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