1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
  25  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2014 RackTop Systems.
  27  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  28  * Copyright (c) 2014 Integros [integros.com]
  29  * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
  30  * Copyright 2017 Nexenta Systems, Inc.
  31  */
  32 
  33 #include <sys/dmu_objset.h>
  34 #include <sys/dsl_dataset.h>
  35 #include <sys/dsl_dir.h>
  36 #include <sys/dsl_prop.h>
  37 #include <sys/dsl_synctask.h>
  38 #include <sys/dmu_traverse.h>
  39 #include <sys/dmu_impl.h>
  40 #include <sys/dmu_tx.h>
  41 #include <sys/arc.h>
  42 #include <sys/zio.h>
  43 #include <sys/zap.h>
  44 #include <sys/zfeature.h>
  45 #include <sys/unique.h>
  46 #include <sys/zfs_context.h>
  47 #include <sys/zfs_ioctl.h>
  48 #include <sys/spa.h>
  49 #include <sys/vdev.h>
  50 #include <sys/zfs_znode.h>
  51 #include <sys/zfs_onexit.h>
  52 #include <sys/zvol.h>
  53 #include <sys/dsl_scan.h>
  54 #include <sys/dsl_deadlist.h>
  55 #include <sys/dsl_destroy.h>
  56 #include <sys/dsl_userhold.h>
  57 #include <sys/dsl_bookmark.h>
  58 #include <sys/dmu_send.h>
  59 #include <sys/zio_checksum.h>
  60 #include <sys/zio_compress.h>
  61 #include <zfs_fletcher.h>
  62 
  63 /*
  64  * The SPA supports block sizes up to 16MB.  However, very large blocks
  65  * can have an impact on i/o latency (e.g. tying up a spinning disk for
  66  * ~300ms), and also potentially on the memory allocator.  Therefore,
  67  * we do not allow the recordsize to be set larger than zfs_max_recordsize
  68  * (default 1MB).  Larger blocks can be created by changing this tunable,
  69  * and pools with larger blocks can always be imported and used, regardless
  70  * of this setting.
  71  */
  72 int zfs_max_recordsize = 1 * 1024 * 1024;
  73 
  74 #define SWITCH64(x, y) \
  75         { \
  76                 uint64_t __tmp = (x); \
  77                 (x) = (y); \
  78                 (y) = __tmp; \
  79         }
  80 
  81 #define DS_REF_MAX      (1ULL << 62)
  82 
  83 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
  84 
  85 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds,
  86     uint64_t obj, dmu_tx_t *tx);
  87 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds,
  88     dmu_tx_t *tx);
  89 
  90 extern int spa_asize_inflation;
  91 
  92 static zil_header_t zero_zil;
  93 
  94 /*
  95  * Figure out how much of this delta should be propogated to the dsl_dir
  96  * layer.  If there's a refreservation, that space has already been
  97  * partially accounted for in our ancestors.
  98  */
  99 static int64_t
 100 parent_delta(dsl_dataset_t *ds, int64_t delta)
 101 {
 102         dsl_dataset_phys_t *ds_phys;
 103         uint64_t old_bytes, new_bytes;
 104 
 105         if (ds->ds_reserved == 0)
 106                 return (delta);
 107 
 108         ds_phys = dsl_dataset_phys(ds);
 109         old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
 110         new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
 111 
 112         ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
 113         return (new_bytes - old_bytes);
 114 }
 115 
 116 void
 117 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
 118 {
 119         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 120         int compressed = BP_GET_PSIZE(bp);
 121         int uncompressed = BP_GET_UCSIZE(bp);
 122         int64_t delta;
 123 
 124         dprintf_bp(bp, "ds=%p", ds);
 125 
 126         ASSERT(dmu_tx_is_syncing(tx));
 127         /* It could have been compressed away to nothing */
 128         if (BP_IS_HOLE(bp))
 129                 return;
 130         ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
 131         ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
 132         if (ds == NULL) {
 133                 dsl_pool_mos_diduse_space(tx->tx_pool,
 134                     used, compressed, uncompressed);
 135                 return;
 136         }
 137 
 138         ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
 139         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 140         mutex_enter(&ds->ds_lock);
 141         delta = parent_delta(ds, used);
 142         dsl_dataset_phys(ds)->ds_referenced_bytes += used;
 143         dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
 144         dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
 145         dsl_dataset_phys(ds)->ds_unique_bytes += used;
 146 
 147         if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
 148                 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
 149                     B_TRUE;
 150         }
 151 
 152         spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
 153         if (f != SPA_FEATURE_NONE)
 154                 ds->ds_feature_activation_needed[f] = B_TRUE;
 155 
 156         mutex_exit(&ds->ds_lock);
 157         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
 158             compressed, uncompressed, tx);
 159         dsl_dir_transfer_space(ds->ds_dir, used - delta,
 160             DD_USED_REFRSRV, DD_USED_HEAD, tx);
 161 }
 162 
 163 /*
 164  * Called when the specified segment has been remapped, and is thus no
 165  * longer referenced in the head dataset.  The vdev must be indirect.
 166  *
 167  * If the segment is referenced by a snapshot, put it on the remap deadlist.
 168  * Otherwise, add this segment to the obsolete spacemap.
 169  */
 170 void
 171 dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, uint64_t offset,
 172     uint64_t size, uint64_t birth, dmu_tx_t *tx)
 173 {
 174         spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
 175 
 176         ASSERT(dmu_tx_is_syncing(tx));
 177         ASSERT(birth <= tx->tx_txg);
 178         ASSERT(!ds->ds_is_snapshot);
 179 
 180         if (birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
 181                 spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx);
 182         } else {
 183                 blkptr_t fakebp;
 184                 dva_t *dva = &fakebp.blk_dva[0];
 185 
 186                 ASSERT(ds != NULL);
 187 
 188                 mutex_enter(&ds->ds_remap_deadlist_lock);
 189                 if (!dsl_dataset_remap_deadlist_exists(ds)) {
 190                         dsl_dataset_create_remap_deadlist(ds, tx);
 191                 }
 192                 mutex_exit(&ds->ds_remap_deadlist_lock);
 193 
 194                 BP_ZERO(&fakebp);
 195                 fakebp.blk_birth = birth;
 196                 DVA_SET_VDEV(dva, vdev);
 197                 DVA_SET_OFFSET(dva, offset);
 198                 DVA_SET_ASIZE(dva, size);
 199 
 200                 dsl_deadlist_insert(&ds->ds_remap_deadlist, &fakebp, tx);
 201         }
 202 }
 203 
 204 int
 205 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 206     boolean_t async)
 207 {
 208         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 209         int compressed = BP_GET_PSIZE(bp);
 210         int uncompressed = BP_GET_UCSIZE(bp);
 211 
 212         if (BP_IS_HOLE(bp))
 213                 return (0);
 214 
 215         ASSERT(dmu_tx_is_syncing(tx));
 216         ASSERT(bp->blk_birth <= tx->tx_txg);
 217 
 218         if (ds == NULL) {
 219                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 220                 dsl_pool_mos_diduse_space(tx->tx_pool,
 221                     -used, -compressed, -uncompressed);
 222                 return (used);
 223         }
 224         ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
 225 
 226         ASSERT(!ds->ds_is_snapshot);
 227         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 228 
 229         if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
 230                 int64_t delta;
 231 
 232                 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
 233                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 234 
 235                 mutex_enter(&ds->ds_lock);
 236                 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
 237                     !DS_UNIQUE_IS_ACCURATE(ds));
 238                 delta = parent_delta(ds, -used);
 239                 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
 240                 mutex_exit(&ds->ds_lock);
 241                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 242                     delta, -compressed, -uncompressed, tx);
 243                 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
 244                     DD_USED_REFRSRV, DD_USED_HEAD, tx);
 245         } else {
 246                 dprintf_bp(bp, "putting on dead list: %s", "");
 247                 if (async) {
 248                         /*
 249                          * We are here as part of zio's write done callback,
 250                          * which means we're a zio interrupt thread.  We can't
 251                          * call dsl_deadlist_insert() now because it may block
 252                          * waiting for I/O.  Instead, put bp on the deferred
 253                          * queue and let dsl_pool_sync() finish the job.
 254                          */
 255                         bplist_append(&ds->ds_pending_deadlist, bp);
 256                 } else {
 257                         dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
 258                 }
 259                 ASSERT3U(ds->ds_prev->ds_object, ==,
 260                     dsl_dataset_phys(ds)->ds_prev_snap_obj);
 261                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
 262                 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
 263                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
 264                     ds->ds_object && bp->blk_birth >
 265                     dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
 266                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 267                         mutex_enter(&ds->ds_prev->ds_lock);
 268                         dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
 269                         mutex_exit(&ds->ds_prev->ds_lock);
 270                 }
 271                 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
 272                         dsl_dir_transfer_space(ds->ds_dir, used,
 273                             DD_USED_HEAD, DD_USED_SNAP, tx);
 274                 }
 275         }
 276         mutex_enter(&ds->ds_lock);
 277         ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
 278         dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
 279         ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
 280         dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
 281         ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
 282         dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
 283         mutex_exit(&ds->ds_lock);
 284 
 285         return (used);
 286 }
 287 
 288 /*
 289  * We have to release the fsid syncronously or we risk that a subsequent
 290  * mount of the same dataset will fail to unique_insert the fsid.  This
 291  * failure would manifest itself as the fsid of this dataset changing
 292  * between mounts which makes NFS clients quite unhappy.
 293  */
 294 static void
 295 dsl_dataset_evict_sync(void *dbu)
 296 {
 297         dsl_dataset_t *ds = dbu;
 298 
 299         ASSERT(ds->ds_owner == NULL);
 300 
 301         unique_remove(ds->ds_fsid_guid);
 302 }
 303 
 304 static void
 305 dsl_dataset_evict_async(void *dbu)
 306 {
 307         dsl_dataset_t *ds = dbu;
 308 
 309         ASSERT(ds->ds_owner == NULL);
 310 
 311         ds->ds_dbuf = NULL;
 312 
 313         if (ds->ds_objset != NULL)
 314                 dmu_objset_evict(ds->ds_objset);
 315 
 316         if (ds->ds_prev) {
 317                 dsl_dataset_rele(ds->ds_prev, ds);
 318                 ds->ds_prev = NULL;
 319         }
 320 
 321         bplist_destroy(&ds->ds_pending_deadlist);
 322         if (dsl_deadlist_is_open(&ds->ds_deadlist))
 323                 dsl_deadlist_close(&ds->ds_deadlist);
 324         if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
 325                 dsl_deadlist_close(&ds->ds_remap_deadlist);
 326         if (ds->ds_dir)
 327                 dsl_dir_async_rele(ds->ds_dir, ds);
 328 
 329         ASSERT(!list_link_active(&ds->ds_synced_link));
 330 
 331         list_destroy(&ds->ds_prop_cbs);
 332         mutex_destroy(&ds->ds_lock);
 333         mutex_destroy(&ds->ds_opening_lock);
 334         mutex_destroy(&ds->ds_sendstream_lock);
 335         mutex_destroy(&ds->ds_remap_deadlist_lock);
 336         refcount_destroy(&ds->ds_longholds);
 337         rrw_destroy(&ds->ds_bp_rwlock);
 338 
 339         kmem_free(ds, sizeof (dsl_dataset_t));
 340 }
 341 
 342 int
 343 dsl_dataset_get_snapname(dsl_dataset_t *ds)
 344 {
 345         dsl_dataset_phys_t *headphys;
 346         int err;
 347         dmu_buf_t *headdbuf;
 348         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 349         objset_t *mos = dp->dp_meta_objset;
 350 
 351         if (ds->ds_snapname[0])
 352                 return (0);
 353         if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
 354                 return (0);
 355 
 356         err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
 357             FTAG, &headdbuf);
 358         if (err != 0)
 359                 return (err);
 360         headphys = headdbuf->db_data;
 361         err = zap_value_search(dp->dp_meta_objset,
 362             headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
 363         dmu_buf_rele(headdbuf, FTAG);
 364         return (err);
 365 }
 366 
 367 int
 368 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
 369 {
 370         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 371         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 372         matchtype_t mt = 0;
 373         int err;
 374 
 375         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 376                 mt = MT_NORMALIZE;
 377 
 378         err = zap_lookup_norm(mos, snapobj, name, 8, 1,
 379             value, mt, NULL, 0, NULL);
 380         if (err == ENOTSUP && (mt & MT_NORMALIZE))
 381                 err = zap_lookup(mos, snapobj, name, 8, 1, value);
 382         return (err);
 383 }
 384 
 385 int
 386 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
 387     boolean_t adj_cnt)
 388 {
 389         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 390         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 391         matchtype_t mt = 0;
 392         int err;
 393 
 394         dsl_dir_snap_cmtime_update(ds->ds_dir);
 395 
 396         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 397                 mt = MT_NORMALIZE;
 398 
 399         err = zap_remove_norm(mos, snapobj, name, mt, tx);
 400         if (err == ENOTSUP && (mt & MT_NORMALIZE))
 401                 err = zap_remove(mos, snapobj, name, tx);
 402 
 403         if (err == 0 && adj_cnt)
 404                 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
 405                     DD_FIELD_SNAPSHOT_COUNT, tx);
 406 
 407         return (err);
 408 }
 409 
 410 boolean_t
 411 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
 412 {
 413         dmu_buf_t *dbuf = ds->ds_dbuf;
 414         boolean_t result = B_FALSE;
 415 
 416         if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
 417             ds->ds_object, DMU_BONUS_BLKID, tag)) {
 418 
 419                 if (ds == dmu_buf_get_user(dbuf))
 420                         result = B_TRUE;
 421                 else
 422                         dmu_buf_rele(dbuf, tag);
 423         }
 424 
 425         return (result);
 426 }
 427 
 428 int
 429 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 430     dsl_dataset_t **dsp)
 431 {
 432         objset_t *mos = dp->dp_meta_objset;
 433         dmu_buf_t *dbuf;
 434         dsl_dataset_t *ds;
 435         int err;
 436         dmu_object_info_t doi;
 437 
 438         ASSERT(dsl_pool_config_held(dp));
 439 
 440         err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
 441         if (err != 0)
 442                 return (err);
 443 
 444         /* Make sure dsobj has the correct object type. */
 445         dmu_object_info_from_db(dbuf, &doi);
 446         if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
 447                 dmu_buf_rele(dbuf, tag);
 448                 return (SET_ERROR(EINVAL));
 449         }
 450 
 451         ds = dmu_buf_get_user(dbuf);
 452         if (ds == NULL) {
 453                 dsl_dataset_t *winner = NULL;
 454 
 455                 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
 456                 ds->ds_dbuf = dbuf;
 457                 ds->ds_object = dsobj;
 458                 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
 459 
 460                 err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj,
 461                     NULL, ds, &ds->ds_dir);
 462                 if (err != 0) {
 463                         kmem_free(ds, sizeof (dsl_dataset_t));
 464                         dmu_buf_rele(dbuf, tag);
 465                         return (err);
 466                 }
 467 
 468                 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
 469                 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 470                 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
 471                 mutex_init(&ds->ds_remap_deadlist_lock,
 472                     NULL, MUTEX_DEFAULT, NULL);
 473                 rrw_init(&ds->ds_bp_rwlock, B_FALSE);
 474                 refcount_create(&ds->ds_longholds);
 475 
 476                 bplist_create(&ds->ds_pending_deadlist);
 477 
 478                 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
 479                     offsetof(dmu_sendarg_t, dsa_link));
 480 
 481                 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
 482                     offsetof(dsl_prop_cb_record_t, cbr_ds_node));
 483 
 484                 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 485                         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 486                                 if (!(spa_feature_table[f].fi_flags &
 487                                     ZFEATURE_FLAG_PER_DATASET))
 488                                         continue;
 489                                 err = zap_contains(mos, dsobj,
 490                                     spa_feature_table[f].fi_guid);
 491                                 if (err == 0) {
 492                                         ds->ds_feature_inuse[f] = B_TRUE;
 493                                 } else {
 494                                         ASSERT3U(err, ==, ENOENT);
 495                                         err = 0;
 496                                 }
 497                         }
 498                 }
 499 
 500                 if (!ds->ds_is_snapshot) {
 501                         ds->ds_snapname[0] = '\0';
 502                         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 503                                 err = dsl_dataset_hold_obj(dp,
 504                                     dsl_dataset_phys(ds)->ds_prev_snap_obj,
 505                                     ds, &ds->ds_prev);
 506                         }
 507                         if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 508                                 int zaperr = zap_lookup(mos, ds->ds_object,
 509                                     DS_FIELD_BOOKMARK_NAMES,
 510                                     sizeof (ds->ds_bookmarks), 1,
 511                                     &ds->ds_bookmarks);
 512                                 if (zaperr != ENOENT)
 513                                         VERIFY0(zaperr);
 514                         }
 515                 } else {
 516                         if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
