sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dataset.c

   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  * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
  26  * All rights reserved.
  27  * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org>
  28  */
  29
  30 #include <sys/dmu_objset.h>
  31 #include <sys/dsl_dataset.h>
  32 #include <sys/dsl_dir.h>
  33 #include <sys/dsl_prop.h>
  34 #include <sys/dsl_synctask.h>
  35 #include <sys/dmu_traverse.h>
  36 #include <sys/dmu_impl.h>
  37 #include <sys/dmu_tx.h>
  38 #include <sys/arc.h>
  39 #include <sys/zio.h>
  40 #include <sys/zap.h>
  41 #include <sys/zfeature.h>
  42 #include <sys/unique.h>
  43 #include <sys/zfs_context.h>
  44 #include <sys/zfs_ioctl.h>
  45 #include <sys/spa.h>
  46 #include <sys/zfs_znode.h>
  47 #include <sys/zfs_onexit.h>
  48 #include <sys/zvol.h>
  49 #include <sys/dsl_scan.h>
  50 #include <sys/dsl_deadlist.h>
  51
  52 static char *dsl_reaper = "the grim reaper";
  53
  54 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
  55 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
  56 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
  57
  58 #define SWITCH64(x, y) \
  59         { \
  60                 uint64_t __tmp = (x); \
  61                 (x) = (y); \
  62                 (y) = __tmp; \
  63         }
  64
  65 #define DS_REF_MAX      (1ULL << 62)
  66
  67 #define DSL_DEADLIST_BLOCKSIZE  SPA_MAXBLOCKSIZE
  68
  69 #define DSL_DATASET_IS_DESTROYED(ds)    ((ds)->ds_owner == dsl_reaper)
  70
  71
  72 /*
  73  * Figure out how much of this delta should be propogated to the dsl_dir
  74  * layer.  If there's a refreservation, that space has already been
  75  * partially accounted for in our ancestors.
  76  */
  77 static int64_t
  78 parent_delta(dsl_dataset_t *ds, int64_t delta)
  79 {
  80         uint64_t old_bytes, new_bytes;
  81
  82         if (ds->ds_reserved == 0)
  83                 return (delta);
  84
  85         old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
  86         new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
  87
  88         ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
  89         return (new_bytes - old_bytes);
  90 }
  91
  92 void
  93 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
  94 {
  95         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
  96         int compressed = BP_GET_PSIZE(bp);
  97         int uncompressed = BP_GET_UCSIZE(bp);
  98         int64_t delta;
  99
 100         dprintf_bp(bp, "ds=%p", ds);
 101
 102         ASSERT(dmu_tx_is_syncing(tx));
 103         /* It could have been compressed away to nothing */
 104         if (BP_IS_HOLE(bp))
 105                 return;
 106         ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
 107         ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
 108         if (ds == NULL) {
 109                 /*
 110                  * Account for the meta-objset space in its placeholder
 111                  * dsl_dir.
 112                  */
 113                 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
 114                 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
 115                     used, compressed, uncompressed, tx);
 116                 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
 117                 return;
 118         }
 119         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 120
 121         mutex_enter(&ds->ds_dir->dd_lock);
 122         mutex_enter(&ds->ds_lock);
 123         delta = parent_delta(ds, used);
 124         ds->ds_phys->ds_referenced_bytes += used;
 125         ds->ds_phys->ds_compressed_bytes += compressed;
 126         ds->ds_phys->ds_uncompressed_bytes += uncompressed;
 127         ds->ds_phys->ds_unique_bytes += used;
 128         mutex_exit(&ds->ds_lock);
 129         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
 130             compressed, uncompressed, tx);
 131         dsl_dir_transfer_space(ds->ds_dir, used - delta,
 132             DD_USED_REFRSRV, DD_USED_HEAD, tx);
 133         mutex_exit(&ds->ds_dir->dd_lock);
 134 }
 135
 136 int
 137 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 138     boolean_t async)
 139 {
 140         if (BP_IS_HOLE(bp))
 141                 return (0);
 142
 143         ASSERT(dmu_tx_is_syncing(tx));
 144         ASSERT(bp->blk_birth <= tx->tx_txg);
 145
 146         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 147         int compressed = BP_GET_PSIZE(bp);
 148         int uncompressed = BP_GET_UCSIZE(bp);
 149
 150         ASSERT(used > 0);
 151         if (ds == NULL) {
 152                 /*
 153                  * Account for the meta-objset space in its placeholder
 154                  * dataset.
 155                  */
 156                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 157
 158                 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
 159                     -used, -compressed, -uncompressed, tx);
 160                 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
 161                 return (used);
 162         }
 163         ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
 164
 165         ASSERT(!dsl_dataset_is_snapshot(ds));
 166         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 167
 168         if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
 169                 int64_t delta;
 170
 171                 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
 172                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 173
 174                 mutex_enter(&ds->ds_dir->dd_lock);
 175                 mutex_enter(&ds->ds_lock);
 176                 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
 177                     !DS_UNIQUE_IS_ACCURATE(ds));
 178                 delta = parent_delta(ds, -used);
 179                 ds->ds_phys->ds_unique_bytes -= used;
 180                 mutex_exit(&ds->ds_lock);
 181                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 182                     delta, -compressed, -uncompressed, tx);
 183                 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
 184                     DD_USED_REFRSRV, DD_USED_HEAD, tx);
 185                 mutex_exit(&ds->ds_dir->dd_lock);
 186         } else {
 187                 dprintf_bp(bp, "putting on dead list: %s", "");
 188                 if (async) {
 189                         /*
 190                          * We are here as part of zio's write done callback,
 191                          * which means we're a zio interrupt thread.  We can't
 192                          * call dsl_deadlist_insert() now because it may block
 193                          * waiting for I/O.  Instead, put bp on the deferred
 194                          * queue and let dsl_pool_sync() finish the job.
 195                          */
 196                         bplist_append(&ds->ds_pending_deadlist, bp);
 197                 } else {
 198                         dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
 199                 }
 200                 ASSERT3U(ds->ds_prev->ds_object, ==,
 201                     ds->ds_phys->ds_prev_snap_obj);
 202                 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
 203                 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
 204                 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
 205                     ds->ds_object && bp->blk_birth >
 206                     ds->ds_prev->ds_phys->ds_prev_snap_txg) {
 207                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 208                         mutex_enter(&ds->ds_prev->ds_lock);
 209                         ds->ds_prev->ds_phys->ds_unique_bytes += used;
 210                         mutex_exit(&ds->ds_prev->ds_lock);
 211                 }
 212                 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
 213                         dsl_dir_transfer_space(ds->ds_dir, used,
 214                             DD_USED_HEAD, DD_USED_SNAP, tx);
 215                 }
 216         }
 217         mutex_enter(&ds->ds_lock);
 218         ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
 219         ds->ds_phys->ds_referenced_bytes -= used;
 220         ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
 221         ds->ds_phys->ds_compressed_bytes -= compressed;
 222         ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
 223         ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
 224         mutex_exit(&ds->ds_lock);
 225
 226         return (used);
 227 }
 228
 229 uint64_t
 230 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
 231 {
 232         uint64_t trysnap = 0;
 233
 234         if (ds == NULL)
 235                 return (0);
 236         /*
 237          * The snapshot creation could fail, but that would cause an
 238          * incorrect FALSE return, which would only result in an
 239          * overestimation of the amount of space that an operation would
 240          * consume, which is OK.
 241          *
 242          * There's also a small window where we could miss a pending
 243          * snapshot, because we could set the sync task in the quiescing
 244          * phase.  So this should only be used as a guess.
 245          */
 246         if (ds->ds_trysnap_txg >
 247             spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
 248                 trysnap = ds->ds_trysnap_txg;
 249         return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
 250 }
 251
 252 boolean_t
 253 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
 254     uint64_t blk_birth)
 255 {
 256         if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
 257                 return (B_FALSE);
 258
 259         ddt_prefetch(dsl_dataset_get_spa(ds), bp);
 260
 261         return (B_TRUE);
 262 }
 263
 264 /* ARGSUSED */
 265 static void
 266 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
 267 {
 268         dsl_dataset_t *ds = dsv;
 269
 270         ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
 271
 272         unique_remove(ds->ds_fsid_guid);
 273
 274         if (ds->ds_objset != NULL)
 275                 dmu_objset_evict(ds->ds_objset);
 276
 277         if (ds->ds_prev) {
 278                 dsl_dataset_drop_ref(ds->ds_prev, ds);
 279                 ds->ds_prev = NULL;
 280         }
 281
 282         bplist_destroy(&ds->ds_pending_deadlist);
 283         if (db != NULL) {
 284                 dsl_deadlist_close(&ds->ds_deadlist);
 285         } else {
 286                 ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
 287                 ASSERT(!ds->ds_deadlist.dl_oldfmt);
 288         }
 289         if (ds->ds_dir)
 290                 dsl_dir_close(ds->ds_dir, ds);
 291
 292         ASSERT(!list_link_active(&ds->ds_synced_link));
 293
 294         if (mutex_owned(&ds->ds_lock))
 295                 mutex_exit(&ds->ds_lock);
 296         mutex_destroy(&ds->ds_lock);
 297         mutex_destroy(&ds->ds_recvlock);
 298         if (mutex_owned(&ds->ds_opening_lock))
 299                 mutex_exit(&ds->ds_opening_lock);
 300         mutex_destroy(&ds->ds_opening_lock);
 301         rw_destroy(&ds->ds_rwlock);
 302         cv_destroy(&ds->ds_exclusive_cv);
 303
 304         kmem_free(ds, sizeof (dsl_dataset_t));
 305 }
 306
 307 static int
 308 dsl_dataset_get_snapname(dsl_dataset_t *ds)
 309 {
 310         dsl_dataset_phys_t *headphys;
 311         int err;
 312         dmu_buf_t *headdbuf;
 313         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 314         objset_t *mos = dp->dp_meta_objset;
 315
 316         if (ds->ds_snapname[0])
 317                 return (0);
 318         if (ds->ds_phys->ds_next_snap_obj == 0)
 319                 return (0);
 320
 321         err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
 322             FTAG, &headdbuf);
 323         if (err)
 324                 return (err);
 325         headphys = headdbuf->db_data;
 326         err = zap_value_search(dp->dp_meta_objset,
 327             headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
 328         dmu_buf_rele(headdbuf, FTAG);
 329         return (err);
 330 }
 331
 332 static int
 333 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
 334 {
 335         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 336         uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
 337         matchtype_t mt;
 338         int err;
 339
 340         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 341                 mt = MT_FIRST;
 342         else
 343                 mt = MT_EXACT;
 344
 345         err = zap_lookup_norm(mos, snapobj, name, 8, 1,
 346             value, mt, NULL, 0, NULL);
 347         if (err == ENOTSUP && mt == MT_FIRST)
 348                 err = zap_lookup(mos, snapobj, name, 8, 1, value);
 349         return (err);
 350 }
 351
 352 static int
 353 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
 354 {
 355         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 356         uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
 357         matchtype_t mt;
 358         int err;
 359
 360         dsl_dir_snap_cmtime_update(ds->ds_dir);
 361
 362         if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 363                 mt = MT_FIRST;
 364         else
 365                 mt = MT_EXACT;
 366
 367         err = zap_remove_norm(mos, snapobj, name, mt, tx);
 368         if (err == ENOTSUP && mt == MT_FIRST)
 369                 err = zap_remove(mos, snapobj, name, tx);
 370         return (err);
 371 }
 372
 373 static int
 374 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 375     dsl_dataset_t **dsp)
 376 {
 377         objset_t *mos = dp->dp_meta_objset;
 378         dmu_buf_t *dbuf;
 379         dsl_dataset_t *ds;
 380         int err;
 381         dmu_object_info_t doi;
 382
 383         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
 384             dsl_pool_sync_context(dp));
 385
 386         err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
 387         if (err)
 388                 return (err);
 389
 390         /* Make sure dsobj has the correct object type. */
 391         dmu_object_info_from_db(dbuf, &doi);
 392         if (doi.doi_type != DMU_OT_DSL_DATASET)
 393                 return (EINVAL);
 394
 395         ds = dmu_buf_get_user(dbuf);
 396         if (ds == NULL) {
 397                 dsl_dataset_t *winner;
 398
 399                 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
 400                 ds->ds_dbuf = dbuf;
 401                 ds->ds_object = dsobj;
 402                 ds->ds_phys = dbuf->db_data;
 403
 404                 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
 405                 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
 406                 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 407                 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
 408
 409                 rw_init(&ds->ds_rwlock, 0, 0, 0);
 410                 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
 411
 412                 bplist_create(&ds->ds_pending_deadlist);
 413                 dsl_deadlist_open(&ds->ds_deadlist,
 414                     mos, ds->ds_phys->ds_deadlist_obj);
 415
 416                 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
 417                     offsetof(dmu_sendarg_t, dsa_link));
 418
 419                 if (err == 0) {
 420                         err = dsl_dir_open_obj(dp,
 421                             ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
 422                 }
 423                 if (err) {
 424                         mutex_destroy(&ds->ds_lock);
 425                         mutex_destroy(&ds->ds_recvlock);
 426                         mutex_destroy(&ds->ds_opening_lock);
 427                         rw_destroy(&ds->ds_rwlock);
 428                         cv_destroy(&ds->ds_exclusive_cv);
 429                         bplist_destroy(&ds->ds_pending_deadlist);
 430                         dsl_deadlist_close(&ds->ds_deadlist);
 431                         kmem_free(ds, sizeof (dsl_dataset_t));
 432                         dmu_buf_rele(dbuf, tag);
 433                         return (err);
 434                 }
 435
 436                 if (!dsl_dataset_is_snapshot(ds)) {
 437                         ds->ds_snapname[0] = '\0';
 438                         if (ds->ds_phys->ds_prev_snap_obj) {
 439                                 err = dsl_dataset_get_ref(dp,
 440                                     ds->ds_phys->ds_prev_snap_obj,
 441                                     ds, &ds->ds_prev);
 442                         }
 443                 } else {
 444                         if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
 445                                 err = dsl_dataset_get_snapname(ds);
 446                         if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
 447                                 err = zap_count(
 448                                     ds->ds_dir->dd_pool->dp_meta_objset,
 449                                     ds->ds_phys->ds_userrefs_obj,
 450                                     &ds->ds_userrefs);
 451                         }
 452                 }
 453
 454                 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
 455                         /*
 456                          * In sync context, we're called with either no lock
 457                          * or with the write lock.  If we're not syncing,
 458                          * we're always called with the read lock held.
 459                          */
 460                         boolean_t need_lock =
 461                             !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
 462                             dsl_pool_sync_context(dp);
 463
 464                         if (need_lock)
 465                                 rw_enter(&dp->dp_config_rwlock, RW_READER);
 466
 467                         err = dsl_prop_get_ds(ds,
 468                             "refreservation", sizeof (uint64_t), 1,
 469                             &ds->ds_reserved, NULL);
 470                         if (err == 0) {
 471                                 err = dsl_prop_get_ds(ds,
 472                                     "refquota", sizeof (uint64_t), 1,
 473                                     &ds->ds_quota, NULL);
 474                         }
 475
 476                         if (need_lock)
 477                                 rw_exit(&dp->dp_config_rwlock);
 478                 } else {
 479                         ds->ds_reserved = ds->ds_quota = 0;
 480                 }
 481
 482                 if (err == 0) {
 483                         winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
 484                             dsl_dataset_evict);
 485                 }
 486                 if (err || winner) {
 487                         bplist_destroy(&ds->ds_pending_deadlist);
 488                         dsl_deadlist_close(&ds->ds_deadlist);
 489                         if (ds->ds_prev)
 490                                 dsl_dataset_drop_ref(ds->ds_prev, ds);
 491                         dsl_dir_close(ds->ds_dir, ds);
 492                         mutex_destroy(&ds->ds_lock);
 493                         mutex_destroy(&ds->ds_recvlock);
 494                         mutex_destroy(&ds->ds_opening_lock);
 495                         rw_destroy(&ds->ds_rwlock);
 496                         cv_destroy(&ds->ds_exclusive_cv);
 497                         kmem_free(ds, sizeof (dsl_dataset_t));
 498                         if (err) {
 499                                 dmu_buf_rele(dbuf, tag);
 500                                 return (err);
 501                         }
 502                         ds = winner;
 503                 } else {
 504                         ds->ds_fsid_guid =
 505                             unique_insert(ds->ds_phys->ds_fsid_guid);
 506                 }
 507         }
 508         ASSERT3P(ds->ds_dbuf, ==, dbuf);
 509         ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
 510         ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
 511             spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
 512             dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
 513         mutex_enter(&ds->ds_lock);
 514         if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
 515                 mutex_exit(&ds->ds_lock);
 516                 dmu_buf_rele(ds->ds_dbuf, tag);
 517                 return (ENOENT);
 518         }
 519         mutex_exit(&ds->ds_lock);
 520         *dsp = ds;
 521         return (0);
 522 }
 523
 524 static int
 525 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
 526 {
 527         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 528
 529         /*
 530          * In syncing context we don't want the rwlock lock: there
 531          * may be an existing writer waiting for sync phase to
 532          * finish.  We don't need to worry about such writers, since
 533          * sync phase is single-threaded, so the writer can't be
 534          * doing anything while we are active.
