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

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org>
  24  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
  25  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  26  * Copyright (c) 2014 RackTop Systems.
  27  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  28  * Copyright (c) 2014 Integros [integros.com]
  29  * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
  30  */
  31
  32 #include <sys/dmu_objset.h>
  33 #include <sys/dsl_dataset.h>
  34 #include <sys/dsl_dir.h>
  35 #include <sys/dsl_prop.h>
  36 #include <sys/dsl_synctask.h>
  37 #include <sys/dmu_traverse.h>
  38 #include <sys/dmu_impl.h>
  39 #include <sys/dmu_send.h>
  40 #include <sys/dmu_tx.h>
  41 #include <sys/arc.h>
  42 #include <sys/zio.h>
  43 #include <sys/zap.h>
  44 #include <sys/zfeature.h>
  45 #include <sys/unique.h>
  46 #include <sys/zfs_context.h>
  47 #include <sys/zfs_ioctl.h>
  48 #include <sys/spa.h>
  49 #include <sys/zfs_znode.h>
  50 #include <sys/zfs_onexit.h>
  51 #include <sys/zvol.h>
  52 #include <sys/dsl_scan.h>
  53 #include <sys/dsl_deadlist.h>
  54 #include <sys/dsl_destroy.h>
  55 #include <sys/dsl_userhold.h>
  56 #include <sys/dsl_bookmark.h>
  57 #include <sys/dmu_send.h>
  58 #include <sys/zio_checksum.h>
  59 #include <sys/zio_compress.h>
  60 #include <zfs_fletcher.h>
  61
  62 SYSCTL_DECL(_vfs_zfs);
  63
  64 /*
  65  * The SPA supports block sizes up to 16MB.  However, very large blocks
  66  * can have an impact on i/o latency (e.g. tying up a spinning disk for
  67  * ~300ms), and also potentially on the memory allocator.  Therefore,
  68  * we do not allow the recordsize to be set larger than zfs_max_recordsize
  69  * (default 1MB).  Larger blocks can be created by changing this tunable,
  70  * and pools with larger blocks can always be imported and used, regardless
  71  * of this setting.
  72  */
  73 int zfs_max_recordsize = 1 * 1024 * 1024;
  74 SYSCTL_INT(_vfs_zfs, OID_AUTO, max_recordsize, CTLFLAG_RWTUN,
  75     &zfs_max_recordsize, 0,
  76     "Maximum block size.  Expect dragons when tuning this.");
  77
  78 #define SWITCH64(x, y) \
  79         { \
  80                 uint64_t __tmp = (x); \
  81                 (x) = (y); \
  82                 (y) = __tmp; \
  83         }
  84
  85 #define DS_REF_MAX      (1ULL << 62)
  86
  87 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
  88
  89 extern int spa_asize_inflation;
  90
  91 /*
  92  * Figure out how much of this delta should be propogated to the dsl_dir
  93  * layer.  If there's a refreservation, that space has already been
  94  * partially accounted for in our ancestors.
  95  */
  96 static int64_t
  97 parent_delta(dsl_dataset_t *ds, int64_t delta)
  98 {
  99         dsl_dataset_phys_t *ds_phys;
 100         uint64_t old_bytes, new_bytes;
 101
 102         if (ds->ds_reserved == 0)
 103                 return (delta);
 104
 105         ds_phys = dsl_dataset_phys(ds);
 106         old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
 107         new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
 108
 109         ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
 110         return (new_bytes - old_bytes);
 111 }
 112
 113 void
 114 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
 115 {
 116         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 117         int compressed = BP_GET_PSIZE(bp);
 118         int uncompressed = BP_GET_UCSIZE(bp);
 119         int64_t delta;
 120
 121         dprintf_bp(bp, "ds=%p", ds);
 122
 123         ASSERT(dmu_tx_is_syncing(tx));
 124         /* It could have been compressed away to nothing */
 125         if (BP_IS_HOLE(bp))
 126                 return;
 127         ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
 128         ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
 129         if (ds == NULL) {
 130                 dsl_pool_mos_diduse_space(tx->tx_pool,
 131                     used, compressed, uncompressed);
 132                 return;
 133         }
 134
 135         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 136         mutex_enter(&ds->ds_lock);
 137         delta = parent_delta(ds, used);
 138         dsl_dataset_phys(ds)->ds_referenced_bytes += used;
 139         dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
 140         dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
 141         dsl_dataset_phys(ds)->ds_unique_bytes += used;
 142
 143         if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
 144                 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
 145                     B_TRUE;
 146         }
 147
 148         spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
 149         if (f != SPA_FEATURE_NONE)
 150                 ds->ds_feature_activation_needed[f] = B_TRUE;
 151
 152         mutex_exit(&ds->ds_lock);
 153         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
 154             compressed, uncompressed, tx);