 517                                 err = dsl_dataset_get_snapname(ds);
 518                         if (err == 0 &&
 519                             dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
 520                                 err = zap_count(
 521                                     ds->ds_dir->dd_pool->dp_meta_objset,
 522                                     dsl_dataset_phys(ds)->ds_userrefs_obj,
 523                                     &ds->ds_userrefs);
 524                         }
 525                 }
 526 
 527                 if (err == 0 && !ds->ds_is_snapshot) {
 528                         err = dsl_prop_get_int_ds(ds,
 529                             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
 530                             &ds->ds_reserved);
 531                         if (err == 0) {
 532                                 err = dsl_prop_get_int_ds(ds,
 533                                     zfs_prop_to_name(ZFS_PROP_REFQUOTA),
 534                                     &ds->ds_quota);
 535                         }
 536                 } else {
 537                         ds->ds_reserved = ds->ds_quota = 0;
 538                 }
 539 
 540                 dsl_deadlist_open(&ds->ds_deadlist,
 541                     mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
 542                 uint64_t remap_deadlist_obj =
 543                     dsl_dataset_get_remap_deadlist_object(ds);
 544                 if (remap_deadlist_obj != 0) {
 545                         dsl_deadlist_open(&ds->ds_remap_deadlist, mos,
 546                             remap_deadlist_obj);
 547                 }
 548 
 549                 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync,
 550                     dsl_dataset_evict_async, &ds->ds_dbuf);
 551                 if (err == 0)
 552                         winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
 553 
 554                 if (err != 0 || winner != NULL) {
 555                         bplist_destroy(&ds->ds_pending_deadlist);
 556                         dsl_deadlist_close(&ds->ds_deadlist);
 557                         if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
 558                                 dsl_deadlist_close(&ds->ds_remap_deadlist);
 559                         if (ds->ds_prev)
 560                                 dsl_dataset_rele(ds->ds_prev, ds);
 561                         dsl_dir_rele(ds->ds_dir, ds);
 562                         mutex_destroy(&ds->ds_lock);
 563                         mutex_destroy(&ds->ds_opening_lock);
 564                         mutex_destroy(&ds->ds_sendstream_lock);
 565                         refcount_destroy(&ds->ds_longholds);
 566                         kmem_free(ds, sizeof (dsl_dataset_t));
 567                         if (err != 0) {
 568                                 dmu_buf_rele(dbuf, tag);
 569                                 return (err);
 570                         }
 571                         ds = winner;
 572                 } else {
 573                         ds->ds_fsid_guid =
 574                             unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
 575                         if (ds->ds_fsid_guid !=
 576                             dsl_dataset_phys(ds)->ds_fsid_guid) {
 577                                 zfs_dbgmsg("ds_fsid_guid changed from "
 578                                     "%llx to %llx for pool %s dataset id %llu",
 579                                     (long long)
 580                                     dsl_dataset_phys(ds)->ds_fsid_guid,
 581                                     (long long)ds->ds_fsid_guid,
 582                                     spa_name(dp->dp_spa),
 583                                     dsobj);
 584                         }
 585                 }
 586         }
 587         ASSERT3P(ds->ds_dbuf, ==, dbuf);
 588         ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
 589         ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
 590             spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
 591             dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
 592         *dsp = ds;
 593         return (0);
 594 }
 595 
 596 int
 597 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
 598     void *tag, dsl_dataset_t **dsp)
 599 {
 600         dsl_dir_t *dd;
 601         const char *snapname;
 602         uint64_t obj;
 603         int err = 0;
 604         dsl_dataset_t *ds;
 605 
 606         err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
 607         if (err != 0)
 608                 return (err);
 609 
 610         ASSERT(dsl_pool_config_held(dp));
 611         obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
 612         if (obj != 0)
 613                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
 614         else
 615                 err = SET_ERROR(ENOENT);
 616 
 617         /* we may be looking for a snapshot */
 618         if (err == 0 && snapname != NULL) {
 619                 dsl_dataset_t *snap_ds;
 620 
 621                 if (*snapname++ != '@') {
 622                         dsl_dataset_rele(ds, tag);
 623                         dsl_dir_rele(dd, FTAG);
 624                         return (SET_ERROR(ENOENT));
 625                 }
 626 
 627                 dprintf("looking for snapshot '%s'\n", snapname);
 628                 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
 629                 if (err == 0)
 630                         err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
 631                 dsl_dataset_rele(ds, tag);
 632 
 633                 if (err == 0) {
 634                         mutex_enter(&snap_ds->ds_lock);
 635                         if (snap_ds->ds_snapname[0] == 0)
 636                                 (void) strlcpy(snap_ds->ds_snapname, snapname,
 637                                     sizeof (snap_ds->ds_snapname));
 638                         mutex_exit(&snap_ds->ds_lock);
 639                         ds = snap_ds;
 640                 }
 641         }
 642         if (err == 0)
 643                 *dsp = ds;
 644         dsl_dir_rele(dd, FTAG);
 645         return (err);
 646 }
 647 
 648 int
 649 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
 650     void *tag, dsl_dataset_t **dsp)
 651 {
 652         int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
 653         if (err != 0)
 654                 return (err);
 655         if (!dsl_dataset_tryown(*dsp, tag)) {
 656                 dsl_dataset_rele(*dsp, tag);
 657                 *dsp = NULL;
 658                 return (SET_ERROR(EBUSY));
 659         }
 660         return (0);
 661 }
 662 
 663 int
 664 dsl_dataset_own(dsl_pool_t *dp, const char *name,
 665     void *tag, dsl_dataset_t **dsp)
 666 {
 667         int err = dsl_dataset_hold(dp, name, tag, dsp);
 668         if (err != 0)
 669                 return (err);
 670         if (!dsl_dataset_tryown(*dsp, tag)) {
 671                 dsl_dataset_rele(*dsp, tag);
 672                 return (SET_ERROR(EBUSY));
 673         }
 674         return (0);
 675 }
 676 
 677 /*
 678  * See the comment above dsl_pool_hold() for details.  In summary, a long
 679  * hold is used to prevent destruction of a dataset while the pool hold
 680  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
 681  *
 682  * The dataset and pool must be held when this function is called.  After it
 683  * is called, the pool hold may be released while the dataset is still held
 684  * and accessed.
 685  */
 686 void
 687 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
 688 {
 689         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 690         (void) refcount_add(&ds->ds_longholds, tag);
 691 }
 692 
 693 void
 694 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
 695 {
 696         (void) refcount_remove(&ds->ds_longholds, tag);
 697 }
 698 
 699 /* Return B_TRUE if there are any long holds on this dataset. */
 700 boolean_t
 701 dsl_dataset_long_held(dsl_dataset_t *ds)
 702 {
 703         return (!refcount_is_zero(&ds->ds_longholds));
 704 }
 705 
 706 void
 707 dsl_dataset_name(dsl_dataset_t *ds, char *name)
 708 {
 709         if (ds == NULL) {
 710                 (void) strcpy(name, "mos");
 711         } else {
 712                 dsl_dir_name(ds->ds_dir, name);
 713                 VERIFY0(dsl_dataset_get_snapname(ds));
 714                 if (ds->ds_snapname[0]) {
 715                         VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
 716                             <, ZFS_MAX_DATASET_NAME_LEN);
 717                         /*
 718                          * We use a "recursive" mutex so that we
 719                          * can call dprintf_ds() with ds_lock held.
 720                          */
 721                         if (!MUTEX_HELD(&ds->ds_lock)) {
 722                                 mutex_enter(&ds->ds_lock);
 723                                 VERIFY3U(strlcat(name, ds->ds_snapname,
 724                                     ZFS_MAX_DATASET_NAME_LEN), <,
 725                                     ZFS_MAX_DATASET_NAME_LEN);
 726                                 mutex_exit(&ds->ds_lock);
 727                         } else {
 728                                 VERIFY3U(strlcat(name, ds->ds_snapname,
 729                                     ZFS_MAX_DATASET_NAME_LEN), <,
 730                                     ZFS_MAX_DATASET_NAME_LEN);
 731                         }
 732                 }
 733         }
 734 }
 735 
 736 int
 737 dsl_dataset_namelen(dsl_dataset_t *ds)
 738 {
 739         VERIFY0(dsl_dataset_get_snapname(ds));
 740         mutex_enter(&ds->ds_lock);
 741         int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
 742         mutex_exit(&ds->ds_lock);
 743         return (len);
 744 }
 745 
 746 void
 747 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
 748 {
 749         dmu_buf_rele(ds->ds_dbuf, tag);
 750 }
 751 
 752 void
 753 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
 754 {
 755         ASSERT3P(ds->ds_owner, ==, tag);
 756         ASSERT(ds->ds_dbuf != NULL);
 757 
 758         mutex_enter(&ds->ds_lock);
 759         ds->ds_owner = NULL;
 760         mutex_exit(&ds->ds_lock);
 761         dsl_dataset_long_rele(ds, tag);
 762         dsl_dataset_rele(ds, tag);
 763 }
 764 
 765 boolean_t
 766 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
 767 {
 768         boolean_t gotit = FALSE;
 769 
 770         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 771         mutex_enter(&ds->ds_lock);
 772         if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
 773                 ds->ds_owner = tag;
 774                 dsl_dataset_long_hold(ds, tag);
 775                 gotit = TRUE;
 776         }
 777         mutex_exit(&ds->ds_lock);
 778         return (gotit);
 779 }
 780 
 781 boolean_t
 782 dsl_dataset_has_owner(dsl_dataset_t *ds)
 783 {
 784         boolean_t rv;
 785         mutex_enter(&ds->ds_lock);
 786         rv = (ds->ds_owner != NULL);
 787         mutex_exit(&ds->ds_lock);
 788         return (rv);
 789 }
 790 
 791 static void
 792 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
 793 {
 794         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 795         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
 796         uint64_t zero = 0;
 797 
 798         VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
 799 
 800         spa_feature_incr(spa, f, tx);
 801         dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
 802 
 803         VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
 804             sizeof (zero), 1, &zero, tx));
 805 }
 806 
 807 void
 808 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
 809 {
 810         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 811         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
 812 
 813         VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
 814 
 815         VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
 816         spa_feature_decr(spa, f, tx);
 817 }
 818 
 819 uint64_t
 820 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
 821     uint64_t flags, dmu_tx_t *tx)
 822 {
 823         dsl_pool_t *dp = dd->dd_pool;
 824         dmu_buf_t *dbuf;
 825         dsl_dataset_phys_t *dsphys;
 826         uint64_t dsobj;
 827         objset_t *mos = dp->dp_meta_objset;
 828 
 829         if (origin == NULL)
 830                 origin = dp->dp_origin_snap;
 831 
 832         ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
 833         ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
 834         ASSERT(dmu_tx_is_syncing(tx));
 835         ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
 836 
 837         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
 838             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
 839         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
 840         dmu_buf_will_dirty(dbuf, tx);
 841         dsphys = dbuf->db_data;
 842         bzero(dsphys, sizeof (dsl_dataset_phys_t));
 843         dsphys->ds_dir_obj = dd->dd_object;
 844         dsphys->ds_flags = flags;
 845         dsphys->ds_fsid_guid = unique_create();
 846         (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
 847             sizeof (dsphys->ds_guid));
 848         dsphys->ds_snapnames_zapobj =
 849             zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
 850             DMU_OT_NONE, 0, tx);
 851         dsphys->ds_creation_time = gethrestime_sec();
 852         dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
 853 
 854         if (origin == NULL) {
 855                 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
 856         } else {
 857                 dsl_dataset_t *ohds; /* head of the origin snapshot */
 858 
 859                 dsphys->ds_prev_snap_obj = origin->ds_object;
 860                 dsphys->ds_prev_snap_txg =
 861                     dsl_dataset_phys(origin)->ds_creation_txg;
 862                 dsphys->ds_referenced_bytes =
 863                     dsl_dataset_phys(origin)->ds_referenced_bytes;
 864                 dsphys->ds_compressed_bytes =
 865                     dsl_dataset_phys(origin)->ds_compressed_bytes;
 866                 dsphys->ds_uncompressed_bytes =
 867                     dsl_dataset_phys(origin)->ds_uncompressed_bytes;
 868                 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG);
 869                 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
 870                 rrw_exit(&origin->ds_bp_rwlock, FTAG);
 871 
 872                 /*
 873                  * Inherit flags that describe the dataset's contents
 874                  * (INCONSISTENT) or properties (Case Insensitive).
 875                  */
 876                 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
 877                     (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
 878 
 879                 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 880                         if (origin->ds_feature_inuse[f])
 881                                 dsl_dataset_activate_feature(dsobj, f, tx);
 882                 }
 883 
 884                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
 885                 dsl_dataset_phys(origin)->ds_num_children++;
 886 
 887                 VERIFY0(dsl_dataset_hold_obj(dp,
 888                     dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
 889                     FTAG, &ohds));
 890                 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
 891                     dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
 892                 dsl_dataset_rele(ohds, FTAG);
 893 
 894                 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
 895                         if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
 896                                 dsl_dataset_phys(origin)->ds_next_clones_obj =
 897                                     zap_create(mos,
 898                                     DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
 899                         }
 900                         VERIFY0(zap_add_int(mos,
 901                             dsl_dataset_phys(origin)->ds_next_clones_obj,
 902                             dsobj, tx));
 903                 }
 904 
 905                 dmu_buf_will_dirty(dd->dd_dbuf, tx);
 906                 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
 907                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 908                         if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
 909                                 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
 910                                 dsl_dir_phys(origin->ds_dir)->dd_clones =
 911                                     zap_create(mos,
 912                                     DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
 913                         }
 914                         VERIFY0(zap_add_int(mos,
 915                             dsl_dir_phys(origin->ds_dir)->dd_clones,
 916                             dsobj, tx));
 917                 }
 918         }
 919 
 920         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
 921                 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 922 
 923         dmu_buf_rele(dbuf, FTAG);
 924 
 925         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 926         dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
 927 
 928         return (dsobj);
 929 }
 930 
 931 static void
 932 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
 933 {
 934         objset_t *os;
 935 
 936         VERIFY0(dmu_objset_from_ds(ds, &os));
 937         if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) {
 938                 dsl_pool_t *dp = ds->ds_dir->dd_pool;
 939                 zio_t *zio;
 940 
 941                 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
 942 
 943                 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
 944                 dsl_dataset_sync(ds, zio, tx);
 945                 VERIFY0(zio_wait(zio));
 946 
 947                 /* dsl_dataset_sync_done will drop this reference. */
 948                 dmu_buf_add_ref(ds->ds_dbuf, ds);
 949                 dsl_dataset_sync_done(ds, tx);
 950         }
 951 }
 952 
 953 uint64_t
 954 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
 955     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
 956 {
 957         dsl_pool_t *dp = pdd->dd_pool;
 958         uint64_t dsobj, ddobj;
 959         dsl_dir_t *dd;
 960 
 961         ASSERT(dmu_tx_is_syncing(tx));
 962         ASSERT(lastname[0] != '@');
 963 
 964         ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
 965         VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
 966 
 967         dsobj = dsl_dataset_create_sync_dd(dd, origin,
 968             flags & ~DS_CREATE_FLAG_NODIRTY, tx);
 969 
 970         dsl_deleg_set_create_perms(dd, tx, cr);
 971 
 972         /*
 973          * Since we're creating a new node we know it's a leaf, so we can
 974          * initialize the counts if the limit feature is active.
 975          */
 976         if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
 977                 uint64_t cnt = 0;
 978                 objset_t *os = dd->dd_pool->dp_meta_objset;
 979 
 980                 dsl_dir_zapify(dd, tx);
 981                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
 982                     sizeof (cnt), 1, &cnt, tx));
 983                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
 984                     sizeof (cnt), 1, &cnt, tx));
 985         }
 986 
 987         dsl_dir_rele(dd, FTAG);
 988 
 989         /*
 990          * If we are creating a clone, make sure we zero out any stale
 991          * data from the origin snapshots zil header.
