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