 155         dsl_dir_transfer_space(ds->ds_dir, used - delta,
 156             DD_USED_REFRSRV, DD_USED_HEAD, NULL);
 157 }
 158
 159 int
 160 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 161     boolean_t async)
 162 {
 163         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 164         int compressed = BP_GET_PSIZE(bp);
 165         int uncompressed = BP_GET_UCSIZE(bp);
 166
 167         if (BP_IS_HOLE(bp))
 168                 return (0);
 169
 170         ASSERT(dmu_tx_is_syncing(tx));
 171         ASSERT(bp->blk_birth <= tx->tx_txg);
 172
 173         if (ds == NULL) {
 174                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 175                 dsl_pool_mos_diduse_space(tx->tx_pool,
 176                     -used, -compressed, -uncompressed);
 177                 return (used);
 178         }
 179         ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
 180
 181         ASSERT(!ds->ds_is_snapshot);
 182         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 183
 184         if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
 185                 int64_t delta;
 186
 187                 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
 188                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 189
 190                 mutex_enter(&ds->ds_lock);
 191                 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
 192                     !DS_UNIQUE_IS_ACCURATE(ds));
 193                 delta = parent_delta(ds, -used);
 194                 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
 195                 mutex_exit(&ds->ds_lock);
 196                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 197                     delta, -compressed, -uncompressed, tx);
 198                 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
 199                     DD_USED_REFRSRV, DD_USED_HEAD, NULL);
 200         } else {
 201                 dprintf_bp(bp, "putting on dead list: %s", "");
 202                 if (async) {
 203                         /*
 204                          * We are here as part of zio's write done callback,
 205                          * which means we're a zio interrupt thread.  We can't
 206                          * call dsl_deadlist_insert() now because it may block
 207                          * waiting for I/O.  Instead, put bp on the deferred
 208                          * queue and let dsl_pool_sync() finish the job.
 209                          */
 210                         bplist_append(&ds->ds_pending_deadlist, bp);
 211                 } else {
 212                         dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
 213                 }
 214                 ASSERT3U(ds->ds_prev->ds_object, ==,
 215                     dsl_dataset_phys(ds)->ds_prev_snap_obj);
 216                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
 217                 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
 218                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
 219                     ds->ds_object && bp->blk_birth >
 220                     dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
 221                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 222                         mutex_enter(&ds->ds_prev->ds_lock);
 223                         dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
 224                         mutex_exit(&ds->ds_prev->ds_lock);
 225                 }
 226                 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
 227                         dsl_dir_transfer_space(ds->ds_dir, used,
 228                             DD_USED_HEAD, DD_USED_SNAP, tx);
 229                 }
 230         }
 231         mutex_enter(&ds->ds_lock);
 232         ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
 233         dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
 234         ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
 235         dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
 236         ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
 237         dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
 238         mutex_exit(&ds->ds_lock);
 239
 240         return (used);
 241 }
 242
 243 uint64_t
 244 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
 245 {
 246         uint64_t trysnap = 0;
 247
 248         if (ds == NULL)
 249                 return (0);
 250         /*
 251          * The snapshot creation could fail, but that would cause an
 252          * incorrect FALSE return, which would only result in an
 253          * overestimation of the amount of space that an operation would
 254          * consume, which is OK.
 255          *
 256          * There's also a small window where we could miss a pending
 257          * snapshot, because we could set the sync task in the quiescing
 258          * phase.  So this should only be used as a guess.