 992          */
 993         if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
 994                 dsl_dataset_t *ds;
 995 
 996                 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
 997                 dsl_dataset_zero_zil(ds, tx);
 998                 dsl_dataset_rele(ds, FTAG);
 999         }
1000 
1001         return (dsobj);
1002 }
1003 
1004 /*
1005  * The unique space in the head dataset can be calculated by subtracting
1006  * the space used in the most recent snapshot, that is still being used
1007  * in this file system, from the space currently in use.  To figure out
1008  * the space in the most recent snapshot still in use, we need to take
1009  * the total space used in the snapshot and subtract out the space that
1010  * has been freed up since the snapshot was taken.
1011  */
1012 void
1013 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1014 {
1015         uint64_t mrs_used;
1016         uint64_t dlused, dlcomp, dluncomp;
1017 
1018         ASSERT(!ds->ds_is_snapshot);
1019 
1020         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1021                 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1022         else
1023                 mrs_used = 0;
1024 
1025         dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1026 
1027         ASSERT3U(dlused, <=, mrs_used);
1028         dsl_dataset_phys(ds)->ds_unique_bytes =
1029             dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1030 
1031         if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1032             SPA_VERSION_UNIQUE_ACCURATE)
1033                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1034 }
1035 
1036 void
1037 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1038     dmu_tx_t *tx)
1039 {
1040         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1041         uint64_t count;
1042         int err;
1043 
1044         ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1045         err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1046             obj, tx);
1047         /*
1048          * The err should not be ENOENT, but a bug in a previous version
1049          * of the code could cause upgrade_clones_cb() to not set
1050          * ds_next_snap_obj when it should, leading to a missing entry.
1051          * If we knew that the pool was created after
1052          * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1053          * ENOENT.  However, at least we can check that we don't have
1054          * too many entries in the next_clones_obj even after failing to
1055          * remove this one.
1056          */
1057         if (err != ENOENT)
1058                 VERIFY0(err);
1059         ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1060             &count));
1061         ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1062 }
1063 
1064 
1065 blkptr_t *
1066 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1067 {
1068         return (&dsl_dataset_phys(ds)->ds_bp);
1069 }
1070 
1071 spa_t *
1072 dsl_dataset_get_spa(dsl_dataset_t *ds)
1073 {
1074         return (ds->ds_dir->dd_pool->dp_spa);
1075 }
1076 
1077 void
1078 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1079 {
1080         dsl_pool_t *dp;
1081 
1082         if (ds == NULL) /* this is the meta-objset */
1083                 return;
1084 
1085         ASSERT(ds->ds_objset != NULL);
1086 
1087         if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1088                 panic("dirtying snapshot!");
1089 
1090         /* Must not dirty a dataset in the same txg where it got snapshotted. */
1091         ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
1092 
1093         dp = ds->ds_dir->dd_pool;
1094         if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1095                 /* up the hold count until we can be written out */
1096                 dmu_buf_add_ref(ds->ds_dbuf, ds);
1097         }
1098 }
1099 
1100 boolean_t
1101 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1102 {
1103         for (int t = 0; t < TXG_SIZE; t++) {
1104                 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1105                     ds, t))
1106                         return (B_TRUE);
1107         }
1108         return (B_FALSE);
1109 }
1110 
1111 static int
1112 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1113 {
1114         uint64_t asize;
1115 
1116         if (!dmu_tx_is_syncing(tx))
1117                 return (0);
1118 
1119         /*
1120          * If there's an fs-only reservation, any blocks that might become
1121          * owned by the snapshot dataset must be accommodated by space
1122          * outside of the reservation.
1123          */
1124         ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1125         asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1126         if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1127                 return (SET_ERROR(ENOSPC));
1128 
1129         /*
1130          * Propagate any reserved space for this snapshot to other
1131          * snapshot checks in this sync group.
1132          */
1133         if (asize > 0)
1134                 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1135 
1136         return (0);
1137 }
1138 
1139 int
1140 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1141     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1142 {
1143         int error;
1144         uint64_t value;
1145 
1146         ds->ds_trysnap_txg = tx->tx_txg;
1147 
1148         if (!dmu_tx_is_syncing(tx))
1149                 return (0);
1150 
1151         /*
1152          * We don't allow multiple snapshots of the same txg.  If there
1153          * is already one, try again.
1154          */
1155         if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1156                 return (SET_ERROR(EAGAIN));
1157 
1158         /*
1159          * Check for conflicting snapshot name.
1160          */
1161         error = dsl_dataset_snap_lookup(ds, snapname, &value);
1162         if (error == 0)
1163                 return (SET_ERROR(EEXIST));
1164         if (error != ENOENT)
1165                 return (error);
1166 
1167         /*
1168          * We don't allow taking snapshots of inconsistent datasets, such as
1169          * those into which we are currently receiving.  However, if we are
1170          * creating this snapshot as part of a receive, this check will be
1171          * executed atomically with respect to the completion of the receive
1172          * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1173          * case we ignore this, knowing it will be fixed up for us shortly in
1174          * dmu_recv_end_sync().
1175          */
1176         if (!recv && DS_IS_INCONSISTENT(ds))
1177                 return (SET_ERROR(EBUSY));
1178 
1179         /*
1180          * Skip the check for temporary snapshots or if we have already checked
1181          * the counts in dsl_dataset_snapshot_check. This means we really only
1182          * check the count here when we're receiving a stream.
1183          */
1184         if (cnt != 0 && cr != NULL) {
1185                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1186                     ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1187                 if (error != 0)
1188                         return (error);
1189         }
1190 
1191         error = dsl_dataset_snapshot_reserve_space(ds, tx);
1192         if (error != 0)
1193                 return (error);
1194 
1195         return (0);
1196 }
1197 
1198 int
1199 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1200 {
1201         dsl_dataset_snapshot_arg_t *ddsa = arg;
1202         dsl_pool_t *dp = dmu_tx_pool(tx);
1203         nvpair_t *pair;
1204         int rv = 0;
1205 
1206         /*
1207          * Pre-compute how many total new snapshots will be created for each
1208          * level in the tree and below. This is needed for validating the
1209          * snapshot limit when either taking a recursive snapshot or when
1210          * taking multiple snapshots.
1211          *
1212          * The problem is that the counts are not actually adjusted when
1213          * we are checking, only when we finally sync. For a single snapshot,
1214          * this is easy, the count will increase by 1 at each node up the tree,
1215          * but its more complicated for the recursive/multiple snapshot case.
1216          *
1217          * The dsl_fs_ss_limit_check function does recursively check the count
1218          * at each level up the tree but since it is validating each snapshot
1219          * independently we need to be sure that we are validating the complete
1220          * count for the entire set of snapshots. We do this by rolling up the
1221          * counts for each component of the name into an nvlist and then
1222          * checking each of those cases with the aggregated count.
1223          *
1224          * This approach properly handles not only the recursive snapshot
1225          * case (where we get all of those on the ddsa_snaps list) but also
1226          * the sibling case (e.g. snapshot a/b and a/c so that we will also
1227          * validate the limit on 'a' using a count of 2).
1228          *
1229          * We validate the snapshot names in the third loop and only report
1230          * name errors once.
1231          */
1232         if (dmu_tx_is_syncing(tx)) {
1233                 nvlist_t *cnt_track = NULL;
1234                 cnt_track = fnvlist_alloc();
1235 
1236                 /* Rollup aggregated counts into the cnt_track list */
1237                 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1238                     pair != NULL;
1239                     pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1240                         char *pdelim;
1241                         uint64_t val;
1242                         char nm[MAXPATHLEN];
1243 
1244                         (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1245                         pdelim = strchr(nm, '@');
1246                         if (pdelim == NULL)
1247                                 continue;
1248                         *pdelim = '\0';
1249 
1250                         do {
1251                                 if (nvlist_lookup_uint64(cnt_track, nm,
1252                                     &val) == 0) {
1253                                         /* update existing entry */
1254                                         fnvlist_add_uint64(cnt_track, nm,
1255                                             val + 1);
1256                                 } else {
1257                                         /* add to list */
1258                                         fnvlist_add_uint64(cnt_track, nm, 1);
1259                                 }
1260 
1261                                 pdelim = strrchr(nm, '/');
1262                                 if (pdelim != NULL)
1263                                         *pdelim = '\0';
1264                         } while (pdelim != NULL);
1265                 }
1266 
1267                 /* Check aggregated counts at each level */
1268                 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1269                     pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1270                         int error = 0;
1271                         char *name;
1272                         uint64_t cnt = 0;
1273                         dsl_dataset_t *ds;
1274 
1275                         name = nvpair_name(pair);
1276                         cnt = fnvpair_value_uint64(pair);
1277                         ASSERT(cnt > 0);
1278 
1279                         error = dsl_dataset_hold(dp, name, FTAG, &ds);
1280                         if (error == 0) {
1281                                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1282                                     ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1283                                     ddsa->ddsa_cr);
1284                                 dsl_dataset_rele(ds, FTAG);
1285                         }
1286 
1287                         if (error != 0) {
1288                                 if (ddsa->ddsa_errors != NULL)
1289                                         fnvlist_add_int32(ddsa->ddsa_errors,
1290                                             name, error);
1291                                 rv = error;
1292                                 /* only report one error for this check */
1293                                 break;
1294                         }
1295                 }
1296                 nvlist_free(cnt_track);
1297         }
1298 
1299         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1300             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1301                 int error = 0;
1302                 dsl_dataset_t *ds;
1303                 char *name, *atp;
1304                 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1305 
1306                 name = nvpair_name(pair);
1307                 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1308                         error = SET_ERROR(ENAMETOOLONG);
1309                 if (error == 0) {
1310                         atp = strchr(name, '@');
1311                         if (atp == NULL)
1312                                 error = SET_ERROR(EINVAL);
1313                         if (error == 0)
1314                                 (void) strlcpy(dsname, name, atp - name + 1);
1315                 }
1316                 if (error == 0)
1317                         error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1318                 if (error == 0) {
1319                         /* passing 0/NULL skips dsl_fs_ss_limit_check */
1320                         error = dsl_dataset_snapshot_check_impl(ds,
1321                             atp + 1, tx, B_FALSE, 0, NULL);
1322                         dsl_dataset_rele(ds, FTAG);
1323                 }
1324 
1325                 if (error != 0) {
1326                         if (ddsa->ddsa_errors != NULL) {
1327                                 fnvlist_add_int32(ddsa->ddsa_errors,
1328                                     name, error);
1329                         }
1330                         rv = error;
1331                 }
1332         }
1333 
1334         return (rv);
1335 }
1336 
1337 void
1338 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1339     dmu_tx_t *tx)
1340 {
1341         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1342         dmu_buf_t *dbuf;
1343         dsl_dataset_phys_t *dsphys;
1344         uint64_t dsobj, crtxg;
1345         objset_t *mos = dp->dp_meta_objset;
1346         objset_t *os;
1347 
1348         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1349 
1350         /*
1351          * If we are on an old pool, the zil must not be active, in which
1352          * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1353          */
1354         ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1355             dmu_objset_from_ds(ds, &os) != 0 ||
1356             bcmp(&os->os_phys->os_zil_header, &zero_zil,
1357             sizeof (zero_zil)) == 0);
1358 
1359         /* Should not snapshot a dirty dataset. */
1360         ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1361             ds, tx->tx_txg));
1362 
1363         dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1364 
1365         /*
1366          * The origin's ds_creation_txg has to be < TXG_INITIAL
1367          */
1368         if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1369                 crtxg = 1;
1370         else
1371                 crtxg = tx->tx_txg;
1372 
1373         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1374             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1375         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1376         dmu_buf_will_dirty(dbuf, tx);
1377         dsphys = dbuf->db_data;
1378         bzero(dsphys, sizeof (dsl_dataset_phys_t));
1379         dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1380         dsphys->ds_fsid_guid = unique_create();
1381         (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1382             sizeof (dsphys->ds_guid));
1383         dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1384         dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1385         dsphys->ds_next_snap_obj = ds->ds_object;
1386         dsphys->ds_num_children = 1;
1387         dsphys->ds_creation_time = gethrestime_sec();
1388         dsphys->ds_creation_txg = crtxg;
1389         dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1390         dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1391         dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1392         dsphys->ds_uncompressed_bytes =
1393             dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1394         dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1395         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1396         dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1397         rrw_exit(&ds->ds_bp_rwlock, FTAG);
1398         dmu_buf_rele(dbuf, FTAG);
1399 
1400         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1401                 if (ds->ds_feature_inuse[f])
1402                         dsl_dataset_activate_feature(dsobj, f, tx);
1403         }
1404 
1405         ASSERT3U(ds->ds_prev != 0, ==,
1406             dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1407         if (ds->ds_prev) {
1408                 uint64_t next_clones_obj =
1409                     dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1410                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1411                     ds->ds_object ||
1412                     dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1413                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1414                     ds->ds_object) {
1415                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1416                         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1417                             dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1418                         dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1419                 } else if (next_clones_obj != 0) {
1420                         dsl_dataset_remove_from_next_clones(ds->ds_prev,
1421                             dsphys->ds_next_snap_obj, tx);
1422                         VERIFY0(zap_add_int(mos,
1423                             next_clones_obj, dsobj, tx));
1424                 }
1425         }
1426 
1427         /*
1428          * If we have a reference-reservation on this dataset, we will
1429          * need to increase the amount of refreservation being charged
1430          * since our unique space is going to zero.
1431          */
1432         if (ds->ds_reserved) {
1433                 int64_t delta;
1434                 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1435                 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1436                     ds->ds_reserved);
1437                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1438                     delta, 0, 0, tx);
1439         }
1440 
1441         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1442         dsl_dataset_phys(ds)->ds_deadlist_obj =
1443             dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1444             dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1445         dsl_deadlist_close(&ds->ds_deadlist);
1446         dsl_deadlist_open(&ds->ds_deadlist, mos,
1447             dsl_dataset_phys(ds)->ds_deadlist_obj);
1448         dsl_deadlist_add_key(&ds->ds_deadlist,
1449             dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1450 
1451         if (dsl_dataset_remap_deadlist_exists(ds)) {
1452                 uint64_t remap_deadlist_obj =
1453                     dsl_dataset_get_remap_deadlist_object(ds);
1454                 /*
1455                  * Move the remap_deadlist to the snapshot.  The head
1456                  * will create a new remap deadlist on demand, from
1457                  * dsl_dataset_block_remapped().
1458                  */
1459                 dsl_dataset_unset_remap_deadlist_object(ds, tx);
1460                 dsl_deadlist_close(&ds->ds_remap_deadlist);
1461 
1462                 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1463                 VERIFY0(zap_add(mos, dsobj, DS_FIELD_REMAP_DEADLIST,
1464                     sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj, tx));
1465         }
1466 
1467         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1468         dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1469         dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1470         dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1471 
1472         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1473                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1474 
1475         VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1476             snapname, 8, 1, &dsobj, tx));
1477 
1478         if (ds->ds_prev)
1479                 dsl_dataset_rele(ds->ds_prev, ds);
1480         VERIFY0(dsl_dataset_hold_obj(dp,
1481             dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1482 
1483         dsl_scan_ds_snapshotted(ds, tx);
1484 
1485         dsl_dir_snap_cmtime_update(ds->ds_dir);
1486 
1487         spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1488 }
1489 
1490 void
1491 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1492 {
1493         dsl_dataset_snapshot_arg_t *ddsa = arg;
1494         dsl_pool_t *dp = dmu_tx_pool(tx);
1495         nvpair_t *pair;
1496 
1497         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1498             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1499                 dsl_dataset_t *ds;
1500                 char *name, *atp;
1501                 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1502 
1503                 name = nvpair_name(pair);
1504                 atp = strchr(name, '@');
1505                 (void) strlcpy(dsname, name, atp - name + 1);
1506                 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1507 
1508                 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1509                 if (ddsa->ddsa_props != NULL) {
1510                         dsl_props_set_sync_impl(ds->ds_prev,
1511                             ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1512                 }
1513                 dsl_dataset_rele(ds, FTAG);
1514         }
1515 }
1516 
1517 /*
1518  * The snapshots must all be in the same pool.
1519  * All-or-nothing: if there are any failures, nothing will be modified.
