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  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2012 by Delphix. All rights reserved.
  24  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  25  */
  26 
  27 #include <sys/dmu_objset.h>
  28 #include <sys/dsl_dataset.h>
  29 #include <sys/dsl_dir.h>
  30 #include <sys/dsl_prop.h>
  31 #include <sys/dsl_synctask.h>
  32 #include <sys/dmu_traverse.h>
  33 #include <sys/dmu_impl.h>
  34 #include <sys/dmu_tx.h>
  35 #include <sys/arc.h>
  36 #include <sys/zio.h>
  37 #include <sys/zap.h>
  38 #include <sys/zfeature.h>
  39 #include <sys/unique.h>
  40 #include <sys/zfs_context.h>
  41 #include <sys/zfs_ioctl.h>
  42 #include <sys/spa.h>
  43 #include <sys/zfs_znode.h>
  44 #include <sys/zfs_onexit.h>
  45 #include <sys/zvol.h>
  46 #include <sys/dsl_scan.h>
  47 #include <sys/dsl_deadlist.h>
  48 
  49 static char *dsl_reaper = "the grim reaper";
  50 
  51 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
  52 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
  53 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
  54 
  55 #define SWITCH64(x, y) \
  56         { \
  57                 uint64_t __tmp = (x); \
  58                 (x) = (y); \
  59                 (y) = __tmp; \
  60         }
  61 
  62 #define DS_REF_MAX      (1ULL << 62)
  63 
  64 #define DSL_DEADLIST_BLOCKSIZE  SPA_MAXBLOCKSIZE
  65 
  66 #define DSL_DATASET_IS_DESTROYED(ds)    ((ds)->ds_owner == dsl_reaper)
  67 
  68 
  69 /*
  70  * Figure out how much of this delta should be propogated to the dsl_dir
  71  * layer.  If there's a refreservation, that space has already been
  72  * partially accounted for in our ancestors.
  73  */
  74 static int64_t
  75 parent_delta(dsl_dataset_t *ds, int64_t delta)
  76 {
  77         uint64_t old_bytes, new_bytes;
  78 
  79         if (ds->ds_reserved == 0)
  80                 return (delta);
  81 
  82         old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
  83         new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
  84 
  85         ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
  86         return (new_bytes - old_bytes);
  87 }
  88 
  89 void
  90 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
  91 {
  92         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
  93         int compressed = BP_GET_PSIZE(bp);
  94         int uncompressed = BP_GET_UCSIZE(bp);
  95         int64_t delta;
  96 
  97         dprintf_bp(bp, "ds=%p", ds);
  98 
  99         ASSERT(dmu_tx_is_syncing(tx));
 100         /* It could have been compressed away to nothing */
 101         if (BP_IS_HOLE(bp))
 102                 return;
 103         ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
 104         ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
 105         if (ds == NULL) {
 106                 dsl_pool_mos_diduse_space(tx->tx_pool,
 107                     used, compressed, uncompressed);
 108                 return;
 109         }
 110         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 111 
 112         mutex_enter(&ds->ds_dir->dd_lock);
 113         mutex_enter(&ds->ds_lock);
 114         delta = parent_delta(ds, used);
 115         ds->ds_phys->ds_referenced_bytes += used;
 116         ds->ds_phys->ds_compressed_bytes += compressed;
 117         ds->ds_phys->ds_uncompressed_bytes += uncompressed;
 118         ds->ds_phys->ds_unique_bytes += used;
 119         mutex_exit(&ds->ds_lock);
 120         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
 121             compressed, uncompressed, tx);
 122         dsl_dir_transfer_space(ds->ds_dir, used - delta,
 123             DD_USED_REFRSRV, DD_USED_HEAD, tx);
 124         mutex_exit(&ds->ds_dir->dd_lock);
 125 }
 126 
 127 int
 128 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 129     boolean_t async)
 130 {
 131         if (BP_IS_HOLE(bp))
 132                 return (0);
 133 
 134         ASSERT(dmu_tx_is_syncing(tx));
 135         ASSERT(bp->blk_birth <= tx->tx_txg);
 136 
 137         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 138         int compressed = BP_GET_PSIZE(bp);
 139         int uncompressed = BP_GET_UCSIZE(bp);
 140 
 141         ASSERT(used > 0);
 142         if (ds == NULL) {
 143                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 144                 dsl_pool_mos_diduse_space(tx->tx_pool,
 145                     -used, -compressed, -uncompressed);
 146                 return (used);
 147         }
 148         ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
 149 
 150         ASSERT(!dsl_dataset_is_snapshot(ds));
 151         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 152 
 153         if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
 154                 int64_t delta;
 155 
 156                 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
 157                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 158 
 159                 mutex_enter(&ds->ds_dir->dd_lock);
 160                 mutex_enter(&ds->ds_lock);
 161                 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
 162                     !DS_UNIQUE_IS_ACCURATE(ds));
 163                 delta = parent_delta(ds, -used);
 164                 ds->ds_phys->ds_unique_bytes -= used;
 165                 mutex_exit(&ds->ds_lock);
 166                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 167                     delta, -compressed, -uncompressed, tx);
 168                 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
 169                     DD_USED_REFRSRV, DD_USED_HEAD, tx);
 170                 mutex_exit(&ds->ds_dir->dd_lock);
 171         } else {
 172                 dprintf_bp(bp, "putting on dead list: %s", "");
 173                 if (async) {
 174                         /*
 175                          * We are here as part of zio's write done callback,
 176                          * which means we're a zio interrupt thread.  We can't
 177                          * call dsl_deadlist_insert() now because it may block
 178                          * waiting for I/O.  Instead, put bp on the deferred
 179                          * queue and let dsl_pool_sync() finish the job.
 180                          */
 181                         bplist_append(&ds->ds_pending_deadlist, bp);
 182                 } else {
 183                         dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
 184                 }
 185                 ASSERT3U(ds->ds_prev->ds_object, ==,
 186                     ds->ds_phys->ds_prev_snap_obj);
 187                 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
 188                 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
 189                 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
 190                     ds->ds_object && bp->blk_birth >
 191                     ds->ds_prev->ds_phys->ds_prev_snap_txg) {
 192                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 193                         mutex_enter(&ds->ds_prev->ds_lock);
 194                         ds->ds_prev->ds_phys->ds_unique_bytes += used;
 195                         mutex_exit(&ds->ds_prev->ds_lock);
 196                 }
 197                 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
 198                         dsl_dir_transfer_space(ds->ds_dir, used,
 199                             DD_USED_HEAD, DD_USED_SNAP, tx);
 200                 }
 201         }
 202         mutex_enter(&ds->ds_lock);
 203         ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
 204         ds->ds_phys->ds_referenced_bytes -= used;
 205         ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
 206         ds->ds_phys->ds_compressed_bytes -= compressed;
 207         ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
 208         ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
 209         mutex_exit(&ds->ds_lock);
 210 
 211         return (used);
 212 }
 213 
 214 uint64_t
 215 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
 216 {
 217         uint64_t trysnap = 0;
 218 
 219         if (ds == NULL)
 220                 return (0);
 221         /*
 222          * The snapshot creation could fail, but that would cause an
 223          * incorrect FALSE return, which would only result in an
 224          * overestimation of the amount of space that an operation would
 225          * consume, which is OK.
 226          *
 227          * There's also a small window where we could miss a pending
 228          * snapshot, because we could set the sync task in the quiescing
 229          * phase.  So this should only be used as a guess.
 230          */
 231         if (ds->ds_trysnap_txg >
 232             spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
 233                 trysnap = ds->ds_trysnap_txg;
 234         return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
 235 }
 236 
 237 boolean_t
 238 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
 239     uint64_t blk_birth)
 240 {
 241         if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
 242                 return (B_FALSE);
 243 
 244         ddt_prefetch(dsl_dataset_get_spa(ds), bp);
 245 
 246         return (B_TRUE);
 247 }
 248 
 249 /* ARGSUSED */
 250 static void
 251 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
 252 {
 253         dsl_dataset_t *ds = dsv;
 254 
 255         ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
 256 
 257         unique_remove(ds->ds_fsid_guid);
 258 
 259         if (ds->ds_objset != NULL)
 260                 dmu_objset_evict(ds->ds_objset);
 261 
 262         if (ds->ds_prev) {
 263                 dsl_dataset_drop_ref(ds->ds_prev, ds);
 264                 ds->ds_prev = NULL;
 265         }
 266 
 267         bplist_destroy(&ds->ds_pending_deadlist);
 268         if (db != NULL) {
 269                 dsl_deadlist_close(&ds->ds_deadlist);
 270         } else {
 271                 ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
 272                 ASSERT(!ds->ds_deadlist.dl_oldfmt);
 273         }
 274         if (ds->ds_dir)
 275                 dsl_dir_close(ds->ds_dir, ds);
 276 
 277         ASSERT(!list_link_active(&ds->ds_synced_link));
 278 
 279         mutex_destroy(&ds->ds_lock);
 280         mutex_destroy(&ds->ds_recvlock);
 281         mutex_destroy(&ds->ds_opening_lock);
 282         rw_destroy(&ds->ds_rwlock);
 283         cv_destroy(&ds->ds_exclusive_cv);
 284 
 285         kmem_free(ds, sizeof (dsl_dataset_t));
 286 }
 287 
 288 static int
 289 dsl_dataset_get_snapname(dsl_dataset_t *ds)
 290 {
 291         dsl_dataset_phys_t *headphys;
 292         int err;
 293         dmu_buf_t *headdbuf;
 294         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 295         objset_t *mos = dp->dp_meta_objset;
 296 
 297         if (ds->ds_snapname[0])
 298                 return (0);
 299         if (ds->ds_phys->ds_next_snap_obj == 0)
 300                 return (0);
 301 
 302         err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
 303             FTAG, &headdbuf);
 304         if (err)
 305                 return (err);
 306         headphys = headdbuf->db_data;
 307         err = zap_value_search(dp->dp_meta_objset,
 308             headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
 309         dmu_buf_rele(headdbuf, FTAG);
 310         return (err);
 311 }
 312 
 313 static int
 314 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
 315 {
 316         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 317         uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
 318         matchtype_t mt;
 319         int err;
 320 
 321         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 322                 mt = MT_FIRST;
 323         else
 324                 mt = MT_EXACT;
 325 
 326         err = zap_lookup_norm(mos, snapobj, name, 8, 1,
 327             value, mt, NULL, 0, NULL);
 328         if (err == ENOTSUP && mt == MT_FIRST)
 329                 err = zap_lookup(mos, snapobj, name, 8, 1, value);
 330         return (err);
 331 }
 332 
 333 static int
 334 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
 335 {
 336         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 337         uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
 338         matchtype_t mt;
 339         int err;
 340 
 341         dsl_dir_snap_cmtime_update(ds->ds_dir);
 342 
 343         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 344                 mt = MT_FIRST;
 345         else
 346                 mt = MT_EXACT;
 347 
 348         err = zap_remove_norm(mos, snapobj, name, mt, tx);
 349         if (err == ENOTSUP && mt == MT_FIRST)
 350                 err = zap_remove(mos, snapobj, name, tx);
 351         return (err);
 352 }
 353 
 354 static int
 355 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 356     dsl_dataset_t **dsp)
 357 {
 358         objset_t *mos = dp->dp_meta_objset;
 359         dmu_buf_t *dbuf;
 360         dsl_dataset_t *ds;
 361         int err;
 362         dmu_object_info_t doi;
 363 
 364         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
 365             dsl_pool_sync_context(dp));
 366 
 367         err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
 368         if (err)
 369                 return (err);
 370 
 371         /* Make sure dsobj has the correct object type. */
 372         dmu_object_info_from_db(dbuf, &doi);
 373         if (doi.doi_type != DMU_OT_DSL_DATASET)
 374                 return (EINVAL);
 375 
 376         ds = dmu_buf_get_user(dbuf);
 377         if (ds == NULL) {
 378                 dsl_dataset_t *winner;
 379 
 380                 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
 381                 ds->ds_dbuf = dbuf;
 382                 ds->ds_object = dsobj;
 383                 ds->ds_phys = dbuf->db_data;
 384 
 385                 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
 386                 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
 387                 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 388                 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
 389 
 390                 rw_init(&ds->ds_rwlock, 0, 0, 0);
 391                 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
 392 
 393                 bplist_create(&ds->ds_pending_deadlist);
 394                 dsl_deadlist_open(&ds->ds_deadlist,
 395                     mos, ds->ds_phys->ds_deadlist_obj);
 396 
 397                 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
 398                     offsetof(dmu_sendarg_t, dsa_link));
 399 
 400                 if (err == 0) {
 401                         err = dsl_dir_open_obj(dp,
 402                             ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
 403                 }
 404                 if (err) {
 405                         mutex_destroy(&ds->ds_lock);
 406                         mutex_destroy(&ds->ds_recvlock);
 407                         mutex_destroy(&ds->ds_opening_lock);
 408                         rw_destroy(&ds->ds_rwlock);
 409                         cv_destroy(&ds->ds_exclusive_cv);
 410                         bplist_destroy(&ds->ds_pending_deadlist);
 411                         dsl_deadlist_close(&ds->ds_deadlist);
 412                         kmem_free(ds, sizeof (dsl_dataset_t));
 413                         dmu_buf_rele(dbuf, tag);
 414                         return (err);
 415                 }
 416 
 417                 if (!dsl_dataset_is_snapshot(ds)) {
 418                         ds->ds_snapname[0] = '\0';
 419                         if (ds->ds_phys->ds_prev_snap_obj) {
 420                                 err = dsl_dataset_get_ref(dp,
 421                                     ds->ds_phys->ds_prev_snap_obj,
 422                                     ds, &ds->ds_prev);
 423                         }
 424                 } else {
 425                         if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
 426                                 err = dsl_dataset_get_snapname(ds);
 427                         if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
 428                                 err = zap_count(
 429                                     ds->ds_dir->dd_pool->dp_meta_objset,
 430                                     ds->ds_phys->ds_userrefs_obj,
 431                                     &ds->ds_userrefs);
 432                         }
 433                 }
 434 
 435                 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
 436                         /*
 437                          * In sync context, we're called with either no lock
 438                          * or with the write lock.  If we're not syncing,
 439                          * we're always called with the read lock held.
 440                          */
 441                         boolean_t need_lock =
 442                             !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
 443                             dsl_pool_sync_context(dp);
 444 
 445                         if (need_lock)
 446                                 rw_enter(&dp->dp_config_rwlock, RW_READER);
 447 
 448                         err = dsl_prop_get_ds(ds,
 449                             "refreservation", sizeof (uint64_t), 1,
 450                             &ds->ds_reserved, NULL);
 451                         if (err == 0) {
 452                                 err = dsl_prop_get_ds(ds,
 453                                     "refquota", sizeof (uint64_t), 1,
 454                                     &ds->ds_quota, NULL);
 455                         }
 456 
 457                         if (need_lock)
 458                                 rw_exit(&dp->dp_config_rwlock);
 459                 } else {
 460                         ds->ds_reserved = ds->ds_quota = 0;
 461                 }
 462 
 463                 if (err == 0) {
 464                         winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
 465                             dsl_dataset_evict);
 466                 }
 467                 if (err || winner) {
 468                         bplist_destroy(&ds->ds_pending_deadlist);
 469                         dsl_deadlist_close(&ds->ds_deadlist);
 470                         if (ds->ds_prev)
 471                                 dsl_dataset_drop_ref(ds->ds_prev, ds);
 472                         dsl_dir_close(ds->ds_dir, ds);
 473                         mutex_destroy(&ds->ds_lock);
 474                         mutex_destroy(&ds->ds_recvlock);
 475                         mutex_destroy(&ds->ds_opening_lock);
 476                         rw_destroy(&ds->ds_rwlock);
 477                         cv_destroy(&ds->ds_exclusive_cv);
 478                         kmem_free(ds, sizeof (dsl_dataset_t));
 479                         if (err) {
 480                                 dmu_buf_rele(dbuf, tag);
 481                                 return (err);
 482                         }
 483                         ds = winner;
 484                 } else {
 485                         ds->ds_fsid_guid =
 486                             unique_insert(ds->ds_phys->ds_fsid_guid);
 487                 }
 488         }
 489         ASSERT3P(ds->ds_dbuf, ==, dbuf);
 490         ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
 491         ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
 492             spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
 493             dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
 494         mutex_enter(&ds->ds_lock);
 495         if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
 496                 mutex_exit(&ds->ds_lock);
 497                 dmu_buf_rele(ds->ds_dbuf, tag);
 498                 return (ENOENT);
 499         }
 500         mutex_exit(&ds->ds_lock);
 501         *dsp = ds;
 502         return (0);
 503 }
 504 
 505 static int
 506 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
 507 {
 508         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 509 
 510         /*
 511          * In syncing context we don't want the rwlock lock: there
 512          * may be an existing writer waiting for sync phase to
 513          * finish.  We don't need to worry about such writers, since
 514          * sync phase is single-threaded, so the writer can't be
 515          * doing anything while we are active.
 516          */
 517         if (dsl_pool_sync_context(dp)) {
 518                 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
 519                 return (0);
 520         }
 521 
 522         /*
 523          * Normal users will hold the ds_rwlock as a READER until they
 524          * are finished (i.e., call dsl_dataset_rele()).  "Owners" will
 525          * drop their READER lock after they set the ds_owner field.
