmu_impl.h

@@ -294,12 +294,15 @@ typedef struct dmu_sendarg {
 	int dsa_err;
 	dmu_pendop_t dsa_pending_op;
 	boolean_t dsa_incremental;
+	uint64_t dsa_featureflags;
 	uint64_t dsa_last_data_object;
 	uint64_t dsa_last_data_offset;
 } dmu_sendarg_t;
 
 void dmu_object_zapify(objset_t *, uint64_t, dmu_object_type_t, dmu_tx_t *);
 void dmu_object_free_zapified(objset_t *, uint64_t, dmu_tx_t *);
+int dmu_buf_hold_noread(objset_t *, uint64_t, uint64_t,
+    void *, dmu_buf_t **);
 
 #ifdef	__cplusplus
 }

uts/common/fs/zfs/sys/dmu_send.h

@@ -37,12 +37,12 @@ struct dsl_dataset;
 struct drr_begin;
 struct avl_tree;
 
-int dmu_send(const char *tosnap, const char *fromsnap, int outfd,
-    struct vnode *vp, offset_t *off);
+int dmu_send(const char *tosnap, const char *fromsnap, boolean_t embedok,
+    int outfd, struct vnode *vp, offset_t *off);
 int dmu_send_estimate(struct dsl_dataset *ds, struct dsl_dataset *fromds,
     uint64_t *sizep);
 int dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
-    int outfd, struct vnode *vp, offset_t *off);
+    boolean_t embedok, int outfd, vnode_t *vp, offset_t *off);
 
 typedef struct dmu_recv_cookie {
 	struct dsl_dataset *drc_ds;