 535          */
 536         if (dsl_pool_sync_context(dp)) {
 537                 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
 538                 return (0);
 539         }
 540
 541         /*
 542          * Normal users will hold the ds_rwlock as a READER until they
 543          * are finished (i.e., call dsl_dataset_rele()).  "Owners" will
 544          * drop their READER lock after they set the ds_owner field.
 545          *
 546          * If the dataset is being destroyed, the destroy thread will
 547          * obtain a WRITER lock for exclusive access after it's done its
 548          * open-context work and then change the ds_owner to
 549          * dsl_reaper once destruction is assured.  So threads
 550          * may block here temporarily, until the "destructability" of
 551          * the dataset is determined.
 552          */
 553         ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
 554         mutex_enter(&ds->ds_lock);
 555         while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
 556                 rw_exit(&dp->dp_config_rwlock);
 557                 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
 558                 if (DSL_DATASET_IS_DESTROYED(ds)) {
 559                         mutex_exit(&ds->ds_lock);
 560                         dsl_dataset_drop_ref(ds, tag);
 561                         rw_enter(&dp->dp_config_rwlock, RW_READER);
 562                         return (ENOENT);
 563                 }
 564                 /*
 565                  * The dp_config_rwlock lives above the ds_lock. And
 566                  * we need to check DSL_DATASET_IS_DESTROYED() while
 567                  * holding the ds_lock, so we have to drop and reacquire
 568                  * the ds_lock here.
 569                  */
 570                 mutex_exit(&ds->ds_lock);
 571                 rw_enter(&dp->dp_config_rwlock, RW_READER);
 572                 mutex_enter(&ds->ds_lock);
 573         }
 574         mutex_exit(&ds->ds_lock);
 575         return (0);
 576 }
 577
 578 int
 579 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 580     dsl_dataset_t **dsp)
 581 {
 582         int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
 583
 584         if (err)
 585                 return (err);
 586         return (dsl_dataset_hold_ref(*dsp, tag));
 587 }
 588
 589 int
 590 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
 591     void *tag, dsl_dataset_t **dsp)
 592 {
 593         int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
 594         if (err)
 595                 return (err);
 596         if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
 597                 dsl_dataset_rele(*dsp, tag);
 598                 *dsp = NULL;
 599                 return (EBUSY);
 600         }
 601         return (0);
 602 }
 603
 604 int
 605 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
 606 {
 607         dsl_dir_t *dd;
 608         dsl_pool_t *dp;
 609         const char *snapname;
 610         uint64_t obj;
 611         int err = 0;
 612
 613         err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
 614         if (err)
 615                 return (err);
 616
 617         dp = dd->dd_pool;
 618         obj = dd->dd_phys->dd_head_dataset_obj;
 619         rw_enter(&dp->dp_config_rwlock, RW_READER);
 620         if (obj)
 621                 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
 622         else
 623                 err = ENOENT;
 624         if (err)
 625                 goto out;
 626
 627         err = dsl_dataset_hold_ref(*dsp, tag);
 628
 629         /* we may be looking for a snapshot */
 630         if (err == 0 && snapname != NULL) {
 631                 dsl_dataset_t *ds = NULL;
 632
 633                 if (*snapname++ != '@') {
 634                         dsl_dataset_rele(*dsp, tag);
 635                         err = ENOENT;
 636                         goto out;
 637                 }
 638
 639                 dprintf("looking for snapshot '%s'\n", snapname);
 640                 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
 641                 if (err == 0)
 642                         err = dsl_dataset_get_ref(dp, obj, tag, &ds);
 643                 dsl_dataset_rele(*dsp, tag);
 644
 645                 ASSERT3U((err == 0), ==, (ds != NULL));
 646
 647                 if (ds) {
 648                         mutex_enter(&ds->ds_lock);
 649                         if (ds->ds_snapname[0] == 0)
 650                                 (void) strlcpy(ds->ds_snapname, snapname,
 651                                     sizeof (ds->ds_snapname));
 652                         mutex_exit(&ds->ds_lock);
 653                         err = dsl_dataset_hold_ref(ds, tag);
 654                         *dsp = err ? NULL : ds;
 655                 }
 656         }
 657 out:
 658         rw_exit(&dp->dp_config_rwlock);
 659         dsl_dir_close(dd, FTAG);
 660         return (err);
 661 }
 662
 663 int
 664 dsl_dataset_own(const char *name, boolean_t inconsistentok,
 665     void *tag, dsl_dataset_t **dsp)
 666 {
 667         int err = dsl_dataset_hold(name, tag, dsp);
 668         if (err)
 669                 return (err);
 670         if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
 671                 dsl_dataset_rele(*dsp, tag);
 672                 return (EBUSY);
 673         }
 674         return (0);
 675 }
 676
 677 void
 678 dsl_dataset_name(dsl_dataset_t *ds, char *name)
 679 {
 680         if (ds == NULL) {
 681                 (void) strcpy(name, "mos");
 682         } else {
 683                 dsl_dir_name(ds->ds_dir, name);
 684                 VERIFY(0 == dsl_dataset_get_snapname(ds));
 685                 if (ds->ds_snapname[0]) {
 686                         (void) strcat(name, "@");
 687                         /*
 688                          * We use a "recursive" mutex so that we
 689                          * can call dprintf_ds() with ds_lock held.
 690                          */
 691                         if (!MUTEX_HELD(&ds->ds_lock)) {
 692                                 mutex_enter(&ds->ds_lock);
 693                                 (void) strcat(name, ds->ds_snapname);
 694                                 mutex_exit(&ds->ds_lock);
 695                         } else {
 696                                 (void) strcat(name, ds->ds_snapname);
 697                         }
 698                 }
 699         }
 700 }
 701
 702 static int
 703 dsl_dataset_namelen(dsl_dataset_t *ds)
 704 {
 705         int result;
 706
 707         if (ds == NULL) {
 708                 result = 3;     /* "mos" */
 709         } else {
 710                 result = dsl_dir_namelen(ds->ds_dir);
 711                 VERIFY(0 == dsl_dataset_get_snapname(ds));
 712                 if (ds->ds_snapname[0]) {
 713                         ++result;       /* adding one for the @-sign */
 714                         if (!MUTEX_HELD(&ds->ds_lock)) {
 715                                 mutex_enter(&ds->ds_lock);
 716                                 result += strlen(ds->ds_snapname);
 717                                 mutex_exit(&ds->ds_lock);
 718                         } else {
 719                                 result += strlen(ds->ds_snapname);
 720                         }
 721                 }
 722         }
 723
 724         return (result);
 725 }
 726
 727 void
 728 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
 729 {
 730         dmu_buf_rele(ds->ds_dbuf, tag);
 731 }
 732
 733 void
 734 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
 735 {
 736         if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
 737                 rw_exit(&ds->ds_rwlock);
 738         }
 739         dsl_dataset_drop_ref(ds, tag);
 740 }
 741
 742 void
 743 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
 744 {
 745         ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
 746             (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
 747
 748         mutex_enter(&ds->ds_lock);
 749         ds->ds_owner = NULL;
 750         if (RW_WRITE_HELD(&ds->ds_rwlock)) {
 751                 rw_exit(&ds->ds_rwlock);
 752                 cv_broadcast(&ds->ds_exclusive_cv);
 753         }
 754         mutex_exit(&ds->ds_lock);
 755         if (ds->ds_dbuf)
 756                 dsl_dataset_drop_ref(ds, tag);
 757         else
 758                 dsl_dataset_evict(NULL, ds);
 759 }
 760
 761 boolean_t
 762 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
 763 {
 764         boolean_t gotit = FALSE;
 765
 766         mutex_enter(&ds->ds_lock);
 767         if (ds->ds_owner == NULL &&
 768             (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
 769                 ds->ds_owner = tag;
 770                 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
 771                         rw_exit(&ds->ds_rwlock);
 772                 gotit = TRUE;
 773         }
 774         mutex_exit(&ds->ds_lock);
 775         return (gotit);
 776 }
 777
 778 void
 779 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
 780 {
 781         ASSERT3P(owner, ==, ds->ds_owner);
 782         if (!RW_WRITE_HELD(&ds->ds_rwlock))
 783                 rw_enter(&ds->ds_rwlock, RW_WRITER);
 784 }
 785
 786 uint64_t
 787 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
 788     uint64_t flags, dmu_tx_t *tx)
 789 {
 790         dsl_pool_t *dp = dd->dd_pool;
 791         dmu_buf_t *dbuf;
 792         dsl_dataset_phys_t *dsphys;
 793         uint64_t dsobj;
 794         objset_t *mos = dp->dp_meta_objset;
 795
 796         if (origin == NULL)
 797                 origin = dp->dp_origin_snap;
 798
 799         ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
 800         ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
 801         ASSERT(dmu_tx_is_syncing(tx));
 802         ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
 803
 804         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
 805             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
 806         VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
 807         dmu_buf_will_dirty(dbuf, tx);
 808         dsphys = dbuf->db_data;
 809         bzero(dsphys, sizeof (dsl_dataset_phys_t));
 810         dsphys->ds_dir_obj = dd->dd_object;
 811         dsphys->ds_flags = flags;
 812         dsphys->ds_fsid_guid = unique_create();
 813         do {
 814                 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
 815                     sizeof (dsphys->ds_guid));
 816         } while (dsphys->ds_guid == 0);
 817         dsphys->ds_snapnames_zapobj =
 818             zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
 819             DMU_OT_NONE, 0, tx);
 820         dsphys->ds_creation_time = gethrestime_sec();
 821         dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
 822
 823         if (origin == NULL) {
 824                 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
 825         } else {
 826                 dsl_dataset_t *ohds;
 827
 828                 dsphys->ds_prev_snap_obj = origin->ds_object;
 829                 dsphys->ds_prev_snap_txg =
 830                     origin->ds_phys->ds_creation_txg;
 831                 dsphys->ds_referenced_bytes =
 832                     origin->ds_phys->ds_referenced_bytes;
 833                 dsphys->ds_compressed_bytes =
 834                     origin->ds_phys->ds_compressed_bytes;
 835                 dsphys->ds_uncompressed_bytes =
 836                     origin->ds_phys->ds_uncompressed_bytes;
 837                 dsphys->ds_bp = origin->ds_phys->ds_bp;
 838                 dsphys->ds_flags |= origin->ds_phys->ds_flags;
 839
 840                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
 841                 origin->ds_phys->ds_num_children++;
 842
 843                 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
 844                     origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
 845                 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
 846                     dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
 847                 dsl_dataset_rele(ohds, FTAG);
 848
 849                 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
 850                         if (origin->ds_phys->ds_next_clones_obj == 0) {
 851                                 origin->ds_phys->ds_next_clones_obj =
 852                                     zap_create(mos,
 853                                     DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
 854                         }
 855                         VERIFY(0 == zap_add_int(mos,
 856                             origin->ds_phys->ds_next_clones_obj,
 857                             dsobj, tx));
 858                 }
 859
 860                 dmu_buf_will_dirty(dd->dd_dbuf, tx);
 861                 dd->dd_phys->dd_origin_obj = origin->ds_object;
 862                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 863                         if (origin->ds_dir->dd_phys->dd_clones == 0) {
 864                                 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
 865                                 origin->ds_dir->dd_phys->dd_clones =
 866                                     zap_create(mos,
 867                                     DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
 868                         }
 869                         VERIFY3U(0, ==, zap_add_int(mos,
 870                             origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
 871                 }
 872         }
 873
 874         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
 875                 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 876
 877         dmu_buf_rele(dbuf, FTAG);
 878
 879         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 880         dd->dd_phys->dd_head_dataset_obj = dsobj;
 881
 882         return (dsobj);
 883 }
 884
 885 uint64_t
 886 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
 887     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
 888 {
 889         dsl_pool_t *dp = pdd->dd_pool;
 890         uint64_t dsobj, ddobj;
 891         dsl_dir_t *dd;
 892
 893         ASSERT(lastname[0] != '@');
 894
 895         ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
 896         VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
 897
 898         dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
 899
 900         dsl_deleg_set_create_perms(dd, tx, cr);
 901
 902         dsl_dir_close(dd, FTAG);
 903
 904         /*
 905          * If we are creating a clone, make sure we zero out any stale
 906          * data from the origin snapshots zil header.
 907          */
 908         if (origin != NULL) {
 909                 dsl_dataset_t *ds;
 910                 objset_t *os;
 911
 912                 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
 913                 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
 914                 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
 915                 dsl_dataset_dirty(ds, tx);
 916                 dsl_dataset_rele(ds, FTAG);
 917         }
 918
 919         return (dsobj);
 920 }
 921
 922 #ifdef __FreeBSD__
 923 /* FreeBSD ioctl compat begin */
 924 struct destroyarg {
 925         nvlist_t *nvl;
 926         const char *snapname;
 927 };
 928
 929 static int
 930 dsl_check_snap_cb(const char *name, void *arg)
 931 {
 932         struct destroyarg *da = arg;
 933         dsl_dataset_t *ds;
 934         char *dsname;
 935
 936         dsname = kmem_asprintf("%s@%s", name, da->snapname);
 937         VERIFY(nvlist_add_boolean(da->nvl, dsname) == 0);
 938
 939         return (0);
 940 }
 941
 942 int
 943 dmu_get_recursive_snaps_nvl(const char *fsname, const char *snapname,
 944     nvlist_t *snaps)
 945 {
 946         struct destroyarg *da;
 947         int err;
 948
 949         da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP);
 950         da->nvl = snaps;
 951         da->snapname = snapname;
 952         err = dmu_objset_find(fsname, dsl_check_snap_cb, da,
 953             DS_FIND_CHILDREN);
 954         kmem_free(da, sizeof (struct destroyarg));
 955
 956         return (err);
 957 }
 958 /* FreeBSD ioctl compat end */
 959 #endif /* __FreeBSD__ */
 960
 961 /*
 962  * The snapshots must all be in the same pool.
 963  */
 964 int
 965 dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer, char *failed)
 966 {
 967         int err;
 968         dsl_sync_task_t *dst;
 969         spa_t *spa;
 970         nvpair_t *pair;
 971         dsl_sync_task_group_t *dstg;
 972
 973         pair = nvlist_next_nvpair(snaps, NULL);
 974         if (pair == NULL)
 975                 return (0);
 976
 977         err = spa_open(nvpair_name(pair), &spa, FTAG);
 978         if (err)
 979                 return (err);
 980         dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
 981
 982         for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
 983             pair = nvlist_next_nvpair(snaps, pair)) {
 984                 dsl_dataset_t *ds;
 985
 986                 err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds);
 987                 if (err == 0) {
 988                         struct dsl_ds_destroyarg *dsda;
 989
 990                         dsl_dataset_make_exclusive(ds, dstg);
 991                         dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg),
 992                             KM_SLEEP);
 993                         dsda->ds = ds;
 994                         dsda->defer = defer;
 995                         dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
 996                             dsl_dataset_destroy_sync, dsda, dstg, 0);
 997                 } else if (err == ENOENT) {
 998                         err = 0;
 999                 } else {
1000                         (void) strcpy(failed, nvpair_name(pair));
1001                         break;
1002                 }
1003         }
1004
1005         if (err == 0)
1006                 err = dsl_sync_task_group_wait(dstg);
1007
1008         for (dst = list_head(&dstg->dstg_tasks); dst;
1009             dst = list_next(&dstg->dstg_tasks, dst)) {
1010                 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
1011                 dsl_dataset_t *ds = dsda->ds;
1012
1013                 /*
1014                  * Return the file system name that triggered the error
1015                  */
1016                 if (dst->dst_err) {
1017                         dsl_dataset_name(ds, failed);
1018                 }
1019                 ASSERT3P(dsda->rm_origin, ==, NULL);
1020                 dsl_dataset_disown(ds, dstg);
1021                 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
1022         }
1023
1024         dsl_sync_task_group_destroy(dstg);
1025         spa_close(spa, FTAG);
1026         return (err);
1027
1028 }
1029
1030 static boolean_t
1031 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
1032 {
1033         boolean_t might_destroy = B_FALSE;
1034
1035         mutex_enter(&ds->ds_lock);
1036         if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
1037             DS_IS_DEFER_DESTROY(ds))
1038                 might_destroy = B_TRUE;
1039         mutex_exit(&ds->ds_lock);
1040
1041         return (might_destroy);
1042 }
1043
1044 /*
1045  * If we're removing a clone, and these three conditions are true:
1046  *      1) the clone's origin has no other children
1047  *      2) the clone's origin has no user references
1048  *      3) the clone's origin has been marked for deferred destruction
1049  * Then, prepare to remove the origin as part of this sync task group.
1050  */
1051 static int
1052 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1053 {
1054         dsl_dataset_t *ds = dsda->ds;
1055         dsl_dataset_t *origin = ds->ds_prev;
1056
1057         if (dsl_dataset_might_destroy_origin(origin)) {
1058                 char *name;
1059                 int namelen;
1060                 int error;
1061
1062                 namelen = dsl_dataset_namelen(origin) + 1;
1063                 name = kmem_alloc(namelen, KM_SLEEP);
1064                 dsl_dataset_name(origin, name);
1065 #ifdef _KERNEL
1066                 error = zfs_unmount_snap(name, NULL);
1067                 if (error) {
1068                         kmem_free(name, namelen);
1069                         return (error);
1070                 }
1071 #endif
1072                 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1073                 kmem_free(name, namelen);
1074                 if (error)
1075                         return (error);
1076                 dsda->rm_origin = origin;
1077                 dsl_dataset_make_exclusive(origin, tag);
1078         }
1079
1080         return (0);
1081 }
1082
1083 /*
1084  * ds must be opened as OWNER.  On return (whether successful or not),
1085  * ds will be closed and caller can no longer dereference it.