 526          *
 527          * If the dataset is being destroyed, the destroy thread will
 528          * obtain a WRITER lock for exclusive access after it's done its
 529          * open-context work and then change the ds_owner to
 530          * dsl_reaper once destruction is assured.  So threads
 531          * may block here temporarily, until the "destructability" of
 532          * the dataset is determined.
 533          */
 534         ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
 535         mutex_enter(&ds->ds_lock);
 536         while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
 537                 rw_exit(&dp->dp_config_rwlock);
 538                 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
 539                 if (DSL_DATASET_IS_DESTROYED(ds)) {
 540                         mutex_exit(&ds->ds_lock);
 541                         dsl_dataset_drop_ref(ds, tag);
 542                         rw_enter(&dp->dp_config_rwlock, RW_READER);
 543                         return (ENOENT);
 544                 }
 545                 /*
 546                  * The dp_config_rwlock lives above the ds_lock. And
 547                  * we need to check DSL_DATASET_IS_DESTROYED() while
 548                  * holding the ds_lock, so we have to drop and reacquire
 549                  * the ds_lock here.
 550                  */
 551                 mutex_exit(&ds->ds_lock);
 552                 rw_enter(&dp->dp_config_rwlock, RW_READER);
 553                 mutex_enter(&ds->ds_lock);
 554         }
 555         mutex_exit(&ds->ds_lock);
 556         return (0);
 557 }
 558 
 559 int
 560 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 561     dsl_dataset_t **dsp)
 562 {
 563         int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
 564 
 565         if (err)
 566                 return (err);
 567         return (dsl_dataset_hold_ref(*dsp, tag));
 568 }
 569 
 570 int
 571 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
 572     void *tag, dsl_dataset_t **dsp)
 573 {
 574         int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
 575         if (err)
 576                 return (err);
 577         if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
 578                 dsl_dataset_rele(*dsp, tag);
 579                 *dsp = NULL;
 580                 return (EBUSY);
 581         }
 582         return (0);
 583 }
 584 
 585 int
 586 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
 587 {
 588         dsl_dir_t *dd;
 589         dsl_pool_t *dp;
 590         const char *snapname;
 591         uint64_t obj;
 592         int err = 0;
 593 
 594         err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
 595         if (err)
 596                 return (err);
 597 
 598         dp = dd->dd_pool;
 599         obj = dd->dd_phys->dd_head_dataset_obj;
 600         rw_enter(&dp->dp_config_rwlock, RW_READER);
 601         if (obj)
 602                 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
 603         else
 604                 err = ENOENT;
 605         if (err)
 606                 goto out;
 607 
 608         err = dsl_dataset_hold_ref(*dsp, tag);
 609 
 610         /* we may be looking for a snapshot */
 611         if (err == 0 && snapname != NULL) {
 612                 dsl_dataset_t *ds = NULL;
 613 
 614                 if (*snapname++ != '@') {
 615                         dsl_dataset_rele(*dsp, tag);
 616                         err = ENOENT;
 617                         goto out;
 618                 }
 619 
 620                 dprintf("looking for snapshot '%s'\n", snapname);
 621                 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
 622                 if (err == 0)
 623                         err = dsl_dataset_get_ref(dp, obj, tag, &ds);
 624                 dsl_dataset_rele(*dsp, tag);
 625 
 626                 ASSERT3U((err == 0), ==, (ds != NULL));
 627 
 628                 if (ds) {
 629                         mutex_enter(&ds->ds_lock);
 630                         if (ds->ds_snapname[0] == 0)
 631                                 (void) strlcpy(ds->ds_snapname, snapname,
 632                                     sizeof (ds->ds_snapname));
 633                         mutex_exit(&ds->ds_lock);
 634                         err = dsl_dataset_hold_ref(ds, tag);
 635                         *dsp = err ? NULL : ds;
 636                 }
 637         }
 638 out:
 639         rw_exit(&dp->dp_config_rwlock);
 640         dsl_dir_close(dd, FTAG);
 641         return (err);
 642 }
 643 
 644 int
 645 dsl_dataset_own(const char *name, boolean_t inconsistentok,
 646     void *tag, dsl_dataset_t **dsp)
 647 {
 648         int err = dsl_dataset_hold(name, tag, dsp);
 649         if (err)
 650                 return (err);
 651         if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
 652                 dsl_dataset_rele(*dsp, tag);
 653                 return (EBUSY);
 654         }
 655         return (0);
 656 }
 657 
 658 void
 659 dsl_dataset_name(dsl_dataset_t *ds, char *name)
 660 {
 661         if (ds == NULL) {
 662                 (void) strcpy(name, "mos");
 663         } else {
 664                 dsl_dir_name(ds->ds_dir, name);
 665                 VERIFY(0 == dsl_dataset_get_snapname(ds));
 666                 if (ds->ds_snapname[0]) {
 667                         (void) strcat(name, "@");
 668                         /*
 669                          * We use a "recursive" mutex so that we
 670                          * can call dprintf_ds() with ds_lock held.
 671                          */
 672                         if (!MUTEX_HELD(&ds->ds_lock)) {
 673                                 mutex_enter(&ds->ds_lock);
 674                                 (void) strcat(name, ds->ds_snapname);
 675                                 mutex_exit(&ds->ds_lock);
 676                         } else {
 677                                 (void) strcat(name, ds->ds_snapname);
 678                         }
 679                 }
 680         }
 681 }
 682 
 683 int
 684 dsl_dataset_namelen(dsl_dataset_t *ds)
 685 {
 686         int result;
 687 
 688         if (ds == NULL) {
 689                 result = 3;     /* "mos" */
 690         } else {
 691                 result = dsl_dir_namelen(ds->ds_dir);
 692                 VERIFY(0 == dsl_dataset_get_snapname(ds));
 693                 if (ds->ds_snapname[0]) {
 694                         ++result;       /* adding one for the @-sign */
 695                         if (!MUTEX_HELD(&ds->ds_lock)) {
 696                                 mutex_enter(&ds->ds_lock);
 697                                 result += strlen(ds->ds_snapname);
 698                                 mutex_exit(&ds->ds_lock);
 699                         } else {
 700                                 result += strlen(ds->ds_snapname);
 701                         }
 702                 }
 703         }
 704 
 705         return (result);
 706 }
 707 
 708 void
 709 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
 710 {
 711         dmu_buf_rele(ds->ds_dbuf, tag);
 712 }
 713 
 714 void
 715 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
 716 {
 717         if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
 718                 rw_exit(&ds->ds_rwlock);
 719         }
 720         dsl_dataset_drop_ref(ds, tag);
 721 }
 722 
 723 void
 724 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
 725 {
 726         ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
 727             (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
 728 
 729         mutex_enter(&ds->ds_lock);
 730         ds->ds_owner = NULL;
 731         if (RW_WRITE_HELD(&ds->ds_rwlock)) {
 732                 rw_exit(&ds->ds_rwlock);
 733                 cv_broadcast(&ds->ds_exclusive_cv);
 734         }
 735         mutex_exit(&ds->ds_lock);
 736         if (ds->ds_dbuf)
 737                 dsl_dataset_drop_ref(ds, tag);
 738         else
 739                 dsl_dataset_evict(NULL, ds);
 740 }
 741 
 742 boolean_t
 743 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
 744 {
 745         boolean_t gotit = FALSE;
 746 
 747         mutex_enter(&ds->ds_lock);
 748         if (ds->ds_owner == NULL &&
 749             (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
 750                 ds->ds_owner = tag;
 751                 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
 752                         rw_exit(&ds->ds_rwlock);
 753                 gotit = TRUE;
 754         }
 755         mutex_exit(&ds->ds_lock);
 756         return (gotit);
 757 }
 758 
 759 void
 760 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
 761 {
 762         ASSERT3P(owner, ==, ds->ds_owner);
 763         if (!RW_WRITE_HELD(&ds->ds_rwlock))
 764                 rw_enter(&ds->ds_rwlock, RW_WRITER);
 765 }
 766 
 767 uint64_t
 768 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
 769     uint64_t flags, dmu_tx_t *tx)
 770 {
 771         dsl_pool_t *dp = dd->dd_pool;
 772         dmu_buf_t *dbuf;
 773         dsl_dataset_phys_t *dsphys;
 774         uint64_t dsobj;
 775         objset_t *mos = dp->dp_meta_objset;
 776 
 777         if (origin == NULL)
 778                 origin = dp->dp_origin_snap;
 779 
 780         ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
 781         ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
 782         ASSERT(dmu_tx_is_syncing(tx));
 783         ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
 784 
 785         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
 786             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
 787         VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
 788         dmu_buf_will_dirty(dbuf, tx);
 789         dsphys = dbuf->db_data;
 790         bzero(dsphys, sizeof (dsl_dataset_phys_t));
 791         dsphys->ds_dir_obj = dd->dd_object;
 792         dsphys->ds_flags = flags;
 793         dsphys->ds_fsid_guid = unique_create();
 794         (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
 795             sizeof (dsphys->ds_guid));
 796         dsphys->ds_snapnames_zapobj =
 797             zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
 798             DMU_OT_NONE, 0, tx);
 799         dsphys->ds_creation_time = gethrestime_sec();
 800         dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
 801 
 802         if (origin == NULL) {
 803                 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
 804         } else {
 805                 dsl_dataset_t *ohds;
 806 
 807                 dsphys->ds_prev_snap_obj = origin->ds_object;
 808                 dsphys->ds_prev_snap_txg =
 809                     origin->ds_phys->ds_creation_txg;
 810                 dsphys->ds_referenced_bytes =
 811                     origin->ds_phys->ds_referenced_bytes;
 812                 dsphys->ds_compressed_bytes =
 813                     origin->ds_phys->ds_compressed_bytes;
 814                 dsphys->ds_uncompressed_bytes =
 815                     origin->ds_phys->ds_uncompressed_bytes;
 816                 dsphys->ds_bp = origin->ds_phys->ds_bp;
 817                 dsphys->ds_flags |= origin->ds_phys->ds_flags;
 818 
 819                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
 820                 origin->ds_phys->ds_num_children++;
 821 
 822                 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
 823                     origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
 824                 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
 825                     dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
 826                 dsl_dataset_rele(ohds, FTAG);
 827 
 828                 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
 829                         if (origin->ds_phys->ds_next_clones_obj == 0) {
 830                                 origin->ds_phys->ds_next_clones_obj =
 831                                     zap_create(mos,
 832                                     DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
 833                         }
 834                         VERIFY(0 == zap_add_int(mos,
 835                             origin->ds_phys->ds_next_clones_obj,
 836                             dsobj, tx));
 837                 }
 838 
 839                 dmu_buf_will_dirty(dd->dd_dbuf, tx);
 840                 dd->dd_phys->dd_origin_obj = origin->ds_object;
 841                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 842                         if (origin->ds_dir->dd_phys->dd_clones == 0) {
 843                                 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
 844                                 origin->ds_dir->dd_phys->dd_clones =
 845                                     zap_create(mos,
 846                                     DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
 847                         }
 848                         VERIFY3U(0, ==, zap_add_int(mos,
 849                             origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
 850                 }
 851         }
 852 
 853         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
 854                 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 855 
 856         dmu_buf_rele(dbuf, FTAG);
 857 
 858         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 859         dd->dd_phys->dd_head_dataset_obj = dsobj;
 860 
 861         return (dsobj);
 862 }
 863 
 864 uint64_t
 865 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
 866     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
 867 {
 868         dsl_pool_t *dp = pdd->dd_pool;
 869         uint64_t dsobj, ddobj;
 870         dsl_dir_t *dd;
 871 
 872         ASSERT(lastname[0] != '@');
 873 
 874         ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
 875         VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
 876 
 877         dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
 878 
 879         dsl_deleg_set_create_perms(dd, tx, cr);
 880 
 881         dsl_dir_close(dd, FTAG);
 882 
 883         /*
 884          * If we are creating a clone, make sure we zero out any stale
 885          * data from the origin snapshots zil header.
 886          */
 887         if (origin != NULL) {
 888                 dsl_dataset_t *ds;
 889                 objset_t *os;
 890 
 891                 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
 892                 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
 893                 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
 894                 dsl_dataset_dirty(ds, tx);
 895                 dsl_dataset_rele(ds, FTAG);
 896         }
 897 
 898         return (dsobj);
 899 }
 900 
 901 /*
 902  * The snapshots must all be in the same pool.
 903  */
 904 int
 905 dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer,
 906     nvlist_t *errlist)
 907 {
 908         int err;
 909         dsl_sync_task_t *dst;
 910         spa_t *spa;
 911         nvpair_t *pair;
 912         dsl_sync_task_group_t *dstg;
 913 
 914         pair = nvlist_next_nvpair(snaps, NULL);
 915         if (pair == NULL)
 916                 return (0);
 917 
 918         err = spa_open(nvpair_name(pair), &spa, FTAG);
 919         if (err)
 920                 return (err);
 921         dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
 922 
 923         for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
 924             pair = nvlist_next_nvpair(snaps, pair)) {
 925                 dsl_dataset_t *ds;
 926 
 927                 err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds);
 928                 if (err == 0) {
 929                         struct dsl_ds_destroyarg *dsda;
 930 
 931                         dsl_dataset_make_exclusive(ds, dstg);
 932                         dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg),
 933                             KM_SLEEP);
 934                         dsda->ds = ds;
 935                         dsda->defer = defer;
 936                         dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
 937                             dsl_dataset_destroy_sync, dsda, dstg, 0);
 938                 } else if (err == ENOENT) {
 939                         err = 0;
 940                 } else {
 941                         fnvlist_add_int32(errlist, nvpair_name(pair), err);
 942                         break;
 943                 }
 944         }
 945 
 946         if (err == 0)
 947                 err = dsl_sync_task_group_wait(dstg);
 948 
 949         for (dst = list_head(&dstg->dstg_tasks); dst;
 950             dst = list_next(&dstg->dstg_tasks, dst)) {
 951                 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
 952                 dsl_dataset_t *ds = dsda->ds;
 953 
 954                 /*
 955                  * Return the snapshots that triggered the error.
 956                  */
 957                 if (dst->dst_err != 0) {
 958                         char name[ZFS_MAXNAMELEN];
 959                         dsl_dataset_name(ds, name);
 960                         fnvlist_add_int32(errlist, name, dst->dst_err);
 961                 }
 962                 ASSERT3P(dsda->rm_origin, ==, NULL);
 963                 dsl_dataset_disown(ds, dstg);
 964                 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
 965         }
 966 
 967         dsl_sync_task_group_destroy(dstg);
 968         spa_close(spa, FTAG);
 969         return (err);
 970 
 971 }
 972 
 973 static boolean_t
 974 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
 975 {
 976         boolean_t might_destroy = B_FALSE;
 977 
 978         mutex_enter(&ds->ds_lock);
 979         if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
 980             DS_IS_DEFER_DESTROY(ds))
 981                 might_destroy = B_TRUE;
 982         mutex_exit(&ds->ds_lock);
 983 
 984         return (might_destroy);
 985 }
 986 
 987 /*
 988  * If we're removing a clone, and these three conditions are true:
 989  *      1) the clone's origin has no other children
 990  *      2) the clone's origin has no user references
 991  *      3) the clone's origin has been marked for deferred destruction
 992  * Then, prepare to remove the origin as part of this sync task group.
 993  */
 994 static int
 995 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
 996 {
 997         dsl_dataset_t *ds = dsda->ds;
 998         dsl_dataset_t *origin = ds->ds_prev;
 999 
1000         if (dsl_dataset_might_destroy_origin(origin)) {
1001                 char *name;
1002                 int namelen;
1003                 int error;
1004 
1005                 namelen = dsl_dataset_namelen(origin) + 1;
1006                 name = kmem_alloc(namelen, KM_SLEEP);
1007                 dsl_dataset_name(origin, name);
1008 #ifdef _KERNEL
1009                 error = zfs_unmount_snap(name, NULL);
1010                 if (error) {
1011                         kmem_free(name, namelen);
1012                         return (error);
1013                 }
1014 #endif
1015                 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1016                 kmem_free(name, namelen);
1017                 if (error)
1018                         return (error);
1019                 dsda->rm_origin = origin;
1020                 dsl_dataset_make_exclusive(origin, tag);
1021         }
1022 
1023         return (0);
1024 }
1025 
1026 /*
1027  * ds must be opened as OWNER.  On return (whether successful or not),
1028  * ds will be closed and caller can no longer dereference it.
1029  */
1030 int
1031 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1032 {
1033         int err;
1034         dsl_sync_task_group_t *dstg;
1035         objset_t *os;
1036         dsl_dir_t *dd;
1037         uint64_t obj;
1038         struct dsl_ds_destroyarg dsda = { 0 };
1039 
1040         dsda.ds = ds;
1041 
1042         if (dsl_dataset_is_snapshot(ds)) {
1043                 /* Destroying a snapshot is simpler */
1044                 dsl_dataset_make_exclusive(ds, tag);
1045 
1046                 dsda.defer = defer;
1047                 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1048                     dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1049                     &dsda, tag, 0);
1050                 ASSERT3P(dsda.rm_origin, ==, NULL);
1051                 goto out;
1052         } else if (defer) {
1053                 err = EINVAL;
1054                 goto out;
1055         }
1056 
1057         dd = ds->ds_dir;
1058 
1059         if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds),
1060             &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
1061                 /*
1062                  * Check for errors and mark this ds as inconsistent, in
1063                  * case we crash while freeing the objects.