uts/common/fs/zfs/sys/spa.h

@@ -154,7 +154,7 @@ typedef struct zio_cksum {
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
  * 5	|G|			 offset3				|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
- * 6	|BDX|lvl| type	| cksum | comp	|     PSIZE	|     LSIZE	|
+ * 6	|BDX|lvl| type	| cksum |E| comp|    PSIZE	|     LSIZE	|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
  * 7	|			padding					|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
@@ -188,7 +188,8 @@ typedef struct zio_cksum {
  * G		gang block indicator
  * B		byteorder (endianness)
  * D		dedup
- * X		unused
+ * X		encryption (on version 30, which is not supported)
+ * E		blkptr_t contains embedded data (see below)
  * lvl		level of indirection
  * type		DMU object type
  * phys birth	txg of block allocation; zero if same as logical birth txg
@@ -196,6 +197,100 @@ typedef struct zio_cksum {
  * fill count	number of non-zero blocks under this bp
  * checksum[4]	256-bit checksum of the data this bp describes
  */
+
+/*
+ * "Embedded" blkptr_t's don't actually point to a block, instead they
+ * have a data payload embedded in the blkptr_t itself.  See the comment
+ * in blkptr.c for more details.
+ *
+ * The blkptr_t is laid out as follows:
+ *
+ *	64	56	48	40	32	24	16	8	0
+ *	+-------+-------+-------+-------+-------+-------+-------+-------+
+ * 0	|      payload                                                  |
+ * 1	|      payload                                                  |
+ * 2	|      payload                                                  |
+ * 3	|      payload                                                  |
+ * 4	|      payload                                                  |
+ * 5	|      payload                                                  |
+ *	+-------+-------+-------+-------+-------+-------+-------+-------+
+ * 6	|BDX|lvl| type	| etype |E| comp| PSIZE|              LSIZE	|
+ *	+-------+-------+-------+-------+-------+-------+-------+-------+
+ * 7	|      payload                                                  |
+ * 8	|      payload                                                  |
+ * 9	|      payload                                                  |
+ *	+-------+-------+-------+-------+-------+-------+-------+-------+
+ * a	|			logical birth txg			|
+ *	+-------+-------+-------+-------+-------+-------+-------+-------+
+ * b	|      payload                                                  |
+ * c	|      payload                                                  |
+ * d	|      payload                                                  |
+ * e	|      payload                                                  |
+ * f	|      payload                                                  |
+ *	+-------+-------+-------+-------+-------+-------+-------+-------+
+ *
+ * Legend:
+ *
+ * payload		contains the embedded data
+ * B (byteorder)	byteorder (endianness)
+ * D (dedup)		padding (set to zero)
+ * X			encryption (set to zero; see above)
+ * E (embedded)		set to one
+ * lvl			indirection level
+ * type			DMU object type
+ * etype		how to interpret embedded data (BP_EMBEDDED_TYPE_*)
+ * comp			compression function of payload
+ * PSIZE		size of payload after compression, in bytes
+ * LSIZE		logical size of payload, in bytes
+ *			note that 25 bits is enough to store the largest
+ *			"normal" BP's LSIZE (2^16 * 2^9) in bytes
+ * log. birth		transaction group in which the block was logically born
+ *
+ * Note that LSIZE and PSIZE are stored in bytes, whereas for non-embedded
+ * bp's they are stored in units of SPA_MINBLOCKSHIFT.
+ * Generally, the generic BP_GET_*() macros can be used on embedded BP's.
+ * The B, D, X, lvl, type, and comp fields are stored the same as with normal
+ * BP's so the BP_SET_* macros can be used with them.  etype, PSIZE, LSIZE must
+ * be set with the BPE_SET_* macros.  BP_SET_EMBEDDED() should be called before
+ * other macros, as they assert that they are only used on BP's of the correct
+ * "embedded-ness".
+ */
+
+#define	BPE_GET_ETYPE(bp)	\
+	(ASSERT(BP_IS_EMBEDDED(bp)), \
+	BF64_GET((bp)->blk_prop, 40, 8))
+#define	BPE_SET_ETYPE(bp, t)	do { \
+	ASSERT(BP_IS_EMBEDDED(bp)); \
+	BF64_SET((bp)->blk_prop, 40, 8, t); \
+_NOTE(CONSTCOND) } while (0)
+
+#define	BPE_GET_LSIZE(bp)	\
+	(ASSERT(BP_IS_EMBEDDED(bp)), \
+	BF64_GET_SB((bp)->blk_prop, 0, 25, 0, 1))
+#define	BPE_SET_LSIZE(bp, x)	do { \
+	ASSERT(BP_IS_EMBEDDED(bp)); \
+	BF64_SET_SB((bp)->blk_prop, 0, 25, 0, 1, x); \
+_NOTE(CONSTCOND) } while (0)
+
+#define	BPE_GET_PSIZE(bp)	\
+	(ASSERT(BP_IS_EMBEDDED(bp)), \
+	BF64_GET_SB((bp)->blk_prop, 25, 7, 0, 1))
+#define	BPE_SET_PSIZE(bp, x)	do { \
+	ASSERT(BP_IS_EMBEDDED(bp)); \
+	BF64_SET_SB((bp)->blk_prop, 25, 7, 0, 1, x); \
+_NOTE(CONSTCOND) } while (0)
+
+typedef enum bp_embedded_type {
+	BP_EMBEDDED_TYPE_DATA,
+	BP_EMBEDDED_TYPE_RESERVED, /* Reserved for an unintegrated feature. */
+	NUM_BP_EMBEDDED_TYPES = BP_EMBEDDED_TYPE_RESERVED
+} bp_embedded_type_t;
+
+#define	BPE_NUM_WORDS 14
+#define	BPE_PAYLOAD_SIZE (BPE_NUM_WORDS * sizeof (uint64_t))
+#define	BPE_IS_PAYLOADWORD(bp, wp) \
+	((wp) != &(bp)->blk_prop && (wp) != &(bp)->blk_birth)
+
 #define	SPA_BLKPTRSHIFT	7		/* blkptr_t is 128 bytes	*/
 #define	SPA_DVAS_PER_BP	3		/* Number of DVAs in a bp	*/
 
@@ -242,20 +337,37 @@ typedef struct blkptr {
 #define	DVA_SET_GANG(dva, x)	BF64_SET((dva)->dva_word[1], 63, 1, x)
 
 #define	BP_GET_LSIZE(bp)	\
-	BF64_GET_SB((bp)->blk_prop, 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1)
-#define	BP_SET_LSIZE(bp, x)	\
-	BF64_SET_SB((bp)->blk_prop, 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1, x)
+	(BP_IS_EMBEDDED(bp) ?	\
+	(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA ? BPE_GET_LSIZE(bp) : 0): \
+	BF64_GET_SB((bp)->blk_prop, 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1))
+#define	BP_SET_LSIZE(bp, x)	do { \
+	ASSERT(!BP_IS_EMBEDDED(bp)); \
+	BF64_SET_SB((bp)->blk_prop, \
+	    0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1, x); \
+_NOTE(CONSTCOND) } while (0)
 
 #define	BP_GET_PSIZE(bp)	\
-	BF64_GET_SB((bp)->blk_prop, 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1)
-#define	BP_SET_PSIZE(bp, x)	\
-	BF64_SET_SB((bp)->blk_prop, 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1, x)
+	(BP_IS_EMBEDDED(bp) ? 0 : \
+	BF64_GET_SB((bp)->blk_prop, 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1))
+#define	BP_SET_PSIZE(bp, x)	do { \
+	ASSERT(!BP_IS_EMBEDDED(bp)); \
+	BF64_SET_SB((bp)->blk_prop, \
+	    16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1, x); \
+_NOTE(CONSTCOND) } while (0)
 