1520  */
1521 int
1522 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1523 {
1524         dsl_dataset_snapshot_arg_t ddsa;
1525         nvpair_t *pair;
1526         boolean_t needsuspend;
1527         int error;
1528         spa_t *spa;
1529         char *firstname;
1530         nvlist_t *suspended = NULL;
1531 
1532         pair = nvlist_next_nvpair(snaps, NULL);
1533         if (pair == NULL)
1534                 return (0);
1535         firstname = nvpair_name(pair);
1536 
1537         error = spa_open(firstname, &spa, FTAG);
1538         if (error != 0)
1539                 return (error);
1540         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1541         spa_close(spa, FTAG);
1542 
1543         if (needsuspend) {
1544                 suspended = fnvlist_alloc();
1545                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1546                     pair = nvlist_next_nvpair(snaps, pair)) {
1547                         char fsname[ZFS_MAX_DATASET_NAME_LEN];
1548                         char *snapname = nvpair_name(pair);
1549                         char *atp;
1550                         void *cookie;
1551 
1552                         atp = strchr(snapname, '@');
1553                         if (atp == NULL) {
1554                                 error = SET_ERROR(EINVAL);
1555                                 break;
1556                         }
1557                         (void) strlcpy(fsname, snapname, atp - snapname + 1);
1558 
1559                         error = zil_suspend(fsname, &cookie);
1560                         if (error != 0)
1561                                 break;
1562                         fnvlist_add_uint64(suspended, fsname,
1563                             (uintptr_t)cookie);
1564                 }
1565         }
1566 
1567         ddsa.ddsa_snaps = snaps;
1568         ddsa.ddsa_props = props;
1569         ddsa.ddsa_errors = errors;
1570         ddsa.ddsa_cr = CRED();
1571 
1572         if (error == 0) {
1573                 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1574                     dsl_dataset_snapshot_sync, &ddsa,
1575                     fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1576         }
1577 
1578         if (suspended != NULL) {
1579                 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1580                     pair = nvlist_next_nvpair(suspended, pair)) {
1581                         zil_resume((void *)(uintptr_t)
1582                             fnvpair_value_uint64(pair));
1583                 }
1584                 fnvlist_free(suspended);
1585         }
1586 
1587         return (error);
1588 }
1589 
1590 typedef struct dsl_dataset_snapshot_tmp_arg {
1591         const char *ddsta_fsname;
1592         const char *ddsta_snapname;
1593         minor_t ddsta_cleanup_minor;
1594         const char *ddsta_htag;
1595 } dsl_dataset_snapshot_tmp_arg_t;
1596 
1597 static int
1598 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1599 {
1600         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1601         dsl_pool_t *dp = dmu_tx_pool(tx);
1602         dsl_dataset_t *ds;
1603         int error;
1604 
1605         error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1606         if (error != 0)
1607                 return (error);
1608 
1609         /* NULL cred means no limit check for tmp snapshot */
1610         error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1611             tx, B_FALSE, 0, NULL);
1612         if (error != 0) {
1613                 dsl_dataset_rele(ds, FTAG);
1614                 return (error);
1615         }
1616 
1617         if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1618                 dsl_dataset_rele(ds, FTAG);
1619                 return (SET_ERROR(ENOTSUP));
1620         }
1621         error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1622             B_TRUE, tx);
1623         if (error != 0) {
1624                 dsl_dataset_rele(ds, FTAG);
1625                 return (error);
1626         }
1627 
1628         dsl_dataset_rele(ds, FTAG);
1629         return (0);
1630 }
1631 
1632 static void
1633 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1634 {
1635         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1636         dsl_pool_t *dp = dmu_tx_pool(tx);
1637         dsl_dataset_t *ds;
1638 
1639         VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1640 
1641         dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1642         dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1643             ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1644         dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1645 
1646         dsl_dataset_rele(ds, FTAG);
1647 }
1648 
1649 int
1650 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1651     minor_t cleanup_minor, const char *htag)
1652 {
1653         dsl_dataset_snapshot_tmp_arg_t ddsta;
1654         int error;
1655         spa_t *spa;
1656         boolean_t needsuspend;
1657         void *cookie;
1658 
1659         ddsta.ddsta_fsname = fsname;
1660         ddsta.ddsta_snapname = snapname;
1661         ddsta.ddsta_cleanup_minor = cleanup_minor;
1662         ddsta.ddsta_htag = htag;
1663 
1664         error = spa_open(fsname, &spa, FTAG);
1665         if (error != 0)
1666                 return (error);
1667         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1668         spa_close(spa, FTAG);
1669 
1670         if (needsuspend) {
1671                 error = zil_suspend(fsname, &cookie);
1672                 if (error != 0)
1673                         return (error);
1674         }
1675 
1676         error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1677             dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1678 
1679         if (needsuspend)
1680                 zil_resume(cookie);
1681         return (error);
1682 }
1683 
1684 void
1685 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1686 {
1687         ASSERT(dmu_tx_is_syncing(tx));
1688         ASSERT(ds->ds_objset != NULL);
1689         ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1690 
1691         /*
1692          * in case we had to change ds_fsid_guid when we opened it,
1693          * sync it out now.
1694          */
1695         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1696         dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1697 
1698         if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1699                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1700                     ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1701                     &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1702                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1703                     ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1704                     &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1705                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1706                     ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1707                     &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1708                 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1709                 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1710                 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1711         }
1712 
1713         dmu_objset_sync(ds->ds_objset, zio, tx);
1714 
1715         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1716                 if (ds->ds_feature_activation_needed[f]) {
1717                         if (ds->ds_feature_inuse[f])
1718                                 continue;
1719                         dsl_dataset_activate_feature(ds->ds_object, f, tx);
1720                         ds->ds_feature_inuse[f] = B_TRUE;
1721                 }
1722         }
1723 }
1724 
1725 static int
1726 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1727 {
1728         dsl_deadlist_t *dl = arg;
1729         dsl_deadlist_insert(dl, bp, tx);
1730         return (0);
1731 }
1732 
1733 void
1734 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx)
1735 {
1736         objset_t *os = ds->ds_objset;
1737 
1738         bplist_iterate(&ds->ds_pending_deadlist,
1739             deadlist_enqueue_cb, &ds->ds_deadlist, tx);
1740 
1741         if (os->os_synced_dnodes != NULL) {
1742                 multilist_destroy(os->os_synced_dnodes);
1743                 os->os_synced_dnodes = NULL;
1744         }
1745 
1746         ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx)));
1747 
1748         dmu_buf_rele(ds->ds_dbuf, ds);
1749 }
1750 
1751 int
1752 get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val)
1753 {
1754         uint64_t count = 0;
1755         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1756         zap_cursor_t zc;
1757         zap_attribute_t za;
1758 
1759         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1760 
1761         /*
1762          * There may be missing entries in ds_next_clones_obj
1763          * due to a bug in a previous version of the code.
1764          * Only trust it if it has the right number of entries.
1765          */
1766         if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1767                 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1768                     &count));
1769         }
1770         if (count != dsl_dataset_phys(ds)->ds_num_children - 1) {
1771                 return (ENOENT);
1772         }
1773         for (zap_cursor_init(&zc, mos,
1774             dsl_dataset_phys(ds)->ds_next_clones_obj);
1775             zap_cursor_retrieve(&zc, &za) == 0;
1776             zap_cursor_advance(&zc)) {
1777                 dsl_dataset_t *clone;
1778                 char buf[ZFS_MAX_DATASET_NAME_LEN];
1779                 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1780                     za.za_first_integer, FTAG, &clone));
1781                 dsl_dir_name(clone->ds_dir, buf);
1782                 fnvlist_add_boolean(val, buf);
1783                 dsl_dataset_rele(clone, FTAG);
1784         }
1785         zap_cursor_fini(&zc);
1786         return (0);
1787 }
1788 
1789 void
1790 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1791 {
1792         nvlist_t *propval = fnvlist_alloc();
1793         nvlist_t *val;
1794 
1795         /*
1796          * We use nvlist_alloc() instead of fnvlist_alloc() because the
1797          * latter would allocate the list with NV_UNIQUE_NAME flag.
1798          * As a result, every time a clone name is appended to the list
1799          * it would be (linearly) searched for for a duplicate name.
1800          * We already know that all clone names must be unique and we
1801          * want avoid the quadratic complexity of double-checking that
1802          * because we can have a large number of clones.
1803          */
1804         VERIFY0(nvlist_alloc(&val, 0, KM_SLEEP));
1805 
1806         if (get_clones_stat_impl(ds, val) == 0) {
1807                 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1808                 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
1809                     propval);
1810         }
1811 
1812         nvlist_free(val);
1813         nvlist_free(propval);
1814 }
1815 
1816 /*
1817  * Returns a string that represents the receive resume stats token. It should
1818  * be freed with strfree().
1819  */
1820 char *
1821 get_receive_resume_stats_impl(dsl_dataset_t *ds)
1822 {
1823         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1824 
1825         if (dsl_dataset_has_resume_receive_state(ds)) {
1826                 char *str;
1827                 void *packed;
1828                 uint8_t *compressed;
1829                 uint64_t val;
1830                 nvlist_t *token_nv = fnvlist_alloc();
1831                 size_t packed_size, compressed_size;
1832 
1833                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1834                     DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1835                         fnvlist_add_uint64(token_nv, "fromguid", val);
1836                 }
1837                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1838                     DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1839                         fnvlist_add_uint64(token_nv, "object", val);
1840                 }
1841                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1842                     DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1843                         fnvlist_add_uint64(token_nv, "offset", val);
1844                 }
1845                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1846                     DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1847                         fnvlist_add_uint64(token_nv, "bytes", val);
1848                 }
1849                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1850                     DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1851                         fnvlist_add_uint64(token_nv, "toguid", val);
1852                 }
1853                 char buf[256];
1854                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1855                     DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1856                         fnvlist_add_string(token_nv, "toname", buf);
1857                 }
1858                 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1859                     DS_FIELD_RESUME_LARGEBLOCK) == 0) {
1860                         fnvlist_add_boolean(token_nv, "largeblockok");
1861                 }
1862                 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1863                     DS_FIELD_RESUME_EMBEDOK) == 0) {
1864                         fnvlist_add_boolean(token_nv, "embedok");
1865                 }
1866                 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1867                     DS_FIELD_RESUME_COMPRESSOK) == 0) {
1868                         fnvlist_add_boolean(token_nv, "compressok");
1869                 }
1870                 packed = fnvlist_pack(token_nv, &packed_size);
1871                 fnvlist_free(token_nv);
1872                 compressed = kmem_alloc(packed_size, KM_SLEEP);
1873 
1874                 compressed_size = gzip_compress(packed, compressed,
1875                     packed_size, packed_size, 6);
1876 
1877                 zio_cksum_t cksum;
1878                 fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1879 
1880                 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1881                 for (int i = 0; i < compressed_size; i++) {
1882                         (void) sprintf(str + i * 2, "%02x", compressed[i]);
1883                 }
1884                 str[compressed_size * 2] = '\0';
1885                 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1886                     ZFS_SEND_RESUME_TOKEN_VERSION,
1887                     (longlong_t)cksum.zc_word[0],
1888                     (longlong_t)packed_size, str);
1889                 kmem_free(packed, packed_size);
1890                 kmem_free(str, compressed_size * 2 + 1);
1891                 kmem_free(compressed, packed_size);
1892                 return (propval);
1893         }
1894         return (strdup(""));
1895 }
1896 
1897 /*
1898  * Returns a string that represents the receive resume stats token of the
1899  * dataset's child. It should be freed with strfree().
1900  */
1901 char *
1902 get_child_receive_stats(dsl_dataset_t *ds)
1903 {
1904         char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1905         dsl_dataset_t *recv_ds;
1906         dsl_dataset_name(ds, recvname);
1907         if (strlcat(recvname, "/", sizeof (recvname)) <
1908             sizeof (recvname) &&
1909             strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1910             sizeof (recvname) &&
1911             dsl_dataset_hold(ds->ds_dir->dd_pool, recvname, FTAG,
1912             &recv_ds)  == 0) {
1913                 char *propval = get_receive_resume_stats_impl(recv_ds);
1914                 dsl_dataset_rele(recv_ds, FTAG);
1915                 return (propval);
1916         }
1917         return (strdup(""));
1918 }
1919 
1920 static void
1921 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1922 {
1923         char *propval = get_receive_resume_stats_impl(ds);
1924         if (strcmp(propval, "") != 0) {
1925                 dsl_prop_nvlist_add_string(nv,
1926                     ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1927         } else {
1928                 char *childval = get_child_receive_stats(ds);
1929                 if (strcmp(childval, "") != 0) {
1930                         dsl_prop_nvlist_add_string(nv,
1931                             ZFS_PROP_RECEIVE_RESUME_TOKEN, childval);
1932                 }
1933                 strfree(childval);
1934         }
1935         strfree(propval);
1936 }
1937 
1938 uint64_t
1939 dsl_get_refratio(dsl_dataset_t *ds)
1940 {
1941         uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1942             (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1943             dsl_dataset_phys(ds)->ds_compressed_bytes);
1944         return (ratio);
1945 }
1946 
1947 uint64_t
1948 dsl_get_logicalreferenced(dsl_dataset_t *ds)
1949 {
1950         return (dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1951 }
1952 
1953 uint64_t
1954 dsl_get_compressratio(dsl_dataset_t *ds)
1955 {
1956         if (ds->ds_is_snapshot) {
1957                 return (dsl_get_refratio(ds));
1958         } else {
1959                 dsl_dir_t *dd = ds->ds_dir;
1960                 mutex_enter(&dd->dd_lock);
1961                 uint64_t val = dsl_dir_get_compressratio(dd);
1962                 mutex_exit(&dd->dd_lock);
1963                 return (val);
1964         }
1965 }
1966 
1967 uint64_t
1968 dsl_get_used(dsl_dataset_t *ds)
1969 {
1970         if (ds->ds_is_snapshot) {
1971                 return (dsl_dataset_phys(ds)->ds_unique_bytes);
1972         } else {
1973                 dsl_dir_t *dd = ds->ds_dir;
1974                 mutex_enter(&dd->dd_lock);
1975                 uint64_t val = dsl_dir_get_used(dd);
1976                 mutex_exit(&dd->dd_lock);
1977                 return (val);
1978         }
1979 }
1980 
1981 uint64_t
1982 dsl_get_creation(dsl_dataset_t *ds)
1983 {
1984         return (dsl_dataset_phys(ds)->ds_creation_time);
1985 }
1986 
1987 uint64_t
1988 dsl_get_creationtxg(dsl_dataset_t *ds)
1989 {
1990         return (dsl_dataset_phys(ds)->ds_creation_txg);
1991 }
1992 
1993 uint64_t
1994 dsl_get_refquota(dsl_dataset_t *ds)
1995 {
1996         return (ds->ds_quota);
1997 }
1998 
1999 uint64_t
2000 dsl_get_refreservation(dsl_dataset_t *ds)
2001 {
2002         return (ds->ds_reserved);
2003 }
2004 
2005 uint64_t
2006 dsl_get_guid(dsl_dataset_t *ds)
2007 {
2008         return (dsl_dataset_phys(ds)->ds_guid);
2009 }
2010 
2011 uint64_t
2012 dsl_get_unique(dsl_dataset_t *ds)
2013 {
2014         return (dsl_dataset_phys(ds)->ds_unique_bytes);
2015 }
2016 
2017 uint64_t
2018 dsl_get_objsetid(dsl_dataset_t *ds)
2019 {
2020         return (ds->ds_object);
2021 }
2022 
2023 uint64_t
2024 dsl_get_userrefs(dsl_dataset_t *ds)
2025 {
2026         return (ds->ds_userrefs);
2027 }
2028 
2029 uint64_t
2030 dsl_get_defer_destroy(dsl_dataset_t *ds)
2031 {
2032         return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2033 }
2034 
2035 uint64_t
2036 dsl_get_referenced(dsl_dataset_t *ds)
2037 {
2038         return (dsl_dataset_phys(ds)->ds_referenced_bytes);
2039 }
2040 
2041 uint64_t
2042 dsl_get_numclones(dsl_dataset_t *ds)
2043 {
2044         ASSERT(ds->ds_is_snapshot);
2045         return (dsl_dataset_phys(ds)->ds_num_children - 1);
2046 }
2047 
2048 uint64_t
2049 dsl_get_inconsistent(dsl_dataset_t *ds)
2050 {
2051         return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ?
2052             1 : 0);
2053 }
2054 
2055 uint64_t
2056 dsl_get_available(dsl_dataset_t *ds)
2057 {
2058         uint64_t refdbytes = dsl_get_referenced(ds);
2059         uint64_t availbytes = dsl_dir_space_available(ds->ds_dir,
2060             NULL, 0, TRUE);
2061         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2062                 availbytes +=
2063                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2064         }
2065         if (ds->ds_quota != 0) {
2066                 /*
2067                  * Adjust available bytes according to refquota
2068                  */
2069                 if (refdbytes < ds->ds_quota) {
2070                         availbytes = MIN(availbytes,
2071                             ds->ds_quota - refdbytes);
2072                 } else {
2073                         availbytes = 0;
2074                 }
2075         }
2076         return (availbytes);
2077 }
2078 
2079 int
2080 dsl_get_written(dsl_dataset_t *ds, uint64_t *written)
2081 {
2082         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2083         dsl_dataset_t *prev;
2084         int err = dsl_dataset_hold_obj(dp,
2085             dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
2086         if (err == 0) {
2087                 uint64_t comp, uncomp;
2088                 err = dsl_dataset_space_written(prev, ds, written,
2089                     &comp, &uncomp);
2090                 dsl_dataset_rele(prev, FTAG);
2091         }
2092         return (err);
2093 }
2094 
2095 /*
2096  * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN.