1064                  */
1065                 err = dsl_sync_task_do(dd->dd_pool,
1066                     dsl_dataset_destroy_begin_check,
1067                     dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1068                 if (err)
1069                         goto out;
1070 
1071                 err = dmu_objset_from_ds(ds, &os);
1072                 if (err)
1073                         goto out;
1074 
1075                 /*
1076                  * Remove all objects while in the open context so that
1077                  * there is less work to do in the syncing context.
1078                  */
1079                 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1080                     ds->ds_phys->ds_prev_snap_txg)) {
1081                         /*
1082                          * Ignore errors, if there is not enough disk space
1083                          * we will deal with it in dsl_dataset_destroy_sync().
1084                          */
1085                         (void) dmu_free_object(os, obj);
1086                 }
1087                 if (err != ESRCH)
1088                         goto out;
1089 
1090                 /*
1091                  * Sync out all in-flight IO.
1092                  */
1093                 txg_wait_synced(dd->dd_pool, 0);
1094 
1095                 /*
1096                  * If we managed to free all the objects in open
1097                  * context, the user space accounting should be zero.
1098                  */
1099                 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1100                     dmu_objset_userused_enabled(os)) {
1101                         uint64_t count;
1102 
1103                         ASSERT(zap_count(os, DMU_USERUSED_OBJECT,
1104                             &count) != 0 || count == 0);
1105                         ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT,
1106                             &count) != 0 || count == 0);
1107                 }
1108         }
1109 
1110         rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1111         err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1112         rw_exit(&dd->dd_pool->dp_config_rwlock);
1113 
1114         if (err)
1115                 goto out;
1116 
1117         /*
1118          * Blow away the dsl_dir + head dataset.
1119          */
1120         dsl_dataset_make_exclusive(ds, tag);
1121         /*
1122          * If we're removing a clone, we might also need to remove its
1123          * origin.
1124          */
1125         do {
1126                 dsda.need_prep = B_FALSE;
1127                 if (dsl_dir_is_clone(dd)) {
1128                         err = dsl_dataset_origin_rm_prep(&dsda, tag);
1129                         if (err) {
1130                                 dsl_dir_close(dd, FTAG);
1131                                 goto out;
1132                         }
1133                 }
1134 
1135                 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1136                 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1137                     dsl_dataset_destroy_sync, &dsda, tag, 0);
1138                 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1139                     dsl_dir_destroy_sync, dd, FTAG, 0);
1140                 err = dsl_sync_task_group_wait(dstg);
1141                 dsl_sync_task_group_destroy(dstg);
1142 
1143                 /*
1144                  * We could be racing against 'zfs release' or 'zfs destroy -d'
1145                  * on the origin snap, in which case we can get EBUSY if we
1146                  * needed to destroy the origin snap but were not ready to
1147                  * do so.
1148                  */
1149                 if (dsda.need_prep) {
1150                         ASSERT(err == EBUSY);
1151                         ASSERT(dsl_dir_is_clone(dd));
1152                         ASSERT(dsda.rm_origin == NULL);
1153                 }
1154         } while (dsda.need_prep);
1155 
1156         if (dsda.rm_origin != NULL)
1157                 dsl_dataset_disown(dsda.rm_origin, tag);
1158 
1159         /* if it is successful, dsl_dir_destroy_sync will close the dd */
1160         if (err)
1161                 dsl_dir_close(dd, FTAG);
1162 out:
1163         dsl_dataset_disown(ds, tag);
1164         return (err);
1165 }
1166 
1167 blkptr_t *
1168 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1169 {
1170         return (&ds->ds_phys->ds_bp);
1171 }
1172 
1173 void
1174 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1175 {
1176         ASSERT(dmu_tx_is_syncing(tx));
1177         /* If it's the meta-objset, set dp_meta_rootbp */
1178         if (ds == NULL) {
1179                 tx->tx_pool->dp_meta_rootbp = *bp;
1180         } else {
1181                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1182                 ds->ds_phys->ds_bp = *bp;
1183         }
1184 }
1185 
1186 spa_t *
1187 dsl_dataset_get_spa(dsl_dataset_t *ds)
1188 {
1189         return (ds->ds_dir->dd_pool->dp_spa);
1190 }
1191 
1192 void
1193 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1194 {
1195         dsl_pool_t *dp;
1196 
1197         if (ds == NULL) /* this is the meta-objset */
1198                 return;
1199 
1200         ASSERT(ds->ds_objset != NULL);
1201 
1202         if (ds->ds_phys->ds_next_snap_obj != 0)
1203                 panic("dirtying snapshot!");
1204 
1205         dp = ds->ds_dir->dd_pool;
1206 
1207         if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1208                 /* up the hold count until we can be written out */
1209                 dmu_buf_add_ref(ds->ds_dbuf, ds);
1210         }
1211 }
1212 
1213 boolean_t
1214 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1215 {
1216         for (int t = 0; t < TXG_SIZE; t++) {
1217                 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1218                     ds, t))
1219                         return (B_TRUE);
1220         }
1221         return (B_FALSE);
1222 }
1223 
1224 /*
1225  * The unique space in the head dataset can be calculated by subtracting
1226  * the space used in the most recent snapshot, that is still being used
1227  * in this file system, from the space currently in use.  To figure out
1228  * the space in the most recent snapshot still in use, we need to take
1229  * the total space used in the snapshot and subtract out the space that
1230  * has been freed up since the snapshot was taken.
1231  */
1232 static void
1233 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1234 {
1235         uint64_t mrs_used;
1236         uint64_t dlused, dlcomp, dluncomp;
1237 
1238         ASSERT(!dsl_dataset_is_snapshot(ds));
1239 
1240         if (ds->ds_phys->ds_prev_snap_obj != 0)
1241                 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
1242         else
1243                 mrs_used = 0;
1244 
1245         dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1246 
1247         ASSERT3U(dlused, <=, mrs_used);
1248         ds->ds_phys->ds_unique_bytes =
1249             ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
1250 
1251         if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1252             SPA_VERSION_UNIQUE_ACCURATE)
1253                 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1254 }
1255 
1256 struct killarg {
1257         dsl_dataset_t *ds;
1258         dmu_tx_t *tx;
1259 };
1260 
1261 /* ARGSUSED */
1262 static int
1263 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1264     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1265 {
1266         struct killarg *ka = arg;
1267         dmu_tx_t *tx = ka->tx;
1268 
1269         if (bp == NULL)
1270                 return (0);
1271 
1272         if (zb->zb_level == ZB_ZIL_LEVEL) {
1273                 ASSERT(zilog != NULL);
1274                 /*
1275                  * It's a block in the intent log.  It has no
1276                  * accounting, so just free it.
1277                  */
1278                 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1279         } else {
1280                 ASSERT(zilog == NULL);
1281                 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1282                 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1283         }
1284 
1285         return (0);
1286 }
1287 
1288 /* ARGSUSED */
1289 static int
1290 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1291 {
1292         dsl_dataset_t *ds = arg1;
1293         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1294         uint64_t count;
1295         int err;
1296 
1297         /*
1298          * Can't delete a head dataset if there are snapshots of it.
1299          * (Except if the only snapshots are from the branch we cloned
1300          * from.)
1301          */
1302         if (ds->ds_prev != NULL &&
1303             ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1304                 return (EBUSY);
1305 
1306         /*
1307          * This is really a dsl_dir thing, but check it here so that
1308          * we'll be less likely to leave this dataset inconsistent &
1309          * nearly destroyed.
1310          */
1311         err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1312         if (err)
1313                 return (err);
1314         if (count != 0)
1315                 return (EEXIST);
1316 
1317         return (0);
1318 }
1319 
1320 /* ARGSUSED */
1321 static void
1322 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1323 {
1324         dsl_dataset_t *ds = arg1;
1325 
1326         /* Mark it as inconsistent on-disk, in case we crash */
1327         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1328         ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1329 
1330         spa_history_log_internal_ds(ds, "destroy begin", tx, "");
1331 }
1332 
1333 static int
1334 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1335     dmu_tx_t *tx)
1336 {
1337         dsl_dataset_t *ds = dsda->ds;
1338         dsl_dataset_t *ds_prev = ds->ds_prev;
1339 
1340         if (dsl_dataset_might_destroy_origin(ds_prev)) {
1341                 struct dsl_ds_destroyarg ndsda = {0};
1342 
1343                 /*
1344                  * If we're not prepared to remove the origin, don't remove
1345                  * the clone either.
1346                  */
1347                 if (dsda->rm_origin == NULL) {
1348                         dsda->need_prep = B_TRUE;
1349                         return (EBUSY);
1350                 }
1351 
1352                 ndsda.ds = ds_prev;
1353                 ndsda.is_origin_rm = B_TRUE;
1354                 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1355         }
1356 
1357         /*
1358          * If we're not going to remove the origin after all,
1359          * undo the open context setup.
1360          */
1361         if (dsda->rm_origin != NULL) {
1362                 dsl_dataset_disown(dsda->rm_origin, tag);
1363                 dsda->rm_origin = NULL;
1364         }
1365 
1366         return (0);
1367 }
1368 
1369 /*
1370  * If you add new checks here, you may need to add
1371  * additional checks to the "temporary" case in
1372  * snapshot_check() in dmu_objset.c.
1373  */
1374 /* ARGSUSED */
1375 int
1376 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1377 {
1378         struct dsl_ds_destroyarg *dsda = arg1;
1379         dsl_dataset_t *ds = dsda->ds;
1380 
1381         /* we have an owner hold, so noone else can destroy us */
1382         ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1383 
1384         /*
1385          * Only allow deferred destroy on pools that support it.
1386          * NOTE: deferred destroy is only supported on snapshots.
1387          */
1388         if (dsda->defer) {
1389                 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1390                     SPA_VERSION_USERREFS)
1391                         return (ENOTSUP);
1392                 ASSERT(dsl_dataset_is_snapshot(ds));
1393                 return (0);
1394         }
1395 
1396         /*
1397          * Can't delete a head dataset if there are snapshots of it.
1398          * (Except if the only snapshots are from the branch we cloned
1399          * from.)
1400          */
1401         if (ds->ds_prev != NULL &&
1402             ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1403                 return (EBUSY);
1404 
1405         /*
1406          * If we made changes this txg, traverse_dsl_dataset won't find
1407          * them.  Try again.
1408          */
1409         if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1410                 return (EAGAIN);
1411 
1412         if (dsl_dataset_is_snapshot(ds)) {
1413                 /*
1414                  * If this snapshot has an elevated user reference count,
1415                  * we can't destroy it yet.
1416                  */
1417                 if (ds->ds_userrefs > 0 && !dsda->releasing)
1418                         return (EBUSY);
1419 
1420                 mutex_enter(&ds->ds_lock);
1421                 /*
1422                  * Can't delete a branch point. However, if we're destroying
1423                  * a clone and removing its origin due to it having a user
1424                  * hold count of 0 and having been marked for deferred destroy,
1425                  * it's OK for the origin to have a single clone.
1426                  */
1427                 if (ds->ds_phys->ds_num_children >
1428                     (dsda->is_origin_rm ? 2 : 1)) {
1429                         mutex_exit(&ds->ds_lock);
1430                         return (EEXIST);
1431                 }
1432                 mutex_exit(&ds->ds_lock);
1433         } else if (dsl_dir_is_clone(ds->ds_dir)) {
1434                 return (dsl_dataset_origin_check(dsda, arg2, tx));
1435         }
1436 
1437         /* XXX we should do some i/o error checking... */
1438         return (0);
1439 }
1440 
1441 struct refsarg {
1442         kmutex_t lock;
1443         boolean_t gone;
1444         kcondvar_t cv;
1445 };
1446 
1447 /* ARGSUSED */
1448 static void
1449 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1450 {
1451         struct refsarg *arg = argv;
1452 
1453         mutex_enter(&arg->lock);
1454         arg->gone = TRUE;
1455         cv_signal(&arg->cv);
1456         mutex_exit(&arg->lock);
1457 }
1458 
1459 static void
1460 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1461 {
1462         struct refsarg arg;
1463 
1464         mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1465         cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1466         arg.gone = FALSE;
1467         (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1468             dsl_dataset_refs_gone);
1469         dmu_buf_rele(ds->ds_dbuf, tag);
1470         mutex_enter(&arg.lock);
1471         while (!arg.gone)
1472                 cv_wait(&arg.cv, &arg.lock);
1473         ASSERT(arg.gone);
1474         mutex_exit(&arg.lock);
1475         ds->ds_dbuf = NULL;
1476         ds->ds_phys = NULL;
1477         mutex_destroy(&arg.lock);
1478         cv_destroy(&arg.cv);
1479 }
1480 
1481 static void
1482 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1483 {
1484         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1485         uint64_t count;
1486         int err;
1487 
1488         ASSERT(ds->ds_phys->ds_num_children >= 2);
1489         err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1490         /*
1491          * The err should not be ENOENT, but a bug in a previous version
1492          * of the code could cause upgrade_clones_cb() to not set
1493          * ds_next_snap_obj when it should, leading to a missing entry.
1494          * If we knew that the pool was created after
1495          * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1496          * ENOENT.  However, at least we can check that we don't have
1497          * too many entries in the next_clones_obj even after failing to
1498          * remove this one.
1499          */
1500         if (err != ENOENT) {
1501                 VERIFY0(err);
1502         }
1503         ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1504             &count));
1505         ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1506 }
1507 
1508 static void
1509 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1510 {
1511         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1512         zap_cursor_t zc;
1513         zap_attribute_t za;
1514 
1515         /*
1516          * If it is the old version, dd_clones doesn't exist so we can't
1517          * find the clones, but deadlist_remove_key() is a no-op so it
1518          * doesn't matter.
1519          */
1520         if (ds->ds_dir->dd_phys->dd_clones == 0)
1521                 return;
1522 
1523         for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1524             zap_cursor_retrieve(&zc, &za) == 0;
1525             zap_cursor_advance(&zc)) {
1526                 dsl_dataset_t *clone;
1527 
1528                 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1529                     za.za_first_integer, FTAG, &clone));
1530                 if (clone->ds_dir->dd_origin_txg > mintxg) {
1531                         dsl_deadlist_remove_key(&clone->ds_deadlist,
1532                             mintxg, tx);
1533                         dsl_dataset_remove_clones_key(clone, mintxg, tx);
1534                 }
1535                 dsl_dataset_rele(clone, FTAG);
1536         }
1537         zap_cursor_fini(&zc);
1538 }
1539 
1540 struct process_old_arg {
1541         dsl_dataset_t *ds;
1542         dsl_dataset_t *ds_prev;
1543         boolean_t after_branch_point;
1544         zio_t *pio;
1545         uint64_t used, comp, uncomp;
1546 };
1547 
1548 static int
1549 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1550 {
1551         struct process_old_arg *poa = arg;
1552         dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1553 
1554         if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1555                 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1556                 if (poa->ds_prev && !poa->after_branch_point &&
1557                     bp->blk_birth >
1558                     poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1559                         poa->ds_prev->ds_phys->ds_unique_bytes +=
1560                             bp_get_dsize_sync(dp->dp_spa, bp);
1561                 }
1562         } else {
1563                 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1564                 poa->comp += BP_GET_PSIZE(bp);
1565                 poa->uncomp += BP_GET_UCSIZE(bp);
1566                 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1567         }
1568         return (0);
1569 }
1570 
1571 static void
1572 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1573     dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1574 {
1575         struct process_old_arg poa = { 0 };
1576         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1577         objset_t *mos = dp->dp_meta_objset;
1578 
1579         ASSERT(ds->ds_deadlist.dl_oldfmt);
1580         ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1581 
1582         poa.ds = ds;
1583         poa.ds_prev = ds_prev;
1584         poa.after_branch_point = after_branch_point;
1585         poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1586         VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1587             process_old_cb, &poa, tx));
1588         VERIFY0(zio_wait(poa.pio));
1589         ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1590 
1591         /* change snapused */
1592         dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1593             -poa.used, -poa.comp, -poa.uncomp, tx);
1594 
1595         /* swap next's deadlist to our deadlist */
1596         dsl_deadlist_close(&ds->ds_deadlist);
1597         dsl_deadlist_close(&ds_next->ds_deadlist);
1598         SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1599             ds->ds_phys->ds_deadlist_obj);
1600         dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1601         dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1602             ds_next->ds_phys->ds_deadlist_obj);
1603 }
1604 
1605 static int
1606 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
1607 {
1608         int err;
1609         struct killarg ka;
1610 
1611         /*
1612          * Free everything that we point to (that's born after
1613          * the previous snapshot, if we are a clone)
1614          *
1615          * NB: this should be very quick, because we already
1616          * freed all the objects in open context.