1086  */
1087 int
1088 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1089 {
1090         int err;
1091         dsl_sync_task_group_t *dstg;
1092         objset_t *os;
1093         dsl_dir_t *dd;
1094         uint64_t obj;
1095         struct dsl_ds_destroyarg dsda = { 0 };
1096         dsl_dataset_t dummy_ds = { 0 };
1097
1098         dsda.ds = ds;
1099
1100         if (dsl_dataset_is_snapshot(ds)) {
1101                 /* Destroying a snapshot is simpler */
1102                 dsl_dataset_make_exclusive(ds, tag);
1103
1104                 dsda.defer = defer;
1105                 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1106                     dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1107                     &dsda, tag, 0);
1108                 ASSERT3P(dsda.rm_origin, ==, NULL);
1109                 goto out;
1110         } else if (defer) {
1111                 err = EINVAL;
1112                 goto out;
1113         }
1114
1115         dd = ds->ds_dir;
1116         dummy_ds.ds_dir = dd;
1117         dummy_ds.ds_object = ds->ds_object;
1118
1119         /*
1120          * Check for errors and mark this ds as inconsistent, in
1121          * case we crash while freeing the objects.
1122          */
1123         err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1124             dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1125         if (err)
1126                 goto out;
1127
1128         err = dmu_objset_from_ds(ds, &os);
1129         if (err)
1130                 goto out;
1131
1132         /*
1133          * If async destruction is not enabled try to remove all objects
1134          * while in the open context so that there is less work to do in
1135          * the syncing context.
1136          */
1137         if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds),
1138             &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
1139                 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1140                     ds->ds_phys->ds_prev_snap_txg)) {
1141                         /*
1142                          * Ignore errors, if there is not enough disk space
1143                          * we will deal with it in dsl_dataset_destroy_sync().
1144                          */
1145                         (void) dmu_free_object(os, obj);
1146                 }
1147                 if (err != ESRCH)
1148                         goto out;
1149         }
1150
1151         /*
1152          * Only the ZIL knows how to free log blocks.
1153          */
1154         zil_destroy(dmu_objset_zil(os), B_FALSE);
1155
1156         /*
1157          * Sync out all in-flight IO.
1158          */
1159         txg_wait_synced(dd->dd_pool, 0);
1160
1161         /*
1162          * If we managed to free all the objects in open
1163          * context, the user space accounting should be zero.
1164          */
1165         if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1166             dmu_objset_userused_enabled(os)) {
1167                 uint64_t count;
1168
1169                 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1170                     count == 0);
1171                 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1172                     count == 0);
1173         }
1174
1175         rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1176         err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1177         rw_exit(&dd->dd_pool->dp_config_rwlock);
1178
1179         if (err)
1180                 goto out;
1181
1182         /*
1183          * Blow away the dsl_dir + head dataset.
1184          */
1185         dsl_dataset_make_exclusive(ds, tag);
1186         /*
1187          * If we're removing a clone, we might also need to remove its
1188          * origin.
1189          */
1190         do {
1191                 dsda.need_prep = B_FALSE;
1192                 if (dsl_dir_is_clone(dd)) {
1193                         err = dsl_dataset_origin_rm_prep(&dsda, tag);
1194                         if (err) {
1195                                 dsl_dir_close(dd, FTAG);
1196                                 goto out;
1197                         }
1198                 }
1199
1200                 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1201                 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1202                     dsl_dataset_destroy_sync, &dsda, tag, 0);
1203                 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1204                     dsl_dir_destroy_sync, &dummy_ds, FTAG, 0);
1205                 err = dsl_sync_task_group_wait(dstg);
1206                 dsl_sync_task_group_destroy(dstg);
1207
1208                 /*
1209                  * We could be racing against 'zfs release' or 'zfs destroy -d'
1210                  * on the origin snap, in which case we can get EBUSY if we
1211                  * needed to destroy the origin snap but were not ready to
1212                  * do so.
1213                  */
1214                 if (dsda.need_prep) {
1215                         ASSERT(err == EBUSY);
1216                         ASSERT(dsl_dir_is_clone(dd));
1217                         ASSERT(dsda.rm_origin == NULL);
1218                 }
1219         } while (dsda.need_prep);
1220
1221         if (dsda.rm_origin != NULL)
1222                 dsl_dataset_disown(dsda.rm_origin, tag);
1223
1224         /* if it is successful, dsl_dir_destroy_sync will close the dd */
1225         if (err)
1226                 dsl_dir_close(dd, FTAG);
1227 out:
1228         dsl_dataset_disown(ds, tag);
1229         return (err);
1230 }
1231
1232 blkptr_t *
1233 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1234 {
1235         return (&ds->ds_phys->ds_bp);
1236 }
1237
1238 void
1239 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1240 {
1241         ASSERT(dmu_tx_is_syncing(tx));
1242         /* If it's the meta-objset, set dp_meta_rootbp */
1243         if (ds == NULL) {
1244                 tx->tx_pool->dp_meta_rootbp = *bp;
1245         } else {
1246                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1247                 ds->ds_phys->ds_bp = *bp;
1248         }
1249 }
1250
1251 spa_t *
1252 dsl_dataset_get_spa(dsl_dataset_t *ds)
1253 {
1254         return (ds->ds_dir->dd_pool->dp_spa);
1255 }
1256
1257 void
1258 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1259 {
1260         dsl_pool_t *dp;
1261
1262         if (ds == NULL) /* this is the meta-objset */
1263                 return;
1264
1265         ASSERT(ds->ds_objset != NULL);
1266
1267         if (ds->ds_phys->ds_next_snap_obj != 0)
1268                 panic("dirtying snapshot!");
1269
1270         dp = ds->ds_dir->dd_pool;
1271
1272         if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1273                 /* up the hold count until we can be written out */
1274                 dmu_buf_add_ref(ds->ds_dbuf, ds);
1275         }
1276 }
1277
1278 /*
1279  * The unique space in the head dataset can be calculated by subtracting
1280  * the space used in the most recent snapshot, that is still being used
1281  * in this file system, from the space currently in use.  To figure out
1282  * the space in the most recent snapshot still in use, we need to take
1283  * the total space used in the snapshot and subtract out the space that
1284  * has been freed up since the snapshot was taken.
1285  */
1286 static void
1287 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1288 {
1289         uint64_t mrs_used;
1290         uint64_t dlused, dlcomp, dluncomp;
1291
1292         ASSERT(!dsl_dataset_is_snapshot(ds));
1293
1294         if (ds->ds_phys->ds_prev_snap_obj != 0)
1295                 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
1296         else
1297                 mrs_used = 0;
1298
1299         dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1300
1301         ASSERT3U(dlused, <=, mrs_used);
1302         ds->ds_phys->ds_unique_bytes =
1303             ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
1304
1305         if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1306             SPA_VERSION_UNIQUE_ACCURATE)
1307                 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1308 }
1309
1310 struct killarg {
1311         dsl_dataset_t *ds;
1312         dmu_tx_t *tx;
1313 };
1314
1315 /* ARGSUSED */
1316 static int
1317 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1318     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1319 {
1320         struct killarg *ka = arg;
1321         dmu_tx_t *tx = ka->tx;
1322
1323         if (bp == NULL)
1324                 return (0);
1325
1326         if (zb->zb_level == ZB_ZIL_LEVEL) {
1327                 ASSERT(zilog != NULL);
1328                 /*
1329                  * It's a block in the intent log.  It has no
1330                  * accounting, so just free it.
1331                  */
1332                 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1333         } else {
1334                 ASSERT(zilog == NULL);
1335                 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1336                 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1337         }
1338
1339         return (0);
1340 }
1341
1342 /* ARGSUSED */
1343 static int
1344 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1345 {
1346         dsl_dataset_t *ds = arg1;
1347         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1348         uint64_t count;
1349         int err;
1350
1351         /*
1352          * Can't delete a head dataset if there are snapshots of it.
1353          * (Except if the only snapshots are from the branch we cloned
1354          * from.)
1355          */
1356         if (ds->ds_prev != NULL &&
1357             ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1358                 return (EBUSY);
1359
1360         /*
1361          * This is really a dsl_dir thing, but check it here so that
1362          * we'll be less likely to leave this dataset inconsistent &
1363          * nearly destroyed.
1364          */
1365         err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1366         if (err)
1367                 return (err);
1368         if (count != 0)
1369                 return (EEXIST);
1370
1371         return (0);
1372 }
1373
1374 /* ARGSUSED */
1375 static void
1376 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1377 {
1378         dsl_dataset_t *ds = arg1;
1379         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1380
1381         /* Mark it as inconsistent on-disk, in case we crash */
1382         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1383         ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1384
1385         spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1386             "dataset = %llu", ds->ds_object);
1387 }
1388
1389 static int
1390 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1391     dmu_tx_t *tx)
1392 {
1393         dsl_dataset_t *ds = dsda->ds;
1394         dsl_dataset_t *ds_prev = ds->ds_prev;
1395
1396         if (dsl_dataset_might_destroy_origin(ds_prev)) {
1397                 struct dsl_ds_destroyarg ndsda = {0};
1398
1399                 /*
1400                  * If we're not prepared to remove the origin, don't remove
1401                  * the clone either.
1402                  */
1403                 if (dsda->rm_origin == NULL) {
1404                         dsda->need_prep = B_TRUE;
1405                         return (EBUSY);
1406                 }
1407
1408                 ndsda.ds = ds_prev;
1409                 ndsda.is_origin_rm = B_TRUE;
1410                 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1411         }
1412
1413         /*
1414          * If we're not going to remove the origin after all,
1415          * undo the open context setup.
1416          */
1417         if (dsda->rm_origin != NULL) {
1418                 dsl_dataset_disown(dsda->rm_origin, tag);
1419                 dsda->rm_origin = NULL;
1420         }
1421
1422         return (0);
1423 }
1424
1425 /*
1426  * If you add new checks here, you may need to add
1427  * additional checks to the "temporary" case in
1428  * snapshot_check() in dmu_objset.c.
1429  */
1430 /* ARGSUSED */
1431 int
1432 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1433 {
1434         struct dsl_ds_destroyarg *dsda = arg1;
1435         dsl_dataset_t *ds = dsda->ds;
1436
1437         /* we have an owner hold, so noone else can destroy us */
1438         ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1439
1440         /*
1441          * Only allow deferred destroy on pools that support it.
1442          * NOTE: deferred destroy is only supported on snapshots.
1443          */
1444         if (dsda->defer) {
1445                 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1446                     SPA_VERSION_USERREFS)
1447                         return (ENOTSUP);
1448                 ASSERT(dsl_dataset_is_snapshot(ds));
1449                 return (0);
1450         }
1451
1452         /*
1453          * Can't delete a head dataset if there are snapshots of it.
1454          * (Except if the only snapshots are from the branch we cloned
1455          * from.)
1456          */
1457         if (ds->ds_prev != NULL &&
1458             ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1459                 return (EBUSY);
1460
1461         /*
1462          * If we made changes this txg, traverse_dsl_dataset won't find
1463          * them.  Try again.
1464          */
1465         if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1466                 return (EAGAIN);
1467
1468         if (dsl_dataset_is_snapshot(ds)) {
1469                 /*
1470                  * If this snapshot has an elevated user reference count,
1471                  * we can't destroy it yet.
1472                  */
1473                 if (ds->ds_userrefs > 0 && !dsda->releasing)
1474                         return (EBUSY);
1475
1476                 mutex_enter(&ds->ds_lock);
1477                 /*
1478                  * Can't delete a branch point. However, if we're destroying
1479                  * a clone and removing its origin due to it having a user
1480                  * hold count of 0 and having been marked for deferred destroy,
1481                  * it's OK for the origin to have a single clone.
1482                  */
1483                 if (ds->ds_phys->ds_num_children >
1484                     (dsda->is_origin_rm ? 2 : 1)) {
1485                         mutex_exit(&ds->ds_lock);
1486                         return (EEXIST);
1487                 }
1488                 mutex_exit(&ds->ds_lock);
1489         } else if (dsl_dir_is_clone(ds->ds_dir)) {
1490                 return (dsl_dataset_origin_check(dsda, arg2, tx));
1491         }
1492
1493         /* XXX we should do some i/o error checking... */
1494         return (0);
1495 }
1496
1497 struct refsarg {
1498         kmutex_t lock;
1499         boolean_t gone;
1500         kcondvar_t cv;
1501 };
1502
1503 /* ARGSUSED */
1504 static void
1505 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1506 {
1507         struct refsarg *arg = argv;
1508
1509         mutex_enter(&arg->lock);
1510         arg->gone = TRUE;
1511         cv_signal(&arg->cv);
1512         mutex_exit(&arg->lock);
1513 }
1514
1515 static void
1516 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1517 {
1518         struct refsarg arg;
1519
1520         bzero(&arg, sizeof(arg));
1521         mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1522         cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1523         arg.gone = FALSE;
1524         (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1525             dsl_dataset_refs_gone);
1526         dmu_buf_rele(ds->ds_dbuf, tag);
1527         mutex_enter(&arg.lock);
1528         while (!arg.gone)
1529                 cv_wait(&arg.cv, &arg.lock);
1530         ASSERT(arg.gone);
1531         mutex_exit(&arg.lock);
1532         ds->ds_dbuf = NULL;
1533         ds->ds_phys = NULL;
1534         mutex_destroy(&arg.lock);
1535         cv_destroy(&arg.cv);
1536 }
1537
1538 static void
1539 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1540 {
1541         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1542         uint64_t count;
1543         int err;
1544
1545         ASSERT(ds->ds_phys->ds_num_children >= 2);
1546         err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1547         /*
1548          * The err should not be ENOENT, but a bug in a previous version
1549          * of the code could cause upgrade_clones_cb() to not set
1550          * ds_next_snap_obj when it should, leading to a missing entry.
1551          * If we knew that the pool was created after
1552          * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1553          * ENOENT.  However, at least we can check that we don't have
1554          * too many entries in the next_clones_obj even after failing to
1555          * remove this one.
1556          */
1557         if (err != ENOENT) {
1558                 VERIFY3U(err, ==, 0);
1559         }
1560         ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1561             &count));
1562         ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1563 }
1564
1565 static void
1566 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1567 {
1568         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1569         zap_cursor_t zc;
1570         zap_attribute_t za;
1571
1572         /*
1573          * If it is the old version, dd_clones doesn't exist so we can't
1574          * find the clones, but deadlist_remove_key() is a no-op so it
1575          * doesn't matter.
1576          */
1577         if (ds->ds_dir->dd_phys->dd_clones == 0)
1578                 return;
1579
1580         for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1581             zap_cursor_retrieve(&zc, &za) == 0;
1582             zap_cursor_advance(&zc)) {
1583                 dsl_dataset_t *clone;
1584
1585                 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1586                     za.za_first_integer, FTAG, &clone));
1587                 if (clone->ds_dir->dd_origin_txg > mintxg) {
1588                         dsl_deadlist_remove_key(&clone->ds_deadlist,
1589                             mintxg, tx);
1590                         dsl_dataset_remove_clones_key(clone, mintxg, tx);
1591                 }
1592                 dsl_dataset_rele(clone, FTAG);
1593         }
1594         zap_cursor_fini(&zc);
1595 }
1596
1597 struct process_old_arg {
1598         dsl_dataset_t *ds;
1599         dsl_dataset_t *ds_prev;
1600         boolean_t after_branch_point;
1601         zio_t *pio;
1602         uint64_t used, comp, uncomp;
1603 };
1604
1605 static int
1606 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1607 {
1608         struct process_old_arg *poa = arg;
1609         dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1610
1611         if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1612                 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1613                 if (poa->ds_prev && !poa->after_branch_point &&
1614                     bp->blk_birth >
1615                     poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1616                         poa->ds_prev->ds_phys->ds_unique_bytes +=
1617                             bp_get_dsize_sync(dp->dp_spa, bp);
1618                 }
1619         } else {
1620                 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1621                 poa->comp += BP_GET_PSIZE(bp);
1622                 poa->uncomp += BP_GET_UCSIZE(bp);
1623                 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1624         }
1625         return (0);
1626 }
1627
1628 static void
1629 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1630     dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1631 {
1632         struct process_old_arg poa = { 0 };
1633         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1634         objset_t *mos = dp->dp_meta_objset;
1635
1636         ASSERT(ds->ds_deadlist.dl_oldfmt);
1637         ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1638
1639         poa.ds = ds;
1640         poa.ds_prev = ds_prev;
1641         poa.after_branch_point = after_branch_point;
1642         poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1643         VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1644             process_old_cb, &poa, tx));
1645         VERIFY3U(zio_wait(poa.pio), ==, 0);
1646         ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1647
1648         /* change snapused */
1649         dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1650             -poa.used, -poa.comp, -poa.uncomp, tx);
1651
1652         /* swap next's deadlist to our deadlist */
1653         dsl_deadlist_close(&ds->ds_deadlist);
1654         dsl_deadlist_close(&ds_next->ds_deadlist);
1655         SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1656             ds->ds_phys->ds_deadlist_obj);
1657         dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1658         dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1659             ds_next->ds_phys->ds_deadlist_obj);
1660 }
1661
1662 static int
1663 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
1664 {
1665         int err;
1666         struct killarg ka;
1667
1668         /*
1669          * Free everything that we point to (that's born after
1670          * the previous snapshot, if we are a clone)
1671          *
1672          * NB: this should be very quick, because we already
1673          * freed all the objects in open context.