2097  */
2098 int
2099 dsl_get_prev_snap(dsl_dataset_t *ds, char *snap)
2100 {
2101         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2102         if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
2103                 dsl_dataset_name(ds->ds_prev, snap);
2104                 return (0);
2105         } else {
2106                 return (ENOENT);
2107         }
2108 }
2109 
2110 /*
2111  * Returns the mountpoint property and source for the given dataset in the value
2112  * and source buffers. The value buffer must be at least as large as MAXPATHLEN
2113  * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN.
2114  * Returns 0 on success and an error on failure.
2115  */
2116 int
2117 dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value,
2118     char *source)
2119 {
2120         int error;
2121         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2122 
2123         /* Retrieve the mountpoint value stored in the zap opbject */
2124         error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1,
2125             ZAP_MAXVALUELEN, value, source);
2126         if (error != 0) {
2127                 return (error);
2128         }
2129 
2130         /* Process the dsname and source to find the full mountpoint string */
2131         if (value[0] == '/') {
2132                 char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP);
2133                 char *root = buf;
2134                 const char *relpath;
2135 
2136                 /*
2137                  * If we inherit the mountpoint, even from a dataset
2138                  * with a received value, the source will be the path of
2139                  * the dataset we inherit from. If source is
2140                  * ZPROP_SOURCE_VAL_RECVD, the received value is not
2141                  * inherited.
2142                  */
2143                 if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) {
2144                         relpath = "";
2145                 } else {
2146                         ASSERT0(strncmp(dsname, source, strlen(source)));
2147                         relpath = dsname + strlen(source);
2148                         if (relpath[0] == '/')
2149                                 relpath++;
2150                 }
2151 
2152                 spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN);
2153 
2154                 /*
2155                  * Special case an alternate root of '/'. This will
2156                  * avoid having multiple leading slashes in the
2157                  * mountpoint path.
2158                  */
2159                 if (strcmp(root, "/") == 0)
2160                         root++;
2161 
2162                 /*
2163                  * If the mountpoint is '/' then skip over this
2164                  * if we are obtaining either an alternate root or
2165                  * an inherited mountpoint.
2166                  */
2167                 char *mnt = value;
2168                 if (value[1] == '\0' && (root[0] != '\0' ||
2169                     relpath[0] != '\0'))
2170                         mnt = value + 1;
2171 
2172                 if (relpath[0] == '\0') {
2173                         (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s",
2174                             root, mnt);
2175                 } else {
2176                         (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s",
2177                             root, mnt, relpath[0] == '@' ? "" : "/",
2178                             relpath);
2179                 }
2180                 kmem_free(buf, ZAP_MAXVALUELEN);
2181         } else {
2182                 /* 'legacy' or 'none' */
2183                 (void) snprintf(value, ZAP_MAXVALUELEN, "%s", value);
2184         }
2185         return (0);
2186 }
2187 
2188 void
2189 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2190 {
2191         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2192 
2193         ASSERT(dsl_pool_config_held(dp));
2194 
2195         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO,
2196             dsl_get_refratio(ds));
2197         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
2198             dsl_get_logicalreferenced(ds));
2199         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2200             dsl_get_compressratio(ds));
2201         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2202             dsl_get_used(ds));
2203 
2204         if (ds->ds_is_snapshot) {
2205                 get_clones_stat(ds, nv);
2206         } else {
2207                 char buf[ZFS_MAX_DATASET_NAME_LEN];
2208                 if (dsl_get_prev_snap(ds, buf) == 0)
2209                         dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP,
2210                             buf);
2211                 dsl_dir_stats(ds->ds_dir, nv);
2212         }
2213 
2214         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE,
2215             dsl_get_available(ds));
2216         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED,
2217             dsl_get_referenced(ds));
2218         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2219             dsl_get_creation(ds));
2220         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2221             dsl_get_creationtxg(ds));
2222         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2223             dsl_get_refquota(ds));
2224         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2225             dsl_get_refreservation(ds));
2226         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2227             dsl_get_guid(ds));
2228         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2229             dsl_get_unique(ds));
2230         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2231             dsl_get_objsetid(ds));
2232         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2233             dsl_get_userrefs(ds));
2234         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2235             dsl_get_defer_destroy(ds));
2236 
2237         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
2238                 uint64_t written;
2239                 if (dsl_get_written(ds, &written) == 0) {
2240                         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2241                             written);
2242                 }
2243         }
2244 
2245         if (!dsl_dataset_is_snapshot(ds)) {
2246                 /*
2247                  * A failed "newfs" (e.g. full) resumable receive leaves
2248                  * the stats set on this dataset.  Check here for the prop.
2249                  */
2250                 get_receive_resume_stats(ds, nv);
2251 
2252                 /*
2253                  * A failed incremental resumable receive leaves the
2254                  * stats set on our child named "%recv".  Check the child
2255                  * for the prop.
2256                  */
2257                 /* 6 extra bytes for /%recv */
2258                 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
2259                 dsl_dataset_t *recv_ds;
2260                 dsl_dataset_name(ds, recvname);
2261                 if (strlcat(recvname, "/", sizeof (recvname)) <
2262                     sizeof (recvname) &&
2263                     strlcat(recvname, recv_clone_name, sizeof (recvname)) <
2264                     sizeof (recvname) &&
2265                     dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
2266                         get_receive_resume_stats(recv_ds, nv);
2267                         dsl_dataset_rele(recv_ds, FTAG);
2268                 }
2269         }
2270 }
2271 
2272 void
2273 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2274 {
2275         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2276         ASSERT(dsl_pool_config_held(dp));
2277 
2278         stat->dds_creation_txg = dsl_get_creationtxg(ds);
2279         stat->dds_inconsistent = dsl_get_inconsistent(ds);
2280         stat->dds_guid = dsl_get_guid(ds);
2281         stat->dds_origin[0] = '\0';
2282         if (ds->ds_is_snapshot) {
2283                 stat->dds_is_snapshot = B_TRUE;
2284                 stat->dds_num_clones = dsl_get_numclones(ds);
2285         } else {
2286                 stat->dds_is_snapshot = B_FALSE;
2287                 stat->dds_num_clones = 0;
2288 
2289                 if (dsl_dir_is_clone(ds->ds_dir)) {
2290                         dsl_dir_get_origin(ds->ds_dir, stat->dds_origin);
2291                 }
2292         }
2293 }
2294 
2295 uint64_t
2296 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2297 {
2298         return (ds->ds_fsid_guid);
2299 }
2300 
2301 void
2302 dsl_dataset_space(dsl_dataset_t *ds,
2303     uint64_t *refdbytesp, uint64_t *availbytesp,
2304     uint64_t *usedobjsp, uint64_t *availobjsp)
2305 {
2306         *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
2307         *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2308         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
2309                 *availbytesp +=
2310                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2311         if (ds->ds_quota != 0) {
2312                 /*
2313                  * Adjust available bytes according to refquota
2314                  */
2315                 if (*refdbytesp < ds->ds_quota)
2316                         *availbytesp = MIN(*availbytesp,
2317                             ds->ds_quota - *refdbytesp);
2318                 else
2319                         *availbytesp = 0;
2320         }
2321         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2322         *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
2323         rrw_exit(&ds->ds_bp_rwlock, FTAG);
2324         *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2325 }
2326 
2327 boolean_t
2328 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
2329 {
2330         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2331         uint64_t birth;
2332 
2333         ASSERT(dsl_pool_config_held(dp));
2334         if (snap == NULL)
2335                 return (B_FALSE);
2336         rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2337         birth = dsl_dataset_get_blkptr(ds)->blk_birth;
2338         rrw_exit(&ds->ds_bp_rwlock, FTAG);
2339         if (birth > dsl_dataset_phys(snap)->ds_creation_txg) {
2340                 objset_t *os, *os_snap;
2341                 /*
2342                  * It may be that only the ZIL differs, because it was
2343                  * reset in the head.  Don't count that as being
2344                  * modified.
2345                  */
2346                 if (dmu_objset_from_ds(ds, &os) != 0)
2347                         return (B_TRUE);
2348                 if (dmu_objset_from_ds(snap, &os_snap) != 0)
2349                         return (B_TRUE);
2350                 return (bcmp(&os->os_phys->os_meta_dnode,
2351                     &os_snap->os_phys->os_meta_dnode,
2352                     sizeof (os->os_phys->os_meta_dnode)) != 0);
2353         }
2354         return (B_FALSE);
2355 }
2356 
2357 typedef struct dsl_dataset_rename_snapshot_arg {
2358         const char *ddrsa_fsname;
2359         const char *ddrsa_oldsnapname;
2360         const char *ddrsa_newsnapname;
2361         boolean_t ddrsa_recursive;
2362         dmu_tx_t *ddrsa_tx;
2363 } dsl_dataset_rename_snapshot_arg_t;
2364 
2365 /* ARGSUSED */
2366 static int
2367 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2368     dsl_dataset_t *hds, void *arg)
2369 {
2370         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2371         int error;
2372         uint64_t val;
2373 
2374         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2375         if (error != 0) {
2376                 /* ignore nonexistent snapshots */
2377                 return (error == ENOENT ? 0 : error);
2378         }
2379 
2380         /* new name should not exist */
2381         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2382         if (error == 0)
2383                 error = SET_ERROR(EEXIST);
2384         else if (error == ENOENT)
2385                 error = 0;
2386 
2387         /* dataset name + 1 for the "@" + the new snapshot name must fit */
2388         if (dsl_dir_namelen(hds->ds_dir) + 1 +
2389             strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2390                 error = SET_ERROR(ENAMETOOLONG);
2391 
2392         return (error);
2393 }
2394 
2395 static int
2396 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2397 {
2398         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2399         dsl_pool_t *dp = dmu_tx_pool(tx);
2400         dsl_dataset_t *hds;
2401         int error;
2402 
2403         error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2404         if (error != 0)
2405                 return (error);
2406 
2407         if (ddrsa->ddrsa_recursive) {
2408                 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2409                     dsl_dataset_rename_snapshot_check_impl, ddrsa,
2410                     DS_FIND_CHILDREN);
2411         } else {
2412                 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2413         }
2414         dsl_dataset_rele(hds, FTAG);
2415         return (error);
2416 }
2417 
2418 static int
2419 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2420     dsl_dataset_t *hds, void *arg)
2421 {
2422         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2423         dsl_dataset_t *ds;
2424         uint64_t val;
2425         dmu_tx_t *tx = ddrsa->ddrsa_tx;
2426         int error;
2427 
2428         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2429         ASSERT(error == 0 || error == ENOENT);
2430         if (error == ENOENT) {
2431                 /* ignore nonexistent snapshots */
2432                 return (0);
2433         }
2434 
2435         VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2436 
2437         /* log before we change the name */
2438         spa_history_log_internal_ds(ds, "rename", tx,
2439             "-> @%s", ddrsa->ddrsa_newsnapname);
2440 
2441         VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2442             B_FALSE));
2443         mutex_enter(&ds->ds_lock);
2444         (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2445         mutex_exit(&ds->ds_lock);
2446         VERIFY0(zap_add(dp->dp_meta_objset,
2447             dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2448             ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2449 
2450         dsl_dataset_rele(ds, FTAG);
2451         return (0);
2452 }
2453 
2454 static void
2455 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2456 {
2457         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2458         dsl_pool_t *dp = dmu_tx_pool(tx);
2459         dsl_dataset_t *hds;
2460 
2461         VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2462         ddrsa->ddrsa_tx = tx;
2463         if (ddrsa->ddrsa_recursive) {
2464                 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2465                     dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2466                     DS_FIND_CHILDREN));
2467         } else {
2468                 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2469         }
2470         dsl_dataset_rele(hds, FTAG);
2471 }
2472 
2473 int
2474 dsl_dataset_rename_snapshot(const char *fsname,
2475     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2476 {
2477         dsl_dataset_rename_snapshot_arg_t ddrsa;
2478 
2479         ddrsa.ddrsa_fsname = fsname;
2480         ddrsa.ddrsa_oldsnapname = oldsnapname;
2481         ddrsa.ddrsa_newsnapname = newsnapname;
2482         ddrsa.ddrsa_recursive = recursive;
2483 
2484         return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2485             dsl_dataset_rename_snapshot_sync, &ddrsa,
2486             1, ZFS_SPACE_CHECK_RESERVED));
2487 }
2488 
2489 /*
2490  * If we're doing an ownership handoff, we need to make sure that there is
2491  * only one long hold on the dataset.  We're not allowed to change anything here
2492  * so we don't permanently release the long hold or regular hold here.  We want
2493  * to do this only when syncing to avoid the dataset unexpectedly going away
2494  * when we release the long hold.
2495  */
2496 static int
2497 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2498 {
2499         boolean_t held;
2500 
2501         if (!dmu_tx_is_syncing(tx))
2502                 return (0);
2503 
2504         if (owner != NULL) {
2505                 VERIFY3P(ds->ds_owner, ==, owner);
2506                 dsl_dataset_long_rele(ds, owner);
2507         }
2508 
2509         held = dsl_dataset_long_held(ds);
2510 
2511         if (owner != NULL)
2512                 dsl_dataset_long_hold(ds, owner);
2513 
2514         if (held)
2515                 return (SET_ERROR(EBUSY));
2516 
2517         return (0);
2518 }
2519 
2520 int
2521 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2522 {
2523         dsl_dataset_rollback_arg_t *ddra = arg;
2524         dsl_pool_t *dp = dmu_tx_pool(tx);
2525         dsl_dataset_t *ds;
2526         int64_t unused_refres_delta;
2527         int error;
2528 
2529         error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2530         if (error != 0)
2531                 return (error);
2532 
2533         /* must not be a snapshot */
2534         if (ds->ds_is_snapshot) {
2535                 dsl_dataset_rele(ds, FTAG);
2536                 return (SET_ERROR(EINVAL));
2537         }
2538 
2539         /* must have a most recent snapshot */
2540         if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2541                 dsl_dataset_rele(ds, FTAG);
2542                 return (SET_ERROR(ESRCH));
2543         }
2544 
2545         /*
2546          * No rollback to a snapshot created in the current txg, because
2547          * the rollback may dirty the dataset and create blocks that are
2548          * not reachable from the rootbp while having a birth txg that
2549          * falls into the snapshot's range.
2550          */
2551         if (dmu_tx_is_syncing(tx) &&
2552             dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) {
2553                 dsl_dataset_rele(ds, FTAG);
2554                 return (SET_ERROR(EAGAIN));
2555         }
2556 
2557         /*
2558          * If the expected target snapshot is specified, then check that
2559          * the latest snapshot is it.
2560          */
2561         if (ddra->ddra_tosnap != NULL) {
2562                 dsl_dataset_t *snapds;
2563 
2564                 /* Check if the target snapshot exists at all. */
2565                 error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds);
2566                 if (error != 0) {
2567                         /*
2568                          * ESRCH is used to signal that the target snapshot does
2569                          * not exist, while ENOENT is used to report that
2570                          * the rolled back dataset does not exist.
2571                          * ESRCH is also used to cover other cases where the
2572                          * target snapshot is not related to the dataset being
2573                          * rolled back such as being in a different pool.
2574                          */
2575                         if (error == ENOENT || error == EXDEV)
2576                                 error = SET_ERROR(ESRCH);
2577                         dsl_dataset_rele(ds, FTAG);
2578                         return (error);
2579                 }
2580                 ASSERT(snapds->ds_is_snapshot);
2581 
2582                 /* Check if the snapshot is the latest snapshot indeed. */
2583                 if (snapds != ds->ds_prev) {
2584                         /*
2585                          * Distinguish between the case where the only problem
2586                          * is intervening snapshots (EEXIST) vs the snapshot
2587                          * not being a valid target for rollback (ESRCH).