-#define	BP_GET_COMPRESS(bp)		BF64_GET((bp)->blk_prop, 32, 8)
-#define	BP_SET_COMPRESS(bp, x)		BF64_SET((bp)->blk_prop, 32, 8, x)
+#define	BP_GET_COMPRESS(bp)		BF64_GET((bp)->blk_prop, 32, 7)
+#define	BP_SET_COMPRESS(bp, x)		BF64_SET((bp)->blk_prop, 32, 7, x)
 
-#define	BP_GET_CHECKSUM(bp)		BF64_GET((bp)->blk_prop, 40, 8)
-#define	BP_SET_CHECKSUM(bp, x)		BF64_SET((bp)->blk_prop, 40, 8, x)
+#define	BP_IS_EMBEDDED(bp)		BF64_GET((bp)->blk_prop, 39, 1)
+#define	BP_SET_EMBEDDED(bp, x)		BF64_SET((bp)->blk_prop, 39, 1, x)
+
+#define	BP_GET_CHECKSUM(bp)		\
+	(BP_IS_EMBEDDED(bp) ? ZIO_CHECKSUM_OFF : \
+	BF64_GET((bp)->blk_prop, 40, 8))
+#define	BP_SET_CHECKSUM(bp, x)		do { \
+	ASSERT(!BP_IS_EMBEDDED(bp)); \
+	BF64_SET((bp)->blk_prop, 40, 8, x); \
+_NOTE(CONSTCOND) } while (0)
 
 #define	BP_GET_TYPE(bp)			BF64_GET((bp)->blk_prop, 48, 8)
 #define	BP_SET_TYPE(bp, x)		BF64_SET((bp)->blk_prop, 48, 8, x)
@@ -263,9 +375,6 @@ typedef struct blkptr {
 #define	BP_GET_LEVEL(bp)		BF64_GET((bp)->blk_prop, 56, 5)
 #define	BP_SET_LEVEL(bp, x)		BF64_SET((bp)->blk_prop, 56, 5, x)
 
-#define	BP_GET_PROP_BIT_61(bp)		BF64_GET((bp)->blk_prop, 61, 1)
-#define	BP_SET_PROP_BIT_61(bp, x)	BF64_SET((bp)->blk_prop, 61, 1, x)
-
 #define	BP_GET_DEDUP(bp)		BF64_GET((bp)->blk_prop, 62, 1)
 #define	BP_SET_DEDUP(bp, x)		BF64_SET((bp)->blk_prop, 62, 1, x)
 
@@ -273,31 +382,39 @@ typedef struct blkptr {
 #define	BP_SET_BYTEORDER(bp, x)		BF64_SET((bp)->blk_prop, 63, 1, x)
 
 #define	BP_PHYSICAL_BIRTH(bp)		\
-	((bp)->blk_phys_birth ? (bp)->blk_phys_birth : (bp)->blk_birth)
+	(BP_IS_EMBEDDED(bp) ? 0 : \
+	(bp)->blk_phys_birth ? (bp)->blk_phys_birth : (bp)->blk_birth)
 
 #define	BP_SET_BIRTH(bp, logical, physical)	\
 {						\
+	ASSERT(!BP_IS_EMBEDDED(bp));		\
 	(bp)->blk_birth = (logical);		\
 	(bp)->blk_phys_birth = ((logical) == (physical) ? 0 : (physical)); \
 }
 
+#define	BP_GET_FILL(bp) (BP_IS_EMBEDDED(bp) ? 1 : (bp)->blk_fill)
+
 #define	BP_GET_ASIZE(bp)	\
-	(DVA_GET_ASIZE(&(bp)->blk_dva[0]) + DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \
-		DVA_GET_ASIZE(&(bp)->blk_dva[2]))
+	(BP_IS_EMBEDDED(bp) ? 0 : \
+	DVA_GET_ASIZE(&(bp)->blk_dva[0]) + \
+	DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \
+	DVA_GET_ASIZE(&(bp)->blk_dva[2]))
 
 #define	BP_GET_UCSIZE(bp) \
 	((BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp))) ? \
 	BP_GET_PSIZE(bp) : BP_GET_LSIZE(bp))
 
 #define	BP_GET_NDVAS(bp)	\
-	(!!DVA_GET_ASIZE(&(bp)->blk_dva[0]) + \
+	(BP_IS_EMBEDDED(bp) ? 0 : \
+	!!DVA_GET_ASIZE(&(bp)->blk_dva[0]) + \
 	!!DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \
 	!!DVA_GET_ASIZE(&(bp)->blk_dva[2]))
 