1617          */
1618         ka.ds = ds;
1619         ka.tx = tx;
1620         err = traverse_dataset(ds,
1621             ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST,
1622             kill_blkptr, &ka);
1623         ASSERT0(err);
1624         ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0);
1625 
1626         return (err);
1627 }
1628 
1629 void
1630 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1631 {
1632         struct dsl_ds_destroyarg *dsda = arg1;
1633         dsl_dataset_t *ds = dsda->ds;
1634         int err;
1635         int after_branch_point = FALSE;
1636         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1637         objset_t *mos = dp->dp_meta_objset;
1638         dsl_dataset_t *ds_prev = NULL;
1639         boolean_t wont_destroy;
1640         uint64_t obj;
1641 
1642         wont_destroy = (dsda->defer &&
1643             (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1644 
1645         ASSERT(ds->ds_owner || wont_destroy);
1646         ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1647         ASSERT(ds->ds_prev == NULL ||
1648             ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1649         ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1650 
1651         if (wont_destroy) {
1652                 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1653                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1654                 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1655                 spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
1656                 return;
1657         }
1658 
1659         /* We need to log before removing it from the namespace. */
1660         spa_history_log_internal_ds(ds, "destroy", tx, "");
1661 
1662         /* signal any waiters that this dataset is going away */
1663         mutex_enter(&ds->ds_lock);
1664         ds->ds_owner = dsl_reaper;
1665         cv_broadcast(&ds->ds_exclusive_cv);
1666         mutex_exit(&ds->ds_lock);
1667 
1668         /* Remove our reservation */
1669         if (ds->ds_reserved != 0) {
1670                 dsl_prop_setarg_t psa;
1671                 uint64_t value = 0;
1672 
1673                 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1674                     (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1675                     &value);
1676                 psa.psa_effective_value = 0;    /* predict default value */
1677 
1678                 dsl_dataset_set_reservation_sync(ds, &psa, tx);
1679                 ASSERT0(ds->ds_reserved);
1680         }
1681 
1682         ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1683 
1684         dsl_scan_ds_destroyed(ds, tx);
1685 
1686         obj = ds->ds_object;
1687 
1688         if (ds->ds_phys->ds_prev_snap_obj != 0) {
1689                 if (ds->ds_prev) {
1690                         ds_prev = ds->ds_prev;
1691                 } else {
1692                         VERIFY(0 == dsl_dataset_hold_obj(dp,
1693                             ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1694                 }
1695                 after_branch_point =
1696                     (ds_prev->ds_phys->ds_next_snap_obj != obj);
1697 
1698                 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1699                 if (after_branch_point &&
1700                     ds_prev->ds_phys->ds_next_clones_obj != 0) {
1701                         remove_from_next_clones(ds_prev, obj, tx);
1702                         if (ds->ds_phys->ds_next_snap_obj != 0) {
1703                                 VERIFY(0 == zap_add_int(mos,
1704                                     ds_prev->ds_phys->ds_next_clones_obj,
1705                                     ds->ds_phys->ds_next_snap_obj, tx));
1706                         }
1707                 }
1708                 if (after_branch_point &&
1709                     ds->ds_phys->ds_next_snap_obj == 0) {
1710                         /* This clone is toast. */
1711                         ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1712                         ds_prev->ds_phys->ds_num_children--;
1713 
1714                         /*
1715                          * If the clone's origin has no other clones, no
1716                          * user holds, and has been marked for deferred
1717                          * deletion, then we should have done the necessary
1718                          * destroy setup for it.
1719                          */
1720                         if (ds_prev->ds_phys->ds_num_children == 1 &&
1721                             ds_prev->ds_userrefs == 0 &&
1722                             DS_IS_DEFER_DESTROY(ds_prev)) {
1723                                 ASSERT3P(dsda->rm_origin, !=, NULL);
1724                         } else {
1725                                 ASSERT3P(dsda->rm_origin, ==, NULL);
1726                         }
1727                 } else if (!after_branch_point) {
1728                         ds_prev->ds_phys->ds_next_snap_obj =
1729                             ds->ds_phys->ds_next_snap_obj;
1730                 }
1731         }
1732 
1733         if (dsl_dataset_is_snapshot(ds)) {
1734                 dsl_dataset_t *ds_next;
1735                 uint64_t old_unique;
1736                 uint64_t used = 0, comp = 0, uncomp = 0;
1737 
1738                 VERIFY(0 == dsl_dataset_hold_obj(dp,
1739                     ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1740                 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1741 
1742                 old_unique = ds_next->ds_phys->ds_unique_bytes;
1743 
1744                 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1745                 ds_next->ds_phys->ds_prev_snap_obj =
1746                     ds->ds_phys->ds_prev_snap_obj;
1747                 ds_next->ds_phys->ds_prev_snap_txg =
1748                     ds->ds_phys->ds_prev_snap_txg;
1749                 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1750                     ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1751 
1752 
1753                 if (ds_next->ds_deadlist.dl_oldfmt) {
1754                         process_old_deadlist(ds, ds_prev, ds_next,
1755                             after_branch_point, tx);
1756                 } else {
1757                         /* Adjust prev's unique space. */
1758                         if (ds_prev && !after_branch_point) {
1759                                 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1760                                     ds_prev->ds_phys->ds_prev_snap_txg,
1761                                     ds->ds_phys->ds_prev_snap_txg,
1762                                     &used, &comp, &uncomp);
1763                                 ds_prev->ds_phys->ds_unique_bytes += used;
1764                         }
1765 
1766                         /* Adjust snapused. */
1767                         dsl_deadlist_space_range(&ds_next->ds_deadlist,
1768                             ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1769                             &used, &comp, &uncomp);
1770                         dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1771                             -used, -comp, -uncomp, tx);
1772 
1773                         /* Move blocks to be freed to pool's free list. */
1774                         dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1775                             &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1776                             tx);
1777                         dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1778                             DD_USED_HEAD, used, comp, uncomp, tx);
1779 
1780                         /* Merge our deadlist into next's and free it. */
1781                         dsl_deadlist_merge(&ds_next->ds_deadlist,
1782                             ds->ds_phys->ds_deadlist_obj, tx);
1783                 }
1784                 dsl_deadlist_close(&ds->ds_deadlist);
1785                 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1786 
1787                 /* Collapse range in clone heads */
1788                 dsl_dataset_remove_clones_key(ds,
1789                     ds->ds_phys->ds_creation_txg, tx);
1790 
1791                 if (dsl_dataset_is_snapshot(ds_next)) {
1792                         dsl_dataset_t *ds_nextnext;
1793 
1794                         /*
1795                          * Update next's unique to include blocks which
1796                          * were previously shared by only this snapshot
1797                          * and it.  Those blocks will be born after the
1798                          * prev snap and before this snap, and will have
1799                          * died after the next snap and before the one
1800                          * after that (ie. be on the snap after next's
1801                          * deadlist).
1802                          */
1803                         VERIFY(0 == dsl_dataset_hold_obj(dp,
1804                             ds_next->ds_phys->ds_next_snap_obj,
1805                             FTAG, &ds_nextnext));
1806                         dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1807                             ds->ds_phys->ds_prev_snap_txg,
1808                             ds->ds_phys->ds_creation_txg,
1809                             &used, &comp, &uncomp);
1810                         ds_next->ds_phys->ds_unique_bytes += used;
1811                         dsl_dataset_rele(ds_nextnext, FTAG);
1812                         ASSERT3P(ds_next->ds_prev, ==, NULL);
1813 
1814                         /* Collapse range in this head. */
1815                         dsl_dataset_t *hds;
1816                         VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1817                             ds->ds_dir->dd_phys->dd_head_dataset_obj,
1818                             FTAG, &hds));
1819                         dsl_deadlist_remove_key(&hds->ds_deadlist,
1820                             ds->ds_phys->ds_creation_txg, tx);
1821                         dsl_dataset_rele(hds, FTAG);
1822 
1823                 } else {
1824                         ASSERT3P(ds_next->ds_prev, ==, ds);
1825                         dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1826                         ds_next->ds_prev = NULL;
1827                         if (ds_prev) {
1828                                 VERIFY(0 == dsl_dataset_get_ref(dp,
1829                                     ds->ds_phys->ds_prev_snap_obj,
1830                                     ds_next, &ds_next->ds_prev));
1831                         }
1832 
1833                         dsl_dataset_recalc_head_uniq(ds_next);
1834 
1835                         /*
1836                          * Reduce the amount of our unconsmed refreservation
1837                          * being charged to our parent by the amount of
1838                          * new unique data we have gained.
1839                          */
1840                         if (old_unique < ds_next->ds_reserved) {
1841                                 int64_t mrsdelta;
1842                                 uint64_t new_unique =
1843                                     ds_next->ds_phys->ds_unique_bytes;
1844 
1845                                 ASSERT(old_unique <= new_unique);
1846                                 mrsdelta = MIN(new_unique - old_unique,
1847                                     ds_next->ds_reserved - old_unique);
1848                                 dsl_dir_diduse_space(ds->ds_dir,
1849                                     DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1850                         }
1851                 }
1852                 dsl_dataset_rele(ds_next, FTAG);
1853         } else {
1854                 zfeature_info_t *async_destroy =
1855                     &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY];
1856                 objset_t *os;
1857 
1858                 /*
1859                  * There's no next snapshot, so this is a head dataset.
1860                  * Destroy the deadlist.  Unless it's a clone, the
1861                  * deadlist should be empty.  (If it's a clone, it's
1862                  * safe to ignore the deadlist contents.)
1863                  */
1864                 dsl_deadlist_close(&ds->ds_deadlist);
1865                 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1866                 ds->ds_phys->ds_deadlist_obj = 0;
1867 
1868                 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
1869 
1870                 if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) {
1871                         err = old_synchronous_dataset_destroy(ds, tx);
1872                 } else {
1873                         /*
1874                          * Move the bptree into the pool's list of trees to
1875                          * clean up and update space accounting information.
1876                          */
1877                         uint64_t used, comp, uncomp;
1878 
1879                         zil_destroy_sync(dmu_objset_zil(os), tx);
1880 
1881                         if (!spa_feature_is_active(dp->dp_spa, async_destroy)) {
1882                                 spa_feature_incr(dp->dp_spa, async_destroy, tx);
1883                                 dp->dp_bptree_obj = bptree_alloc(mos, tx);
1884                                 VERIFY(zap_add(mos,
1885                                     DMU_POOL_DIRECTORY_OBJECT,
1886                                     DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
1887                                     &dp->dp_bptree_obj, tx) == 0);
1888                         }
1889 
1890                         used = ds->ds_dir->dd_phys->dd_used_bytes;
1891                         comp = ds->ds_dir->dd_phys->dd_compressed_bytes;
1892                         uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes;
1893 
1894                         ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1895                             ds->ds_phys->ds_unique_bytes == used);
1896 
1897                         bptree_add(mos, dp->dp_bptree_obj,
1898                             &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg,
1899                             used, comp, uncomp, tx);
1900                         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
1901                             -used, -comp, -uncomp, tx);
1902                         dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
1903                             used, comp, uncomp, tx);
1904                 }
1905 
1906                 if (ds->ds_prev != NULL) {
1907                         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1908                                 VERIFY3U(0, ==, zap_remove_int(mos,
1909                                     ds->ds_prev->ds_dir->dd_phys->dd_clones,
1910                                     ds->ds_object, tx));
1911                         }
1912                         dsl_dataset_rele(ds->ds_prev, ds);
1913                         ds->ds_prev = ds_prev = NULL;
1914                 }
1915         }
1916 
1917         /*
1918          * This must be done after the dsl_traverse(), because it will
1919          * re-open the objset.
1920          */
1921         if (ds->ds_objset) {
1922                 dmu_objset_evict(ds->ds_objset);
1923                 ds->ds_objset = NULL;
1924         }
1925 
1926         if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1927                 /* Erase the link in the dir */
1928                 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1929                 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1930                 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1931                 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1932                 ASSERT(err == 0);
1933         } else {
1934                 /* remove from snapshot namespace */
1935                 dsl_dataset_t *ds_head;
1936                 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1937                 VERIFY(0 == dsl_dataset_hold_obj(dp,
1938                     ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1939                 VERIFY(0 == dsl_dataset_get_snapname(ds));
1940 #ifdef ZFS_DEBUG
1941                 {
1942                         uint64_t val;
1943 
1944                         err = dsl_dataset_snap_lookup(ds_head,
1945                             ds->ds_snapname, &val);
1946                         ASSERT0(err);
1947                         ASSERT3U(val, ==, obj);
1948                 }
1949 #endif
1950                 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1951                 ASSERT(err == 0);
1952                 dsl_dataset_rele(ds_head, FTAG);
1953         }
1954 
1955         if (ds_prev && ds->ds_prev != ds_prev)
1956                 dsl_dataset_rele(ds_prev, FTAG);
1957 
1958         spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1959 
1960         if (ds->ds_phys->ds_next_clones_obj != 0) {
1961                 uint64_t count;
1962                 ASSERT(0 == zap_count(mos,
1963                     ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1964                 VERIFY(0 == dmu_object_free(mos,
1965                     ds->ds_phys->ds_next_clones_obj, tx));
1966         }
1967         if (ds->ds_phys->ds_props_obj != 0)
1968                 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1969         if (ds->ds_phys->ds_userrefs_obj != 0)
1970                 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1971         dsl_dir_close(ds->ds_dir, ds);
1972         ds->ds_dir = NULL;
1973         dsl_dataset_drain_refs(ds, tag);
1974         VERIFY(0 == dmu_object_free(mos, obj, tx));
1975 
1976         if (dsda->rm_origin) {
1977                 /*
1978                  * Remove the origin of the clone we just destroyed.
1979                  */
1980                 struct dsl_ds_destroyarg ndsda = {0};
1981 
1982                 ndsda.ds = dsda->rm_origin;
1983                 dsl_dataset_destroy_sync(&ndsda, tag, tx);
1984         }
1985 }
1986 
1987 static int
1988 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1989 {
1990         uint64_t asize;
1991 
1992         if (!dmu_tx_is_syncing(tx))
1993                 return (0);
1994 
1995         /*
1996          * If there's an fs-only reservation, any blocks that might become
1997          * owned by the snapshot dataset must be accommodated by space
1998          * outside of the reservation.
1999          */
2000         ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
2001         asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2002         if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2003                 return (ENOSPC);
2004 
2005         /*
2006          * Propagate any reserved space for this snapshot to other
2007          * snapshot checks in this sync group.
2008          */
2009         if (asize > 0)
2010                 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
2011 
2012         return (0);
2013 }
2014 
2015 int
2016 dsl_dataset_snapshot_check(dsl_dataset_t *ds, const char *snapname,
2017     dmu_tx_t *tx)
2018 {
2019         int err;
2020         uint64_t value;
2021 
2022         /*
2023          * We don't allow multiple snapshots of the same txg.  If there
2024          * is already one, try again.
2025          */
2026         if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
2027                 return (EAGAIN);
2028 
2029         /*
2030          * Check for conflicting snapshot name.
2031          */
2032         err = dsl_dataset_snap_lookup(ds, snapname, &value);
2033         if (err == 0)
2034                 return (EEXIST);
2035         if (err != ENOENT)
2036                 return (err);
2037 
2038         /*
2039          * Check that the dataset's name is not too long.  Name consists
2040          * of the dataset's length + 1 for the @-sign + snapshot name's length
2041          */
2042         if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2043                 return (ENAMETOOLONG);
2044 
2045         err = dsl_dataset_snapshot_reserve_space(ds, tx);
2046         if (err)
2047                 return (err);
2048 
2049         ds->ds_trysnap_txg = tx->tx_txg;
2050         return (0);
2051 }
2052 
2053 void
2054 dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *snapname,
2055     dmu_tx_t *tx)
2056 {
2057         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2058         dmu_buf_t *dbuf;
2059         dsl_dataset_phys_t *dsphys;
2060         uint64_t dsobj, crtxg;
2061         objset_t *mos = dp->dp_meta_objset;
2062         int err;
2063 
2064         ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2065 
2066         /*
2067          * The origin's ds_creation_txg has to be < TXG_INITIAL
2068          */
2069         if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2070                 crtxg = 1;
2071         else
2072                 crtxg = tx->tx_txg;
2073 
2074         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2075             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2076         VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2077         dmu_buf_will_dirty(dbuf, tx);
2078         dsphys = dbuf->db_data;
2079         bzero(dsphys, sizeof (dsl_dataset_phys_t));
2080         dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2081         dsphys->ds_fsid_guid = unique_create();
2082         (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2083             sizeof (dsphys->ds_guid));
2084         dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2085         dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2086         dsphys->ds_next_snap_obj = ds->ds_object;
2087         dsphys->ds_num_children = 1;
2088         dsphys->ds_creation_time = gethrestime_sec();
2089         dsphys->ds_creation_txg = crtxg;
2090         dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2091         dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
2092         dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2093         dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2094         dsphys->ds_flags = ds->ds_phys->ds_flags;
2095         dsphys->ds_bp = ds->ds_phys->ds_bp;
2096         dmu_buf_rele(dbuf, FTAG);
2097 
2098         ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2099         if (ds->ds_prev) {
2100                 uint64_t next_clones_obj =
2101                     ds->ds_prev->ds_phys->ds_next_clones_obj;
2102                 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2103                     ds->ds_object ||
2104                     ds->ds_prev->ds_phys->ds_num_children > 1);
2105                 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2106                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2107                         ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2108                             ds->ds_prev->ds_phys->ds_creation_txg);
2109                         ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2110                 } else if (next_clones_obj != 0) {
2111                         remove_from_next_clones(ds->ds_prev,
2112                             dsphys->ds_next_snap_obj, tx);
2113                         VERIFY3U(0, ==, zap_add_int(mos,
2114                             next_clones_obj, dsobj, tx));
2115                 }
2116         }
2117 
2118         /*
2119          * If we have a reference-reservation on this dataset, we will
2120          * need to increase the amount of refreservation being charged
2121          * since our unique space is going to zero.