 259          */
 260         if (ds->ds_trysnap_txg >
 261             spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
 262                 trysnap = ds->ds_trysnap_txg;
 263         return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
 264 }
 265
 266 boolean_t
 267 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
 268     uint64_t blk_birth)
 269 {
 270         if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
 271             (bp != NULL && BP_IS_HOLE(bp)))
 272                 return (B_FALSE);
 273
 274         ddt_prefetch(dsl_dataset_get_spa(ds), bp);
 275
 276         return (B_TRUE);
 277 }
 278
 279 static void
 280 dsl_dataset_evict(void *dbu)
 281 {
 282         dsl_dataset_t *ds = dbu;
 283
 284         ASSERT(ds->ds_owner == NULL);
 285
 286         ds->ds_dbuf = NULL;
 287
 288         unique_remove(ds->ds_fsid_guid);
 289
 290         if (ds->ds_objset != NULL)
 291                 dmu_objset_evict(ds->ds_objset);
 292
 293         if (ds->ds_prev) {
 294                 dsl_dataset_rele(ds->ds_prev, ds);
 295                 ds->ds_prev = NULL;
 296         }
 297
 298         bplist_destroy(&ds->ds_pending_deadlist);
 299         if (ds->ds_deadlist.dl_os != NULL)
 300                 dsl_deadlist_close(&ds->ds_deadlist);
 301         if (ds->ds_dir)
 302                 dsl_dir_async_rele(ds->ds_dir, ds);
 303
 304         ASSERT(!list_link_active(&ds->ds_synced_link));
 305
 306         list_destroy(&ds->ds_prop_cbs);
 307         if (mutex_owned(&ds->ds_lock))
 308                 mutex_exit(&ds->ds_lock);
 309         mutex_destroy(&ds->ds_lock);
 310         if (mutex_owned(&ds->ds_opening_lock))
 311                 mutex_exit(&ds->ds_opening_lock);
 312         mutex_destroy(&ds->ds_opening_lock);
 313         mutex_destroy(&ds->ds_sendstream_lock);
 314         refcount_destroy(&ds->ds_longholds);
 315
 316         kmem_free(ds, sizeof (dsl_dataset_t));
 317 }
 318
 319 int
 320 dsl_dataset_get_snapname(dsl_dataset_t *ds)
 321 {
 322         dsl_dataset_phys_t *headphys;
 323         int err;
 324         dmu_buf_t *headdbuf;
 325         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 326         objset_t *mos = dp->dp_meta_objset;
 327
 328         if (ds->ds_snapname[0])
 329                 return (0);
 330         if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
 331                 return (0);
 332
 333         err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
 334             FTAG, &headdbuf);
 335         if (err != 0)
 336                 return (err);
 337         headphys = headdbuf->db_data;
 338         err = zap_value_search(dp->dp_meta_objset,
 339             headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
 340         dmu_buf_rele(headdbuf, FTAG);
 341         return (err);
 342 }
 343
 344 int
 345 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
 346 {
 347         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 348         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 349         matchtype_t mt;
 350         int err;
 351
 352         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 353                 mt = MT_FIRST;
 354         else
 355                 mt = MT_EXACT;
 356
 357         err = zap_lookup_norm(mos, snapobj, name, 8, 1,
 358             value, mt, NULL, 0, NULL);
 359         if (err == ENOTSUP && mt == MT_FIRST)
 360                 err = zap_lookup(mos, snapobj, name, 8, 1, value);
 361         return (err);
 362 }
 363
 364 int
 365 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
 366     boolean_t adj_cnt)
 367 {
 368         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 369         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 370         matchtype_t mt;
 371         int err;
 372
 373         dsl_dir_snap_cmtime_update(ds->ds_dir);
 374
 375         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 376                 mt = MT_FIRST;
 377         else
 378                 mt = MT_EXACT;
 379
 380         err = zap_remove_norm(mos, snapobj, name, mt, tx);
 381         if (err == ENOTSUP && mt == MT_FIRST)
 382                 err = zap_remove(mos, snapobj, name, tx);
 383
 384         if (err == 0 && adj_cnt)
 385                 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
 386                     DD_FIELD_SNAPSHOT_COUNT, tx);
 387
 388         return (err);
 389 }
 390
 391 boolean_t
 392 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
 393 {
 394         dmu_buf_t *dbuf = ds->ds_dbuf;
 395         boolean_t result = B_FALSE;
 396
 397         if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
 398             ds->ds_object, DMU_BONUS_BLKID, tag)) {
 399
 400                 if (ds == dmu_buf_get_user(dbuf))
 401                         result = B_TRUE;
 402                 else
 403                         dmu_buf_rele(dbuf, tag);
 404         }
 405
 406         return (result);
 407 }
 408
 409 int
 410 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 411     dsl_dataset_t **dsp)
 412 {
 413         objset_t *mos = dp->dp_meta_objset;
 414         dmu_buf_t *dbuf;