1674          */
1675         ka.ds = ds;
1676         ka.tx = tx;
1677         err = traverse_dataset(ds,
1678             ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST,
1679             kill_blkptr, &ka);
1680         ASSERT3U(err, ==, 0);
1681         ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0);
1682
1683         return (err);
1684 }
1685
1686 void
1687 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1688 {
1689         struct dsl_ds_destroyarg *dsda = arg1;
1690         dsl_dataset_t *ds = dsda->ds;
1691         int err;
1692         int after_branch_point = FALSE;
1693         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1694         objset_t *mos = dp->dp_meta_objset;
1695         dsl_dataset_t *ds_prev = NULL;
1696         boolean_t wont_destroy;
1697         uint64_t obj;
1698
1699         wont_destroy = (dsda->defer &&
1700             (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1701
1702         ASSERT(ds->ds_owner || wont_destroy);
1703         ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1704         ASSERT(ds->ds_prev == NULL ||
1705             ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1706         ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1707
1708         if (wont_destroy) {
1709                 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1710                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1711                 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1712                 return;
1713         }
1714
1715         /* signal any waiters that this dataset is going away */
1716         mutex_enter(&ds->ds_lock);
1717         ds->ds_owner = dsl_reaper;
1718         cv_broadcast(&ds->ds_exclusive_cv);
1719         mutex_exit(&ds->ds_lock);
1720
1721         /* Remove our reservation */
1722         if (ds->ds_reserved != 0) {
1723                 dsl_prop_setarg_t psa;
1724                 uint64_t value = 0;
1725
1726                 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1727                     (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1728                     &value);
1729                 psa.psa_effective_value = 0;    /* predict default value */
1730
1731                 dsl_dataset_set_reservation_sync(ds, &psa, tx);
1732                 ASSERT3U(ds->ds_reserved, ==, 0);
1733         }
1734
1735         ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1736
1737         dsl_scan_ds_destroyed(ds, tx);
1738
1739         obj = ds->ds_object;
1740
1741         if (ds->ds_phys->ds_prev_snap_obj != 0) {
1742                 if (ds->ds_prev) {
1743                         ds_prev = ds->ds_prev;
1744                 } else {
1745                         VERIFY(0 == dsl_dataset_hold_obj(dp,
1746                             ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1747                 }
1748                 after_branch_point =
1749                     (ds_prev->ds_phys->ds_next_snap_obj != obj);
1750
1751                 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1752                 if (after_branch_point &&
1753                     ds_prev->ds_phys->ds_next_clones_obj != 0) {
1754                         remove_from_next_clones(ds_prev, obj, tx);
1755                         if (ds->ds_phys->ds_next_snap_obj != 0) {
1756                                 VERIFY(0 == zap_add_int(mos,
1757                                     ds_prev->ds_phys->ds_next_clones_obj,
1758                                     ds->ds_phys->ds_next_snap_obj, tx));
1759                         }
1760                 }
1761                 if (after_branch_point &&
1762                     ds->ds_phys->ds_next_snap_obj == 0) {
1763                         /* This clone is toast. */
1764                         ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1765                         ds_prev->ds_phys->ds_num_children--;
1766
1767                         /*
1768                          * If the clone's origin has no other clones, no
1769                          * user holds, and has been marked for deferred
1770                          * deletion, then we should have done the necessary
1771                          * destroy setup for it.
1772                          */
1773                         if (ds_prev->ds_phys->ds_num_children == 1 &&
1774                             ds_prev->ds_userrefs == 0 &&
1775                             DS_IS_DEFER_DESTROY(ds_prev)) {
1776                                 ASSERT3P(dsda->rm_origin, !=, NULL);
1777                         } else {
1778                                 ASSERT3P(dsda->rm_origin, ==, NULL);
1779                         }
1780                 } else if (!after_branch_point) {
1781                         ds_prev->ds_phys->ds_next_snap_obj =
1782                             ds->ds_phys->ds_next_snap_obj;
1783                 }
1784         }
1785
1786         if (dsl_dataset_is_snapshot(ds)) {
1787                 dsl_dataset_t *ds_next;
1788                 uint64_t old_unique;
1789                 uint64_t used = 0, comp = 0, uncomp = 0;
1790
1791                 VERIFY(0 == dsl_dataset_hold_obj(dp,
1792                     ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1793                 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1794
1795                 old_unique = ds_next->ds_phys->ds_unique_bytes;
1796
1797                 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1798                 ds_next->ds_phys->ds_prev_snap_obj =
1799                     ds->ds_phys->ds_prev_snap_obj;
1800                 ds_next->ds_phys->ds_prev_snap_txg =
1801                     ds->ds_phys->ds_prev_snap_txg;
1802                 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1803                     ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1804
1805
1806                 if (ds_next->ds_deadlist.dl_oldfmt) {
1807                         process_old_deadlist(ds, ds_prev, ds_next,
1808                             after_branch_point, tx);
1809                 } else {
1810                         /* Adjust prev's unique space. */
1811                         if (ds_prev && !after_branch_point) {
1812                                 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1813                                     ds_prev->ds_phys->ds_prev_snap_txg,
1814                                     ds->ds_phys->ds_prev_snap_txg,
1815                                     &used, &comp, &uncomp);
1816                                 ds_prev->ds_phys->ds_unique_bytes += used;
1817                         }
1818
1819                         /* Adjust snapused. */
1820                         dsl_deadlist_space_range(&ds_next->ds_deadlist,
1821                             ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1822                             &used, &comp, &uncomp);
1823                         dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1824                             -used, -comp, -uncomp, tx);
1825
1826                         /* Move blocks to be freed to pool's free list. */
1827                         dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1828                             &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1829                             tx);
1830                         dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1831                             DD_USED_HEAD, used, comp, uncomp, tx);
1832
1833                         /* Merge our deadlist into next's and free it. */
1834                         dsl_deadlist_merge(&ds_next->ds_deadlist,
1835                             ds->ds_phys->ds_deadlist_obj, tx);
1836                 }
1837                 dsl_deadlist_close(&ds->ds_deadlist);
1838                 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1839
1840                 /* Collapse range in clone heads */
1841                 dsl_dataset_remove_clones_key(ds,
1842                     ds->ds_phys->ds_creation_txg, tx);
1843
1844                 if (dsl_dataset_is_snapshot(ds_next)) {
1845                         dsl_dataset_t *ds_nextnext;
1846
1847                         /*
1848                          * Update next's unique to include blocks which
1849                          * were previously shared by only this snapshot
1850                          * and it.  Those blocks will be born after the
1851                          * prev snap and before this snap, and will have
1852                          * died after the next snap and before the one
1853                          * after that (ie. be on the snap after next's
1854                          * deadlist).
1855                          */
1856                         VERIFY(0 == dsl_dataset_hold_obj(dp,
1857                             ds_next->ds_phys->ds_next_snap_obj,
1858                             FTAG, &ds_nextnext));
1859                         dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1860                             ds->ds_phys->ds_prev_snap_txg,
1861                             ds->ds_phys->ds_creation_txg,
1862                             &used, &comp, &uncomp);
1863                         ds_next->ds_phys->ds_unique_bytes += used;
1864                         dsl_dataset_rele(ds_nextnext, FTAG);
1865                         ASSERT3P(ds_next->ds_prev, ==, NULL);
1866
1867                         /* Collapse range in this head. */
1868                         dsl_dataset_t *hds;
1869                         VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1870                             ds->ds_dir->dd_phys->dd_head_dataset_obj,
1871                             FTAG, &hds));
1872                         dsl_deadlist_remove_key(&hds->ds_deadlist,
1873                             ds->ds_phys->ds_creation_txg, tx);
1874                         dsl_dataset_rele(hds, FTAG);
1875
1876                 } else {
1877                         ASSERT3P(ds_next->ds_prev, ==, ds);
1878                         dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1879                         ds_next->ds_prev = NULL;
1880                         if (ds_prev) {
1881                                 VERIFY(0 == dsl_dataset_get_ref(dp,
1882                                     ds->ds_phys->ds_prev_snap_obj,
1883                                     ds_next, &ds_next->ds_prev));
1884                         }
1885
1886                         dsl_dataset_recalc_head_uniq(ds_next);
1887
1888                         /*
1889                          * Reduce the amount of our unconsmed refreservation
1890                          * being charged to our parent by the amount of
1891                          * new unique data we have gained.
1892                          */
1893                         if (old_unique < ds_next->ds_reserved) {
1894                                 int64_t mrsdelta;
1895                                 uint64_t new_unique =
1896                                     ds_next->ds_phys->ds_unique_bytes;
1897
1898                                 ASSERT(old_unique <= new_unique);
1899                                 mrsdelta = MIN(new_unique - old_unique,
1900                                     ds_next->ds_reserved - old_unique);
1901                                 dsl_dir_diduse_space(ds->ds_dir,
1902                                     DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1903                         }
1904                 }
1905                 dsl_dataset_rele(ds_next, FTAG);
1906         } else {
1907                 zfeature_info_t *async_destroy =
1908                     &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY];
1909
1910                 /*
1911                  * There's no next snapshot, so this is a head dataset.
1912                  * Destroy the deadlist.  Unless it's a clone, the
1913                  * deadlist should be empty.  (If it's a clone, it's
1914                  * safe to ignore the deadlist contents.)
1915                  */
1916                 dsl_deadlist_close(&ds->ds_deadlist);
1917                 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1918                 ds->ds_phys->ds_deadlist_obj = 0;
1919
1920                 if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) {
1921                         err = old_synchronous_dataset_destroy(ds, tx);
1922                 } else {
1923                         /*
1924                          * Move the bptree into the pool's list of trees to
1925                          * clean up and update space accounting information.
1926                          */
1927                         uint64_t used, comp, uncomp;
1928
1929                         ASSERT(err == 0 || err == EBUSY);
1930                         if (!spa_feature_is_active(dp->dp_spa, async_destroy)) {
1931                                 spa_feature_incr(dp->dp_spa, async_destroy, tx);
1932                                 dp->dp_bptree_obj = bptree_alloc(
1933                                     dp->dp_meta_objset, tx);
1934                                 VERIFY(zap_add(dp->dp_meta_objset,
1935                                     DMU_POOL_DIRECTORY_OBJECT,
1936                                     DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
1937                                     &dp->dp_bptree_obj, tx) == 0);
1938                         }
1939
1940                         used = ds->ds_dir->dd_phys->dd_used_bytes;
1941                         comp = ds->ds_dir->dd_phys->dd_compressed_bytes;
1942                         uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes;
1943
1944                         ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1945                             ds->ds_phys->ds_unique_bytes == used);
1946
1947                         bptree_add(dp->dp_meta_objset, dp->dp_bptree_obj,
1948                             &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg,
1949                             used, comp, uncomp, tx);
1950                         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
1951                             -used, -comp, -uncomp, tx);
1952                         dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
1953                             used, comp, uncomp, tx);
1954                 }
1955
1956                 if (ds->ds_prev != NULL) {
1957                         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1958                                 VERIFY3U(0, ==, zap_remove_int(mos,
1959                                     ds->ds_prev->ds_dir->dd_phys->dd_clones,
1960                                     ds->ds_object, tx));
1961                         }
1962                         dsl_dataset_rele(ds->ds_prev, ds);
1963                         ds->ds_prev = ds_prev = NULL;
1964                 }
1965         }
1966
1967         /*
1968          * This must be done after the dsl_traverse(), because it will
1969          * re-open the objset.
1970          */
1971         if (ds->ds_objset) {
1972                 dmu_objset_evict(ds->ds_objset);
1973                 ds->ds_objset = NULL;
1974         }
1975
1976         if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1977                 /* Erase the link in the dir */
1978                 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1979                 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1980                 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1981                 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1982                 ASSERT(err == 0);
1983         } else {
1984                 /* remove from snapshot namespace */
1985                 dsl_dataset_t *ds_head;
1986                 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1987                 VERIFY(0 == dsl_dataset_hold_obj(dp,
1988                     ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1989                 VERIFY(0 == dsl_dataset_get_snapname(ds));
1990 #ifdef ZFS_DEBUG
1991                 {
1992                         uint64_t val;
1993
1994                         err = dsl_dataset_snap_lookup(ds_head,
1995                             ds->ds_snapname, &val);
1996                         ASSERT3U(err, ==, 0);
1997                         ASSERT3U(val, ==, obj);
1998                 }
1999 #endif
2000                 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
2001                 ASSERT(err == 0);
2002                 dsl_dataset_rele(ds_head, FTAG);
2003         }
2004
2005         if (ds_prev && ds->ds_prev != ds_prev)
2006                 dsl_dataset_rele(ds_prev, FTAG);
2007
2008         spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
2009         spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx,
2010             "dataset = %llu", ds->ds_object);
2011
2012         if (ds->ds_phys->ds_next_clones_obj != 0) {
2013                 uint64_t count;
2014                 ASSERT(0 == zap_count(mos,
2015                     ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
2016                 VERIFY(0 == dmu_object_free(mos,
2017                     ds->ds_phys->ds_next_clones_obj, tx));
2018         }
2019         if (ds->ds_phys->ds_props_obj != 0)
2020                 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
2021         if (ds->ds_phys->ds_userrefs_obj != 0)
2022                 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
2023         dsl_dir_close(ds->ds_dir, ds);
2024         ds->ds_dir = NULL;
2025         dsl_dataset_drain_refs(ds, tag);
2026         VERIFY(0 == dmu_object_free(mos, obj, tx));
2027
2028         if (dsda->rm_origin) {
2029                 /*
2030                  * Remove the origin of the clone we just destroyed.
2031                  */
2032                 struct dsl_ds_destroyarg ndsda = {0};
2033
2034                 ndsda.ds = dsda->rm_origin;
2035                 dsl_dataset_destroy_sync(&ndsda, tag, tx);
2036         }
2037 }
2038
2039 static int
2040 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
2041 {
2042         uint64_t asize;
2043
2044         if (!dmu_tx_is_syncing(tx))
2045                 return (0);
2046
2047         /*
2048          * If there's an fs-only reservation, any blocks that might become
2049          * owned by the snapshot dataset must be accommodated by space
2050          * outside of the reservation.
2051          */
2052         ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
2053         asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2054         if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2055                 return (ENOSPC);
2056
2057         /*
2058          * Propogate any reserved space for this snapshot to other
2059          * snapshot checks in this sync group.
2060          */
2061         if (asize > 0)
2062                 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
2063
2064         return (0);
2065 }
2066
2067 int
2068 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
2069 {
2070         dsl_dataset_t *ds = arg1;
2071         const char *snapname = arg2;
2072         int err;
2073         uint64_t value;
2074
2075         /*
2076          * We don't allow multiple snapshots of the same txg.  If there
2077          * is already one, try again.
2078          */
2079         if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
2080                 return (EAGAIN);
2081
2082         /*
2083          * Check for conflicting name snapshot name.
2084          */
2085         err = dsl_dataset_snap_lookup(ds, snapname, &value);
2086         if (err == 0)
2087                 return (EEXIST);
2088         if (err != ENOENT)
2089                 return (err);
2090
2091         /*
2092          * Check that the dataset's name is not too long.  Name consists
2093          * of the dataset's length + 1 for the @-sign + snapshot name's length
2094          */
2095         if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2096                 return (ENAMETOOLONG);
2097
2098         err = dsl_dataset_snapshot_reserve_space(ds, tx);
2099         if (err)
2100                 return (err);
2101
2102         ds->ds_trysnap_txg = tx->tx_txg;
2103         return (0);
2104 }
2105
2106 void
2107 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2108 {
2109         dsl_dataset_t *ds = arg1;
2110         const char *snapname = arg2;
2111         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2112         dmu_buf_t *dbuf;
2113         dsl_dataset_phys_t *dsphys;
2114         uint64_t dsobj, crtxg;
2115         objset_t *mos = dp->dp_meta_objset;
2116         int err;
2117
2118         ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2119
2120         /*
2121          * The origin's ds_creation_txg has to be < TXG_INITIAL
2122          */
2123         if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2124                 crtxg = 1;
2125         else
2126                 crtxg = tx->tx_txg;
2127
2128         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2129             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2130         VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2131         dmu_buf_will_dirty(dbuf, tx);
2132         dsphys = dbuf->db_data;
2133         bzero(dsphys, sizeof (dsl_dataset_phys_t));
2134         dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2135         dsphys->ds_fsid_guid = unique_create();
2136         do {
2137                 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2138                     sizeof (dsphys->ds_guid));
2139         } while (dsphys->ds_guid == 0);
2140         dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2141         dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2142         dsphys->ds_next_snap_obj = ds->ds_object;
2143         dsphys->ds_num_children = 1;
2144         dsphys->ds_creation_time = gethrestime_sec();
2145         dsphys->ds_creation_txg = crtxg;
2146         dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2147         dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
2148         dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2149         dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2150         dsphys->ds_flags = ds->ds_phys->ds_flags;
2151         dsphys->ds_bp = ds->ds_phys->ds_bp;
2152         dmu_buf_rele(dbuf, FTAG);
2153
2154         ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2155         if (ds->ds_prev) {
2156                 uint64_t next_clones_obj =
2157                     ds->ds_prev->ds_phys->ds_next_clones_obj;
2158                 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2159                     ds->ds_object ||
2160                     ds->ds_prev->ds_phys->ds_num_children > 1);
2161                 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2162                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2163                         ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2164                             ds->ds_prev->ds_phys->ds_creation_txg);
2165                         ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2166                 } else if (next_clones_obj != 0) {
2167                         remove_from_next_clones(ds->ds_prev,
2168                             dsphys->ds_next_snap_obj, tx);
2169                         VERIFY3U(0, ==, zap_add_int(mos,
2170                             next_clones_obj, dsobj, tx));
2171                 }
2172         }
2173
2174         /*
2175          * If we have a reference-reservation on this dataset, we will
2176          * need to increase the amount of refreservation being charged
2177          * since our unique space is going to zero.