 #define	BP_COUNT_GANG(bp)	\
+	(BP_IS_EMBEDDED(bp) ? 0 : \
 	(DVA_GET_GANG(&(bp)->blk_dva[0]) + \
 	DVA_GET_GANG(&(bp)->blk_dva[1]) + \
-	DVA_GET_GANG(&(bp)->blk_dva[2]))
+	DVA_GET_GANG(&(bp)->blk_dva[2])))
 
 #define	DVA_EQUAL(dva1, dva2)	\
 	((dva1)->dva_word[1] == (dva2)->dva_word[1] && \
@@ -305,6 +422,7 @@ typedef struct blkptr {
 
 #define	BP_EQUAL(bp1, bp2)	\
 	(BP_PHYSICAL_BIRTH(bp1) == BP_PHYSICAL_BIRTH(bp2) &&	\
+	(bp1)->blk_birth == (bp2)->blk_birth &&			\
 	DVA_EQUAL(&(bp1)->blk_dva[0], &(bp2)->blk_dva[0]) &&	\
 	DVA_EQUAL(&(bp1)->blk_dva[1], &(bp2)->blk_dva[1]) &&	\
 	DVA_EQUAL(&(bp1)->blk_dva[2], &(bp2)->blk_dva[2]))
@@ -325,11 +443,13 @@ typedef struct blkptr {
 	(zcp)->zc_word[3] = w3;			\
 }
 
-#define	BP_IDENTITY(bp)		(&(bp)->blk_dva[0])
-#define	BP_IS_GANG(bp)		DVA_GET_GANG(BP_IDENTITY(bp))
+#define	BP_IDENTITY(bp)		(ASSERT(!BP_IS_EMBEDDED(bp)), &(bp)->blk_dva[0])
+#define	BP_IS_GANG(bp)		\
+	(BP_IS_EMBEDDED(bp) ? B_FALSE : DVA_GET_GANG(BP_IDENTITY(bp)))
 #define	DVA_IS_EMPTY(dva)	((dva)->dva_word[0] == 0ULL &&	\
 				(dva)->dva_word[1] == 0ULL)
-#define	BP_IS_HOLE(bp)		DVA_IS_EMPTY(BP_IDENTITY(bp))
+#define	BP_IS_HOLE(bp) \
+	(!BP_IS_EMBEDDED(bp) && DVA_IS_EMPTY(BP_IDENTITY(bp)))
 
 /* BP_IS_RAIDZ(bp) assumes no block compression */
 #define	BP_IS_RAIDZ(bp)		(DVA_GET_ASIZE(&(bp)->blk_dva[0]) > \
@@ -383,6 +503,17 @@ typedef struct blkptr {
 			    " birth=%lluL",				\
 			    (u_longlong_t)bp->blk_birth);		\
 		}							\
+	} else if (BP_IS_EMBEDDED(bp)) {				\
+		len = func(buf + len, size - len,			\
+		    "EMBEDDED [L%llu %s] et=%u %s "			\
+		    "size=%llxL/%llxP birth=%lluL",			\
+		    (u_longlong_t)BP_GET_LEVEL(bp),			\
+		    type,						\
+		    (int)BPE_GET_ETYPE(bp),				\
+		    compress,						\
+		    (u_longlong_t)BPE_GET_LSIZE(bp),			\
+		    (u_longlong_t)BPE_GET_PSIZE(bp),			\
+		    (u_longlong_t)bp->blk_birth);			\
 	} else {							\
 		for (int d = 0; d < BP_GET_NDVAS(bp); d++) {		\
 			const dva_t *dva = &bp->blk_dva[d];		\
@@ -416,7 +547,7 @@ typedef struct blkptr {
 		    (u_longlong_t)BP_GET_PSIZE(bp),			\
 		    (u_longlong_t)bp->blk_birth,			\
 		    (u_longlong_t)BP_PHYSICAL_BIRTH(bp),		\
-		    (u_longlong_t)bp->blk_fill,				\
+		    (u_longlong_t)BP_GET_FILL(bp),			\
 		    ws,							\
 		    (u_longlong_t)bp->blk_cksum.zc_word[0],		\
 		    (u_longlong_t)bp->blk_cksum.zc_word[1],		\

uts/common/fs/zfs/sys/spa_impl.h

@@ -38,6 +38,7 @@
 #include <sys/refcount.h>
 #include <sys/bplist.h>
 #include <sys/bpobj.h>
+#include <sys/zfeature.h>
 #include <zfeature_common.h>
 
 #ifdef	__cplusplus

uts/common/fs/zfs/sys/zfs_ioctl.h

@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2013 by Delphix. All rights reserved.
  */
 