2122          */
2123         if (ds->ds_reserved) {
2124                 int64_t delta;
2125                 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2126                 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2127                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2128                     delta, 0, 0, tx);
2129         }
2130 
2131         dmu_buf_will_dirty(ds->ds_dbuf, tx);
2132         zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2133             ds->ds_dir->dd_myname, snapname, dsobj,
2134             ds->ds_phys->ds_prev_snap_txg);
2135         ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2136             UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2137         dsl_deadlist_close(&ds->ds_deadlist);
2138         dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2139         dsl_deadlist_add_key(&ds->ds_deadlist,
2140             ds->ds_phys->ds_prev_snap_txg, tx);
2141 
2142         ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2143         ds->ds_phys->ds_prev_snap_obj = dsobj;
2144         ds->ds_phys->ds_prev_snap_txg = crtxg;
2145         ds->ds_phys->ds_unique_bytes = 0;
2146         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2147                 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2148 
2149         err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2150             snapname, 8, 1, &dsobj, tx);
2151         ASSERT(err == 0);
2152 
2153         if (ds->ds_prev)
2154                 dsl_dataset_drop_ref(ds->ds_prev, ds);
2155         VERIFY(0 == dsl_dataset_get_ref(dp,
2156             ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2157 
2158         dsl_scan_ds_snapshotted(ds, tx);
2159 
2160         dsl_dir_snap_cmtime_update(ds->ds_dir);
2161 
2162         spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
2163 }
2164 
2165 void
2166 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2167 {
2168         ASSERT(dmu_tx_is_syncing(tx));
2169         ASSERT(ds->ds_objset != NULL);
2170         ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2171 
2172         /*
2173          * in case we had to change ds_fsid_guid when we opened it,
2174          * sync it out now.
2175          */
2176         dmu_buf_will_dirty(ds->ds_dbuf, tx);
2177         ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2178 
2179         dmu_objset_sync(ds->ds_objset, zio, tx);
2180 }
2181 
2182 static void
2183 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2184 {
2185         uint64_t count = 0;
2186         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2187         zap_cursor_t zc;
2188         zap_attribute_t za;
2189         nvlist_t *propval;
2190         nvlist_t *val;
2191 
2192         rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2193         VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2194         VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2195 
2196         /*
2197          * There may me missing entries in ds_next_clones_obj
2198          * due to a bug in a previous version of the code.
2199          * Only trust it if it has the right number of entries.
2200          */
2201         if (ds->ds_phys->ds_next_clones_obj != 0) {
2202                 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
2203                     &count));
2204         }
2205         if (count != ds->ds_phys->ds_num_children - 1) {
2206                 goto fail;
2207         }
2208         for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
2209             zap_cursor_retrieve(&zc, &za) == 0;
2210             zap_cursor_advance(&zc)) {
2211                 dsl_dataset_t *clone;
2212                 char buf[ZFS_MAXNAMELEN];
2213                 /*
2214                  * Even though we hold the dp_config_rwlock, the dataset
2215                  * may fail to open, returning ENOENT.  If there is a
2216                  * thread concurrently attempting to destroy this
2217                  * dataset, it will have the ds_rwlock held for
2218                  * RW_WRITER.  Our call to dsl_dataset_hold_obj() ->
2219                  * dsl_dataset_hold_ref() will fail its
2220                  * rw_tryenter(&ds->ds_rwlock, RW_READER), drop the
2221                  * dp_config_rwlock, and wait for the destroy progress
2222                  * and signal ds_exclusive_cv.  If the destroy was
2223                  * successful, we will see that
2224                  * DSL_DATASET_IS_DESTROYED(), and return ENOENT.
2225                  */
2226                 if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2227                     za.za_first_integer, FTAG, &clone) != 0)
2228                         continue;
2229                 dsl_dir_name(clone->ds_dir, buf);
2230                 VERIFY(nvlist_add_boolean(val, buf) == 0);
2231                 dsl_dataset_rele(clone, FTAG);
2232         }
2233         zap_cursor_fini(&zc);
2234         VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0);
2235         VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2236             propval) == 0);
2237 fail:
2238         nvlist_free(val);
2239         nvlist_free(propval);
2240         rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2241 }
2242 
2243 void
2244 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2245 {
2246         uint64_t refd, avail, uobjs, aobjs, ratio;
2247 
2248         ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2249             (ds->ds_phys->ds_uncompressed_bytes * 100 /
2250             ds->ds_phys->ds_compressed_bytes);
2251 
2252         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2253 
2254         if (dsl_dataset_is_snapshot(ds)) {
2255                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2256                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2257                     ds->ds_phys->ds_unique_bytes);
2258                 get_clones_stat(ds, nv);
2259         } else {
2260                 dsl_dir_stats(ds->ds_dir, nv);
2261         }
2262 
2263         dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2264         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2265         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2266 
2267         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2268             ds->ds_phys->ds_creation_time);
2269         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2270             ds->ds_phys->ds_creation_txg);
2271         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2272             ds->ds_quota);
2273         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2274             ds->ds_reserved);
2275         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2276             ds->ds_phys->ds_guid);
2277         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2278             ds->ds_phys->ds_unique_bytes);
2279         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2280             ds->ds_object);
2281         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2282             ds->ds_userrefs);
2283         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2284             DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2285 
2286         if (ds->ds_phys->ds_prev_snap_obj != 0) {
2287                 uint64_t written, comp, uncomp;
2288                 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2289                 dsl_dataset_t *prev;
2290 
2291                 rw_enter(&dp->dp_config_rwlock, RW_READER);
2292                 int err = dsl_dataset_hold_obj(dp,
2293                     ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
2294                 rw_exit(&dp->dp_config_rwlock);
2295                 if (err == 0) {
2296                         err = dsl_dataset_space_written(prev, ds, &written,
2297                             &comp, &uncomp);
2298                         dsl_dataset_rele(prev, FTAG);
2299                         if (err == 0) {
2300                                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2301                                     written);
2302                         }
2303                 }
2304         }
2305 }
2306 
2307 void
2308 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2309 {
2310         stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2311         stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2312         stat->dds_guid = ds->ds_phys->ds_guid;
2313         stat->dds_origin[0] = '\0';
2314         if (dsl_dataset_is_snapshot(ds)) {
2315                 stat->dds_is_snapshot = B_TRUE;
2316                 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2317         } else {
2318                 stat->dds_is_snapshot = B_FALSE;
2319                 stat->dds_num_clones = 0;
2320 
2321                 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2322                 if (dsl_dir_is_clone(ds->ds_dir)) {
2323                         dsl_dataset_t *ods;
2324 
2325                         VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2326                             ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2327                         dsl_dataset_name(ods, stat->dds_origin);
2328                         dsl_dataset_drop_ref(ods, FTAG);
2329                 }
2330                 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2331         }
2332 }
2333 
2334 uint64_t
2335 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2336 {
2337         return (ds->ds_fsid_guid);
2338 }
2339 
2340 void
2341 dsl_dataset_space(dsl_dataset_t *ds,
2342     uint64_t *refdbytesp, uint64_t *availbytesp,
2343     uint64_t *usedobjsp, uint64_t *availobjsp)
2344 {
2345         *refdbytesp = ds->ds_phys->ds_referenced_bytes;
2346         *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2347         if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2348                 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2349         if (ds->ds_quota != 0) {
2350                 /*
2351                  * Adjust available bytes according to refquota
2352                  */
2353                 if (*refdbytesp < ds->ds_quota)
2354                         *availbytesp = MIN(*availbytesp,
2355                             ds->ds_quota - *refdbytesp);
2356                 else
2357                         *availbytesp = 0;
2358         }
2359         *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2360         *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2361 }
2362 
2363 boolean_t
2364 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2365 {
2366         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2367 
2368         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2369             dsl_pool_sync_context(dp));
2370         if (ds->ds_prev == NULL)
2371                 return (B_FALSE);
2372         if (ds->ds_phys->ds_bp.blk_birth >
2373             ds->ds_prev->ds_phys->ds_creation_txg) {
2374                 objset_t *os, *os_prev;
2375                 /*
2376                  * It may be that only the ZIL differs, because it was
2377                  * reset in the head.  Don't count that as being
2378                  * modified.
2379                  */
2380                 if (dmu_objset_from_ds(ds, &os) != 0)
2381                         return (B_TRUE);
2382                 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2383                         return (B_TRUE);
2384                 return (bcmp(&os->os_phys->os_meta_dnode,
2385                     &os_prev->os_phys->os_meta_dnode,
2386                     sizeof (os->os_phys->os_meta_dnode)) != 0);
2387         }
2388         return (B_FALSE);
2389 }
2390 
2391 /* ARGSUSED */
2392 static int
2393 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2394 {
2395         dsl_dataset_t *ds = arg1;
2396         char *newsnapname = arg2;
2397         dsl_dir_t *dd = ds->ds_dir;
2398         dsl_dataset_t *hds;
2399         uint64_t val;
2400         int err;
2401 
2402         err = dsl_dataset_hold_obj(dd->dd_pool,
2403             dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2404         if (err)
2405                 return (err);
2406 
2407         /* new name better not be in use */
2408         err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2409         dsl_dataset_rele(hds, FTAG);
2410 
2411         if (err == 0)
2412                 err = EEXIST;
2413         else if (err == ENOENT)
2414                 err = 0;
2415 
2416         /* dataset name + 1 for the "@" + the new snapshot name must fit */
2417         if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2418                 err = ENAMETOOLONG;
2419 
2420         return (err);
2421 }
2422 
2423 static void
2424 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2425 {
2426         dsl_dataset_t *ds = arg1;
2427         const char *newsnapname = arg2;
2428         dsl_dir_t *dd = ds->ds_dir;
2429         objset_t *mos = dd->dd_pool->dp_meta_objset;
2430         dsl_dataset_t *hds;
2431         int err;
2432 
2433         ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2434 
2435         VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2436             dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2437 
2438         VERIFY(0 == dsl_dataset_get_snapname(ds));
2439         err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2440         ASSERT0(err);
2441         mutex_enter(&ds->ds_lock);
2442         (void) strcpy(ds->ds_snapname, newsnapname);
2443         mutex_exit(&ds->ds_lock);
2444         err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2445             ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2446         ASSERT0(err);
2447 
2448         spa_history_log_internal_ds(ds, "rename", tx,
2449             "-> @%s", newsnapname);
2450         dsl_dataset_rele(hds, FTAG);
2451 }
2452 
2453 struct renamesnaparg {
2454         dsl_sync_task_group_t *dstg;
2455         char failed[MAXPATHLEN];
2456         char *oldsnap;
2457         char *newsnap;
2458 };
2459 
2460 static int
2461 dsl_snapshot_rename_one(const char *name, void *arg)
2462 {
2463         struct renamesnaparg *ra = arg;
2464         dsl_dataset_t *ds = NULL;
2465         char *snapname;
2466         int err;
2467 
2468         snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2469         (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2470 
2471         /*
2472          * For recursive snapshot renames the parent won't be changing
2473          * so we just pass name for both the to/from argument.
2474          */
2475         err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2476         if (err != 0) {
2477                 strfree(snapname);
2478                 return (err == ENOENT ? 0 : err);
2479         }
2480 
2481 #ifdef _KERNEL
2482         /*
2483          * For all filesystems undergoing rename, we'll need to unmount it.
2484          */
2485         (void) zfs_unmount_snap(snapname, NULL);
2486 #endif
2487         err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2488         strfree(snapname);
2489         if (err != 0)
2490                 return (err == ENOENT ? 0 : err);
2491 
2492         dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2493             dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2494 
2495         return (0);
2496 }
2497 
2498 static int
2499 dsl_recursive_rename(char *oldname, const char *newname)
2500 {
2501         int err;
2502         struct renamesnaparg *ra;
2503         dsl_sync_task_t *dst;
2504         spa_t *spa;
2505         char *cp, *fsname = spa_strdup(oldname);
2506         int len = strlen(oldname) + 1;
2507 
2508         /* truncate the snapshot name to get the fsname */
2509         cp = strchr(fsname, '@');
2510         *cp = '\0';
2511 
2512         err = spa_open(fsname, &spa, FTAG);
2513         if (err) {
2514                 kmem_free(fsname, len);
2515                 return (err);
2516         }
2517         ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2518         ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2519 
2520         ra->oldsnap = strchr(oldname, '@') + 1;
2521         ra->newsnap = strchr(newname, '@') + 1;
2522         *ra->failed = '\0';
2523 
2524         err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2525             DS_FIND_CHILDREN);
2526         kmem_free(fsname, len);
2527 
2528         if (err == 0) {
2529                 err = dsl_sync_task_group_wait(ra->dstg);
2530         }
2531 
2532         for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2533             dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2534                 dsl_dataset_t *ds = dst->dst_arg1;
2535                 if (dst->dst_err) {
2536                         dsl_dir_name(ds->ds_dir, ra->failed);
2537                         (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2538                         (void) strlcat(ra->failed, ra->newsnap,
2539                             sizeof (ra->failed));
2540                 }
2541                 dsl_dataset_rele(ds, ra->dstg);
2542         }
2543 
2544         if (err)
2545                 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2546 
2547         dsl_sync_task_group_destroy(ra->dstg);
2548         kmem_free(ra, sizeof (struct renamesnaparg));
2549         spa_close(spa, FTAG);
2550         return (err);
2551 }
2552 
2553 static int
2554 dsl_valid_rename(const char *oldname, void *arg)
2555 {
2556         int delta = *(int *)arg;
2557 
2558         if (strlen(oldname) + delta >= MAXNAMELEN)
2559                 return (ENAMETOOLONG);
2560 
2561         return (0);
2562 }
2563 
2564 #pragma weak dmu_objset_rename = dsl_dataset_rename
2565 int
2566 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2567 {
2568         dsl_dir_t *dd;
2569         dsl_dataset_t *ds;
2570         const char *tail;
2571         int err;
2572 
2573         err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2574         if (err)
2575                 return (err);
2576 
2577         if (tail == NULL) {
2578                 int delta = strlen(newname) - strlen(oldname);
2579 
2580                 /* if we're growing, validate child name lengths */
2581                 if (delta > 0)
2582                         err = dmu_objset_find(oldname, dsl_valid_rename,
2583                             &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2584 
2585                 if (err == 0)
2586                         err = dsl_dir_rename(dd, newname);
2587                 dsl_dir_close(dd, FTAG);
2588                 return (err);
2589         }
2590 
2591         if (tail[0] != '@') {
2592                 /* the name ended in a nonexistent component */
2593                 dsl_dir_close(dd, FTAG);
2594                 return (ENOENT);
2595         }
2596 
2597         dsl_dir_close(dd, FTAG);
2598 
2599         /* new name must be snapshot in same filesystem */
2600         tail = strchr(newname, '@');
2601         if (tail == NULL)
2602                 return (EINVAL);
2603         tail++;
2604         if (strncmp(oldname, newname, tail - newname) != 0)
2605                 return (EXDEV);
2606 
2607         if (recursive) {
2608                 err = dsl_recursive_rename(oldname, newname);
2609         } else {
2610                 err = dsl_dataset_hold(oldname, FTAG, &ds);
2611                 if (err)
2612                         return (err);
2613 
2614                 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2615                     dsl_dataset_snapshot_rename_check,
2616                     dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2617 
2618                 dsl_dataset_rele(ds, FTAG);
2619         }
2620 
2621         return (err);
2622 }
2623 
2624 struct promotenode {
2625         list_node_t link;
2626         dsl_dataset_t *ds;
2627 };
2628 
2629 struct promotearg {
2630         list_t shared_snaps, origin_snaps, clone_snaps;
2631         dsl_dataset_t *origin_origin;
2632         uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2633         char *err_ds;
2634 };
2635 
2636 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2637 static boolean_t snaplist_unstable(list_t *l);
2638 
2639 static int
2640 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2641 {
2642         dsl_dataset_t *hds = arg1;
2643         struct promotearg *pa = arg2;
2644         struct promotenode *snap = list_head(&pa->shared_snaps);
2645         dsl_dataset_t *origin_ds = snap->ds;
2646         int err;
2647         uint64_t unused;
2648 
2649         /* Check that it is a real clone */
2650         if (!dsl_dir_is_clone(hds->ds_dir))
2651                 return (EINVAL);
2652 
2653         /* Since this is so expensive, don't do the preliminary check */
2654         if (!dmu_tx_is_syncing(tx))
2655                 return (0);
2656 
2657         if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2658                 return (EXDEV);
2659 
2660         /* compute origin's new unique space */
2661         snap = list_tail(&pa->clone_snaps);
2662         ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2663         dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2664             origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2665             &pa->unique, &unused, &unused);
2666 
2667         /*
2668          * Walk the snapshots that we are moving
2669          *
2670          * Compute space to transfer.  Consider the incremental changes
2671          * to used for each snapshot:
2672          * (my used) = (prev's used) + (blocks born) - (blocks killed)
2673          * So each snapshot gave birth to:
2674          * (blocks born) = (my used) - (prev's used) + (blocks killed)
2675          * So a sequence would look like:
2676          * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2677          * Which simplifies to:
2678          * uN + kN + kN-1 + ... + k1 + k0
2679          * Note however, if we stop before we reach the ORIGIN we get:
2680          * uN + kN + kN-1 + ... + kM - uM-1
2681          */
2682         pa->used = origin_ds->ds_phys->ds_referenced_bytes;
2683         pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2684         pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2685         for (snap = list_head(&pa->shared_snaps); snap;
2686             snap = list_next(&pa->shared_snaps, snap)) {
2687                 uint64_t val, dlused, dlcomp, dluncomp;
2688                 dsl_dataset_t *ds = snap->ds;
2689 
2690                 /* Check that the snapshot name does not conflict */
2691                 VERIFY(0 == dsl_dataset_get_snapname(ds));
2692                 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2693                 if (err == 0) {
2694                         err = EEXIST;
2695                         goto out;
2696                 }
2697                 if (err != ENOENT)
2698                         goto out;
2699 
2700                 /* The very first snapshot does not have a deadlist */
2701                 if (ds->ds_phys->ds_prev_snap_obj == 0)
2702                         continue;
2703 
2704                 dsl_deadlist_space(&ds->ds_deadlist,
2705                     &dlused, &dlcomp, &dluncomp);
2706                 pa->used += dlused;
2707                 pa->comp += dlcomp;
2708                 pa->uncomp += dluncomp;
2709         }
2710 
2711         /*
2712          * If we are a clone of a clone then we never reached ORIGIN,
2713          * so we need to subtract out the clone origin's used space.