2178          */
2179         if (ds->ds_reserved) {
2180                 int64_t delta;
2181                 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2182                 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2183                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2184                     delta, 0, 0, tx);
2185         }
2186
2187         dmu_buf_will_dirty(ds->ds_dbuf, tx);
2188         zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2189             ds->ds_dir->dd_myname, snapname, dsobj,
2190             ds->ds_phys->ds_prev_snap_txg);
2191         ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2192             UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2193         dsl_deadlist_close(&ds->ds_deadlist);
2194         dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2195         dsl_deadlist_add_key(&ds->ds_deadlist,
2196             ds->ds_phys->ds_prev_snap_txg, tx);
2197
2198         ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2199         ds->ds_phys->ds_prev_snap_obj = dsobj;
2200         ds->ds_phys->ds_prev_snap_txg = crtxg;
2201         ds->ds_phys->ds_unique_bytes = 0;
2202         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2203                 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2204
2205         err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2206             snapname, 8, 1, &dsobj, tx);
2207         ASSERT(err == 0);
2208
2209         if (ds->ds_prev)
2210                 dsl_dataset_drop_ref(ds->ds_prev, ds);
2211         VERIFY(0 == dsl_dataset_get_ref(dp,
2212             ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2213
2214         dsl_scan_ds_snapshotted(ds, tx);
2215
2216         dsl_dir_snap_cmtime_update(ds->ds_dir);
2217
2218         spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx,
2219             "dataset = %llu", dsobj);
2220 }
2221
2222 void
2223 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2224 {
2225         ASSERT(dmu_tx_is_syncing(tx));
2226         ASSERT(ds->ds_objset != NULL);
2227         ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2228
2229         /*
2230          * in case we had to change ds_fsid_guid when we opened it,
2231          * sync it out now.
2232          */
2233         dmu_buf_will_dirty(ds->ds_dbuf, tx);
2234         ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2235
2236         dsl_dir_dirty(ds->ds_dir, tx);
2237         dmu_objset_sync(ds->ds_objset, zio, tx);
2238 }
2239
2240 static void
2241 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2242 {
2243         uint64_t count = 0;
2244         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2245         zap_cursor_t zc;
2246         zap_attribute_t za;
2247         nvlist_t *propval;
2248         nvlist_t *val;
2249
2250         rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2251         VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2252         VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2253
2254         /*
2255          * There may me missing entries in ds_next_clones_obj
2256          * due to a bug in a previous version of the code.
2257          * Only trust it if it has the right number of entries.
2258          */
2259         if (ds->ds_phys->ds_next_clones_obj != 0) {
2260                 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
2261                     &count));
2262         }
2263         if (count != ds->ds_phys->ds_num_children - 1) {
2264                 goto fail;
2265         }
2266         for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
2267             zap_cursor_retrieve(&zc, &za) == 0;
2268             zap_cursor_advance(&zc)) {
2269                 dsl_dataset_t *clone;
2270                 char buf[ZFS_MAXNAMELEN];
2271                 /*
2272                  * Even though we hold the dp_config_rwlock, the dataset
2273                  * may fail to open, returning ENOENT.  If there is a
2274                  * thread concurrently attempting to destroy this
2275                  * dataset, it will have the ds_rwlock held for
2276                  * RW_WRITER.  Our call to dsl_dataset_hold_obj() ->
2277                  * dsl_dataset_hold_ref() will fail its
2278                  * rw_tryenter(&ds->ds_rwlock, RW_READER), drop the
2279                  * dp_config_rwlock, and wait for the destroy progress
2280                  * and signal ds_exclusive_cv.  If the destroy was
2281                  * successful, we will see that
2282                  * DSL_DATASET_IS_DESTROYED(), and return ENOENT.
2283                  */
2284                 if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2285                     za.za_first_integer, FTAG, &clone) != 0)
2286                         continue;
2287                 dsl_dir_name(clone->ds_dir, buf);
2288                 VERIFY(nvlist_add_boolean(val, buf) == 0);
2289                 dsl_dataset_rele(clone, FTAG);
2290         }
2291         zap_cursor_fini(&zc);
2292         VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0);
2293         VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2294             propval) == 0);
2295 fail:
2296         nvlist_free(val);
2297         nvlist_free(propval);
2298         rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2299 }
2300
2301 void
2302 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2303 {
2304         uint64_t refd, avail, uobjs, aobjs, ratio;
2305
2306         dsl_dir_stats(ds->ds_dir, nv);
2307
2308         dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2309         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2310         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2311
2312         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2313             ds->ds_phys->ds_creation_time);
2314         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2315             ds->ds_phys->ds_creation_txg);
2316         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2317             ds->ds_quota);
2318         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2319             ds->ds_reserved);
2320         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2321             ds->ds_phys->ds_guid);
2322         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2323             ds->ds_phys->ds_unique_bytes);
2324         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2325             ds->ds_object);
2326         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2327             ds->ds_userrefs);
2328         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2329             DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2330
2331         if (ds->ds_phys->ds_prev_snap_obj != 0) {
2332                 uint64_t written, comp, uncomp;
2333                 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2334                 dsl_dataset_t *prev;
2335
2336                 rw_enter(&dp->dp_config_rwlock, RW_READER);
2337                 int err = dsl_dataset_hold_obj(dp,
2338                     ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
2339                 rw_exit(&dp->dp_config_rwlock);
2340                 if (err == 0) {
2341                         err = dsl_dataset_space_written(prev, ds, &written,
2342                             &comp, &uncomp);
2343                         dsl_dataset_rele(prev, FTAG);
2344                         if (err == 0) {
2345                                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2346                                     written);
2347                         }
2348                 }
2349         }
2350
2351         ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2352             (ds->ds_phys->ds_uncompressed_bytes * 100 /
2353             ds->ds_phys->ds_compressed_bytes);
2354         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2355
2356         if (ds->ds_phys->ds_next_snap_obj) {
2357                 /*
2358                  * This is a snapshot; override the dd's space used with
2359                  * our unique space and compression ratio.
2360                  */
2361                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2362                     ds->ds_phys->ds_unique_bytes);
2363                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2364
2365                 get_clones_stat(ds, nv);
2366         }
2367 }
2368
2369 void
2370 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2371 {
2372         stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2373         stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2374         stat->dds_guid = ds->ds_phys->ds_guid;
2375         if (ds->ds_phys->ds_next_snap_obj) {
2376                 stat->dds_is_snapshot = B_TRUE;
2377                 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2378         } else {
2379                 stat->dds_is_snapshot = B_FALSE;
2380                 stat->dds_num_clones = 0;
2381         }
2382
2383         /* clone origin is really a dsl_dir thing... */
2384         rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2385         if (dsl_dir_is_clone(ds->ds_dir)) {
2386                 dsl_dataset_t *ods;
2387
2388                 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2389                     ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2390                 dsl_dataset_name(ods, stat->dds_origin);
2391                 dsl_dataset_drop_ref(ods, FTAG);
2392         } else {
2393                 stat->dds_origin[0] = '\0';
2394         }
2395         rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2396 }
2397
2398 uint64_t
2399 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2400 {
2401         return (ds->ds_fsid_guid);
2402 }
2403
2404 void
2405 dsl_dataset_space(dsl_dataset_t *ds,
2406     uint64_t *refdbytesp, uint64_t *availbytesp,
2407     uint64_t *usedobjsp, uint64_t *availobjsp)
2408 {
2409         *refdbytesp = ds->ds_phys->ds_referenced_bytes;
2410         *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2411         if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2412                 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2413         if (ds->ds_quota != 0) {
2414                 /*
2415                  * Adjust available bytes according to refquota
2416                  */
2417                 if (*refdbytesp < ds->ds_quota)
2418                         *availbytesp = MIN(*availbytesp,
2419                             ds->ds_quota - *refdbytesp);
2420                 else
2421                         *availbytesp = 0;
2422         }
2423         *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2424         *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2425 }
2426
2427 boolean_t
2428 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2429 {
2430         dsl_pool_t *dp = ds->ds_dir->dd_pool;
2431
2432         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2433             dsl_pool_sync_context(dp));
2434         if (ds->ds_prev == NULL)
2435                 return (B_FALSE);
2436         if (ds->ds_phys->ds_bp.blk_birth >
2437             ds->ds_prev->ds_phys->ds_creation_txg) {
2438                 objset_t *os, *os_prev;
2439                 /*
2440                  * It may be that only the ZIL differs, because it was
2441                  * reset in the head.  Don't count that as being
2442                  * modified.
2443                  */
2444                 if (dmu_objset_from_ds(ds, &os) != 0)
2445                         return (B_TRUE);
2446                 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2447                         return (B_TRUE);
2448                 return (bcmp(&os->os_phys->os_meta_dnode,
2449                     &os_prev->os_phys->os_meta_dnode,
2450                     sizeof (os->os_phys->os_meta_dnode)) != 0);
2451         }
2452         return (B_FALSE);
2453 }
2454
2455 /* ARGSUSED */
2456 static int
2457 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2458 {
2459         dsl_dataset_t *ds = arg1;
2460         char *newsnapname = arg2;
2461         dsl_dir_t *dd = ds->ds_dir;
2462         dsl_dataset_t *hds;
2463         uint64_t val;
2464         int err;
2465
2466         err = dsl_dataset_hold_obj(dd->dd_pool,
2467             dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2468         if (err)
2469                 return (err);
2470
2471         /* new name better not be in use */
2472         err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2473         dsl_dataset_rele(hds, FTAG);
2474
2475         if (err == 0)
2476                 err = EEXIST;
2477         else if (err == ENOENT)
2478                 err = 0;
2479
2480         /* dataset name + 1 for the "@" + the new snapshot name must fit */
2481         if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2482                 err = ENAMETOOLONG;
2483
2484         return (err);
2485 }
2486
2487 static void
2488 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2489 {
2490         char oldname[MAXPATHLEN], newname[MAXPATHLEN];
2491         dsl_dataset_t *ds = arg1;
2492         const char *newsnapname = arg2;
2493         dsl_dir_t *dd = ds->ds_dir;
2494         objset_t *mos = dd->dd_pool->dp_meta_objset;
2495         dsl_dataset_t *hds;
2496         int err;
2497
2498         ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2499
2500         VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2501             dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2502
2503         VERIFY(0 == dsl_dataset_get_snapname(ds));
2504         err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2505         ASSERT3U(err, ==, 0);
2506         dsl_dataset_name(ds, oldname);
2507         mutex_enter(&ds->ds_lock);
2508         (void) strcpy(ds->ds_snapname, newsnapname);
2509         mutex_exit(&ds->ds_lock);
2510         err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2511             ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2512         ASSERT3U(err, ==, 0);
2513         dsl_dataset_name(ds, newname);
2514 #ifdef _KERNEL
2515         zvol_rename_minors(oldname, newname);
2516 #endif
2517
2518         spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2519             "dataset = %llu", ds->ds_object);
2520         dsl_dataset_rele(hds, FTAG);
2521 }
2522
2523 struct renamesnaparg {
2524         dsl_sync_task_group_t *dstg;
2525         char failed[MAXPATHLEN];
2526         char *oldsnap;
2527         char *newsnap;
2528 };
2529
2530 static int
2531 dsl_snapshot_rename_one(const char *name, void *arg)
2532 {
2533         struct renamesnaparg *ra = arg;
2534         dsl_dataset_t *ds = NULL;
2535         char *snapname;
2536         int err;
2537
2538         snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2539         (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2540
2541         /*
2542          * For recursive snapshot renames the parent won't be changing
2543          * so we just pass name for both the to/from argument.
2544          */
2545         err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2546         if (err != 0) {
2547                 strfree(snapname);
2548                 return (err == ENOENT ? 0 : err);
2549         }
2550
2551 #ifdef _KERNEL
2552         /*
2553          * For all filesystems undergoing rename, we'll need to unmount it.
2554          */
2555         (void) zfs_unmount_snap(snapname, NULL);
2556 #endif
2557         err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2558         strfree(snapname);
2559         if (err != 0)
2560                 return (err == ENOENT ? 0 : err);
2561
2562         dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2563             dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2564
2565         return (0);
2566 }
2567
2568 static int
2569 dsl_recursive_rename(char *oldname, const char *newname)
2570 {
2571         int err;
2572         struct renamesnaparg *ra;
2573         dsl_sync_task_t *dst;
2574         spa_t *spa;
2575         char *cp, *fsname = spa_strdup(oldname);
2576         int len = strlen(oldname) + 1;
2577
2578         /* truncate the snapshot name to get the fsname */
2579         cp = strchr(fsname, '@');
2580         *cp = '\0';
2581
2582         err = spa_open(fsname, &spa, FTAG);
2583         if (err) {
2584                 kmem_free(fsname, len);
2585                 return (err);
2586         }
2587         ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2588         ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2589
2590         ra->oldsnap = strchr(oldname, '@') + 1;
2591         ra->newsnap = strchr(newname, '@') + 1;
2592         *ra->failed = '\0';
2593
2594         err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2595             DS_FIND_CHILDREN);
2596         kmem_free(fsname, len);
2597
2598         if (err == 0) {
2599                 err = dsl_sync_task_group_wait(ra->dstg);
2600         }
2601
2602         for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2603             dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2604                 dsl_dataset_t *ds = dst->dst_arg1;
2605                 if (dst->dst_err) {
2606                         dsl_dir_name(ds->ds_dir, ra->failed);
2607                         (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2608                         (void) strlcat(ra->failed, ra->newsnap,
2609                             sizeof (ra->failed));
2610                 }
2611                 dsl_dataset_rele(ds, ra->dstg);
2612         }
2613
2614         if (err)
2615                 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2616
2617         dsl_sync_task_group_destroy(ra->dstg);
2618         kmem_free(ra, sizeof (struct renamesnaparg));
2619         spa_close(spa, FTAG);
2620         return (err);
2621 }
2622
2623 static int
2624 dsl_valid_rename(const char *oldname, void *arg)
2625 {
2626         int delta = *(int *)arg;
2627
2628         if (strlen(oldname) + delta >= MAXNAMELEN)
2629                 return (ENAMETOOLONG);
2630
2631         return (0);
2632 }
2633
2634 #pragma weak dmu_objset_rename = dsl_dataset_rename
2635 int
2636 dsl_dataset_rename(char *oldname, const char *newname, int flags)
2637 {
2638         dsl_dir_t *dd;
2639         dsl_dataset_t *ds;
2640         const char *tail;
2641         int err;
2642
2643         err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2644         if (err)
2645                 return (err);
2646
2647         if (tail == NULL) {
2648                 int delta = strlen(newname) - strlen(oldname);
2649
2650                 /* if we're growing, validate child name lengths */
2651                 if (delta > 0)
2652                         err = dmu_objset_find(oldname, dsl_valid_rename,
2653                             &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2654
2655                 if (err == 0)
2656                         err = dsl_dir_rename(dd, newname, flags);
2657                 dsl_dir_close(dd, FTAG);
2658                 return (err);
2659         }
2660
2661         if (tail[0] != '@') {
2662                 /* the name ended in a nonexistent component */
2663                 dsl_dir_close(dd, FTAG);
2664                 return (ENOENT);
2665         }
2666
2667         dsl_dir_close(dd, FTAG);
2668
2669         /* new name must be snapshot in same filesystem */
2670         tail = strchr(newname, '@');
2671         if (tail == NULL)
2672                 return (EINVAL);
2673         tail++;
2674         if (strncmp(oldname, newname, tail - newname) != 0)
2675                 return (EXDEV);
2676
2677         if (flags & ZFS_RENAME_RECURSIVE) {
2678                 err = dsl_recursive_rename(oldname, newname);
2679         } else {
2680                 err = dsl_dataset_hold(oldname, FTAG, &ds);
2681                 if (err)
2682                         return (err);
2683
2684                 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2685                     dsl_dataset_snapshot_rename_check,
2686                     dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2687
2688                 dsl_dataset_rele(ds, FTAG);
2689         }
2690
2691         return (err);
2692 }
2693
2694 struct promotenode {
2695         list_node_t link;
2696         dsl_dataset_t *ds;
2697 };
2698
2699 struct promotearg {
2700         list_t shared_snaps, origin_snaps, clone_snaps;
2701         dsl_dataset_t *origin_origin;
2702         uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2703         char *err_ds;
2704 };
2705
2706 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2707 static boolean_t snaplist_unstable(list_t *l);
2708
2709 static int
2710 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2711 {
2712         dsl_dataset_t *hds = arg1;
2713         struct promotearg *pa = arg2;
2714         struct promotenode *snap = list_head(&pa->shared_snaps);
2715         dsl_dataset_t *origin_ds = snap->ds;
2716         int err;
2717         uint64_t unused;
2718
2719         /* Check that it is a real clone */
2720         if (!dsl_dir_is_clone(hds->ds_dir))
2721                 return (EINVAL);
2722
2723         /* Since this is so expensive, don't do the preliminary check */
2724         if (!dmu_tx_is_syncing(tx))
2725                 return (0);
2726
2727         if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2728                 return (EXDEV);
2729
2730         /* compute origin's new unique space */
2731         snap = list_tail(&pa->clone_snaps);
2732         ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2733         dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2734             origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2735             &pa->unique, &unused, &unused);
2736
2737         /*
2738          * Walk the snapshots that we are moving
2739          *
2740          * Compute space to transfer.  Consider the incremental changes
2741          * to used for each snapshot:
2742          * (my used) = (prev's used) + (blocks born) - (blocks killed)
2743          * So each snapshot gave birth to:
2744          * (blocks born) = (my used) - (prev's used) + (blocks killed)
2745          * So a sequence would look like:
2746          * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2747          * Which simplifies to:
2748          * uN + kN + kN-1 + ... + k1 + k0
2749          * Note however, if we stop before we reach the ORIGIN we get:
2750          * uN + kN + kN-1 + ... + kM - uM-1
2751          */
2752         pa->used = origin_ds->ds_phys->ds_referenced_bytes;
2753         pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2754         pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2755         for (snap = list_head(&pa->shared_snaps); snap;
2756             snap = list_next(&pa->shared_snaps, snap)) {
2757                 uint64_t val, dlused, dlcomp, dluncomp;
2758                 dsl_dataset_t *ds = snap->ds;
2759
2760                 /* Check that the snapshot name does not conflict */
2761                 VERIFY(0 == dsl_dataset_get_snapname(ds));
2762                 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2763                 if (err == 0) {
2764                         err = EEXIST;
2765                         goto out;
2766                 }
2767                 if (err != ENOENT)
2768                         goto out;
2769
2770                 /* The very first snapshot does not have a deadlist */
2771                 if (ds->ds_phys->ds_prev_snap_obj == 0)
2772                         continue;
2773
2774                 dsl_deadlist_space(&ds->ds_deadlist,
2775                     &dlused, &dlcomp, &dluncomp);
2776                 pa->used += dlused;
2777                 pa->comp += dlcomp;
2778                 pa->uncomp += dluncomp;
2779         }
2780
2781         /*
2782          * If we are a clone of a clone then we never reached ORIGIN,
2783          * so we need to subtract out the clone origin's used space.