 #ifndef	_SYS_ZFS_IOCTL_H
@@ -79,15 +79,19 @@ typedef enum drr_headertype {
  * Feature flags for zfs send streams (flags in drr_versioninfo)
  */
 
-#define	DMU_BACKUP_FEATURE_DEDUP	(0x1)
-#define	DMU_BACKUP_FEATURE_DEDUPPROPS	(0x2)
-#define	DMU_BACKUP_FEATURE_SA_SPILL	(0x4)
+#define	DMU_BACKUP_FEATURE_DEDUP		(1<<0)
+#define	DMU_BACKUP_FEATURE_DEDUPPROPS		(1<<1)
+#define	DMU_BACKUP_FEATURE_SA_SPILL		(1<<2)
+/* flags #3 - #15 are reserved for incompatible closed-source implementations */
+#define	DMU_BACKUP_FEATURE_EMBED_DATA		(1<<16)
+#define	DMU_BACKUP_FEATURE_EMBED_DATA_LZ4	(1<<17)
 
 /*
  * Mask of all supported backup features
  */
 #define	DMU_BACKUP_FEATURE_MASK	(DMU_BACKUP_FEATURE_DEDUP | \
-		DMU_BACKUP_FEATURE_DEDUPPROPS | DMU_BACKUP_FEATURE_SA_SPILL)
+    DMU_BACKUP_FEATURE_DEDUPPROPS | DMU_BACKUP_FEATURE_SA_SPILL | \
+    DMU_BACKUP_FEATURE_EMBED_DATA | DMU_BACKUP_FEATURE_EMBED_DATA_LZ4)
 
 /* Are all features in the given flag word currently supported? */
 #define	DMU_STREAM_SUPPORTED(x)	(!((x) & ~DMU_BACKUP_FEATURE_MASK))
@@ -129,7 +133,7 @@ typedef struct dmu_replay_record {
 	enum {
 		DRR_BEGIN, DRR_OBJECT, DRR_FREEOBJECTS,
 		DRR_WRITE, DRR_FREE, DRR_END, DRR_WRITE_BYREF,
-		DRR_SPILL, DRR_NUMTYPES
+		DRR_SPILL, DRR_WRITE_EMBEDDED, DRR_NUMTYPES
 	} drr_type;
 	uint32_t drr_payloadlen;
 	union {
@@ -206,6 +210,19 @@ typedef struct dmu_replay_record {
 			uint64_t drr_pad[4]; /* needed for crypto */
 			/* spill data follows */
 		} drr_spill;
+		struct drr_write_embedded {
+			uint64_t drr_object;
+			uint64_t drr_offset;
+			/* logical length, should equal blocksize */
+			uint64_t drr_length;
+			uint64_t drr_toguid;
+			uint8_t drr_compression;
+			uint8_t drr_etype;
+			uint8_t drr_pad[6];
+			uint32_t drr_lsize; /* uncompressed size of payload */
+			uint32_t drr_psize; /* compr. (real) size of payload */
+			/* (possibly compressed) content follows */
+		} drr_write_embedded;
 	} drr_u;
 } dmu_replay_record_t;
 
@@ -307,8 +324,8 @@ typedef struct zfs_cmd {
 	dmu_objset_stats_t zc_objset_stats;
 	struct drr_begin zc_begin_record;
 	zinject_record_t zc_inject_record;
-	boolean_t	zc_defer_destroy;
-	boolean_t	zc_temphold;
+	uint32_t	zc_defer_destroy;
+	uint32_t	zc_flags;
 	uint64_t	zc_action_handle;
 	int		zc_cleanup_fd;
 	uint8_t		zc_pad[4];		/* alignment */

uts/common/fs/zfs/sys/zio.h

@@ -84,6 +84,12 @@ enum zio_checksum {
 	ZIO_CHECKSUM_FUNCTIONS
 };
 
+/*
+ * The number of "legacy" compression functions which can be set on individual
+ * objects.
+ */
+#define	ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2
+
 #define	ZIO_CHECKSUM_ON_VALUE	ZIO_CHECKSUM_FLETCHER_4
 #define	ZIO_CHECKSUM_DEFAULT	ZIO_CHECKSUM_ON
 
@@ -113,6 +119,12 @@ enum zio_compress {
 	ZIO_COMPRESS_FUNCTIONS
 };
 