2714          */
2715         if (pa->origin_origin) {
2716                 pa->used -= pa->origin_origin->ds_phys->ds_referenced_bytes;
2717                 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2718                 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2719         }
2720 
2721         /* Check that there is enough space here */
2722         err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2723             pa->used);
2724         if (err)
2725                 return (err);
2726 
2727         /*
2728          * Compute the amounts of space that will be used by snapshots
2729          * after the promotion (for both origin and clone).  For each,
2730          * it is the amount of space that will be on all of their
2731          * deadlists (that was not born before their new origin).
2732          */
2733         if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2734                 uint64_t space;
2735 
2736                 /*
2737                  * Note, typically this will not be a clone of a clone,
2738                  * so dd_origin_txg will be < TXG_INITIAL, so
2739                  * these snaplist_space() -> dsl_deadlist_space_range()
2740                  * calls will be fast because they do not have to
2741                  * iterate over all bps.
2742                  */
2743                 snap = list_head(&pa->origin_snaps);
2744                 err = snaplist_space(&pa->shared_snaps,
2745                     snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2746                 if (err)
2747                         return (err);
2748 
2749                 err = snaplist_space(&pa->clone_snaps,
2750                     snap->ds->ds_dir->dd_origin_txg, &space);
2751                 if (err)
2752                         return (err);
2753                 pa->cloneusedsnap += space;
2754         }
2755         if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2756                 err = snaplist_space(&pa->origin_snaps,
2757                     origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2758                 if (err)
2759                         return (err);
2760         }
2761 
2762         return (0);
2763 out:
2764         pa->err_ds =  snap->ds->ds_snapname;
2765         return (err);
2766 }
2767 
2768 static void
2769 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2770 {
2771         dsl_dataset_t *hds = arg1;
2772         struct promotearg *pa = arg2;
2773         struct promotenode *snap = list_head(&pa->shared_snaps);
2774         dsl_dataset_t *origin_ds = snap->ds;
2775         dsl_dataset_t *origin_head;
2776         dsl_dir_t *dd = hds->ds_dir;
2777         dsl_pool_t *dp = hds->ds_dir->dd_pool;
2778         dsl_dir_t *odd = NULL;
2779         uint64_t oldnext_obj;
2780         int64_t delta;
2781 
2782         ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2783 
2784         snap = list_head(&pa->origin_snaps);
2785         origin_head = snap->ds;
2786 
2787         /*
2788          * We need to explicitly open odd, since origin_ds's dd will be
2789          * changing.
2790          */
2791         VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2792             NULL, FTAG, &odd));
2793 
2794         /* change origin's next snap */
2795         dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2796         oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2797         snap = list_tail(&pa->clone_snaps);
2798         ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2799         origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2800 
2801         /* change the origin's next clone */
2802         if (origin_ds->ds_phys->ds_next_clones_obj) {
2803                 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2804                 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2805                     origin_ds->ds_phys->ds_next_clones_obj,
2806                     oldnext_obj, tx));
2807         }
2808 
2809         /* change origin */
2810         dmu_buf_will_dirty(dd->dd_dbuf, tx);
2811         ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2812         dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2813         dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2814         dmu_buf_will_dirty(odd->dd_dbuf, tx);
2815         odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2816         origin_head->ds_dir->dd_origin_txg =
2817             origin_ds->ds_phys->ds_creation_txg;
2818 
2819         /* change dd_clone entries */
2820         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2821                 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2822                     odd->dd_phys->dd_clones, hds->ds_object, tx));
2823                 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2824                     pa->origin_origin->ds_dir->dd_phys->dd_clones,
2825                     hds->ds_object, tx));
2826 
2827                 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2828                     pa->origin_origin->ds_dir->dd_phys->dd_clones,
2829                     origin_head->ds_object, tx));
2830                 if (dd->dd_phys->dd_clones == 0) {
2831                         dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2832                             DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2833                 }
2834                 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2835                     dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2836 
2837         }
2838 
2839         /* move snapshots to this dir */
2840         for (snap = list_head(&pa->shared_snaps); snap;
2841             snap = list_next(&pa->shared_snaps, snap)) {
2842                 dsl_dataset_t *ds = snap->ds;
2843 
2844                 /* unregister props as dsl_dir is changing */
2845                 if (ds->ds_objset) {
2846                         dmu_objset_evict(ds->ds_objset);
2847                         ds->ds_objset = NULL;
2848                 }
2849                 /* move snap name entry */
2850                 VERIFY(0 == dsl_dataset_get_snapname(ds));
2851                 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2852                     ds->ds_snapname, tx));
2853                 VERIFY(0 == zap_add(dp->dp_meta_objset,
2854                     hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2855                     8, 1, &ds->ds_object, tx));
2856 
2857                 /* change containing dsl_dir */
2858                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2859                 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2860                 ds->ds_phys->ds_dir_obj = dd->dd_object;
2861                 ASSERT3P(ds->ds_dir, ==, odd);
2862                 dsl_dir_close(ds->ds_dir, ds);
2863                 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2864                     NULL, ds, &ds->ds_dir));
2865 
2866                 /* move any clone references */
2867                 if (ds->ds_phys->ds_next_clones_obj &&
2868                     spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2869                         zap_cursor_t zc;
2870                         zap_attribute_t za;
2871 
2872                         for (zap_cursor_init(&zc, dp->dp_meta_objset,
2873                             ds->ds_phys->ds_next_clones_obj);
2874                             zap_cursor_retrieve(&zc, &za) == 0;
2875                             zap_cursor_advance(&zc)) {
2876                                 dsl_dataset_t *cnds;
2877                                 uint64_t o;
2878 
2879                                 if (za.za_first_integer == oldnext_obj) {
2880                                         /*
2881                                          * We've already moved the
2882                                          * origin's reference.
2883                                          */
2884                                         continue;
2885                                 }
2886 
2887                                 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2888                                     za.za_first_integer, FTAG, &cnds));
2889                                 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2890 
2891                                 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2892                                     odd->dd_phys->dd_clones, o, tx), ==, 0);
2893                                 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2894                                     dd->dd_phys->dd_clones, o, tx), ==, 0);
2895                                 dsl_dataset_rele(cnds, FTAG);
2896                         }
2897                         zap_cursor_fini(&zc);
2898                 }
2899 
2900                 ASSERT0(dsl_prop_numcb(ds));
2901         }
2902 
2903         /*
2904          * Change space accounting.
2905          * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2906          * both be valid, or both be 0 (resulting in delta == 0).  This
2907          * is true for each of {clone,origin} independently.
2908          */
2909 
2910         delta = pa->cloneusedsnap -
2911             dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2912         ASSERT3S(delta, >=, 0);
2913         ASSERT3U(pa->used, >=, delta);
2914         dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2915         dsl_dir_diduse_space(dd, DD_USED_HEAD,
2916             pa->used - delta, pa->comp, pa->uncomp, tx);
2917 
2918         delta = pa->originusedsnap -
2919             odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2920         ASSERT3S(delta, <=, 0);
2921         ASSERT3U(pa->used, >=, -delta);
2922         dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2923         dsl_dir_diduse_space(odd, DD_USED_HEAD,
2924             -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2925 
2926         origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2927 
2928         /* log history record */
2929         spa_history_log_internal_ds(hds, "promote", tx, "");
2930 
2931         dsl_dir_close(odd, FTAG);
2932 }
2933 
2934 static char *snaplist_tag = "snaplist";
2935 /*
2936  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2937  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2938  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2939  * snapshots back to this dataset's origin.
2940  */
2941 static int
2942 snaplist_make(dsl_pool_t *dp, boolean_t own,
2943     uint64_t first_obj, uint64_t last_obj, list_t *l)
2944 {
2945         uint64_t obj = last_obj;
2946 
2947         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2948 
2949         list_create(l, sizeof (struct promotenode),
2950             offsetof(struct promotenode, link));
2951 
2952         while (obj != first_obj) {
2953                 dsl_dataset_t *ds;
2954                 struct promotenode *snap;
2955                 int err;
2956 
2957                 if (own) {
2958                         err = dsl_dataset_own_obj(dp, obj,
2959                             0, snaplist_tag, &ds);
2960                         if (err == 0)
2961                                 dsl_dataset_make_exclusive(ds, snaplist_tag);
2962                 } else {
2963                         err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2964                 }
2965                 if (err == ENOENT) {
2966                         /* lost race with snapshot destroy */
2967                         struct promotenode *last = list_tail(l);
2968                         ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2969                         obj = last->ds->ds_phys->ds_prev_snap_obj;
2970                         continue;
2971                 } else if (err) {
2972                         return (err);
2973                 }
2974 
2975                 if (first_obj == 0)
2976                         first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2977 
2978                 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2979                 snap->ds = ds;
2980                 list_insert_tail(l, snap);
2981                 obj = ds->ds_phys->ds_prev_snap_obj;
2982         }
2983 
2984         return (0);
2985 }
2986 
2987 static int
2988 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2989 {
2990         struct promotenode *snap;
2991 
2992         *spacep = 0;
2993         for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2994                 uint64_t used, comp, uncomp;
2995                 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2996                     mintxg, UINT64_MAX, &used, &comp, &uncomp);
2997                 *spacep += used;
2998         }
2999         return (0);
3000 }
3001 
3002 static void
3003 snaplist_destroy(list_t *l, boolean_t own)
3004 {
3005         struct promotenode *snap;
3006 
3007         if (!l || !list_link_active(&l->list_head))
3008                 return;
3009 
3010         while ((snap = list_tail(l)) != NULL) {
3011                 list_remove(l, snap);
3012                 if (own)
3013                         dsl_dataset_disown(snap->ds, snaplist_tag);
3014                 else
3015                         dsl_dataset_rele(snap->ds, snaplist_tag);
3016                 kmem_free(snap, sizeof (struct promotenode));
3017         }
3018         list_destroy(l);
3019 }
3020 
3021 /*
3022  * Promote a clone.  Nomenclature note:
3023  * "clone" or "cds": the original clone which is being promoted
3024  * "origin" or "ods": the snapshot which is originally clone's origin
3025  * "origin head" or "ohds": the dataset which is the head
3026  * (filesystem/volume) for the origin
3027  * "origin origin": the origin of the origin's filesystem (typically
3028  * NULL, indicating that the clone is not a clone of a clone).
3029  */
3030 int
3031 dsl_dataset_promote(const char *name, char *conflsnap)
3032 {
3033         dsl_dataset_t *ds;
3034         dsl_dir_t *dd;
3035         dsl_pool_t *dp;
3036         dmu_object_info_t doi;
3037         struct promotearg pa = { 0 };
3038         struct promotenode *snap;
3039         int err;
3040 
3041         err = dsl_dataset_hold(name, FTAG, &ds);
3042         if (err)
3043                 return (err);
3044         dd = ds->ds_dir;
3045         dp = dd->dd_pool;
3046 
3047         err = dmu_object_info(dp->dp_meta_objset,
3048             ds->ds_phys->ds_snapnames_zapobj, &doi);
3049         if (err) {
3050                 dsl_dataset_rele(ds, FTAG);
3051                 return (err);
3052         }
3053 
3054         if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
3055                 dsl_dataset_rele(ds, FTAG);
3056                 return (EINVAL);
3057         }
3058 
3059         /*
3060          * We are going to inherit all the snapshots taken before our
3061          * origin (i.e., our new origin will be our parent's origin).
3062          * Take ownership of them so that we can rename them into our
3063          * namespace.
3064          */
3065         rw_enter(&dp->dp_config_rwlock, RW_READER);
3066 
3067         err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
3068             &pa.shared_snaps);
3069         if (err != 0)
3070                 goto out;
3071 
3072         err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
3073         if (err != 0)
3074                 goto out;
3075 
3076         snap = list_head(&pa.shared_snaps);
3077         ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
3078         err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
3079             snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
3080         if (err != 0)
3081                 goto out;
3082 
3083         if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
3084                 err = dsl_dataset_hold_obj(dp,
3085                     snap->ds->ds_dir->dd_phys->dd_origin_obj,
3086                     FTAG, &pa.origin_origin);
3087                 if (err != 0)
3088                         goto out;
3089         }
3090 
3091 out:
3092         rw_exit(&dp->dp_config_rwlock);
3093 
3094         /*
3095          * Add in 128x the snapnames zapobj size, since we will be moving
3096          * a bunch of snapnames to the promoted ds, and dirtying their
3097          * bonus buffers.
3098          */
3099         if (err == 0) {
3100                 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
3101                     dsl_dataset_promote_sync, ds, &pa,
3102                     2 + 2 * doi.doi_physical_blocks_512);
3103                 if (err && pa.err_ds && conflsnap)
3104                         (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3105         }
3106 
3107         snaplist_destroy(&pa.shared_snaps, B_TRUE);
3108         snaplist_destroy(&pa.clone_snaps, B_FALSE);
3109         snaplist_destroy(&pa.origin_snaps, B_FALSE);
3110         if (pa.origin_origin)
3111                 dsl_dataset_rele(pa.origin_origin, FTAG);
3112         dsl_dataset_rele(ds, FTAG);
3113         return (err);
3114 }
3115 
3116 struct cloneswaparg {
3117         dsl_dataset_t *cds; /* clone dataset */
3118         dsl_dataset_t *ohds; /* origin's head dataset */
3119         boolean_t force;
3120         int64_t unused_refres_delta; /* change in unconsumed refreservation */
3121 };
3122 
3123 /* ARGSUSED */
3124 static int
3125 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3126 {
3127         struct cloneswaparg *csa = arg1;
3128 
3129         /* they should both be heads */
3130         if (dsl_dataset_is_snapshot(csa->cds) ||
3131             dsl_dataset_is_snapshot(csa->ohds))
3132                 return (EINVAL);
3133 
3134         /* the branch point should be just before them */
3135         if (csa->cds->ds_prev != csa->ohds->ds_prev)
3136                 return (EINVAL);
3137 
3138         /* cds should be the clone (unless they are unrelated) */
3139         if (csa->cds->ds_prev != NULL &&
3140             csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3141             csa->ohds->ds_object !=
3142             csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3143                 return (EINVAL);
3144 
3145         /* the clone should be a child of the origin */
3146         if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3147                 return (EINVAL);
3148 
3149         /* ohds shouldn't be modified unless 'force' */
3150         if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3151                 return (ETXTBSY);
3152 
3153         /* adjust amount of any unconsumed refreservation */
3154         csa->unused_refres_delta =
3155             (int64_t)MIN(csa->ohds->ds_reserved,
3156             csa->ohds->ds_phys->ds_unique_bytes) -
3157             (int64_t)MIN(csa->ohds->ds_reserved,
3158             csa->cds->ds_phys->ds_unique_bytes);
3159 
3160         if (csa->unused_refres_delta > 0 &&
3161             csa->unused_refres_delta >
3162             dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3163                 return (ENOSPC);
3164 
3165         if (csa->ohds->ds_quota != 0 &&
3166             csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3167                 return (EDQUOT);
3168 
3169         return (0);
3170 }
3171 
3172 /* ARGSUSED */
3173 static void
3174 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3175 {
3176         struct cloneswaparg *csa = arg1;
3177         dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3178 
3179         ASSERT(csa->cds->ds_reserved == 0);
3180         ASSERT(csa->ohds->ds_quota == 0 ||
3181             csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3182 
3183         dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3184         dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3185 
3186         if (csa->cds->ds_objset != NULL) {
3187                 dmu_objset_evict(csa->cds->ds_objset);
3188                 csa->cds->ds_objset = NULL;
3189         }
3190 
3191         if (csa->ohds->ds_objset != NULL) {
3192                 dmu_objset_evict(csa->ohds->ds_objset);
3193                 csa->ohds->ds_objset = NULL;
3194         }
3195 
3196         /*
3197          * Reset origin's unique bytes, if it exists.