2784          */
2785         if (pa->origin_origin) {
2786                 pa->used -= pa->origin_origin->ds_phys->ds_referenced_bytes;
2787                 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2788                 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2789         }
2790
2791         /* Check that there is enough space here */
2792         err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2793             pa->used);
2794         if (err)
2795                 return (err);
2796
2797         /*
2798          * Compute the amounts of space that will be used by snapshots
2799          * after the promotion (for both origin and clone).  For each,
2800          * it is the amount of space that will be on all of their
2801          * deadlists (that was not born before their new origin).
2802          */
2803         if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2804                 uint64_t space;
2805
2806                 /*
2807                  * Note, typically this will not be a clone of a clone,
2808                  * so dd_origin_txg will be < TXG_INITIAL, so
2809                  * these snaplist_space() -> dsl_deadlist_space_range()
2810                  * calls will be fast because they do not have to
2811                  * iterate over all bps.
2812                  */
2813                 snap = list_head(&pa->origin_snaps);
2814                 err = snaplist_space(&pa->shared_snaps,
2815                     snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2816                 if (err)
2817                         return (err);
2818
2819                 err = snaplist_space(&pa->clone_snaps,
2820                     snap->ds->ds_dir->dd_origin_txg, &space);
2821                 if (err)
2822                         return (err);
2823                 pa->cloneusedsnap += space;
2824         }
2825         if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2826                 err = snaplist_space(&pa->origin_snaps,
2827                     origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2828                 if (err)
2829                         return (err);
2830         }
2831
2832         return (0);
2833 out:
2834         pa->err_ds =  snap->ds->ds_snapname;
2835         return (err);
2836 }
2837
2838 static void
2839 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2840 {
2841         dsl_dataset_t *hds = arg1;
2842         struct promotearg *pa = arg2;
2843         struct promotenode *snap = list_head(&pa->shared_snaps);
2844         dsl_dataset_t *origin_ds = snap->ds;
2845         dsl_dataset_t *origin_head;
2846         dsl_dir_t *dd = hds->ds_dir;
2847         dsl_pool_t *dp = hds->ds_dir->dd_pool;
2848         dsl_dir_t *odd = NULL;
2849         uint64_t oldnext_obj;
2850         int64_t delta;
2851
2852         ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2853
2854         snap = list_head(&pa->origin_snaps);
2855         origin_head = snap->ds;
2856
2857         /*
2858          * We need to explicitly open odd, since origin_ds's dd will be
2859          * changing.
2860          */
2861         VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2862             NULL, FTAG, &odd));
2863
2864         /* change origin's next snap */
2865         dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2866         oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2867         snap = list_tail(&pa->clone_snaps);
2868         ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2869         origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2870
2871         /* change the origin's next clone */
2872         if (origin_ds->ds_phys->ds_next_clones_obj) {
2873                 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2874                 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2875                     origin_ds->ds_phys->ds_next_clones_obj,
2876                     oldnext_obj, tx));
2877         }
2878
2879         /* change origin */
2880         dmu_buf_will_dirty(dd->dd_dbuf, tx);
2881         ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2882         dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2883         dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2884         dmu_buf_will_dirty(odd->dd_dbuf, tx);
2885         odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2886         origin_head->ds_dir->dd_origin_txg =
2887             origin_ds->ds_phys->ds_creation_txg;
2888
2889         /* change dd_clone entries */
2890         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2891                 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2892                     odd->dd_phys->dd_clones, hds->ds_object, tx));
2893                 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2894                     pa->origin_origin->ds_dir->dd_phys->dd_clones,
2895                     hds->ds_object, tx));
2896
2897                 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2898                     pa->origin_origin->ds_dir->dd_phys->dd_clones,
2899                     origin_head->ds_object, tx));
2900                 if (dd->dd_phys->dd_clones == 0) {
2901                         dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2902                             DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2903                 }
2904                 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2905                     dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2906
2907         }
2908
2909         /* move snapshots to this dir */
2910         for (snap = list_head(&pa->shared_snaps); snap;
2911             snap = list_next(&pa->shared_snaps, snap)) {
2912                 dsl_dataset_t *ds = snap->ds;
2913
2914                 /* unregister props as dsl_dir is changing */
2915                 if (ds->ds_objset) {
2916                         dmu_objset_evict(ds->ds_objset);
2917                         ds->ds_objset = NULL;
2918                 }
2919                 /* move snap name entry */
2920                 VERIFY(0 == dsl_dataset_get_snapname(ds));
2921                 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2922                     ds->ds_snapname, tx));
2923                 VERIFY(0 == zap_add(dp->dp_meta_objset,
2924                     hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2925                     8, 1, &ds->ds_object, tx));
2926
2927                 /* change containing dsl_dir */
2928                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2929                 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2930                 ds->ds_phys->ds_dir_obj = dd->dd_object;
2931                 ASSERT3P(ds->ds_dir, ==, odd);
2932                 dsl_dir_close(ds->ds_dir, ds);
2933                 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2934                     NULL, ds, &ds->ds_dir));
2935
2936                 /* move any clone references */
2937                 if (ds->ds_phys->ds_next_clones_obj &&
2938                     spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2939                         zap_cursor_t zc;
2940                         zap_attribute_t za;
2941
2942                         for (zap_cursor_init(&zc, dp->dp_meta_objset,
2943                             ds->ds_phys->ds_next_clones_obj);
2944                             zap_cursor_retrieve(&zc, &za) == 0;
2945                             zap_cursor_advance(&zc)) {
2946                                 dsl_dataset_t *cnds;
2947                                 uint64_t o;
2948
2949                                 if (za.za_first_integer == oldnext_obj) {
2950                                         /*
2951                                          * We've already moved the
2952                                          * origin's reference.
2953                                          */
2954                                         continue;
2955                                 }
2956
2957                                 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2958                                     za.za_first_integer, FTAG, &cnds));
2959                                 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2960
2961                                 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2962                                     odd->dd_phys->dd_clones, o, tx), ==, 0);
2963                                 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2964                                     dd->dd_phys->dd_clones, o, tx), ==, 0);
2965                                 dsl_dataset_rele(cnds, FTAG);
2966                         }
2967                         zap_cursor_fini(&zc);
2968                 }
2969
2970                 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2971         }
2972
2973         /*
2974          * Change space accounting.
2975          * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2976          * both be valid, or both be 0 (resulting in delta == 0).  This
2977          * is true for each of {clone,origin} independently.
2978          */
2979
2980         delta = pa->cloneusedsnap -
2981             dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2982         ASSERT3S(delta, >=, 0);
2983         ASSERT3U(pa->used, >=, delta);
2984         dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2985         dsl_dir_diduse_space(dd, DD_USED_HEAD,
2986             pa->used - delta, pa->comp, pa->uncomp, tx);
2987
2988         delta = pa->originusedsnap -
2989             odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2990         ASSERT3S(delta, <=, 0);
2991         ASSERT3U(pa->used, >=, -delta);
2992         dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2993         dsl_dir_diduse_space(odd, DD_USED_HEAD,
2994             -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2995
2996         origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2997
2998         /* log history record */
2999         spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
3000             "dataset = %llu", hds->ds_object);
3001
3002         dsl_dir_close(odd, FTAG);
3003 }
3004
3005 static char *snaplist_tag = "snaplist";
3006 /*
3007  * Make a list of dsl_dataset_t's for the snapshots between first_obj
3008  * (exclusive) and last_obj (inclusive).  The list will be in reverse
3009  * order (last_obj will be the list_head()).  If first_obj == 0, do all
3010  * snapshots back to this dataset's origin.
3011  */
3012 static int
3013 snaplist_make(dsl_pool_t *dp, boolean_t own,
3014     uint64_t first_obj, uint64_t last_obj, list_t *l)
3015 {
3016         uint64_t obj = last_obj;
3017
3018         ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
3019
3020         list_create(l, sizeof (struct promotenode),
3021             offsetof(struct promotenode, link));
3022
3023         while (obj != first_obj) {
3024                 dsl_dataset_t *ds;
3025                 struct promotenode *snap;
3026                 int err;
3027
3028                 if (own) {
3029                         err = dsl_dataset_own_obj(dp, obj,
3030                             0, snaplist_tag, &ds);
3031                         if (err == 0)
3032                                 dsl_dataset_make_exclusive(ds, snaplist_tag);
3033                 } else {
3034                         err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
3035                 }
3036                 if (err == ENOENT) {
3037                         /* lost race with snapshot destroy */
3038                         struct promotenode *last = list_tail(l);
3039                         ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
3040                         obj = last->ds->ds_phys->ds_prev_snap_obj;
3041                         continue;
3042                 } else if (err) {
3043                         return (err);
3044                 }
3045
3046                 if (first_obj == 0)
3047                         first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
3048
3049                 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
3050                 snap->ds = ds;
3051                 list_insert_tail(l, snap);
3052                 obj = ds->ds_phys->ds_prev_snap_obj;
3053         }
3054
3055         return (0);
3056 }
3057
3058 static int
3059 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3060 {
3061         struct promotenode *snap;
3062
3063         *spacep = 0;
3064         for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3065                 uint64_t used, comp, uncomp;
3066                 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3067                     mintxg, UINT64_MAX, &used, &comp, &uncomp);
3068                 *spacep += used;
3069         }
3070         return (0);
3071 }
3072
3073 static void
3074 snaplist_destroy(list_t *l, boolean_t own)
3075 {
3076         struct promotenode *snap;
3077
3078         if (!l || !list_link_active(&l->list_head))
3079                 return;
3080
3081         while ((snap = list_tail(l)) != NULL) {
3082                 list_remove(l, snap);
3083                 if (own)
3084                         dsl_dataset_disown(snap->ds, snaplist_tag);
3085                 else
3086                         dsl_dataset_rele(snap->ds, snaplist_tag);
3087                 kmem_free(snap, sizeof (struct promotenode));
3088         }
3089         list_destroy(l);
3090 }
3091
3092 /*
3093  * Promote a clone.  Nomenclature note:
3094  * "clone" or "cds": the original clone which is being promoted
3095  * "origin" or "ods": the snapshot which is originally clone's origin
3096  * "origin head" or "ohds": the dataset which is the head
3097  * (filesystem/volume) for the origin
3098  * "origin origin": the origin of the origin's filesystem (typically
3099  * NULL, indicating that the clone is not a clone of a clone).
3100  */
3101 int
3102 dsl_dataset_promote(const char *name, char *conflsnap)
3103 {
3104         dsl_dataset_t *ds;
3105         dsl_dir_t *dd;
3106         dsl_pool_t *dp;
3107         dmu_object_info_t doi;
3108         struct promotearg pa = { 0 };
3109         struct promotenode *snap;
3110         int err;
3111
3112         err = dsl_dataset_hold(name, FTAG, &ds);
3113         if (err)
3114                 return (err);
3115         dd = ds->ds_dir;
3116         dp = dd->dd_pool;
3117
3118         err = dmu_object_info(dp->dp_meta_objset,
3119             ds->ds_phys->ds_snapnames_zapobj, &doi);
3120         if (err) {
3121                 dsl_dataset_rele(ds, FTAG);
3122                 return (err);
3123         }
3124
3125         if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
3126                 dsl_dataset_rele(ds, FTAG);
3127                 return (EINVAL);
3128         }
3129
3130         /*
3131          * We are going to inherit all the snapshots taken before our
3132          * origin (i.e., our new origin will be our parent's origin).
3133          * Take ownership of them so that we can rename them into our
3134          * namespace.
3135          */
3136         rw_enter(&dp->dp_config_rwlock, RW_READER);
3137
3138         err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
3139             &pa.shared_snaps);
3140         if (err != 0)
3141                 goto out;
3142
3143         err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
3144         if (err != 0)
3145                 goto out;
3146
3147         snap = list_head(&pa.shared_snaps);
3148         ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
3149         err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
3150             snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
3151         if (err != 0)
3152                 goto out;
3153
3154         if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
3155                 err = dsl_dataset_hold_obj(dp,
3156                     snap->ds->ds_dir->dd_phys->dd_origin_obj,
3157                     FTAG, &pa.origin_origin);
3158                 if (err != 0)
3159                         goto out;
3160         }
3161
3162 out:
3163         rw_exit(&dp->dp_config_rwlock);
3164
3165         /*
3166          * Add in 128x the snapnames zapobj size, since we will be moving
3167          * a bunch of snapnames to the promoted ds, and dirtying their
3168          * bonus buffers.
3169          */
3170         if (err == 0) {
3171                 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
3172                     dsl_dataset_promote_sync, ds, &pa,
3173                     2 + 2 * doi.doi_physical_blocks_512);
3174                 if (err && pa.err_ds && conflsnap)
3175                         (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3176         }
3177
3178         snaplist_destroy(&pa.shared_snaps, B_TRUE);
3179         snaplist_destroy(&pa.clone_snaps, B_FALSE);
3180         snaplist_destroy(&pa.origin_snaps, B_FALSE);
3181         if (pa.origin_origin)
3182                 dsl_dataset_rele(pa.origin_origin, FTAG);
3183         dsl_dataset_rele(ds, FTAG);
3184         return (err);
3185 }
3186
3187 struct cloneswaparg {
3188         dsl_dataset_t *cds; /* clone dataset */
3189         dsl_dataset_t *ohds; /* origin's head dataset */
3190         boolean_t force;
3191         int64_t unused_refres_delta; /* change in unconsumed refreservation */
3192 };
3193
3194 /* ARGSUSED */
3195 static int
3196 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3197 {
3198         struct cloneswaparg *csa = arg1;
3199
3200         /* they should both be heads */
3201         if (dsl_dataset_is_snapshot(csa->cds) ||
3202             dsl_dataset_is_snapshot(csa->ohds))
3203                 return (EINVAL);
3204
3205         /* the branch point should be just before them */
3206         if (csa->cds->ds_prev != csa->ohds->ds_prev)
3207                 return (EINVAL);
3208
3209         /* cds should be the clone (unless they are unrelated) */
3210         if (csa->cds->ds_prev != NULL &&
3211             csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3212             csa->ohds->ds_object !=
3213             csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3214                 return (EINVAL);
3215
3216         /* the clone should be a child of the origin */
3217         if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3218                 return (EINVAL);
3219
3220         /* ohds shouldn't be modified unless 'force' */
3221         if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3222                 return (ETXTBSY);
3223
3224         /* adjust amount of any unconsumed refreservation */
3225         csa->unused_refres_delta =
3226             (int64_t)MIN(csa->ohds->ds_reserved,
3227             csa->ohds->ds_phys->ds_unique_bytes) -
3228             (int64_t)MIN(csa->ohds->ds_reserved,
3229             csa->cds->ds_phys->ds_unique_bytes);
3230
3231         if (csa->unused_refres_delta > 0 &&
3232             csa->unused_refres_delta >
3233             dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3234                 return (ENOSPC);
3235
3236         if (csa->ohds->ds_quota != 0 &&
3237             csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3238                 return (EDQUOT);
3239
3240         return (0);
3241 }
3242
3243 /* ARGSUSED */
3244 static void
3245 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3246 {
3247         struct cloneswaparg *csa = arg1;
3248         dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3249
3250         ASSERT(csa->cds->ds_reserved == 0);
3251         ASSERT(csa->ohds->ds_quota == 0 ||
3252             csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3253
3254         dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3255         dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3256
3257         if (csa->cds->ds_objset != NULL) {
3258                 dmu_objset_evict(csa->cds->ds_objset);
3259                 csa->cds->ds_objset = NULL;
3260         }
3261
3262         if (csa->ohds->ds_objset != NULL) {
3263                 dmu_objset_evict(csa->ohds->ds_objset);
3264                 csa->ohds->ds_objset = NULL;
3265         }
3266
3267         /*
3268          * Reset origin's unique bytes, if it exists.