+/*
+ * The number of "legacy" compression functions which can be set on individual
+ * objects.
+ */
+#define	ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
+
 /* N.B. when altering this value, also change BOOTFS_COMPRESS_VALID below */
 #define	ZIO_COMPRESS_ON_VALUE	ZIO_COMPRESS_LZJB
 #define	ZIO_COMPRESS_DEFAULT	ZIO_COMPRESS_OFF

uts/common/fs/zfs/zfs_ioctl.c

@@ -4275,6 +4275,7 @@ zfs_ioc_recv(zfs_cmd_t *zc)
  * zc_fromobj	objsetid of incremental fromsnap (may be zero)
  * zc_guid	if set, estimate size of stream only.  zc_cookie is ignored.
  *		output size in zc_objset_type.
+ * zc_flags	if =1, WRITE_EMBEDDED records are permitted
  *
  * outputs:
  * zc_objset_type	estimated size, if zc_guid is set
@@ -4285,6 +4286,7 @@ zfs_ioc_send(zfs_cmd_t *zc)
 	int error;
 	offset_t off;
 	boolean_t estimate = (zc->zc_guid != 0);
+	boolean_t embedok = (zc->zc_flags & 0x1);
 
 	if (zc->zc_obj != 0) {
 		dsl_pool_t *dp;
@@ -4345,7 +4347,7 @@ zfs_ioc_send(zfs_cmd_t *zc)
 
 		off = fp->f_offset;
 		error = dmu_send_obj(zc->zc_name, zc->zc_sendobj,
-		    zc->zc_fromobj, zc->zc_cookie, fp->f_vnode, &off);
+		    zc->zc_fromobj, embedok, zc->zc_cookie, fp->f_vnode, &off);
 
 		if (VOP_SEEK(fp->f_vnode, fp->f_offset, &off, NULL) == 0)
 			fp->f_offset = off;
@@ -5247,6 +5249,8 @@ zfs_ioc_space_snaps(const char *lastsnap, nvlist_t *innvl, nvlist_t *outnvl)
  * innvl: {
  *     "fd" -> file descriptor to write stream to (int32)
  *     (optional) "fromsnap" -> full snap name to send an incremental from
+ *     (optional) "embedok" -> (value ignored)
+ *         presence indicates DRR_WRITE_EMBEDDED records are permitted
  * }
  *
  * outnvl is unused
@@ -5259,6 +5263,7 @@ zfs_ioc_send_new(const char *snapname, nvlist_t *innvl, nvlist_t *outnvl)
 	offset_t off;
 	char *fromname = NULL;
 	int fd;
+	boolean_t embedok;
 
 	error = nvlist_lookup_int32(innvl, "fd", &fd);
 	if (error != 0)
@@ -5266,12 +5271,14 @@ zfs_ioc_send_new(const char *snapname, nvlist_t *innvl, nvlist_t *outnvl)
 
 	(void) nvlist_lookup_string(innvl, "fromsnap", &fromname);
 
+	embedok = nvlist_exists(innvl, "embedok");
+
 	file_t *fp = getf(fd);
 	if (fp == NULL)
 		return (SET_ERROR(EBADF));
 
 	off = fp->f_offset;
-	error = dmu_send(snapname, fromname, fd, fp->f_vnode, &off);
+	error = dmu_send(snapname, fromname, embedok, fd, fp->f_vnode, &off);
 
 	if (VOP_SEEK(fp->f_vnode, fp->f_offset, &off, NULL) == 0)
 		fp->f_offset = off;

uts/common/fs/zfs/zil.c

@@ -137,10 +137,15 @@ int
 zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp)
 {
 	avl_tree_t *t = &zilog->zl_bp_tree;
-	const dva_t *dva = BP_IDENTITY(bp);
+	const dva_t *dva;
 	zil_bp_node_t *zn;
 	avl_index_t where;
 
+	if (BP_IS_EMBEDDED(bp))
+		return (0);
+
+	dva = BP_IDENTITY(bp);
+
 	if (avl_find(t, dva, &where) != NULL)
 		return (SET_ERROR(EEXIST));
 
@@ -831,7 +836,7 @@ zil_lwb_write_done(zio_t *zio)
 	ASSERT(BP_GET_BYTEORDER(zio->io_bp) == ZFS_HOST_BYTEORDER);
 	ASSERT(!BP_IS_GANG(zio->io_bp));
 	ASSERT(!BP_IS_HOLE(zio->io_bp));
-	ASSERT(zio->io_bp->blk_fill == 0);
+	ASSERT(BP_GET_FILL(zio->io_bp) == 0);
 
 	/*
 	 * Ensure the lwb buffer pointer is cleared before releasing

uts/common/fs/zfs/zio.c

@@ -36,6 +36,7 @@
 #include <sys/dmu_objset.h>
 #include <sys/arc.h>
 #include <sys/ddt.h>
+#include <sys/blkptr.h>
 #include <sys/zfeature.h>
 
 /*
@@ -214,7 +215,7 @@ zio_buf_alloc(size_t size)
 {
 	size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
 