3198          */
3199         if (csa->cds->ds_prev) {
3200                 dsl_dataset_t *origin = csa->cds->ds_prev;
3201                 uint64_t comp, uncomp;
3202 
3203                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3204                 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3205                     origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3206                     &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3207         }
3208 
3209         /* swap blkptrs */
3210         {
3211                 blkptr_t tmp;
3212                 tmp = csa->ohds->ds_phys->ds_bp;
3213                 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3214                 csa->cds->ds_phys->ds_bp = tmp;
3215         }
3216 
3217         /* set dd_*_bytes */
3218         {
3219                 int64_t dused, dcomp, duncomp;
3220                 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3221                 uint64_t odl_used, odl_comp, odl_uncomp;
3222 
3223                 ASSERT3U(csa->cds->ds_dir->dd_phys->
3224                     dd_used_breakdown[DD_USED_SNAP], ==, 0);
3225 
3226                 dsl_deadlist_space(&csa->cds->ds_deadlist,
3227                     &cdl_used, &cdl_comp, &cdl_uncomp);
3228                 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3229                     &odl_used, &odl_comp, &odl_uncomp);
3230 
3231                 dused = csa->cds->ds_phys->ds_referenced_bytes + cdl_used -
3232                     (csa->ohds->ds_phys->ds_referenced_bytes + odl_used);
3233                 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3234                     (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3235                 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3236                     cdl_uncomp -
3237                     (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3238 
3239                 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3240                     dused, dcomp, duncomp, tx);
3241                 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3242                     -dused, -dcomp, -duncomp, tx);
3243 
3244                 /*
3245                  * The difference in the space used by snapshots is the
3246                  * difference in snapshot space due to the head's
3247                  * deadlist (since that's the only thing that's
3248                  * changing that affects the snapused).
3249                  */
3250                 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3251                     csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3252                     &cdl_used, &cdl_comp, &cdl_uncomp);
3253                 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3254                     csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3255                     &odl_used, &odl_comp, &odl_uncomp);
3256                 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3257                     DD_USED_HEAD, DD_USED_SNAP, tx);
3258         }
3259 
3260         /* swap ds_*_bytes */
3261         SWITCH64(csa->ohds->ds_phys->ds_referenced_bytes,
3262             csa->cds->ds_phys->ds_referenced_bytes);
3263         SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3264             csa->cds->ds_phys->ds_compressed_bytes);
3265         SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3266             csa->cds->ds_phys->ds_uncompressed_bytes);
3267         SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3268             csa->cds->ds_phys->ds_unique_bytes);
3269 
3270         /* apply any parent delta for change in unconsumed refreservation */
3271         dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3272             csa->unused_refres_delta, 0, 0, tx);
3273 
3274         /*
3275          * Swap deadlists.
3276          */
3277         dsl_deadlist_close(&csa->cds->ds_deadlist);
3278         dsl_deadlist_close(&csa->ohds->ds_deadlist);
3279         SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3280             csa->cds->ds_phys->ds_deadlist_obj);
3281         dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3282             csa->cds->ds_phys->ds_deadlist_obj);
3283         dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3284             csa->ohds->ds_phys->ds_deadlist_obj);
3285 
3286         dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3287 
3288         spa_history_log_internal_ds(csa->cds, "clone swap", tx,
3289             "parent=%s", csa->ohds->ds_dir->dd_myname);
3290 }
3291 
3292 /*
3293  * Swap 'clone' with its origin head datasets.  Used at the end of "zfs
3294  * recv" into an existing fs to swizzle the file system to the new
3295  * version, and by "zfs rollback".  Can also be used to swap two
3296  * independent head datasets if neither has any snapshots.
3297  */
3298 int
3299 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3300     boolean_t force)
3301 {
3302         struct cloneswaparg csa;
3303         int error;
3304 
3305         ASSERT(clone->ds_owner);
3306         ASSERT(origin_head->ds_owner);
3307 retry:
3308         /*
3309          * Need exclusive access for the swap. If we're swapping these
3310          * datasets back after an error, we already hold the locks.
3311          */
3312         if (!RW_WRITE_HELD(&clone->ds_rwlock))
3313                 rw_enter(&clone->ds_rwlock, RW_WRITER);
3314         if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3315             !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3316                 rw_exit(&clone->ds_rwlock);
3317                 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3318                 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3319                         rw_exit(&origin_head->ds_rwlock);
3320                         goto retry;
3321                 }
3322         }
3323         csa.cds = clone;
3324         csa.ohds = origin_head;
3325         csa.force = force;
3326         error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3327             dsl_dataset_clone_swap_check,
3328             dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3329         return (error);
3330 }
3331 
3332 /*
3333  * Given a pool name and a dataset object number in that pool,
3334  * return the name of that dataset.
3335  */
3336 int
3337 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3338 {
3339         spa_t *spa;
3340         dsl_pool_t *dp;
3341         dsl_dataset_t *ds;
3342         int error;
3343 
3344         if ((error = spa_open(pname, &spa, FTAG)) != 0)
3345                 return (error);
3346         dp = spa_get_dsl(spa);
3347         rw_enter(&dp->dp_config_rwlock, RW_READER);
3348         if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3349                 dsl_dataset_name(ds, buf);
3350                 dsl_dataset_rele(ds, FTAG);
3351         }
3352         rw_exit(&dp->dp_config_rwlock);
3353         spa_close(spa, FTAG);
3354 
3355         return (error);
3356 }
3357 
3358 int
3359 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3360     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3361 {
3362         int error = 0;
3363 
3364         ASSERT3S(asize, >, 0);
3365 
3366         /*
3367          * *ref_rsrv is the portion of asize that will come from any
3368          * unconsumed refreservation space.
3369          */
3370         *ref_rsrv = 0;
3371 
3372         mutex_enter(&ds->ds_lock);
3373         /*
3374          * Make a space adjustment for reserved bytes.
3375          */
3376         if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3377                 ASSERT3U(*used, >=,
3378                     ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3379                 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3380                 *ref_rsrv =
3381                     asize - MIN(asize, parent_delta(ds, asize + inflight));
3382         }
3383 
3384         if (!check_quota || ds->ds_quota == 0) {
3385                 mutex_exit(&ds->ds_lock);
3386                 return (0);
3387         }
3388         /*
3389          * If they are requesting more space, and our current estimate
3390          * is over quota, they get to try again unless the actual
3391          * on-disk is over quota and there are no pending changes (which
3392          * may free up space for us).
3393          */
3394         if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
3395                 if (inflight > 0 ||
3396                     ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
3397                         error = ERESTART;
3398                 else
3399                         error = EDQUOT;
3400         }
3401         mutex_exit(&ds->ds_lock);
3402 
3403         return (error);
3404 }
3405 
3406 /* ARGSUSED */
3407 static int
3408 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3409 {
3410         dsl_dataset_t *ds = arg1;
3411         dsl_prop_setarg_t *psa = arg2;
3412         int err;
3413 
3414         if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3415                 return (ENOTSUP);
3416 
3417         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3418                 return (err);
3419 
3420         if (psa->psa_effective_value == 0)
3421                 return (0);
3422 
3423         if (psa->psa_effective_value < ds->ds_phys->ds_referenced_bytes ||
3424             psa->psa_effective_value < ds->ds_reserved)
3425                 return (ENOSPC);
3426 
3427         return (0);
3428 }
3429 
3430 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3431 
3432 void
3433 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3434 {
3435         dsl_dataset_t *ds = arg1;
3436         dsl_prop_setarg_t *psa = arg2;
3437         uint64_t effective_value = psa->psa_effective_value;
3438 
3439         dsl_prop_set_sync(ds, psa, tx);
3440         DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3441 
3442         if (ds->ds_quota != effective_value) {
3443                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3444                 ds->ds_quota = effective_value;
3445         }
3446 }
3447 
3448 int
3449 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3450 {
3451         dsl_dataset_t *ds;
3452         dsl_prop_setarg_t psa;
3453         int err;
3454 
3455         dsl_prop_setarg_init_uint64(&psa, "refquota", source, &quota);
3456 
3457         err = dsl_dataset_hold(dsname, FTAG, &ds);
3458         if (err)
3459                 return (err);
3460 
3461         /*
3462          * If someone removes a file, then tries to set the quota, we
3463          * want to make sure the file freeing takes effect.
3464          */
3465         txg_wait_open(ds->ds_dir->dd_pool, 0);
3466 
3467         err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3468             dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3469             ds, &psa, 0);
3470 
3471         dsl_dataset_rele(ds, FTAG);
3472         return (err);
3473 }
3474 
3475 static int
3476 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3477 {
3478         dsl_dataset_t *ds = arg1;
3479         dsl_prop_setarg_t *psa = arg2;
3480         uint64_t effective_value;
3481         uint64_t unique;
3482         int err;
3483 
3484         if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3485             SPA_VERSION_REFRESERVATION)
3486                 return (ENOTSUP);
3487 
3488         if (dsl_dataset_is_snapshot(ds))
3489                 return (EINVAL);
3490 
3491         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3492                 return (err);
3493 
3494         effective_value = psa->psa_effective_value;
3495 
3496         /*
3497          * If we are doing the preliminary check in open context, the
3498          * space estimates may be inaccurate.
3499          */
3500         if (!dmu_tx_is_syncing(tx))
3501                 return (0);
3502 
3503         mutex_enter(&ds->ds_lock);
3504         if (!DS_UNIQUE_IS_ACCURATE(ds))
3505                 dsl_dataset_recalc_head_uniq(ds);
3506         unique = ds->ds_phys->ds_unique_bytes;
3507         mutex_exit(&ds->ds_lock);
3508 
3509         if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3510                 uint64_t delta = MAX(unique, effective_value) -
3511                     MAX(unique, ds->ds_reserved);
3512 
3513                 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3514                         return (ENOSPC);
3515                 if (ds->ds_quota > 0 &&
3516                     effective_value > ds->ds_quota)
3517                         return (ENOSPC);
3518         }
3519 
3520         return (0);
3521 }
3522 
3523 static void
3524 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3525 {
3526         dsl_dataset_t *ds = arg1;
3527         dsl_prop_setarg_t *psa = arg2;
3528         uint64_t effective_value = psa->psa_effective_value;
3529         uint64_t unique;
3530         int64_t delta;
3531 
3532         dsl_prop_set_sync(ds, psa, tx);
3533         DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3534 
3535         dmu_buf_will_dirty(ds->ds_dbuf, tx);
3536 
3537         mutex_enter(&ds->ds_dir->dd_lock);
3538         mutex_enter(&ds->ds_lock);
3539         ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3540         unique = ds->ds_phys->ds_unique_bytes;
3541         delta = MAX(0, (int64_t)(effective_value - unique)) -
3542             MAX(0, (int64_t)(ds->ds_reserved - unique));
3543         ds->ds_reserved = effective_value;
3544         mutex_exit(&ds->ds_lock);
3545 
3546         dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3547         mutex_exit(&ds->ds_dir->dd_lock);
3548 }
3549 
3550 int
3551 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3552     uint64_t reservation)
3553 {
3554         dsl_dataset_t *ds;
3555         dsl_prop_setarg_t psa;
3556         int err;
3557 
3558         dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3559             &reservation);
3560 
3561         err = dsl_dataset_hold(dsname, FTAG, &ds);
3562         if (err)
3563                 return (err);
3564 
3565         err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3566             dsl_dataset_set_reservation_check,
3567             dsl_dataset_set_reservation_sync, ds, &psa, 0);
3568 
3569         dsl_dataset_rele(ds, FTAG);
3570         return (err);
3571 }
3572 
3573 typedef struct zfs_hold_cleanup_arg {
3574         dsl_pool_t *dp;
3575         uint64_t dsobj;
3576         char htag[MAXNAMELEN];
3577 } zfs_hold_cleanup_arg_t;
3578 
3579 static void
3580 dsl_dataset_user_release_onexit(void *arg)
3581 {
3582         zfs_hold_cleanup_arg_t *ca = arg;
3583 
3584         (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3585             B_TRUE);
3586         kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3587 }
3588 
3589 void
3590 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3591     minor_t minor)
3592 {
3593         zfs_hold_cleanup_arg_t *ca;
3594 
3595         ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3596         ca->dp = ds->ds_dir->dd_pool;
3597         ca->dsobj = ds->ds_object;
3598         (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3599         VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3600             dsl_dataset_user_release_onexit, ca, NULL));
3601 }
3602 
3603 /*
3604  * If you add new checks here, you may need to add
3605  * additional checks to the "temporary" case in
3606  * snapshot_check() in dmu_objset.c.
3607  */
3608 static int
3609 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3610 {
3611         dsl_dataset_t *ds = arg1;
3612         struct dsl_ds_holdarg *ha = arg2;
3613         const char *htag = ha->htag;
3614         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3615         int error = 0;
3616 
3617         if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3618                 return (ENOTSUP);
3619 
3620         if (!dsl_dataset_is_snapshot(ds))
3621                 return (EINVAL);
3622 
3623         /* tags must be unique */
3624         mutex_enter(&ds->ds_lock);
3625         if (ds->ds_phys->ds_userrefs_obj) {
3626                 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3627                     8, 1, tx);
3628                 if (error == 0)
3629                         error = EEXIST;
3630                 else if (error == ENOENT)
3631                         error = 0;
3632         }
3633         mutex_exit(&ds->ds_lock);
3634 
3635         if (error == 0 && ha->temphold &&
3636             strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3637                 error = E2BIG;
3638 
3639         return (error);
3640 }
3641 
3642 void
3643 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3644 {
3645         dsl_dataset_t *ds = arg1;
3646         struct dsl_ds_holdarg *ha = arg2;
3647         const char *htag = ha->htag;
3648         dsl_pool_t *dp = ds->ds_dir->dd_pool;
3649         objset_t *mos = dp->dp_meta_objset;
3650         uint64_t now = gethrestime_sec();
3651         uint64_t zapobj;
3652 
3653         mutex_enter(&ds->ds_lock);
3654         if (ds->ds_phys->ds_userrefs_obj == 0) {
3655                 /*
3656                  * This is the first user hold for this dataset.  Create
3657                  * the userrefs zap object.
3658                  */
3659                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3660                 zapobj = ds->ds_phys->ds_userrefs_obj =
3661                     zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3662         } else {
3663                 zapobj = ds->ds_phys->ds_userrefs_obj;
3664         }
3665         ds->ds_userrefs++;
3666         mutex_exit(&ds->ds_lock);
3667 
3668         VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3669 
3670         if (ha->temphold) {
3671                 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3672                     htag, &now, tx));
3673         }
3674 
3675         spa_history_log_internal_ds(ds, "hold", tx,
3676             "tag = %s temp = %d holds now = %llu",
3677             htag, (int)ha->temphold, ds->ds_userrefs);
3678 }
3679 
3680 static int
3681 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3682 {
3683         struct dsl_ds_holdarg *ha = arg;
3684         dsl_dataset_t *ds;
3685         int error;
3686         char *name;
3687 
3688         /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3689         name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3690         error = dsl_dataset_hold(name, ha->dstg, &ds);
3691         strfree(name);
3692         if (error == 0) {
3693                 ha->gotone = B_TRUE;
3694                 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3695                     dsl_dataset_user_hold_sync, ds, ha, 0);
3696         } else if (error == ENOENT && ha->recursive) {
3697                 error = 0;
3698         } else {
3699                 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3700         }
3701         return (error);
3702 }
3703 
3704 int
3705 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3706     boolean_t temphold)
3707 {
3708         struct dsl_ds_holdarg *ha;
3709         int error;
3710 
3711         ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3712         ha->htag = htag;
3713         ha->temphold = temphold;
3714         error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3715             dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3716             ds, ha, 0);
3717         kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3718 
3719         return (error);
3720 }
3721 
3722 int
3723 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3724     boolean_t recursive, boolean_t temphold, int cleanup_fd)
3725 {
3726         struct dsl_ds_holdarg *ha;
3727         dsl_sync_task_t *dst;
3728         spa_t *spa;
3729         int error;
3730         minor_t minor = 0;
3731 
3732         if (cleanup_fd != -1) {
3733                 /* Currently we only support cleanup-on-exit of tempholds. */
3734                 if (!temphold)
3735                         return (EINVAL);
3736                 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3737                 if (error)
3738                         return (error);
3739         }
3740 
3741         ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3742 
3743         (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3744 
3745         error = spa_open(dsname, &spa, FTAG);
3746         if (error) {
3747                 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3748                 if (cleanup_fd != -1)
3749                         zfs_onexit_fd_rele(cleanup_fd);
3750                 return (error);
3751         }
3752 
3753         ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3754         ha->htag = htag;
3755         ha->snapname = snapname;
3756         ha->recursive = recursive;
3757         ha->temphold = temphold;
3758 
3759         if (recursive) {
3760                 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3761                     ha, DS_FIND_CHILDREN);
3762         } else {
3763                 error = dsl_dataset_user_hold_one(dsname, ha);
3764         }
3765         if (error == 0)
3766                 error = dsl_sync_task_group_wait(ha->dstg);
3767 
3768         for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3769             dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3770                 dsl_dataset_t *ds = dst->dst_arg1;
3771 
3772                 if (dst->dst_err) {
3773                         dsl_dataset_name(ds, ha->failed);
3774                         *strchr(ha->failed, '@') = '\0';
3775                 } else if (error == 0 && minor != 0 && temphold) {
3776                         /*
3777                          * If this hold is to be released upon process exit,
3778                          * register that action now.