2588                          */
2589                         if (snapds->ds_dir == ds->ds_dir ||
2590                             (dsl_dir_is_clone(ds->ds_dir) &&
2591                             dsl_dir_phys(ds->ds_dir)->dd_origin_obj ==
2592                             snapds->ds_object)) {
2593                                 error = SET_ERROR(EEXIST);
2594                         } else {
2595                                 error = SET_ERROR(ESRCH);
2596                         }
2597                         dsl_dataset_rele(snapds, FTAG);
2598                         dsl_dataset_rele(ds, FTAG);
2599                         return (error);
2600                 }
2601                 dsl_dataset_rele(snapds, FTAG);
2602         }
2603 
2604         /* must not have any bookmarks after the most recent snapshot */
2605         nvlist_t *proprequest = fnvlist_alloc();
2606         fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2607         nvlist_t *bookmarks = fnvlist_alloc();
2608         error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2609         fnvlist_free(proprequest);
2610         if (error != 0) {
2611                 dsl_dataset_rele(ds, FTAG);
2612                 return (error);
2613         }
2614         for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2615             pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2616                 nvlist_t *valuenv =
2617                     fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2618                     zfs_prop_to_name(ZFS_PROP_CREATETXG));
2619                 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2620                 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2621                         fnvlist_free(bookmarks);
2622                         dsl_dataset_rele(ds, FTAG);
2623                         return (SET_ERROR(EEXIST));
2624                 }
2625         }
2626         fnvlist_free(bookmarks);
2627 
2628         error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2629         if (error != 0) {
2630                 dsl_dataset_rele(ds, FTAG);
2631                 return (error);
2632         }
2633 
2634         /*
2635          * Check if the snap we are rolling back to uses more than
2636          * the refquota.
2637          */
2638         if (ds->ds_quota != 0 &&
2639             dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2640                 dsl_dataset_rele(ds, FTAG);
2641                 return (SET_ERROR(EDQUOT));
2642         }
2643 
2644         /*
2645          * When we do the clone swap, we will temporarily use more space
2646          * due to the refreservation (the head will no longer have any
2647          * unique space, so the entire amount of the refreservation will need
2648          * to be free).  We will immediately destroy the clone, freeing
2649          * this space, but the freeing happens over many txg's.
2650          */
2651         unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2652             dsl_dataset_phys(ds)->ds_unique_bytes);
2653 
2654         if (unused_refres_delta > 0 &&
2655             unused_refres_delta >
2656             dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2657                 dsl_dataset_rele(ds, FTAG);
2658                 return (SET_ERROR(ENOSPC));
2659         }
2660 
2661         dsl_dataset_rele(ds, FTAG);
2662         return (0);
2663 }
2664 
2665 void
2666 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2667 {
2668         dsl_dataset_rollback_arg_t *ddra = arg;
2669         dsl_pool_t *dp = dmu_tx_pool(tx);
2670         dsl_dataset_t *ds, *clone;
2671         uint64_t cloneobj;
2672         char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2673 
2674         VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2675 
2676         dsl_dataset_name(ds->ds_prev, namebuf);
2677         fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2678 
2679         cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2680             ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2681 
2682         VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2683 
2684         dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2685         dsl_dataset_zero_zil(ds, tx);
2686 
2687         dsl_destroy_head_sync_impl(clone, tx);
2688 
2689         dsl_dataset_rele(clone, FTAG);
2690         dsl_dataset_rele(ds, FTAG);
2691 }
2692 
2693 /*
2694  * Rolls back the given filesystem or volume to the most recent snapshot.
2695  * The name of the most recent snapshot will be returned under key "target"
2696  * in the result nvlist.
2697  *
2698  * If owner != NULL:
2699  * - The existing dataset MUST be owned by the specified owner at entry
2700  * - Upon return, dataset will still be held by the same owner, whether we
2701  *   succeed or not.
2702  *
2703  * This mode is required any time the existing filesystem is mounted.  See
2704  * notes above zfs_suspend_fs() for further details.
2705  */
2706 int
2707 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner,
2708     nvlist_t *result)
2709 {
2710         dsl_dataset_rollback_arg_t ddra;
2711 
2712         ddra.ddra_fsname = fsname;
2713         ddra.ddra_tosnap = tosnap;
2714         ddra.ddra_owner = owner;
2715         ddra.ddra_result = result;
2716 
2717         return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2718             dsl_dataset_rollback_sync, &ddra,
2719             1, ZFS_SPACE_CHECK_RESERVED));
2720 }
2721 
2722 struct promotenode {
2723         list_node_t link;
2724         dsl_dataset_t *ds;
2725 };
2726 
2727 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2728 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2729     void *tag);
2730 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2731 
2732 int
2733 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2734 {
2735         dsl_dataset_promote_arg_t *ddpa = arg;
2736         dsl_pool_t *dp = dmu_tx_pool(tx);
2737         dsl_dataset_t *hds;
2738         struct promotenode *snap;
2739         dsl_dataset_t *origin_ds;
2740         int err;
2741         uint64_t unused;
2742         uint64_t ss_mv_cnt;
2743         size_t max_snap_len;
2744         boolean_t conflicting_snaps;
2745 
2746         err = promote_hold(ddpa, dp, FTAG);
2747         if (err != 0)
2748                 return (err);
2749 
2750         hds = ddpa->ddpa_clone;
2751         snap = list_head(&ddpa->shared_snaps);
2752         origin_ds = snap->ds;
2753         max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2754 
2755         snap = list_head(&ddpa->origin_snaps);
2756 
2757         if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2758                 promote_rele(ddpa, FTAG);
2759                 return (SET_ERROR(EXDEV));
2760         }
2761 
2762         /*
2763          * Compute and check the amount of space to transfer.  Since this is
2764          * so expensive, don't do the preliminary check.
2765          */
2766         if (!dmu_tx_is_syncing(tx)) {
2767                 promote_rele(ddpa, FTAG);
2768                 return (0);
2769         }
2770 
2771         /* compute origin's new unique space */
2772         snap = list_tail(&ddpa->clone_snaps);
2773         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2774             origin_ds->ds_object);
2775         dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2776             dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2777             &ddpa->unique, &unused, &unused);
2778 
2779         /*
2780          * Walk the snapshots that we are moving
2781          *
2782          * Compute space to transfer.  Consider the incremental changes
2783          * to used by each snapshot:
2784          * (my used) = (prev's used) + (blocks born) - (blocks killed)
2785          * So each snapshot gave birth to:
2786          * (blocks born) = (my used) - (prev's used) + (blocks killed)
2787          * So a sequence would look like:
2788          * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2789          * Which simplifies to:
2790          * uN + kN + kN-1 + ... + k1 + k0
2791          * Note however, if we stop before we reach the ORIGIN we get:
2792          * uN + kN + kN-1 + ... + kM - uM-1
2793          */
2794         conflicting_snaps = B_FALSE;
2795         ss_mv_cnt = 0;
2796         ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2797         ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2798         ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2799         for (snap = list_head(&ddpa->shared_snaps); snap;
2800             snap = list_next(&ddpa->shared_snaps, snap)) {
2801                 uint64_t val, dlused, dlcomp, dluncomp;
2802                 dsl_dataset_t *ds = snap->ds;
2803 
2804                 ss_mv_cnt++;
2805 
2806                 /*
2807                  * If there are long holds, we won't be able to evict
2808                  * the objset.
2809                  */
2810                 if (dsl_dataset_long_held(ds)) {
2811                         err = SET_ERROR(EBUSY);
2812                         goto out;
2813                 }
2814 
2815                 /* Check that the snapshot name does not conflict */
2816                 VERIFY0(dsl_dataset_get_snapname(ds));
2817                 if (strlen(ds->ds_snapname) >= max_snap_len) {
2818                         err = SET_ERROR(ENAMETOOLONG);
2819                         goto out;
2820                 }
2821                 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2822                 if (err == 0) {
2823                         fnvlist_add_boolean(ddpa->err_ds,
2824                             snap->ds->ds_snapname);
2825                         conflicting_snaps = B_TRUE;
2826                 } else if (err != ENOENT) {
2827                         goto out;
2828                 }
2829 
2830                 /* The very first snapshot does not have a deadlist */
2831                 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2832                         continue;
2833 
2834                 dsl_deadlist_space(&ds->ds_deadlist,
2835                     &dlused, &dlcomp, &dluncomp);
2836                 ddpa->used += dlused;
2837                 ddpa->comp += dlcomp;
2838                 ddpa->uncomp += dluncomp;
2839         }
2840 
2841         /*
2842          * In order to return the full list of conflicting snapshots, we check
2843          * whether there was a conflict after traversing all of them.
2844          */
2845         if (conflicting_snaps) {
2846                 err = SET_ERROR(EEXIST);
2847                 goto out;
2848         }
2849 
2850         /*
2851          * If we are a clone of a clone then we never reached ORIGIN,
2852          * so we need to subtract out the clone origin's used space.
2853          */
2854         if (ddpa->origin_origin) {
2855                 ddpa->used -=
2856                     dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2857                 ddpa->comp -=
2858                     dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2859                 ddpa->uncomp -=
2860                     dsl_dataset_phys(ddpa->origin_origin)->
2861                     ds_uncompressed_bytes;
2862         }
2863 
2864         /* Check that there is enough space and limit headroom here */
2865         err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2866             0, ss_mv_cnt, ddpa->used, ddpa->cr);
2867         if (err != 0)
2868                 goto out;
2869 
2870         /*
2871          * Compute the amounts of space that will be used by snapshots
2872          * after the promotion (for both origin and clone).  For each,
2873          * it is the amount of space that will be on all of their
2874          * deadlists (that was not born before their new origin).
2875          */
2876         if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2877                 uint64_t space;
2878 
2879                 /*
2880                  * Note, typically this will not be a clone of a clone,
2881                  * so dd_origin_txg will be < TXG_INITIAL, so
2882                  * these snaplist_space() -> dsl_deadlist_space_range()
2883                  * calls will be fast because they do not have to
2884                  * iterate over all bps.
2885                  */
2886                 snap = list_head(&ddpa->origin_snaps);
2887                 err = snaplist_space(&ddpa->shared_snaps,
2888                     snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2889                 if (err != 0)
2890                         goto out;
2891 
2892                 err = snaplist_space(&ddpa->clone_snaps,
2893                     snap->ds->ds_dir->dd_origin_txg, &space);
2894                 if (err != 0)
2895                         goto out;
2896                 ddpa->cloneusedsnap += space;
2897         }
2898         if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2899             DD_FLAG_USED_BREAKDOWN) {
2900                 err = snaplist_space(&ddpa->origin_snaps,
2901                     dsl_dataset_phys(origin_ds)->ds_creation_txg,
2902                     &ddpa->originusedsnap);
2903                 if (err != 0)
2904                         goto out;
2905         }
2906 
2907 out:
2908         promote_rele(ddpa, FTAG);
2909         return (err);
2910 }
2911 
2912 void
2913 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2914 {
2915         dsl_dataset_promote_arg_t *ddpa = arg;
2916         dsl_pool_t *dp = dmu_tx_pool(tx);
2917         dsl_dataset_t *hds;
2918         struct promotenode *snap;
2919         dsl_dataset_t *origin_ds;
2920         dsl_dataset_t *origin_head;
2921         dsl_dir_t *dd;
2922         dsl_dir_t *odd = NULL;
2923         uint64_t oldnext_obj;
2924         int64_t delta;
2925 
2926         VERIFY0(promote_hold(ddpa, dp, FTAG));
2927         hds = ddpa->ddpa_clone;
2928 
2929         ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2930 
2931         snap = list_head(&ddpa->shared_snaps);
2932         origin_ds = snap->ds;
2933         dd = hds->ds_dir;
2934 
2935         snap = list_head(&ddpa->origin_snaps);
2936         origin_head = snap->ds;
2937 
2938         /*
2939          * We need to explicitly open odd, since origin_ds's dd will be
2940          * changing.
2941          */
2942         VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2943             NULL, FTAG, &odd));
2944 
2945         /* change origin's next snap */
2946         dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2947         oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2948         snap = list_tail(&ddpa->clone_snaps);
2949         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2950             origin_ds->ds_object);
2951         dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2952 
2953         /* change the origin's next clone */
2954         if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2955                 dsl_dataset_remove_from_next_clones(origin_ds,
2956                     snap->ds->ds_object, tx);
2957                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2958                     dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2959                     oldnext_obj, tx));
2960         }
2961 
2962         /* change origin */
2963         dmu_buf_will_dirty(dd->dd_dbuf, tx);
2964         ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2965         dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2966         dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2967         dmu_buf_will_dirty(odd->dd_dbuf, tx);
2968         dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2969         origin_head->ds_dir->dd_origin_txg =
2970             dsl_dataset_phys(origin_ds)->ds_creation_txg;
2971 
2972         /* change dd_clone entries */
2973         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2974                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2975                     dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2976                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2977                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2978                     hds->ds_object, tx));
2979 
2980                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2981                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2982                     origin_head->ds_object, tx));
2983                 if (dsl_dir_phys(dd)->dd_clones == 0) {
2984                         dsl_dir_phys(dd)->dd_clones =
2985                             zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2986                             DMU_OT_NONE, 0, tx);
2987                 }
2988                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2989                     dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2990         }
2991 
2992         /* move snapshots to this dir */
2993         for (snap = list_head(&ddpa->shared_snaps); snap;
2994             snap = list_next(&ddpa->shared_snaps, snap)) {
2995                 dsl_dataset_t *ds = snap->ds;
2996 
2997                 /*
2998                  * Property callbacks are registered to a particular
2999                  * dsl_dir.  Since ours is changing, evict the objset
3000                  * so that they will be unregistered from the old dsl_dir.
3001                  */
3002                 if (ds->ds_objset) {
3003                         dmu_objset_evict(ds->ds_objset);
3004                         ds->ds_objset = NULL;
3005                 }
3006 
3007                 /* move snap name entry */
3008                 VERIFY0(dsl_dataset_get_snapname(ds));
3009                 VERIFY0(dsl_dataset_snap_remove(origin_head,
3010                     ds->ds_snapname, tx, B_TRUE));
3011                 VERIFY0(zap_add(dp->dp_meta_objset,
3012                     dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
3013                     8, 1, &ds->ds_object, tx));
3014                 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
3015                     DD_FIELD_SNAPSHOT_COUNT, tx);
3016 
3017                 /* change containing dsl_dir */
3018                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3019                 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
3020                 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
3021                 ASSERT3P(ds->ds_dir, ==, odd);
3022                 dsl_dir_rele(ds->ds_dir, ds);
3023                 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
3024                     NULL, ds, &ds->ds_dir));
3025 
3026                 /* move any clone references */
3027                 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
3028                     spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3029                         zap_cursor_t zc;
3030                         zap_attribute_t za;
3031 
3032                         for (zap_cursor_init(&zc, dp->dp_meta_objset,
3033                             dsl_dataset_phys(ds)->ds_next_clones_obj);
3034                             zap_cursor_retrieve(&zc, &za) == 0;
3035                             zap_cursor_advance(&zc)) {
3036                                 dsl_dataset_t *cnds;
3037                                 uint64_t o;
3038 
3039                                 if (za.za_first_integer == oldnext_obj) {
3040                                         /*
3041                                          * We've already moved the
3042                                          * origin's reference.
3043                                          */
3044                                         continue;
3045                                 }
3046 
3047                                 VERIFY0(dsl_dataset_hold_obj(dp,
3048                                     za.za_first_integer, FTAG, &cnds));
3049                                 o = dsl_dir_phys(cnds->ds_dir)->
3050                                     dd_head_dataset_obj;
3051 
3052                                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
3053                                     dsl_dir_phys(odd)->dd_clones, o, tx));
3054                                 VERIFY0(zap_add_int(dp->dp_meta_objset,
3055                                     dsl_dir_phys(dd)->dd_clones, o, tx));
3056                                 dsl_dataset_rele(cnds, FTAG);
3057                         }
3058                         zap_cursor_fini(&zc);
3059                 }
3060 
3061                 ASSERT(!dsl_prop_hascb(ds));
3062         }
3063 
3064         /*
3065          * Change space accounting.
3066          * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
3067          * both be valid, or both be 0 (resulting in delta == 0).  This
3068          * is true for each of {clone,origin} independently.