 415         dsl_dataset_t *ds;
 416         int err;
 417         dmu_object_info_t doi;
 418
 419         ASSERT(dsl_pool_config_held(dp));
 420
 421         err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
 422         if (err != 0)
 423                 return (err);
 424
 425         /* Make sure dsobj has the correct object type. */
 426         dmu_object_info_from_db(dbuf, &doi);
 427         if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
 428                 dmu_buf_rele(dbuf, tag);
 429                 return (SET_ERROR(EINVAL));
 430         }
 431
 432         ds = dmu_buf_get_user(dbuf);
 433         if (ds == NULL) {
 434                 dsl_dataset_t *winner = NULL;
 435
 436                 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
 437                 ds->ds_dbuf = dbuf;
 438                 ds->ds_object = dsobj;
 439                 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
 440
 441                 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
 442                 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 443                 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
 444                 refcount_create(&ds->ds_longholds);
 445
 446                 bplist_create(&ds->ds_pending_deadlist);
 447                 dsl_deadlist_open(&ds->ds_deadlist,
 448                     mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
 449
 450                 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
 451                     offsetof(dmu_sendarg_t, dsa_link));
 452
 453                 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
 454                     offsetof(dsl_prop_cb_record_t, cbr_ds_node));
 455
 456                 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 457                         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 458                                 if (!(spa_feature_table[f].fi_flags &
 459                                     ZFEATURE_FLAG_PER_DATASET))
 460                                         continue;
 461                                 err = zap_contains(mos, dsobj,
 462                                     spa_feature_table[f].fi_guid);
 463                                 if (err == 0) {
 464                                         ds->ds_feature_inuse[f] = B_TRUE;
 465                                 } else {
 466                                         ASSERT3U(err, ==, ENOENT);
 467                                         err = 0;
 468                                 }
 469                         }
 470                 }
 471
 472                 err = dsl_dir_hold_obj(dp,
 473                     dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
 474                 if (err != 0) {
 475                         mutex_destroy(&ds->ds_lock);
 476                         mutex_destroy(&ds->ds_opening_lock);
 477                         mutex_destroy(&ds->ds_sendstream_lock);
 478                         refcount_destroy(&ds->ds_longholds);
 479                         bplist_destroy(&ds->ds_pending_deadlist);
 480                         dsl_deadlist_close(&ds->ds_deadlist);
 481                         kmem_free(ds, sizeof (dsl_dataset_t));
 482                         dmu_buf_rele(dbuf, tag);
 483                         return (err);
 484                 }
 485
 486                 if (!ds->ds_is_snapshot) {
 487                         ds->ds_snapname[0] = '\0';
 488                         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 489                                 err = dsl_dataset_hold_obj(dp,
 490                                     dsl_dataset_phys(ds)->ds_prev_snap_obj,
 491                                     ds, &ds->ds_prev);
 492                         }
 493                         if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 494                                 int zaperr = zap_lookup(mos, ds->ds_object,
 495                                     DS_FIELD_BOOKMARK_NAMES,
 496                                     sizeof (ds->ds_bookmarks), 1,
 497                                     &ds->ds_bookmarks);
 498                                 if (zaperr != ENOENT)
 499                                         VERIFY0(zaperr);
 500                         }
 501                 } else {
 502                         if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
 503                                 err = dsl_dataset_get_snapname(ds);
 504                         if (err == 0 &&
 505                             dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
 506                                 err = zap_count(
 507                                     ds->ds_dir->dd_pool->dp_meta_objset,
 508                                     dsl_dataset_phys(ds)->ds_userrefs_obj,
 509                                     &ds->ds_userrefs);
 510                         }
 511                 }
 512
 513                 if (err == 0 && !ds->ds_is_snapshot) {
 514                         err = dsl_prop_get_int_ds(ds,
