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

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