-	ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
+	ASSERT3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
 
 	return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE));
 }
@@ -637,6 +638,16 @@ zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 	zio->io_physdone = physdone;
 	zio->io_prop = *zp;
 
+	/*
+	 * Data can be NULL if we are going to call zio_write_override() to
+	 * provide the already-allocated BP.  But we may need the data to
+	 * verify a dedup hit (if requested).  In this case, don't try to
+	 * dedup (just take the already-allocated BP verbatim).
+	 */
+	if (data == NULL && zio->io_prop.zp_dedup_verify) {
+		zio->io_prop.zp_dedup = zio->io_prop.zp_dedup_verify = B_FALSE;
+	}
+
 	return (zio);
 }
 
@@ -676,6 +687,14 @@ zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite)
 void
 zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp)
 {
+
+	/*
+	 * The check for EMBEDDED is a performance optimization.  We
+	 * process the free here (by ignoring it) rather than
+	 * putting it on the list and then processing it in zio_free_sync().
+	 */
+	if (BP_IS_EMBEDDED(bp))
+		return;
 	metaslab_check_free(spa, bp);
 
 	/*
@@ -700,13 +719,13 @@ zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
 	zio_t *zio;
 	enum zio_stage stage = ZIO_FREE_PIPELINE;
 
-	dprintf_bp(bp, "freeing in txg %llu, pass %u",
-	    (longlong_t)txg, spa->spa_sync_pass);
-
 	ASSERT(!BP_IS_HOLE(bp));
 	ASSERT(spa_syncing_txg(spa) == txg);
 	ASSERT(spa_sync_pass(spa) < zfs_sync_pass_deferred_free);
 
+	if (BP_IS_EMBEDDED(bp))
+		return (zio_null(pio, spa, NULL, NULL, NULL, 0));
+
 	metaslab_check_free(spa, bp);
 	arc_freed(spa, bp);
 
@@ -722,7 +741,6 @@ zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
 	    NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_NOW, flags,
 	    NULL, 0, NULL, ZIO_STAGE_OPEN, stage);
 
-
 	return (zio);
 }
 
@@ -732,6 +750,11 @@ zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
 {
 	zio_t *zio;
 
+	dprintf_bp(bp, "claiming in txg %llu", txg);
+
+	if (BP_IS_EMBEDDED(bp))
+		return (zio_null(pio, spa, NULL, NULL, NULL, 0));
+
 	/*
 	 * A claim is an allocation of a specific block.  Claims are needed
 	 * to support immediate writes in the intent log.  The issue is that
@@ -938,12 +961,20 @@ zio_read_bp_init(zio_t *zio)
 	if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF &&
 	    zio->io_child_type == ZIO_CHILD_LOGICAL &&
 	    !(zio->io_flags & ZIO_FLAG_RAW)) {
-		uint64_t psize = BP_GET_PSIZE(bp);
+		uint64_t psize =
+		    BP_IS_EMBEDDED(bp) ? BPE_GET_PSIZE(bp) : BP_GET_PSIZE(bp);
 		void *cbuf = zio_buf_alloc(psize);
 
 		zio_push_transform(zio, cbuf, psize, psize, zio_decompress);
 	}
 
+	if (BP_IS_EMBEDDED(bp) && BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA) {
+		zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
+		decode_embedded_bp_compressed(bp, zio->io_data);
+	} else {
+		ASSERT(!BP_IS_EMBEDDED(bp));
+	}
+
 	if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0)
 		zio->io_flags |= ZIO_FLAG_DONT_CACHE;
 
@@ -987,6 +1018,9 @@ zio_write_bp_init(zio_t *zio)
 		*bp = *zio->io_bp_override;
 		zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
 
+		if (BP_IS_EMBEDDED(bp))
+			return (ZIO_PIPELINE_CONTINUE);
+
 		/*
 		 * If we've been overridden and nopwrite is set then
 		 * set the flag accordingly to indicate that a nopwrite
@@ -1035,7 +1069,7 @@ zio_write_bp_init(zio_t *zio)
 			compress = ZIO_COMPRESS_OFF;
 
 		/* Make sure someone doesn't change their mind on overwrites */
-		ASSERT(MIN(zp->zp_copies + BP_IS_GANG(bp),
+		ASSERT(BP_IS_EMBEDDED(bp) || MIN(zp->zp_copies + BP_IS_GANG(bp),
 		    spa_max_replication(spa)) == BP_GET_NDVAS(bp));
 	}
 