3269          */
3270         if (csa->cds->ds_prev) {
3271                 dsl_dataset_t *origin = csa->cds->ds_prev;
3272                 uint64_t comp, uncomp;
3273
3274                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3275                 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3276                     origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3277                     &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3278         }
3279
3280         /* swap blkptrs */
3281         {
3282                 blkptr_t tmp;
3283                 tmp = csa->ohds->ds_phys->ds_bp;
3284                 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3285                 csa->cds->ds_phys->ds_bp = tmp;
3286         }
3287
3288         /* set dd_*_bytes */
3289         {
3290                 int64_t dused, dcomp, duncomp;
3291                 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3292                 uint64_t odl_used, odl_comp, odl_uncomp;
3293
3294                 ASSERT3U(csa->cds->ds_dir->dd_phys->
3295                     dd_used_breakdown[DD_USED_SNAP], ==, 0);
3296
3297                 dsl_deadlist_space(&csa->cds->ds_deadlist,
3298                     &cdl_used, &cdl_comp, &cdl_uncomp);
3299                 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3300                     &odl_used, &odl_comp, &odl_uncomp);
3301
3302                 dused = csa->cds->ds_phys->ds_referenced_bytes + cdl_used -
3303                     (csa->ohds->ds_phys->ds_referenced_bytes + odl_used);
3304                 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3305                     (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3306                 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3307                     cdl_uncomp -
3308                     (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3309
3310                 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3311                     dused, dcomp, duncomp, tx);
3312                 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3313                     -dused, -dcomp, -duncomp, tx);
3314
3315                 /*
3316                  * The difference in the space used by snapshots is the
3317                  * difference in snapshot space due to the head's
3318                  * deadlist (since that's the only thing that's
3319                  * changing that affects the snapused).
3320                  */
3321                 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3322                     csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3323                     &cdl_used, &cdl_comp, &cdl_uncomp);
3324                 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3325                     csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3326                     &odl_used, &odl_comp, &odl_uncomp);
3327                 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3328                     DD_USED_HEAD, DD_USED_SNAP, tx);
3329         }
3330
3331         /* swap ds_*_bytes */
3332         SWITCH64(csa->ohds->ds_phys->ds_referenced_bytes,
3333             csa->cds->ds_phys->ds_referenced_bytes);
3334         SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3335             csa->cds->ds_phys->ds_compressed_bytes);
3336         SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3337             csa->cds->ds_phys->ds_uncompressed_bytes);
3338         SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3339             csa->cds->ds_phys->ds_unique_bytes);
3340
3341         /* apply any parent delta for change in unconsumed refreservation */
3342         dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3343             csa->unused_refres_delta, 0, 0, tx);
3344
3345         /*
3346          * Swap deadlists.
3347          */
3348         dsl_deadlist_close(&csa->cds->ds_deadlist);
3349         dsl_deadlist_close(&csa->ohds->ds_deadlist);
3350         SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3351             csa->cds->ds_phys->ds_deadlist_obj);
3352         dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3353             csa->cds->ds_phys->ds_deadlist_obj);
3354         dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3355             csa->ohds->ds_phys->ds_deadlist_obj);
3356
3357         dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3358 }
3359
3360 /*
3361  * Swap 'clone' with its origin head datasets.  Used at the end of "zfs
3362  * recv" into an existing fs to swizzle the file system to the new
3363  * version, and by "zfs rollback".  Can also be used to swap two
3364  * independent head datasets if neither has any snapshots.
3365  */
3366 int
3367 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3368     boolean_t force)
3369 {
3370         struct cloneswaparg csa;
3371         int error;
3372
3373         ASSERT(clone->ds_owner);
3374         ASSERT(origin_head->ds_owner);
3375 retry:
3376         /*
3377          * Need exclusive access for the swap. If we're swapping these
3378          * datasets back after an error, we already hold the locks.
3379          */
3380         if (!RW_WRITE_HELD(&clone->ds_rwlock))
3381                 rw_enter(&clone->ds_rwlock, RW_WRITER);
3382         if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3383             !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3384                 rw_exit(&clone->ds_rwlock);
3385                 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3386                 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3387                         rw_exit(&origin_head->ds_rwlock);
3388                         goto retry;
3389                 }
3390         }
3391         csa.cds = clone;
3392         csa.ohds = origin_head;
3393         csa.force = force;
3394         error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3395             dsl_dataset_clone_swap_check,
3396             dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3397         return (error);
3398 }
3399
3400 /*
3401  * Given a pool name and a dataset object number in that pool,
3402  * return the name of that dataset.
3403  */
3404 int
3405 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3406 {
3407         spa_t *spa;
3408         dsl_pool_t *dp;
3409         dsl_dataset_t *ds;
3410         int error;
3411
3412         if ((error = spa_open(pname, &spa, FTAG)) != 0)
3413                 return (error);
3414         dp = spa_get_dsl(spa);
3415         rw_enter(&dp->dp_config_rwlock, RW_READER);
3416         if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3417                 dsl_dataset_name(ds, buf);
3418                 dsl_dataset_rele(ds, FTAG);
3419         }
3420         rw_exit(&dp->dp_config_rwlock);
3421         spa_close(spa, FTAG);
3422
3423         return (error);
3424 }
3425
3426 int
3427 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3428     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3429 {
3430         int error = 0;
3431
3432         ASSERT3S(asize, >, 0);
3433
3434         /*
3435          * *ref_rsrv is the portion of asize that will come from any
3436          * unconsumed refreservation space.
3437          */
3438         *ref_rsrv = 0;
3439
3440         mutex_enter(&ds->ds_lock);
3441         /*
3442          * Make a space adjustment for reserved bytes.
3443          */
3444         if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3445                 ASSERT3U(*used, >=,
3446                     ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3447                 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3448                 *ref_rsrv =
3449                     asize - MIN(asize, parent_delta(ds, asize + inflight));
3450         }
3451
3452         if (!check_quota || ds->ds_quota == 0) {
3453                 mutex_exit(&ds->ds_lock);
3454                 return (0);
3455         }
3456         /*
3457          * If they are requesting more space, and our current estimate
3458          * is over quota, they get to try again unless the actual
3459          * on-disk is over quota and there are no pending changes (which
3460          * may free up space for us).
3461          */
3462         if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
3463                 if (inflight > 0 ||
3464                     ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
3465                         error = ERESTART;
3466                 else
3467                         error = EDQUOT;
3468         }
3469         mutex_exit(&ds->ds_lock);
3470
3471         return (error);
3472 }
3473
3474 /* ARGSUSED */
3475 static int
3476 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3477 {
3478         dsl_dataset_t *ds = arg1;
3479         dsl_prop_setarg_t *psa = arg2;
3480         int err;
3481
3482         if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3483                 return (ENOTSUP);
3484
3485         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3486                 return (err);
3487
3488         if (psa->psa_effective_value == 0)
3489                 return (0);
3490
3491         if (psa->psa_effective_value < ds->ds_phys->ds_referenced_bytes ||
3492             psa->psa_effective_value < ds->ds_reserved)
3493                 return (ENOSPC);
3494
3495         return (0);
3496 }
3497
3498 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3499
3500 void
3501 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3502 {
3503         dsl_dataset_t *ds = arg1;
3504         dsl_prop_setarg_t *psa = arg2;
3505         uint64_t effective_value = psa->psa_effective_value;
3506
3507         dsl_prop_set_sync(ds, psa, tx);
3508         DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3509
3510         if (ds->ds_quota != effective_value) {
3511                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3512                 ds->ds_quota = effective_value;
3513
3514                 spa_history_log_internal(LOG_DS_REFQUOTA,
3515                     ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ",
3516                     (longlong_t)ds->ds_quota, ds->ds_object);
3517         }
3518 }
3519
3520 int
3521 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3522 {
3523         dsl_dataset_t *ds;
3524         dsl_prop_setarg_t psa;
3525         int err;
3526
3527         dsl_prop_setarg_init_uint64(&psa, "refquota", source, &quota);
3528
3529         err = dsl_dataset_hold(dsname, FTAG, &ds);
3530         if (err)
3531                 return (err);
3532
3533         /*
3534          * If someone removes a file, then tries to set the quota, we
3535          * want to make sure the file freeing takes effect.
3536          */
3537         txg_wait_open(ds->ds_dir->dd_pool, 0);
3538
3539         err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3540             dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3541             ds, &psa, 0);
3542
3543         dsl_dataset_rele(ds, FTAG);
3544         return (err);
3545 }
3546
3547 static int
3548 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3549 {
3550         dsl_dataset_t *ds = arg1;
3551         dsl_prop_setarg_t *psa = arg2;
3552         uint64_t effective_value;
3553         uint64_t unique;
3554         int err;
3555
3556         if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3557             SPA_VERSION_REFRESERVATION)
3558                 return (ENOTSUP);
3559
3560         if (dsl_dataset_is_snapshot(ds))
3561                 return (EINVAL);
3562
3563         if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3564                 return (err);
3565
3566         effective_value = psa->psa_effective_value;
3567
3568         /*
3569          * If we are doing the preliminary check in open context, the
3570          * space estimates may be inaccurate.
3571          */
3572         if (!dmu_tx_is_syncing(tx))
3573                 return (0);
3574
3575         mutex_enter(&ds->ds_lock);
3576         if (!DS_UNIQUE_IS_ACCURATE(ds))
3577                 dsl_dataset_recalc_head_uniq(ds);
3578         unique = ds->ds_phys->ds_unique_bytes;
3579         mutex_exit(&ds->ds_lock);
3580
3581         if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3582                 uint64_t delta = MAX(unique, effective_value) -
3583                     MAX(unique, ds->ds_reserved);
3584
3585                 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3586                         return (ENOSPC);
3587                 if (ds->ds_quota > 0 &&
3588                     effective_value > ds->ds_quota)
3589                         return (ENOSPC);
3590         }
3591
3592         return (0);
3593 }
3594
3595 static void
3596 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3597 {
3598         dsl_dataset_t *ds = arg1;
3599         dsl_prop_setarg_t *psa = arg2;
3600         uint64_t effective_value = psa->psa_effective_value;
3601         uint64_t unique;
3602         int64_t delta;
3603
3604         dsl_prop_set_sync(ds, psa, tx);
3605         DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3606
3607         dmu_buf_will_dirty(ds->ds_dbuf, tx);
3608
3609         mutex_enter(&ds->ds_dir->dd_lock);
3610         mutex_enter(&ds->ds_lock);
3611         ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3612         unique = ds->ds_phys->ds_unique_bytes;
3613         delta = MAX(0, (int64_t)(effective_value - unique)) -
3614             MAX(0, (int64_t)(ds->ds_reserved - unique));
3615         ds->ds_reserved = effective_value;
3616         mutex_exit(&ds->ds_lock);
3617
3618         dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3619         mutex_exit(&ds->ds_dir->dd_lock);
3620
3621         spa_history_log_internal(LOG_DS_REFRESERV,
3622             ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu",
3623             (longlong_t)effective_value, ds->ds_object);
3624 }
3625
3626 int
3627 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3628     uint64_t reservation)
3629 {
3630         dsl_dataset_t *ds;
3631         dsl_prop_setarg_t psa;
3632         int err;
3633
3634         dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3635             &reservation);
3636
3637         err = dsl_dataset_hold(dsname, FTAG, &ds);
3638         if (err)
3639                 return (err);
3640
3641         err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3642             dsl_dataset_set_reservation_check,
3643             dsl_dataset_set_reservation_sync, ds, &psa, 0);
3644
3645         dsl_dataset_rele(ds, FTAG);
3646         return (err);
3647 }
3648
3649 typedef struct zfs_hold_cleanup_arg {
3650         dsl_pool_t *dp;
3651         uint64_t dsobj;
3652         char htag[MAXNAMELEN];
3653 } zfs_hold_cleanup_arg_t;
3654
3655 static void
3656 dsl_dataset_user_release_onexit(void *arg)
3657 {
3658         zfs_hold_cleanup_arg_t *ca = arg;
3659
3660         (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3661             B_TRUE);
3662         kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3663 }
3664
3665 void
3666 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3667     minor_t minor)
3668 {
3669         zfs_hold_cleanup_arg_t *ca;
3670
3671         ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3672         ca->dp = ds->ds_dir->dd_pool;
3673         ca->dsobj = ds->ds_object;
3674         (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3675         VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3676             dsl_dataset_user_release_onexit, ca, NULL));
3677 }
3678
3679 /*
3680  * If you add new checks here, you may need to add
3681  * additional checks to the "temporary" case in
3682  * snapshot_check() in dmu_objset.c.
3683  */
3684 static int
3685 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3686 {
3687         dsl_dataset_t *ds = arg1;
3688         struct dsl_ds_holdarg *ha = arg2;
3689         char *htag = ha->htag;
3690         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3691         int error = 0;
3692
3693         if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3694                 return (ENOTSUP);
3695
3696         if (!dsl_dataset_is_snapshot(ds))
3697                 return (EINVAL);
3698
3699         /* tags must be unique */
3700         mutex_enter(&ds->ds_lock);
3701         if (ds->ds_phys->ds_userrefs_obj) {
3702                 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3703                     8, 1, tx);
3704                 if (error == 0)
3705                         error = EEXIST;
3706                 else if (error == ENOENT)
3707                         error = 0;
3708         }
3709         mutex_exit(&ds->ds_lock);
3710
3711         if (error == 0 && ha->temphold &&
3712             strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3713                 error = E2BIG;
3714
3715         return (error);
3716 }
3717
3718 void
3719 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3720 {
3721         dsl_dataset_t *ds = arg1;
3722         struct dsl_ds_holdarg *ha = arg2;
3723         char *htag = ha->htag;
3724         dsl_pool_t *dp = ds->ds_dir->dd_pool;
3725         objset_t *mos = dp->dp_meta_objset;
3726         uint64_t now = gethrestime_sec();
3727         uint64_t zapobj;
3728
3729         mutex_enter(&ds->ds_lock);
3730         if (ds->ds_phys->ds_userrefs_obj == 0) {
3731                 /*
3732                  * This is the first user hold for this dataset.  Create
3733                  * the userrefs zap object.
3734                  */
3735                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3736                 zapobj = ds->ds_phys->ds_userrefs_obj =
3737                     zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3738         } else {
3739                 zapobj = ds->ds_phys->ds_userrefs_obj;
3740         }
3741         ds->ds_userrefs++;
3742         mutex_exit(&ds->ds_lock);
3743
3744         VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3745
3746         if (ha->temphold) {
3747                 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3748                     htag, &now, tx));
3749         }
3750
3751         spa_history_log_internal(LOG_DS_USER_HOLD,
3752             dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag,
3753             (int)ha->temphold, ds->ds_object);
3754 }
3755
3756 static int
3757 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3758 {
3759         struct dsl_ds_holdarg *ha = arg;
3760         dsl_dataset_t *ds;
3761         int error;
3762         char *name;
3763
3764         /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3765         name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3766         error = dsl_dataset_hold(name, ha->dstg, &ds);
3767         strfree(name);
3768         if (error == 0) {
3769                 ha->gotone = B_TRUE;
3770                 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3771                     dsl_dataset_user_hold_sync, ds, ha, 0);
3772         } else if (error == ENOENT && ha->recursive) {
3773                 error = 0;
3774         } else {
3775                 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3776         }
3777         return (error);
3778 }
3779
3780 int
3781 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3782     boolean_t temphold)
3783 {
3784         struct dsl_ds_holdarg *ha;
3785         int error;
3786
3787         ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3788         ha->htag = htag;
3789         ha->temphold = temphold;
3790         error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3791             dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3792             ds, ha, 0);
3793         kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3794
3795         return (error);
3796 }
3797
3798 int
3799 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3800     boolean_t recursive, boolean_t temphold, int cleanup_fd)
3801 {
3802         struct dsl_ds_holdarg *ha;
3803         dsl_sync_task_t *dst;
3804         spa_t *spa;
3805         int error;
3806         minor_t minor = 0;
3807
3808         if (cleanup_fd != -1) {
3809                 /* Currently we only support cleanup-on-exit of tempholds. */
3810                 if (!temphold)
3811                         return (EINVAL);
3812                 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3813                 if (error)
3814                         return (error);
3815         }
3816
3817         ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3818
3819         (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3820
3821         error = spa_open(dsname, &spa, FTAG);
3822         if (error) {
3823                 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3824                 if (cleanup_fd != -1)
3825                         zfs_onexit_fd_rele(cleanup_fd);
3826                 return (error);
3827         }
3828
3829         ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3830         ha->htag = htag;
3831         ha->snapname = snapname;
3832         ha->recursive = recursive;
3833         ha->temphold = temphold;
3834
3835         if (recursive) {
3836                 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3837                     ha, DS_FIND_CHILDREN);
3838         } else {
3839                 error = dsl_dataset_user_hold_one(dsname, ha);
3840         }
3841         if (error == 0)
3842                 error = dsl_sync_task_group_wait(ha->dstg);
3843
3844         for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3845             dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3846                 dsl_dataset_t *ds = dst->dst_arg1;
3847
3848                 if (dst->dst_err) {
3849                         dsl_dataset_name(ds, ha->failed);
3850                         *strchr(ha->failed, '@') = '\0';
3851                 } else if (error == 0 && minor != 0 && temphold) {
3852                         /*
3853                          * If this hold is to be released upon process exit,
3854                          * register that action now.