3779                          */
3780                         dsl_register_onexit_hold_cleanup(ds, htag, minor);
3781                 }
3782                 dsl_dataset_rele(ds, ha->dstg);
3783         }
3784 
3785         if (error == 0 && recursive && !ha->gotone)
3786                 error = ENOENT;
3787 
3788         if (error)
3789                 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3790 
3791         dsl_sync_task_group_destroy(ha->dstg);
3792 
3793         kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3794         spa_close(spa, FTAG);
3795         if (cleanup_fd != -1)
3796                 zfs_onexit_fd_rele(cleanup_fd);
3797         return (error);
3798 }
3799 
3800 struct dsl_ds_releasearg {
3801         dsl_dataset_t *ds;
3802         const char *htag;
3803         boolean_t own;          /* do we own or just hold ds? */
3804 };
3805 
3806 static int
3807 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3808     boolean_t *might_destroy)
3809 {
3810         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3811         uint64_t zapobj;
3812         uint64_t tmp;
3813         int error;
3814 
3815         *might_destroy = B_FALSE;
3816 
3817         mutex_enter(&ds->ds_lock);
3818         zapobj = ds->ds_phys->ds_userrefs_obj;
3819         if (zapobj == 0) {
3820                 /* The tag can't possibly exist */
3821                 mutex_exit(&ds->ds_lock);
3822                 return (ESRCH);
3823         }
3824 
3825         /* Make sure the tag exists */
3826         error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3827         if (error) {
3828                 mutex_exit(&ds->ds_lock);
3829                 if (error == ENOENT)
3830                         error = ESRCH;
3831                 return (error);
3832         }
3833 
3834         if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3835             DS_IS_DEFER_DESTROY(ds))
3836                 *might_destroy = B_TRUE;
3837 
3838         mutex_exit(&ds->ds_lock);
3839         return (0);
3840 }
3841 
3842 static int
3843 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3844 {
3845         struct dsl_ds_releasearg *ra = arg1;
3846         dsl_dataset_t *ds = ra->ds;
3847         boolean_t might_destroy;
3848         int error;
3849 
3850         if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3851                 return (ENOTSUP);
3852 
3853         error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3854         if (error)
3855                 return (error);
3856 
3857         if (might_destroy) {
3858                 struct dsl_ds_destroyarg dsda = {0};
3859 
3860                 if (dmu_tx_is_syncing(tx)) {
3861                         /*
3862                          * If we're not prepared to remove the snapshot,
3863                          * we can't allow the release to happen right now.
3864                          */
3865                         if (!ra->own)
3866                                 return (EBUSY);
3867                 }
3868                 dsda.ds = ds;
3869                 dsda.releasing = B_TRUE;
3870                 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3871         }
3872 
3873         return (0);
3874 }
3875 
3876 static void
3877 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3878 {
3879         struct dsl_ds_releasearg *ra = arg1;
3880         dsl_dataset_t *ds = ra->ds;
3881         dsl_pool_t *dp = ds->ds_dir->dd_pool;
3882         objset_t *mos = dp->dp_meta_objset;
3883         uint64_t zapobj;
3884         uint64_t refs;
3885         int error;
3886 
3887         mutex_enter(&ds->ds_lock);
3888         ds->ds_userrefs--;
3889         refs = ds->ds_userrefs;
3890         mutex_exit(&ds->ds_lock);
3891         error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3892         VERIFY(error == 0 || error == ENOENT);
3893         zapobj = ds->ds_phys->ds_userrefs_obj;
3894         VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3895 
3896         spa_history_log_internal_ds(ds, "release", tx,
3897             "tag = %s refs now = %lld", ra->htag, (longlong_t)refs);
3898 
3899         if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3900             DS_IS_DEFER_DESTROY(ds)) {
3901                 struct dsl_ds_destroyarg dsda = {0};
3902 
3903                 ASSERT(ra->own);
3904                 dsda.ds = ds;
3905                 dsda.releasing = B_TRUE;
3906                 /* We already did the destroy_check */
3907                 dsl_dataset_destroy_sync(&dsda, tag, tx);
3908         }
3909 }
3910 
3911 static int
3912 dsl_dataset_user_release_one(const char *dsname, void *arg)
3913 {
3914         struct dsl_ds_holdarg *ha = arg;
3915         struct dsl_ds_releasearg *ra;
3916         dsl_dataset_t *ds;
3917         int error;
3918         void *dtag = ha->dstg;
3919         char *name;
3920         boolean_t own = B_FALSE;
3921         boolean_t might_destroy;
3922 
3923         /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3924         name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3925         error = dsl_dataset_hold(name, dtag, &ds);
3926         strfree(name);
3927         if (error == ENOENT && ha->recursive)
3928                 return (0);
3929         (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3930         if (error)
3931                 return (error);
3932 
3933         ha->gotone = B_TRUE;
3934 
3935         ASSERT(dsl_dataset_is_snapshot(ds));
3936 
3937         error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3938         if (error) {
3939                 dsl_dataset_rele(ds, dtag);
3940                 return (error);
3941         }
3942 
3943         if (might_destroy) {
3944 #ifdef _KERNEL
3945                 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3946                 error = zfs_unmount_snap(name, NULL);
3947                 strfree(name);
3948                 if (error) {
3949                         dsl_dataset_rele(ds, dtag);
3950                         return (error);
3951                 }
3952 #endif
3953                 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3954                         dsl_dataset_rele(ds, dtag);
3955                         return (EBUSY);
3956                 } else {
3957                         own = B_TRUE;
3958                         dsl_dataset_make_exclusive(ds, dtag);
3959                 }
3960         }
3961 
3962         ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3963         ra->ds = ds;
3964         ra->htag = ha->htag;
3965         ra->own = own;
3966         dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3967             dsl_dataset_user_release_sync, ra, dtag, 0);
3968 
3969         return (0);
3970 }
3971 
3972 int
3973 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3974     boolean_t recursive)
3975 {
3976         struct dsl_ds_holdarg *ha;
3977         dsl_sync_task_t *dst;
3978         spa_t *spa;
3979         int error;
3980 
3981 top:
3982         ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3983 
3984         (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3985 
3986         error = spa_open(dsname, &spa, FTAG);
3987         if (error) {
3988                 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3989                 return (error);
3990         }
3991 
3992         ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3993         ha->htag = htag;
3994         ha->snapname = snapname;
3995         ha->recursive = recursive;
3996         if (recursive) {
3997                 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
3998                     ha, DS_FIND_CHILDREN);
3999         } else {
4000                 error = dsl_dataset_user_release_one(dsname, ha);
4001         }
4002         if (error == 0)
4003                 error = dsl_sync_task_group_wait(ha->dstg);
4004 
4005         for (dst = list_head(&ha->dstg->dstg_tasks); dst;
4006             dst = list_next(&ha->dstg->dstg_tasks, dst)) {
4007                 struct dsl_ds_releasearg *ra = dst->dst_arg1;
4008                 dsl_dataset_t *ds = ra->ds;
4009 
4010                 if (dst->dst_err)
4011                         dsl_dataset_name(ds, ha->failed);
4012 
4013                 if (ra->own)
4014                         dsl_dataset_disown(ds, ha->dstg);
4015                 else
4016                         dsl_dataset_rele(ds, ha->dstg);
4017 
4018                 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
4019         }
4020 
4021         if (error == 0 && recursive && !ha->gotone)
4022                 error = ENOENT;
4023 
4024         if (error && error != EBUSY)
4025                 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
4026 
4027         dsl_sync_task_group_destroy(ha->dstg);
4028         kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4029         spa_close(spa, FTAG);
4030 
4031         /*
4032          * We can get EBUSY if we were racing with deferred destroy and
4033          * dsl_dataset_user_release_check() hadn't done the necessary
4034          * open context setup.  We can also get EBUSY if we're racing
4035          * with destroy and that thread is the ds_owner.  Either way
4036          * the busy condition should be transient, and we should retry
4037          * the release operation.
4038          */
4039         if (error == EBUSY)
4040                 goto top;
4041 
4042         return (error);
4043 }
4044 
4045 /*
4046  * Called at spa_load time (with retry == B_FALSE) to release a stale
4047  * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
4048  */
4049 int
4050 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
4051     boolean_t retry)
4052 {
4053         dsl_dataset_t *ds;
4054         char *snap;
4055         char *name;
4056         int namelen;
4057         int error;
4058 
4059         do {
4060                 rw_enter(&dp->dp_config_rwlock, RW_READER);
4061                 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
4062                 rw_exit(&dp->dp_config_rwlock);
4063                 if (error)
4064                         return (error);
4065                 namelen = dsl_dataset_namelen(ds)+1;
4066                 name = kmem_alloc(namelen, KM_SLEEP);
4067                 dsl_dataset_name(ds, name);
4068                 dsl_dataset_rele(ds, FTAG);
4069 
4070                 snap = strchr(name, '@');
4071                 *snap = '\0';
4072                 ++snap;
4073                 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
4074                 kmem_free(name, namelen);
4075 
4076                 /*
4077                  * The object can't have been destroyed because we have a hold,
4078                  * but it might have been renamed, resulting in ENOENT.  Retry
4079                  * if we've been requested to do so.
4080                  *
4081                  * It would be nice if we could use the dsobj all the way
4082                  * through and avoid ENOENT entirely.  But we might need to
4083                  * unmount the snapshot, and there's currently no way to lookup
4084                  * a vfsp using a ZFS object id.
4085                  */
4086         } while ((error == ENOENT) && retry);
4087 
4088         return (error);
4089 }
4090 
4091 int
4092 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
4093 {
4094         dsl_dataset_t *ds;
4095         int err;
4096 
4097         err = dsl_dataset_hold(dsname, FTAG, &ds);
4098         if (err)
4099                 return (err);
4100 
4101         VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4102         if (ds->ds_phys->ds_userrefs_obj != 0) {
4103                 zap_attribute_t *za;
4104                 zap_cursor_t zc;
4105 
4106                 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4107                 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4108                     ds->ds_phys->ds_userrefs_obj);
4109                     zap_cursor_retrieve(&zc, za) == 0;
4110                     zap_cursor_advance(&zc)) {
4111                         VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4112                             za->za_first_integer));
4113                 }
4114                 zap_cursor_fini(&zc);
4115                 kmem_free(za, sizeof (zap_attribute_t));
4116         }
4117         dsl_dataset_rele(ds, FTAG);
4118         return (0);
4119 }
4120 
4121 /*
4122  * Note, this function is used as the callback for dmu_objset_find().  We
4123  * always return 0 so that we will continue to find and process
4124  * inconsistent datasets, even if we encounter an error trying to
4125  * process one of them.
4126  */
4127 /* ARGSUSED */
4128 int
4129 dsl_destroy_inconsistent(const char *dsname, void *arg)
4130 {
4131         dsl_dataset_t *ds;
4132 
4133         if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4134                 if (DS_IS_INCONSISTENT(ds))
4135                         (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4136                 else
4137                         dsl_dataset_disown(ds, FTAG);
4138         }
4139         return (0);
4140 }
4141 
4142 /*
4143  * Return (in *usedp) the amount of space written in new that is not
4144  * present in oldsnap.  New may be a snapshot or the head.  Old must be
4145  * a snapshot before new, in new's filesystem (or its origin).  If not then
4146  * fail and return EINVAL.
4147  *
4148  * The written space is calculated by considering two components:  First, we
4149  * ignore any freed space, and calculate the written as new's used space
4150  * minus old's used space.  Next, we add in the amount of space that was freed
4151  * between the two snapshots, thus reducing new's used space relative to old's.
4152  * Specifically, this is the space that was born before old->ds_creation_txg,
4153  * and freed before new (ie. on new's deadlist or a previous deadlist).
4154  *
4155  * space freed                         [---------------------]
4156  * snapshots                       ---O-------O--------O-------O------
4157  *                                         oldsnap            new
4158  */
4159 int
4160 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4161     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4162 {
4163         int err = 0;
4164         uint64_t snapobj;
4165         dsl_pool_t *dp = new->ds_dir->dd_pool;
4166 
4167         *usedp = 0;
4168         *usedp += new->ds_phys->ds_referenced_bytes;
4169         *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
4170 
4171         *compp = 0;
4172         *compp += new->ds_phys->ds_compressed_bytes;
4173         *compp -= oldsnap->ds_phys->ds_compressed_bytes;
4174 
4175         *uncompp = 0;
4176         *uncompp += new->ds_phys->ds_uncompressed_bytes;
4177         *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
4178 
4179         rw_enter(&dp->dp_config_rwlock, RW_READER);
4180         snapobj = new->ds_object;
4181         while (snapobj != oldsnap->ds_object) {
4182                 dsl_dataset_t *snap;
4183                 uint64_t used, comp, uncomp;
4184 
4185                 if (snapobj == new->ds_object) {
4186                         snap = new;
4187                 } else {
4188                         err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4189                         if (err != 0)
4190                                 break;
4191                 }
4192 
4193                 if (snap->ds_phys->ds_prev_snap_txg ==
4194                     oldsnap->ds_phys->ds_creation_txg) {
4195                         /*
4196                          * The blocks in the deadlist can not be born after
4197                          * ds_prev_snap_txg, so get the whole deadlist space,
4198                          * which is more efficient (especially for old-format
4199                          * deadlists).  Unfortunately the deadlist code
4200                          * doesn't have enough information to make this
4201                          * optimization itself.
4202                          */
4203                         dsl_deadlist_space(&snap->ds_deadlist,
4204                             &used, &comp, &uncomp);
4205                 } else {
4206                         dsl_deadlist_space_range(&snap->ds_deadlist,
4207                             0, oldsnap->ds_phys->ds_creation_txg,
4208                             &used, &comp, &uncomp);
4209                 }
4210                 *usedp += used;
4211                 *compp += comp;
4212                 *uncompp += uncomp;
4213 
4214                 /*
4215                  * If we get to the beginning of the chain of snapshots
4216                  * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4217                  * was not a snapshot of/before new.
4218                  */
4219                 snapobj = snap->ds_phys->ds_prev_snap_obj;
4220                 if (snap != new)
4221                         dsl_dataset_rele(snap, FTAG);
4222                 if (snapobj == 0) {
4223                         err = EINVAL;
4224                         break;
4225                 }
4226 
4227         }
4228         rw_exit(&dp->dp_config_rwlock);
4229         return (err);
4230 }
4231 
4232 /*
4233  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4234  * lastsnap, and all snapshots in between are deleted.
4235  *
4236  * blocks that would be freed            [---------------------------]
4237  * snapshots                       ---O-------O--------O-------O--------O
4238  *                                        firstsnap        lastsnap
4239  *
4240  * This is the set of blocks that were born after the snap before firstsnap,
4241  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4242  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4243  * We calculate this by iterating over the relevant deadlists (from the snap
4244  * after lastsnap, backward to the snap after firstsnap), summing up the
4245  * space on the deadlist that was born after the snap before firstsnap.
4246  */
4247 int
4248 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4249     dsl_dataset_t *lastsnap,
4250     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4251 {
4252         int err = 0;
4253         uint64_t snapobj;
4254         dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4255 
4256         ASSERT(dsl_dataset_is_snapshot(firstsnap));
4257         ASSERT(dsl_dataset_is_snapshot(lastsnap));
4258 
4259         /*
4260          * Check that the snapshots are in the same dsl_dir, and firstsnap
4261          * is before lastsnap.
4262          */
4263         if (firstsnap->ds_dir != lastsnap->ds_dir ||
4264             firstsnap->ds_phys->ds_creation_txg >
4265             lastsnap->ds_phys->ds_creation_txg)
4266                 return (EINVAL);
4267 
4268         *usedp = *compp = *uncompp = 0;
4269 
4270         rw_enter(&dp->dp_config_rwlock, RW_READER);
4271         snapobj = lastsnap->ds_phys->ds_next_snap_obj;
4272         while (snapobj != firstsnap->ds_object) {
4273                 dsl_dataset_t *ds;
4274                 uint64_t used, comp, uncomp;
4275 
4276                 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4277                 if (err != 0)
4278                         break;
4279 
4280                 dsl_deadlist_space_range(&ds->ds_deadlist,
4281                     firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
4282                     &used, &comp, &uncomp);
4283                 *usedp += used;
4284                 *compp += comp;
4285                 *uncompp += uncomp;
4286 
4287                 snapobj = ds->ds_phys->ds_prev_snap_obj;
4288                 ASSERT3U(snapobj, !=, 0);
4289                 dsl_dataset_rele(ds, FTAG);
4290         }
4291         rw_exit(&dp->dp_config_rwlock);
4292         return (err);
4293 }