3069          */
3070 
3071         delta = ddpa->cloneusedsnap -
3072             dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
3073         ASSERT3S(delta, >=, 0);
3074         ASSERT3U(ddpa->used, >=, delta);
3075         dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
3076         dsl_dir_diduse_space(dd, DD_USED_HEAD,
3077             ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
3078 
3079         delta = ddpa->originusedsnap -
3080             dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
3081         ASSERT3S(delta, <=, 0);
3082         ASSERT3U(ddpa->used, >=, -delta);
3083         dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
3084         dsl_dir_diduse_space(odd, DD_USED_HEAD,
3085             -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
3086 
3087         dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
3088 
3089         /* log history record */
3090         spa_history_log_internal_ds(hds, "promote", tx, "");
3091 
3092         dsl_dir_rele(odd, FTAG);
3093         promote_rele(ddpa, FTAG);
3094 }
3095 
3096 /*
3097  * Make a list of dsl_dataset_t's for the snapshots between first_obj
3098  * (exclusive) and last_obj (inclusive).  The list will be in reverse
3099  * order (last_obj will be the list_head()).  If first_obj == 0, do all
3100  * snapshots back to this dataset's origin.
3101  */
3102 static int
3103 snaplist_make(dsl_pool_t *dp,
3104     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
3105 {
3106         uint64_t obj = last_obj;
3107 
3108         list_create(l, sizeof (struct promotenode),
3109             offsetof(struct promotenode, link));
3110 
3111         while (obj != first_obj) {
3112                 dsl_dataset_t *ds;
3113                 struct promotenode *snap;
3114                 int err;
3115 
3116                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
3117                 ASSERT(err != ENOENT);
3118                 if (err != 0)
3119                         return (err);
3120 
3121                 if (first_obj == 0)
3122                         first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
3123 
3124                 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
3125                 snap->ds = ds;
3126                 list_insert_tail(l, snap);
3127                 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3128         }
3129 
3130         return (0);
3131 }
3132 
3133 static int
3134 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3135 {
3136         struct promotenode *snap;
3137 
3138         *spacep = 0;
3139         for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3140                 uint64_t used, comp, uncomp;
3141                 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3142                     mintxg, UINT64_MAX, &used, &comp, &uncomp);
3143                 *spacep += used;
3144         }
3145         return (0);
3146 }
3147 
3148 static void
3149 snaplist_destroy(list_t *l, void *tag)
3150 {
3151         struct promotenode *snap;
3152 
3153         if (l == NULL || !list_link_active(&l->list_head))
3154                 return;
3155 
3156         while ((snap = list_tail(l)) != NULL) {
3157                 list_remove(l, snap);
3158                 dsl_dataset_rele(snap->ds, tag);
3159                 kmem_free(snap, sizeof (*snap));
3160         }
3161         list_destroy(l);
3162 }
3163 
3164 static int
3165 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
3166 {
3167         int error;
3168         dsl_dir_t *dd;
3169         struct promotenode *snap;
3170 
3171         error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
3172             &ddpa->ddpa_clone);
3173         if (error != 0)
3174                 return (error);
3175         dd = ddpa->ddpa_clone->ds_dir;
3176 
3177         if (ddpa->ddpa_clone->ds_is_snapshot ||
3178             !dsl_dir_is_clone(dd)) {
3179                 dsl_dataset_rele(ddpa->ddpa_clone, tag);
3180                 return (SET_ERROR(EINVAL));
3181         }
3182 
3183         error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
3184             &ddpa->shared_snaps, tag);
3185         if (error != 0)
3186                 goto out;
3187 
3188         error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
3189             &ddpa->clone_snaps, tag);
3190         if (error != 0)
3191                 goto out;
3192 
3193         snap = list_head(&ddpa->shared_snaps);
3194         ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
3195         error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
3196             dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
3197             &ddpa->origin_snaps, tag);
3198         if (error != 0)
3199                 goto out;
3200 
3201         if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
3202                 error = dsl_dataset_hold_obj(dp,
3203                     dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
3204                     tag, &ddpa->origin_origin);
3205                 if (error != 0)
3206                         goto out;
3207         }
3208 out:
3209         if (error != 0)
3210                 promote_rele(ddpa, tag);
3211         return (error);
3212 }
3213 
3214 static void
3215 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
3216 {
3217         snaplist_destroy(&ddpa->shared_snaps, tag);
3218         snaplist_destroy(&ddpa->clone_snaps, tag);
3219         snaplist_destroy(&ddpa->origin_snaps, tag);
3220         if (ddpa->origin_origin != NULL)
3221                 dsl_dataset_rele(ddpa->origin_origin, tag);
3222         dsl_dataset_rele(ddpa->ddpa_clone, tag);
3223 }
3224 
3225 /*
3226  * Promote a clone.
3227  *
3228  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
3229  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
3230  */
3231 int
3232 dsl_dataset_promote(const char *name, char *conflsnap)
3233 {
3234         dsl_dataset_promote_arg_t ddpa = { 0 };
3235         uint64_t numsnaps;
3236         int error;
3237         nvpair_t *snap_pair;
3238         objset_t *os;
3239 
3240         /*
3241          * We will modify space proportional to the number of
3242          * snapshots.  Compute numsnaps.
3243          */
3244         error = dmu_objset_hold(name, FTAG, &os);
3245         if (error != 0)
3246                 return (error);
3247         error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
3248             dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
3249             &numsnaps);
3250         dmu_objset_rele(os, FTAG);
3251         if (error != 0)
3252                 return (error);
3253 
3254         ddpa.ddpa_clonename = name;
3255         ddpa.err_ds = fnvlist_alloc();
3256         ddpa.cr = CRED();
3257 
3258         error = dsl_sync_task(name, dsl_dataset_promote_check,
3259             dsl_dataset_promote_sync, &ddpa,
3260             2 + numsnaps, ZFS_SPACE_CHECK_RESERVED);
3261 
3262         /*
3263          * Return the first conflicting snapshot found.
3264          */
3265         snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL);
3266         if (snap_pair != NULL && conflsnap != NULL)
3267                 (void) strcpy(conflsnap, nvpair_name(snap_pair));
3268 
3269         fnvlist_free(ddpa.err_ds);
3270         return (error);
3271 }
3272 
3273 int
3274 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
3275     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
3276 {
3277         /*
3278          * "slack" factor for received datasets with refquota set on them.
3279          * See the bottom of this function for details on its use.
3280          */
3281         uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
3282         int64_t unused_refres_delta;
3283 
3284         /* they should both be heads */
3285         if (clone->ds_is_snapshot ||
3286             origin_head->ds_is_snapshot)
3287                 return (SET_ERROR(EINVAL));
3288 
3289         /* if we are not forcing, the branch point should be just before them */
3290         if (!force && clone->ds_prev != origin_head->ds_prev)
3291                 return (SET_ERROR(EINVAL));
3292 
3293         /* clone should be the clone (unless they are unrelated) */
3294         if (clone->ds_prev != NULL &&
3295             clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
3296             origin_head->ds_dir != clone->ds_prev->ds_dir)
3297                 return (SET_ERROR(EINVAL));
3298 
3299         /* the clone should be a child of the origin */
3300         if (clone->ds_dir->dd_parent != origin_head->ds_dir)
3301                 return (SET_ERROR(EINVAL));
3302 
3303         /* origin_head shouldn't be modified unless 'force' */
3304         if (!force &&
3305             dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
3306                 return (SET_ERROR(ETXTBSY));
3307 
3308         /* origin_head should have no long holds (e.g. is not mounted) */
3309         if (dsl_dataset_handoff_check(origin_head, owner, tx))
3310                 return (SET_ERROR(EBUSY));
3311 
3312         /* check amount of any unconsumed refreservation */
3313         unused_refres_delta =
3314             (int64_t)MIN(origin_head->ds_reserved,
3315             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3316             (int64_t)MIN(origin_head->ds_reserved,
3317             dsl_dataset_phys(clone)->ds_unique_bytes);
3318 
3319         if (unused_refres_delta > 0 &&
3320             unused_refres_delta >
3321             dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
3322                 return (SET_ERROR(ENOSPC));
3323 
3324         /*
3325          * The clone can't be too much over the head's refquota.
3326          *
3327          * To ensure that the entire refquota can be used, we allow one
3328          * transaction to exceed the the refquota.  Therefore, this check
3329          * needs to also allow for the space referenced to be more than the
3330          * refquota.  The maximum amount of space that one transaction can use
3331          * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
3332          * overage ensures that we are able to receive a filesystem that
3333          * exceeds the refquota on the source system.
3334          *
3335          * So that overage is the refquota_slack we use below.
3336          */
3337         if (origin_head->ds_quota != 0 &&
3338             dsl_dataset_phys(clone)->ds_referenced_bytes >
3339             origin_head->ds_quota + refquota_slack)
3340                 return (SET_ERROR(EDQUOT));
3341 
3342         return (0);
3343 }
3344 
3345 static void
3346 dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone,
3347     dsl_dataset_t *origin, dmu_tx_t *tx)
3348 {
3349         uint64_t clone_remap_dl_obj, origin_remap_dl_obj;
3350         dsl_pool_t *dp = dmu_tx_pool(tx);
3351 
3352         ASSERT(dsl_pool_sync_context(dp));
3353 
3354         clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone);
3355         origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin);
3356 
3357         if (clone_remap_dl_obj != 0) {
3358                 dsl_deadlist_close(&clone->ds_remap_deadlist);
3359                 dsl_dataset_unset_remap_deadlist_object(clone, tx);
3360         }
3361         if (origin_remap_dl_obj != 0) {
3362                 dsl_deadlist_close(&origin->ds_remap_deadlist);
3363                 dsl_dataset_unset_remap_deadlist_object(origin, tx);
3364         }
3365 
3366         if (clone_remap_dl_obj != 0) {
3367                 dsl_dataset_set_remap_deadlist_object(origin,
3368                     clone_remap_dl_obj, tx);
3369                 dsl_deadlist_open(&origin->ds_remap_deadlist,
3370                     dp->dp_meta_objset, clone_remap_dl_obj);
3371         }
3372         if (origin_remap_dl_obj != 0) {
3373                 dsl_dataset_set_remap_deadlist_object(clone,
3374                     origin_remap_dl_obj, tx);
3375                 dsl_deadlist_open(&clone->ds_remap_deadlist,
3376                     dp->dp_meta_objset, origin_remap_dl_obj);
3377         }
3378 }
3379 
3380 void
3381 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
3382     dsl_dataset_t *origin_head, dmu_tx_t *tx)
3383 {
3384         dsl_pool_t *dp = dmu_tx_pool(tx);
3385         int64_t unused_refres_delta;
3386 
3387         ASSERT(clone->ds_reserved == 0);
3388         /*
3389          * NOTE: On DEBUG kernels there could be a race between this and
3390          * the check function if spa_asize_inflation is adjusted...
3391          */
3392         ASSERT(origin_head->ds_quota == 0 ||
3393             dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
3394             DMU_MAX_ACCESS * spa_asize_inflation);
3395         ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
3396 
3397         /*
3398          * Swap per-dataset feature flags.
3399          */
3400         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
3401                 if (!(spa_feature_table[f].fi_flags &
3402                     ZFEATURE_FLAG_PER_DATASET)) {
3403                         ASSERT(!clone->ds_feature_inuse[f]);
3404                         ASSERT(!origin_head->ds_feature_inuse[f]);
3405                         continue;
3406                 }
3407 
3408                 boolean_t clone_inuse = clone->ds_feature_inuse[f];
3409                 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
3410 
3411                 if (clone_inuse) {
3412                         dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
3413                         clone->ds_feature_inuse[f] = B_FALSE;
3414                 }
3415                 if (origin_head_inuse) {
3416                         dsl_dataset_deactivate_feature(origin_head->ds_object,
3417                             f, tx);
3418                         origin_head->ds_feature_inuse[f] = B_FALSE;
3419                 }
3420                 if (clone_inuse) {
3421                         dsl_dataset_activate_feature(origin_head->ds_object,
3422                             f, tx);
3423                         origin_head->ds_feature_inuse[f] = B_TRUE;
3424                 }
3425                 if (origin_head_inuse) {
3426                         dsl_dataset_activate_feature(clone->ds_object, f, tx);
3427                         clone->ds_feature_inuse[f] = B_TRUE;
3428                 }
3429         }
3430 
3431         dmu_buf_will_dirty(clone->ds_dbuf, tx);
3432         dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
3433 
3434         if (clone->ds_objset != NULL) {
3435                 dmu_objset_evict(clone->ds_objset);
3436                 clone->ds_objset = NULL;
3437         }
3438 
3439         if (origin_head->ds_objset != NULL) {
3440                 dmu_objset_evict(origin_head->ds_objset);
3441                 origin_head->ds_objset = NULL;
3442         }
3443 
3444         unused_refres_delta =
3445             (int64_t)MIN(origin_head->ds_reserved,
3446             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3447             (int64_t)MIN(origin_head->ds_reserved,
3448             dsl_dataset_phys(clone)->ds_unique_bytes);
3449 
3450         /*
3451          * Reset origin's unique bytes, if it exists.
3452          */
3453         if (clone->ds_prev) {
3454                 dsl_dataset_t *origin = clone->ds_prev;
3455                 uint64_t comp, uncomp;
3456 
3457                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3458                 dsl_deadlist_space_range(&clone->ds_deadlist,
3459                     dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
3460                     &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
3461         }
3462 
3463         /* swap blkptrs */
3464         {
3465                 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG);
3466                 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG);
3467                 blkptr_t tmp;
3468                 tmp = dsl_dataset_phys(origin_head)->ds_bp;
3469                 dsl_dataset_phys(origin_head)->ds_bp =
3470                     dsl_dataset_phys(clone)->ds_bp;
3471                 dsl_dataset_phys(clone)->ds_bp = tmp;
3472                 rrw_exit(&origin_head->ds_bp_rwlock, FTAG);
3473                 rrw_exit(&clone->ds_bp_rwlock, FTAG);
3474         }
3475 
3476         /* set dd_*_bytes */
3477         {
3478                 int64_t dused, dcomp, duncomp;
3479                 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3480                 uint64_t odl_used, odl_comp, odl_uncomp;
3481 
3482                 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3483                     dd_used_breakdown[DD_USED_SNAP], ==, 0);
3484 
3485                 dsl_deadlist_space(&clone->ds_deadlist,
3486                     &cdl_used, &cdl_comp, &cdl_uncomp);
3487                 dsl_deadlist_space(&origin_head->ds_deadlist,
3488                     &odl_used, &odl_comp, &odl_uncomp);
3489 
3490                 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3491                     cdl_used -
3492                     (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3493                     odl_used);
3494                 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3495                     cdl_comp -
3496                     (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3497                     odl_comp);
3498                 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3499                     cdl_uncomp -
3500                     (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3501                     odl_uncomp);
3502 
3503                 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3504                     dused, dcomp, duncomp, tx);
3505                 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3506                     -dused, -dcomp, -duncomp, tx);
3507 
3508                 /*
3509                  * The difference in the space used by snapshots is the
3510                  * difference in snapshot space due to the head's
3511                  * deadlist (since that's the only thing that's
3512                  * changing that affects the snapused).
3513                  */
3514                 dsl_deadlist_space_range(&clone->ds_deadlist,
3515                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3516                     &cdl_used, &cdl_comp, &cdl_uncomp);
3517                 dsl_deadlist_space_range(&origin_head->ds_deadlist,
3518                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3519                     &odl_used, &odl_comp, &odl_uncomp);
3520                 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3521                     DD_USED_HEAD, DD_USED_SNAP, tx);
3522         }
3523 
3524         /* swap ds_*_bytes */
3525         SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3526             dsl_dataset_phys(clone)->ds_referenced_bytes);
3527         SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3528             dsl_dataset_phys(clone)->ds_compressed_bytes);
3529         SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3530             dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3531         SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3532             dsl_dataset_phys(clone)->ds_unique_bytes);
3533 
3534         /* apply any parent delta for change in unconsumed refreservation */
3535         dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3536             unused_refres_delta, 0, 0, tx);
3537 
3538         /*
3539          * Swap deadlists.
3540          */
3541         dsl_deadlist_close(&clone->ds_deadlist);
3542         dsl_deadlist_close(&origin_head->ds_deadlist);
3543         SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3544             dsl_dataset_phys(clone)->ds_deadlist_obj);
3545         dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3546             dsl_dataset_phys(clone)->ds_deadlist_obj);
3547         dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3548             dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3549         dsl_dataset_swap_remap_deadlists(clone, origin_head, tx);
3550 
3551         dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3552 
3553         spa_history_log_internal_ds(clone, "clone swap", tx,
3554             "parent=%s", origin_head->ds_dir->dd_myname);
3555 }
3556 
3557 /*
3558  * Given a pool name and a dataset object number in that pool,
3559  * return the name of that dataset.