3855                          */
3856                         dsl_register_onexit_hold_cleanup(ds, htag, minor);
3857                 }
3858                 dsl_dataset_rele(ds, ha->dstg);
3859         }
3860
3861         if (error == 0 && recursive && !ha->gotone)
3862                 error = ENOENT;
3863
3864         if (error)
3865                 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3866
3867         dsl_sync_task_group_destroy(ha->dstg);
3868
3869         kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3870         spa_close(spa, FTAG);
3871         if (cleanup_fd != -1)
3872                 zfs_onexit_fd_rele(cleanup_fd);
3873         return (error);
3874 }
3875
3876 struct dsl_ds_releasearg {
3877         dsl_dataset_t *ds;
3878         const char *htag;
3879         boolean_t own;          /* do we own or just hold ds? */
3880 };
3881
3882 static int
3883 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3884     boolean_t *might_destroy)
3885 {
3886         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3887         uint64_t zapobj;
3888         uint64_t tmp;
3889         int error;
3890
3891         *might_destroy = B_FALSE;
3892
3893         mutex_enter(&ds->ds_lock);
3894         zapobj = ds->ds_phys->ds_userrefs_obj;
3895         if (zapobj == 0) {
3896                 /* The tag can't possibly exist */
3897                 mutex_exit(&ds->ds_lock);
3898                 return (ESRCH);
3899         }
3900
3901         /* Make sure the tag exists */
3902         error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3903         if (error) {
3904                 mutex_exit(&ds->ds_lock);
3905                 if (error == ENOENT)
3906                         error = ESRCH;
3907                 return (error);
3908         }
3909
3910         if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3911             DS_IS_DEFER_DESTROY(ds))
3912                 *might_destroy = B_TRUE;
3913
3914         mutex_exit(&ds->ds_lock);
3915         return (0);
3916 }
3917
3918 static int
3919 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3920 {
3921         struct dsl_ds_releasearg *ra = arg1;
3922         dsl_dataset_t *ds = ra->ds;
3923         boolean_t might_destroy;
3924         int error;
3925
3926         if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3927                 return (ENOTSUP);
3928
3929         error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3930         if (error)
3931                 return (error);
3932
3933         if (might_destroy) {
3934                 struct dsl_ds_destroyarg dsda = {0};
3935
3936                 if (dmu_tx_is_syncing(tx)) {
3937                         /*
3938                          * If we're not prepared to remove the snapshot,
3939                          * we can't allow the release to happen right now.
3940                          */
3941                         if (!ra->own)
3942                                 return (EBUSY);
3943                 }
3944                 dsda.ds = ds;
3945                 dsda.releasing = B_TRUE;
3946                 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3947         }
3948
3949         return (0);
3950 }
3951
3952 static void
3953 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3954 {
3955         struct dsl_ds_releasearg *ra = arg1;
3956         dsl_dataset_t *ds = ra->ds;
3957         dsl_pool_t *dp = ds->ds_dir->dd_pool;
3958         objset_t *mos = dp->dp_meta_objset;
3959         uint64_t zapobj;
3960         uint64_t dsobj = ds->ds_object;
3961         uint64_t refs;
3962         int error;
3963
3964         mutex_enter(&ds->ds_lock);
3965         ds->ds_userrefs--;
3966         refs = ds->ds_userrefs;
3967         mutex_exit(&ds->ds_lock);
3968         error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3969         VERIFY(error == 0 || error == ENOENT);
3970         zapobj = ds->ds_phys->ds_userrefs_obj;
3971         VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3972         if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3973             DS_IS_DEFER_DESTROY(ds)) {
3974                 struct dsl_ds_destroyarg dsda = {0};
3975
3976                 ASSERT(ra->own);
3977                 dsda.ds = ds;
3978                 dsda.releasing = B_TRUE;
3979                 /* We already did the destroy_check */
3980                 dsl_dataset_destroy_sync(&dsda, tag, tx);
3981         }
3982
3983         spa_history_log_internal(LOG_DS_USER_RELEASE,
3984             dp->dp_spa, tx, "<%s> %lld dataset = %llu",
3985             ra->htag, (longlong_t)refs, dsobj);
3986 }
3987
3988 static int
3989 dsl_dataset_user_release_one(const char *dsname, void *arg)
3990 {
3991         struct dsl_ds_holdarg *ha = arg;
3992         struct dsl_ds_releasearg *ra;
3993         dsl_dataset_t *ds;
3994         int error;
3995         void *dtag = ha->dstg;
3996         char *name;
3997         boolean_t own = B_FALSE;
3998         boolean_t might_destroy;
3999
4000         /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
4001         name = kmem_asprintf("%s@%s", dsname, ha->snapname);
4002         error = dsl_dataset_hold(name, dtag, &ds);
4003         strfree(name);
4004         if (error == ENOENT && ha->recursive)
4005                 return (0);
4006         (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
4007         if (error)
4008                 return (error);
4009
4010         ha->gotone = B_TRUE;
4011
4012         ASSERT(dsl_dataset_is_snapshot(ds));
4013
4014         error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
4015         if (error) {
4016                 dsl_dataset_rele(ds, dtag);
4017                 return (error);
4018         }
4019
4020         if (might_destroy) {
4021 #ifdef _KERNEL
4022                 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
4023                 error = zfs_unmount_snap(name, NULL);
4024                 strfree(name);
4025                 if (error) {
4026                         dsl_dataset_rele(ds, dtag);
4027                         return (error);
4028                 }
4029 #endif
4030                 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
4031                         dsl_dataset_rele(ds, dtag);
4032                         return (EBUSY);
4033                 } else {
4034                         own = B_TRUE;
4035                         dsl_dataset_make_exclusive(ds, dtag);
4036                 }
4037         }
4038
4039         ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
4040         ra->ds = ds;
4041         ra->htag = ha->htag;
4042         ra->own = own;
4043         dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
4044             dsl_dataset_user_release_sync, ra, dtag, 0);
4045
4046         return (0);
4047 }
4048
4049 int
4050 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
4051     boolean_t recursive)
4052 {
4053         struct dsl_ds_holdarg *ha;
4054         dsl_sync_task_t *dst;
4055         spa_t *spa;
4056         int error;
4057
4058 top:
4059         ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
4060
4061         (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
4062
4063         error = spa_open(dsname, &spa, FTAG);
4064         if (error) {
4065                 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4066                 return (error);
4067         }
4068
4069         ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
4070         ha->htag = htag;
4071         ha->snapname = snapname;
4072         ha->recursive = recursive;
4073         if (recursive) {
4074                 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
4075                     ha, DS_FIND_CHILDREN);
4076         } else {
4077                 error = dsl_dataset_user_release_one(dsname, ha);
4078         }
4079         if (error == 0)
4080                 error = dsl_sync_task_group_wait(ha->dstg);
4081
4082         for (dst = list_head(&ha->dstg->dstg_tasks); dst;
4083             dst = list_next(&ha->dstg->dstg_tasks, dst)) {
4084                 struct dsl_ds_releasearg *ra = dst->dst_arg1;
4085                 dsl_dataset_t *ds = ra->ds;
4086
4087                 if (dst->dst_err)
4088                         dsl_dataset_name(ds, ha->failed);
4089
4090                 if (ra->own)
4091                         dsl_dataset_disown(ds, ha->dstg);
4092                 else
4093                         dsl_dataset_rele(ds, ha->dstg);
4094
4095                 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
4096         }
4097
4098         if (error == 0 && recursive && !ha->gotone)
4099                 error = ENOENT;
4100
4101         if (error && error != EBUSY)
4102                 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
4103
4104         dsl_sync_task_group_destroy(ha->dstg);
4105         kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4106         spa_close(spa, FTAG);
4107
4108         /*
4109          * We can get EBUSY if we were racing with deferred destroy and
4110          * dsl_dataset_user_release_check() hadn't done the necessary
4111          * open context setup.  We can also get EBUSY if we're racing
4112          * with destroy and that thread is the ds_owner.  Either way
4113          * the busy condition should be transient, and we should retry
4114          * the release operation.
4115          */
4116         if (error == EBUSY)
4117                 goto top;
4118
4119         return (error);
4120 }
4121
4122 /*
4123  * Called at spa_load time (with retry == B_FALSE) to release a stale
4124  * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
4125  */
4126 int
4127 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
4128     boolean_t retry)
4129 {
4130         dsl_dataset_t *ds;
4131         char *snap;
4132         char *name;
4133         int namelen;
4134         int error;
4135
4136         do {
4137                 rw_enter(&dp->dp_config_rwlock, RW_READER);
4138                 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
4139                 rw_exit(&dp->dp_config_rwlock);
4140                 if (error)
4141                         return (error);
4142                 namelen = dsl_dataset_namelen(ds)+1;
4143                 name = kmem_alloc(namelen, KM_SLEEP);
4144                 dsl_dataset_name(ds, name);
4145                 dsl_dataset_rele(ds, FTAG);
4146
4147                 snap = strchr(name, '@');
4148                 *snap = '\0';
4149                 ++snap;
4150                 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
4151                 kmem_free(name, namelen);
4152
4153                 /*
4154                  * The object can't have been destroyed because we have a hold,
4155                  * but it might have been renamed, resulting in ENOENT.  Retry
4156                  * if we've been requested to do so.
4157                  *
4158                  * It would be nice if we could use the dsobj all the way
4159                  * through and avoid ENOENT entirely.  But we might need to
4160                  * unmount the snapshot, and there's currently no way to lookup
4161                  * a vfsp using a ZFS object id.
4162                  */
4163         } while ((error == ENOENT) && retry);
4164
4165         return (error);
4166 }
4167
4168 int
4169 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
4170 {
4171         dsl_dataset_t *ds;
4172         int err;
4173
4174         err = dsl_dataset_hold(dsname, FTAG, &ds);
4175         if (err)
4176                 return (err);
4177
4178         VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4179         if (ds->ds_phys->ds_userrefs_obj != 0) {
4180                 zap_attribute_t *za;
4181                 zap_cursor_t zc;
4182
4183                 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4184                 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4185                     ds->ds_phys->ds_userrefs_obj);
4186                     zap_cursor_retrieve(&zc, za) == 0;
4187                     zap_cursor_advance(&zc)) {
4188                         VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4189                             za->za_first_integer));
4190                 }
4191                 zap_cursor_fini(&zc);
4192                 kmem_free(za, sizeof (zap_attribute_t));
4193         }
4194         dsl_dataset_rele(ds, FTAG);
4195         return (0);
4196 }
4197
4198 /*
4199  * Note, this function is used as the callback for dmu_objset_find().  We
4200  * always return 0 so that we will continue to find and process
4201  * inconsistent datasets, even if we encounter an error trying to
4202  * process one of them.
4203  */
4204 /* ARGSUSED */
4205 int
4206 dsl_destroy_inconsistent(const char *dsname, void *arg)
4207 {
4208         dsl_dataset_t *ds;
4209
4210         if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4211                 if (DS_IS_INCONSISTENT(ds))
4212                         (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4213                 else
4214                         dsl_dataset_disown(ds, FTAG);
4215         }
4216         return (0);
4217 }
4218
4219 /*
4220  * Return (in *usedp) the amount of space written in new that is not
4221  * present in oldsnap.  New may be a snapshot or the head.  Old must be
4222  * a snapshot before new, in new's filesystem (or its origin).  If not then
4223  * fail and return EINVAL.
4224  *
4225  * The written space is calculated by considering two components:  First, we
4226  * ignore any freed space, and calculate the written as new's used space
4227  * minus old's used space.  Next, we add in the amount of space that was freed
4228  * between the two snapshots, thus reducing new's used space relative to old's.
4229  * Specifically, this is the space that was born before old->ds_creation_txg,
4230  * and freed before new (ie. on new's deadlist or a previous deadlist).
4231  *
4232  * space freed                         [---------------------]
4233  * snapshots                       ---O-------O--------O-------O------
4234  *                                         oldsnap            new
4235  */
4236 int
4237 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4238     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4239 {
4240         int err = 0;
4241         uint64_t snapobj;
4242         dsl_pool_t *dp = new->ds_dir->dd_pool;
4243
4244         *usedp = 0;
4245         *usedp += new->ds_phys->ds_referenced_bytes;
4246         *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
4247
4248         *compp = 0;
4249         *compp += new->ds_phys->ds_compressed_bytes;
4250         *compp -= oldsnap->ds_phys->ds_compressed_bytes;
4251
4252         *uncompp = 0;
4253         *uncompp += new->ds_phys->ds_uncompressed_bytes;
4254         *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
4255
4256         rw_enter(&dp->dp_config_rwlock, RW_READER);
4257         snapobj = new->ds_object;
4258         while (snapobj != oldsnap->ds_object) {
4259                 dsl_dataset_t *snap;
4260                 uint64_t used, comp, uncomp;
4261
4262                 if (snapobj == new->ds_object) {
4263                         snap = new;
4264                 } else {
4265                         err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4266                         if (err != 0)
4267                                 break;
4268                 }
4269
4270                 if (snap->ds_phys->ds_prev_snap_txg ==
4271                     oldsnap->ds_phys->ds_creation_txg) {
4272                         /*
4273                          * The blocks in the deadlist can not be born after
4274                          * ds_prev_snap_txg, so get the whole deadlist space,
4275                          * which is more efficient (especially for old-format
4276                          * deadlists).  Unfortunately the deadlist code
4277                          * doesn't have enough information to make this
4278                          * optimization itself.
4279                          */
4280                         dsl_deadlist_space(&snap->ds_deadlist,
4281                             &used, &comp, &uncomp);
4282                 } else {
4283                         dsl_deadlist_space_range(&snap->ds_deadlist,
4284                             0, oldsnap->ds_phys->ds_creation_txg,
4285                             &used, &comp, &uncomp);
4286                 }
4287                 *usedp += used;
4288                 *compp += comp;
4289                 *uncompp += uncomp;
4290
4291                 /*
4292                  * If we get to the beginning of the chain of snapshots
4293                  * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4294                  * was not a snapshot of/before new.
4295                  */
4296                 snapobj = snap->ds_phys->ds_prev_snap_obj;
4297                 if (snap != new)
4298                         dsl_dataset_rele(snap, FTAG);
4299                 if (snapobj == 0) {
4300                         err = EINVAL;
4301                         break;
4302                 }
4303
4304         }
4305         rw_exit(&dp->dp_config_rwlock);
4306         return (err);
4307 }
4308
4309 /*
4310  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4311  * lastsnap, and all snapshots in between are deleted.
4312  *
4313  * blocks that would be freed            [---------------------------]
4314  * snapshots                       ---O-------O--------O-------O--------O
4315  *                                        firstsnap        lastsnap
4316  *
4317  * This is the set of blocks that were born after the snap before firstsnap,
4318  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4319  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4320  * We calculate this by iterating over the relevant deadlists (from the snap
4321  * after lastsnap, backward to the snap after firstsnap), summing up the
4322  * space on the deadlist that was born after the snap before firstsnap.
4323  */
4324 int
4325 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4326     dsl_dataset_t *lastsnap,
4327     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4328 {
4329         int err = 0;
4330         uint64_t snapobj;
4331         dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4332
4333         ASSERT(dsl_dataset_is_snapshot(firstsnap));
4334         ASSERT(dsl_dataset_is_snapshot(lastsnap));
4335
4336         /*
4337          * Check that the snapshots are in the same dsl_dir, and firstsnap
4338          * is before lastsnap.
4339          */
4340         if (firstsnap->ds_dir != lastsnap->ds_dir ||
4341             firstsnap->ds_phys->ds_creation_txg >
4342             lastsnap->ds_phys->ds_creation_txg)
4343                 return (EINVAL);
4344
4345         *usedp = *compp = *uncompp = 0;
4346
4347         rw_enter(&dp->dp_config_rwlock, RW_READER);
4348         snapobj = lastsnap->ds_phys->ds_next_snap_obj;
4349         while (snapobj != firstsnap->ds_object) {
4350                 dsl_dataset_t *ds;
4351                 uint64_t used, comp, uncomp;
4352
4353                 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4354                 if (err != 0)
4355                         break;
4356
4357                 dsl_deadlist_space_range(&ds->ds_deadlist,
4358                     firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
4359                     &used, &comp, &uncomp);
4360                 *usedp += used;
4361                 *compp += comp;
4362                 *uncompp += uncomp;
4363
4364                 snapobj = ds->ds_phys->ds_prev_snap_obj;
4365                 ASSERT3U(snapobj, !=, 0);
4366                 dsl_dataset_rele(ds, FTAG);
4367         }
4368         rw_exit(&dp->dp_config_rwlock);
4369         return (err);
4370 }