@@ -1045,9 +1079,38 @@ zio_write_bp_init(zio_t *zio)
 		if (psize == 0 || psize == lsize) {
 			compress = ZIO_COMPRESS_OFF;
 			zio_buf_free(cbuf, lsize);
+		} else if (!zp->zp_dedup && psize <= BPE_PAYLOAD_SIZE &&
+		    zp->zp_level == 0 && !DMU_OT_HAS_FILL(zp->zp_type) &&
+		    spa_feature_is_enabled(spa, SPA_FEATURE_EMBEDDED_DATA)) {
+			encode_embedded_bp_compressed(bp,
+			    cbuf, compress, lsize, psize);
+			BPE_SET_ETYPE(bp, BP_EMBEDDED_TYPE_DATA);
+			BP_SET_TYPE(bp, zio->io_prop.zp_type);
+			BP_SET_LEVEL(bp, zio->io_prop.zp_level);
+			zio_buf_free(cbuf, lsize);
+			bp->blk_birth = zio->io_txg;
+			zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
+			ASSERT(spa_feature_is_active(spa,
+			    SPA_FEATURE_EMBEDDED_DATA));
+			return (ZIO_PIPELINE_CONTINUE);
 		} else {
-			ASSERT(psize < lsize);
-			zio_push_transform(zio, cbuf, psize, lsize, NULL);
+			/*
+			 * Round up compressed size to MINBLOCKSIZE and
+			 * zero the tail.
+			 */
+			size_t rounded =
+			    P2ROUNDUP(psize, (size_t)SPA_MINBLOCKSIZE);
+			if (rounded > psize) {
+				bzero((char *)cbuf + psize, rounded - psize);
+				psize = rounded;
+			}
+			if (psize == lsize) {
+				compress = ZIO_COMPRESS_OFF;
+				zio_buf_free(cbuf, lsize);
+			} else {
+				zio_push_transform(zio, cbuf,
+				    psize, lsize, NULL);
+			}
 		}
 	}
 
@@ -2745,7 +2808,7 @@ zio_checksum_verified(zio_t *zio)
 /*
  * ==========================================================================
  * Error rank.  Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other.
- * An error of 0 indictes success.  ENXIO indicates whole-device failure,
+ * An error of 0 indicates success.  ENXIO indicates whole-device failure,
  * which may be transient (e.g. unplugged) or permament.  ECKSUM and EIO
  * indicate errors that are specific to one I/O, and most likely permanent.
  * Any other error is presumed to be worse because we weren't expecting it.
@@ -2855,7 +2918,7 @@ zio_done(zio_t *zio)
 		for (int w = 0; w < ZIO_WAIT_TYPES; w++)
 			ASSERT(zio->io_children[c][w] == 0);
 
-	if (bp != NULL) {
+	if (bp != NULL && !BP_IS_EMBEDDED(bp)) {
 		ASSERT(bp->blk_pad[0] == 0);
 		ASSERT(bp->blk_pad[1] == 0);
 		ASSERT(bcmp(bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 ||

uts/common/fs/zfs/zio_compress.c

@@ -80,7 +80,7 @@ size_t
 zio_compress_data(enum zio_compress c, void *src, void *dst, size_t s_len)
 {
 	uint64_t *word, *word_end;
-	size_t c_len, d_len, r_len;
+	size_t c_len, d_len;
 	zio_compress_info_t *ci = &zio_compress_table[c];
 
 	ASSERT((uint_t)c < ZIO_COMPRESS_FUNCTIONS);
@@ -102,28 +102,13 @@ zio_compress_data(enum zio_compress c, void *src, void *dst, size_t s_len)
 		return (s_len);
 
 	/* Compress at least 12.5% */
-	d_len = P2ALIGN(s_len - (s_len >> 3), (size_t)SPA_MINBLOCKSIZE);
-	if (d_len == 0)
-		return (s_len);
-
+	d_len = s_len - (s_len >> 3);
 	c_len = ci->ci_compress(src, dst, s_len, d_len, ci->ci_level);
 
 	if (c_len > d_len)
 		return (s_len);
 
-	/*
-	 * Cool.  We compressed at least as much as we were hoping to.
-	 * For both security and repeatability, pad out the last sector.
-	 */
-	r_len = P2ROUNDUP(c_len, (size_t)SPA_MINBLOCKSIZE);
-	if (r_len > c_len) {
-		bzero((char *)dst + c_len, r_len - c_len);
-		c_len = r_len;
-	}
-
 	ASSERT3U(c_len, <=, d_len);
-	ASSERT(P2PHASE(c_len, (size_t)SPA_MINBLOCKSIZE) == 0);
-
 	return (c_len);
 }
 

uts/common/fs/zfs/zvol.c

@@ -254,6 +254,8 @@ zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 	    zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
 		return (0);
 
+	VERIFY(!BP_IS_EMBEDDED(bp));
+
 	VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
 	ma->ma_blks++;