3560  */
3561 int
3562 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3563 {
3564         dsl_pool_t *dp;
3565         dsl_dataset_t *ds;
3566         int error;
3567 
3568         error = dsl_pool_hold(pname, FTAG, &dp);
3569         if (error != 0)
3570                 return (error);
3571 
3572         error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3573         if (error == 0) {
3574                 dsl_dataset_name(ds, buf);
3575                 dsl_dataset_rele(ds, FTAG);
3576         }
3577         dsl_pool_rele(dp, FTAG);
3578 
3579         return (error);
3580 }
3581 
3582 int
3583 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3584     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3585 {
3586         int error = 0;
3587 
3588         ASSERT3S(asize, >, 0);
3589 
3590         /*
3591          * *ref_rsrv is the portion of asize that will come from any
3592          * unconsumed refreservation space.
3593          */
3594         *ref_rsrv = 0;
3595 
3596         mutex_enter(&ds->ds_lock);
3597         /*
3598          * Make a space adjustment for reserved bytes.
3599          */
3600         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3601                 ASSERT3U(*used, >=,
3602                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3603                 *used -=
3604                     (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3605                 *ref_rsrv =
3606                     asize - MIN(asize, parent_delta(ds, asize + inflight));
3607         }
3608 
3609         if (!check_quota || ds->ds_quota == 0) {
3610                 mutex_exit(&ds->ds_lock);
3611                 return (0);
3612         }
3613         /*
3614          * If they are requesting more space, and our current estimate
3615          * is over quota, they get to try again unless the actual
3616          * on-disk is over quota and there are no pending changes (which
3617          * may free up space for us).
3618          */
3619         if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3620             ds->ds_quota) {
3621                 if (inflight > 0 ||
3622                     dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3623                         error = SET_ERROR(ERESTART);
3624                 else
3625                         error = SET_ERROR(EDQUOT);
3626         }
3627         mutex_exit(&ds->ds_lock);
3628 
3629         return (error);
3630 }
3631 
3632 typedef struct dsl_dataset_set_qr_arg {
3633         const char *ddsqra_name;
3634         zprop_source_t ddsqra_source;
3635         uint64_t ddsqra_value;
3636 } dsl_dataset_set_qr_arg_t;
3637 
3638 
3639 /* ARGSUSED */
3640 static int
3641 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3642 {
3643         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3644         dsl_pool_t *dp = dmu_tx_pool(tx);
3645         dsl_dataset_t *ds;
3646         int error;
3647         uint64_t newval;
3648 
3649         if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3650                 return (SET_ERROR(ENOTSUP));
3651 
3652         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3653         if (error != 0)
3654                 return (error);
3655 
3656         if (ds->ds_is_snapshot) {
3657                 dsl_dataset_rele(ds, FTAG);
3658                 return (SET_ERROR(EINVAL));
3659         }
3660 
3661         error = dsl_prop_predict(ds->ds_dir,
3662             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3663             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3664         if (error != 0) {
3665                 dsl_dataset_rele(ds, FTAG);
3666                 return (error);
3667         }
3668 
3669         if (newval == 0) {
3670                 dsl_dataset_rele(ds, FTAG);
3671                 return (0);
3672         }
3673 
3674         if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3675             newval < ds->ds_reserved) {
3676                 dsl_dataset_rele(ds, FTAG);
3677                 return (SET_ERROR(ENOSPC));
3678         }
3679 
3680         dsl_dataset_rele(ds, FTAG);
3681         return (0);
3682 }
3683 
3684 static void
3685 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3686 {
3687         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3688         dsl_pool_t *dp = dmu_tx_pool(tx);
3689         dsl_dataset_t *ds;
3690         uint64_t newval;
3691 
3692         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3693 
3694         dsl_prop_set_sync_impl(ds,
3695             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3696             ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3697             &ddsqra->ddsqra_value, tx);
3698 
3699         VERIFY0(dsl_prop_get_int_ds(ds,
3700             zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3701 
3702         if (ds->ds_quota != newval) {
3703                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3704                 ds->ds_quota = newval;
3705         }
3706         dsl_dataset_rele(ds, FTAG);
3707 }
3708 
3709 int
3710 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3711     uint64_t refquota)
3712 {
3713         dsl_dataset_set_qr_arg_t ddsqra;
3714 
3715         ddsqra.ddsqra_name = dsname;
3716         ddsqra.ddsqra_source = source;
3717         ddsqra.ddsqra_value = refquota;
3718 
3719         return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3720             dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3721 }
3722 
3723 static int
3724 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3725 {
3726         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3727         dsl_pool_t *dp = dmu_tx_pool(tx);
3728         dsl_dataset_t *ds;
3729         int error;
3730         uint64_t newval, unique;
3731 
3732         if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3733                 return (SET_ERROR(ENOTSUP));
3734 
3735         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3736         if (error != 0)
3737                 return (error);
3738 
3739         if (ds->ds_is_snapshot) {
3740                 dsl_dataset_rele(ds, FTAG);
3741                 return (SET_ERROR(EINVAL));
3742         }
3743 
3744         error = dsl_prop_predict(ds->ds_dir,
3745             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3746             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3747         if (error != 0) {
3748                 dsl_dataset_rele(ds, FTAG);
3749                 return (error);
3750         }
3751 
3752         /*
3753          * If we are doing the preliminary check in open context, the
3754          * space estimates may be inaccurate.
3755          */
3756         if (!dmu_tx_is_syncing(tx)) {
3757                 dsl_dataset_rele(ds, FTAG);
3758                 return (0);
3759         }
3760 
3761         mutex_enter(&ds->ds_lock);
3762         if (!DS_UNIQUE_IS_ACCURATE(ds))
3763                 dsl_dataset_recalc_head_uniq(ds);
3764         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3765         mutex_exit(&ds->ds_lock);
3766 
3767         if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3768                 uint64_t delta = MAX(unique, newval) -
3769                     MAX(unique, ds->ds_reserved);
3770 
3771                 if (delta >
3772                     dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3773                     (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3774                         dsl_dataset_rele(ds, FTAG);
3775                         return (SET_ERROR(ENOSPC));
3776                 }
3777         }
3778 
3779         dsl_dataset_rele(ds, FTAG);
3780         return (0);
3781 }
3782 
3783 void
3784 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3785     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3786 {
3787         uint64_t newval;
3788         uint64_t unique;
3789         int64_t delta;
3790 
3791         dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3792             source, sizeof (value), 1, &value, tx);
3793 
3794         VERIFY0(dsl_prop_get_int_ds(ds,
3795             zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3796 
3797         dmu_buf_will_dirty(ds->ds_dbuf, tx);
3798         mutex_enter(&ds->ds_dir->dd_lock);
3799         mutex_enter(&ds->ds_lock);
3800         ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3801         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3802         delta = MAX(0, (int64_t)(newval - unique)) -
3803             MAX(0, (int64_t)(ds->ds_reserved - unique));
3804         ds->ds_reserved = newval;
3805         mutex_exit(&ds->ds_lock);
3806 
3807         dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3808         mutex_exit(&ds->ds_dir->dd_lock);
3809 }
3810 
3811 static void
3812 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3813 {
3814         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3815         dsl_pool_t *dp = dmu_tx_pool(tx);
3816         dsl_dataset_t *ds;
3817 
3818         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3819         dsl_dataset_set_refreservation_sync_impl(ds,
3820             ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3821         dsl_dataset_rele(ds, FTAG);
3822 }
3823 
3824 int
3825 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3826     uint64_t refreservation)
3827 {
3828         dsl_dataset_set_qr_arg_t ddsqra;
3829 
3830         ddsqra.ddsqra_name = dsname;
3831         ddsqra.ddsqra_source = source;
3832         ddsqra.ddsqra_value = refreservation;
3833 
3834         return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3835             dsl_dataset_set_refreservation_sync, &ddsqra,
3836             0, ZFS_SPACE_CHECK_NONE));
3837 }
3838 
3839 /*
3840  * Return (in *usedp) the amount of space written in new that is not
3841  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3842  * a snapshot before new, in new's filesystem (or its origin).  If not then
3843  * fail and return EINVAL.
3844  *
3845  * The written space is calculated by considering two components:  First, we
3846  * ignore any freed space, and calculate the written as new's used space
3847  * minus old's used space.  Next, we add in the amount of space that was freed
3848  * between the two snapshots, thus reducing new's used space relative to old's.
3849  * Specifically, this is the space that was born before old->ds_creation_txg,
3850  * and freed before new (ie. on new's deadlist or a previous deadlist).
3851  *
3852  * space freed                         [---------------------]
3853  * snapshots                       ---O-------O--------O-------O------
3854  *                                         oldsnap            new
3855  */
3856 int
3857 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3858     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3859 {
3860         int err = 0;
3861         uint64_t snapobj;
3862         dsl_pool_t *dp = new->ds_dir->dd_pool;
3863 
3864         ASSERT(dsl_pool_config_held(dp));
3865 
3866         *usedp = 0;
3867         *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3868         *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3869 
3870         *compp = 0;
3871         *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3872         *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3873 
3874         *uncompp = 0;
3875         *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3876         *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3877 
3878         snapobj = new->ds_object;
3879         while (snapobj != oldsnap->ds_object) {
3880                 dsl_dataset_t *snap;
3881                 uint64_t used, comp, uncomp;
3882 
3883                 if (snapobj == new->ds_object) {
3884                         snap = new;
3885                 } else {
3886                         err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3887                         if (err != 0)
3888                                 break;
3889                 }
3890 
3891                 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3892                     dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3893                         /*
3894                          * The blocks in the deadlist can not be born after
3895                          * ds_prev_snap_txg, so get the whole deadlist space,
3896                          * which is more efficient (especially for old-format
3897                          * deadlists).  Unfortunately the deadlist code
3898                          * doesn't have enough information to make this
3899                          * optimization itself.
3900                          */
3901                         dsl_deadlist_space(&snap->ds_deadlist,
3902                             &used, &comp, &uncomp);
3903                 } else {
3904                         dsl_deadlist_space_range(&snap->ds_deadlist,
3905                             0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3906                             &used, &comp, &uncomp);
3907                 }
3908                 *usedp += used;
3909                 *compp += comp;
3910                 *uncompp += uncomp;
3911 
3912                 /*
3913                  * If we get to the beginning of the chain of snapshots
3914                  * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3915                  * was not a snapshot of/before new.
3916                  */
3917                 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3918                 if (snap != new)
3919                         dsl_dataset_rele(snap, FTAG);
3920                 if (snapobj == 0) {
3921                         err = SET_ERROR(EINVAL);
3922                         break;
3923                 }
3924 
3925         }
3926         return (err);
3927 }
3928 
3929 /*
3930  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3931  * lastsnap, and all snapshots in between are deleted.
3932  *
3933  * blocks that would be freed            [---------------------------]
3934  * snapshots                       ---O-------O--------O-------O--------O
3935  *                                        firstsnap        lastsnap
3936  *
3937  * This is the set of blocks that were born after the snap before firstsnap,
3938  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3939  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3940  * We calculate this by iterating over the relevant deadlists (from the snap
3941  * after lastsnap, backward to the snap after firstsnap), summing up the
3942  * space on the deadlist that was born after the snap before firstsnap.
3943  */
3944 int
3945 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3946     dsl_dataset_t *lastsnap,
3947     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3948 {
3949         int err = 0;
3950         uint64_t snapobj;
3951         dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3952 
3953         ASSERT(firstsnap->ds_is_snapshot);
3954         ASSERT(lastsnap->ds_is_snapshot);
3955 
3956         /*
3957          * Check that the snapshots are in the same dsl_dir, and firstsnap
3958          * is before lastsnap.
3959          */
3960         if (firstsnap->ds_dir != lastsnap->ds_dir ||
3961             dsl_dataset_phys(firstsnap)->ds_creation_txg >
3962             dsl_dataset_phys(lastsnap)->ds_creation_txg)
3963                 return (SET_ERROR(EINVAL));
3964 
3965         *usedp = *compp = *uncompp = 0;
3966 
3967         snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3968         while (snapobj != firstsnap->ds_object) {
3969                 dsl_dataset_t *ds;
3970                 uint64_t used, comp, uncomp;
3971 
3972                 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3973                 if (err != 0)
3974                         break;
3975 
3976                 dsl_deadlist_space_range(&ds->ds_deadlist,
3977                     dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3978                     &used, &comp, &uncomp);
3979                 *usedp += used;
3980                 *compp += comp;
3981                 *uncompp += uncomp;
3982 
3983                 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3984                 ASSERT3U(snapobj, !=, 0);
3985                 dsl_dataset_rele(ds, FTAG);
3986         }
3987         return (err);
3988 }
3989 
3990 /*
3991  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3992  * For example, they could both be snapshots of the same filesystem, and
3993  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3994  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3995  * filesystem.  Or 'earlier' could be the origin's origin.
3996  *
3997  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3998  */
3999 boolean_t
4000 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
4001     uint64_t earlier_txg)
4002 {
4003         dsl_pool_t *dp = later->ds_dir->dd_pool;
4004         int error;
4005         boolean_t ret;
4006 
4007         ASSERT(dsl_pool_config_held(dp));
4008         ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
4009 
4010         if (earlier_txg == 0)
4011                 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
4012 
4013         if (later->ds_is_snapshot &&
4014             earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
4015                 return (B_FALSE);
4016 
4017         if (later->ds_dir == earlier->ds_dir)
4018                 return (B_TRUE);
4019         if (!dsl_dir_is_clone(later->ds_dir))
4020                 return (B_FALSE);
4021 
4022         if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
4023                 return (B_TRUE);
4024         dsl_dataset_t *origin;
4025         error = dsl_dataset_hold_obj(dp,
4026             dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
4027         if (error != 0)
4028                 return (B_FALSE);
4029         ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
4030         dsl_dataset_rele(origin, FTAG);
4031         return (ret);
4032 }
4033 
4034 void
4035 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
4036 {
4037         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
4038         dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
4039 }
4040 
4041 boolean_t
4042 dsl_dataset_is_zapified(dsl_dataset_t *ds)
4043 {
4044         dmu_object_info_t doi;
4045 
4046         dmu_object_info_from_db(ds->ds_dbuf, &doi);
4047         return (doi.doi_type == DMU_OTN_ZAP_METADATA);
4048 }
4049 
4050 boolean_t
4051 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
4052 {
4053         return (dsl_dataset_is_zapified(ds) &&
4054             zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
4055             ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
4056 }
4057 
4058 uint64_t
4059 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds)
4060 {
4061         uint64_t remap_deadlist_obj;
4062         int err;
4063 
4064         if (!dsl_dataset_is_zapified(ds))
4065                 return (0);
4066 
4067         err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4068             DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1,
4069             &remap_deadlist_obj);
4070 
4071         if (err != 0) {
4072                 VERIFY3S(err, ==, ENOENT);
4073                 return (0);
4074         }
4075 
4076         ASSERT(remap_deadlist_obj != 0);
4077         return (remap_deadlist_obj);
4078 }
4079 
4080 boolean_t
4081 dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds)
4082 {
4083         EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist),
4084             dsl_dataset_get_remap_deadlist_object(ds) != 0);
4085         return (dsl_deadlist_is_open(&ds->ds_remap_deadlist));
4086 }
4087 
4088 static void
4089 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj,
4090     dmu_tx_t *tx)
4091 {
4092         ASSERT(obj != 0);
4093         dsl_dataset_zapify(ds, tx);
4094         VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4095             DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx));
4096 }
4097 
4098 static void
4099 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx)
4100 {
4101         VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset,
4102             ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx));
4103 }
4104 
4105 void
4106 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4107 {
4108         uint64_t remap_deadlist_object;
4109         spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4110 
4111         ASSERT(dmu_tx_is_syncing(tx));
4112         ASSERT(dsl_dataset_remap_deadlist_exists(ds));
4113 
4114         remap_deadlist_object = ds->ds_remap_deadlist.dl_object;
4115         dsl_deadlist_close(&ds->ds_remap_deadlist);
4116         dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx);
4117         dsl_dataset_unset_remap_deadlist_object(ds, tx);
4118         spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4119 }
4120 
4121 void
4122 dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4123 {
4124         uint64_t remap_deadlist_obj;
4125         spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4126 
4127         ASSERT(dmu_tx_is_syncing(tx));
4128         ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock));
4129         /*
4130          * Currently we only create remap deadlists when there are indirect
4131          * vdevs with referenced mappings.
4132          */
4133         ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
4134 
4135         remap_deadlist_obj = dsl_deadlist_clone(
4136             &ds->ds_deadlist, UINT64_MAX,
4137             dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
4138         dsl_dataset_set_remap_deadlist_object(ds,
4139             remap_deadlist_obj, tx);
4140         dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa),
4141             remap_deadlist_obj);
4142         spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4143 }