 515                             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
 516                             &ds->ds_reserved);
 517                         if (err == 0) {
 518                                 err = dsl_prop_get_int_ds(ds,
 519                                     zfs_prop_to_name(ZFS_PROP_REFQUOTA),
 520                                     &ds->ds_quota);
 521                         }
 522                 } else {
 523                         ds->ds_reserved = ds->ds_quota = 0;
 524                 }
 525
 526                 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
 527                 if (err == 0)
 528                         winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
 529
 530                 if (err != 0 || winner != NULL) {
 531                         bplist_destroy(&ds->ds_pending_deadlist);
 532                         dsl_deadlist_close(&ds->ds_deadlist);
 533                         if (ds->ds_prev)
 534                                 dsl_dataset_rele(ds->ds_prev, ds);
 535                         dsl_dir_rele(ds->ds_dir, ds);
 536                         mutex_destroy(&ds->ds_lock);
 537                         mutex_destroy(&ds->ds_opening_lock);
 538                         mutex_destroy(&ds->ds_sendstream_lock);
 539                         refcount_destroy(&ds->ds_longholds);
 540                         kmem_free(ds, sizeof (dsl_dataset_t));
 541                         if (err != 0) {
 542                                 dmu_buf_rele(dbuf, tag);
 543                                 return (err);
 544                         }
 545                         ds = winner;
 546                 } else {
 547                         ds->ds_fsid_guid =
 548                             unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
 549                 }
 550         }
 551         ASSERT3P(ds->ds_dbuf, ==, dbuf);
 552         ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
 553         ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
 554             spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
 555             dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
 556         *dsp = ds;
 557         return (0);
 558 }
 559
 560 int
 561 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
 562     void *tag, dsl_dataset_t **dsp)
 563 {
 564         dsl_dir_t *dd;
 565         const char *snapname;
 566         uint64_t obj;
 567         int err = 0;
 568         dsl_dataset_t *ds;
 569
 570         err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
 571         if (err != 0)
 572                 return (err);
 573
 574         ASSERT(dsl_pool_config_held(dp));
 575         obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
 576         if (obj != 0)
 577                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
 578         else
 579                 err = SET_ERROR(ENOENT);
 580
 581         /* we may be looking for a snapshot */
 582         if (err == 0 && snapname != NULL) {
 583                 dsl_dataset_t *snap_ds;
 584
 585                 if (*snapname++ != '@') {
 586                         dsl_dataset_rele(ds, tag);
 587                         dsl_dir_rele(dd, FTAG);
 588                         return (SET_ERROR(ENOENT));
 589                 }
 590
 591                 dprintf("looking for snapshot '%s'\n", snapname);
 592                 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
 593                 if (err == 0)
 594                         err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
 595                 dsl_dataset_rele(ds, tag);
 596
 597                 if (err == 0) {
 598                         mutex_enter(&snap_ds->ds_lock);
 599                         if (snap_ds->ds_snapname[0] == 0)
 600                                 (void) strlcpy(snap_ds->ds_snapname, snapname,
 601                                     sizeof (snap_ds->ds_snapname));
 602                         mutex_exit(&snap_ds->ds_lock);
 603                         ds = snap_ds;
 604                 }
 605         }
 606         if (err == 0)
 607                 *dsp = ds;
 608         dsl_dir_rele(dd, FTAG);
 609         return (err);
 610 }
 611
 612 int
 613 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
 614     void *tag, dsl_dataset_t **dsp)
 615 {
 616         int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
 617         if (err != 0)
 618                 return (err);
 619         if (!dsl_dataset_tryown(*dsp, tag)) {
 620                 dsl_dataset_rele(*dsp, tag);
 621                 *dsp = NULL;
 622                 return (SET_ERROR(EBUSY));
 623         }
 624         return (0);
 625 }
 626
 627 int
 628 dsl_dataset_own(dsl_pool_t *dp, const char *name,
 629     void *tag, dsl_dataset_t **dsp)
 630 {
 631         int err = dsl_dataset_hold(dp, name, tag, dsp);
 632         if (err != 0)
 633                 return (err);
 634         if (!dsl_dataset_tryown(*dsp, tag)) {
 635                 dsl_dataset_rele(*dsp, tag);
 636                 return (SET_ERROR(EBUSY));
 637         }
 638         return (0);
 639 }
 640
 641 /*
 642  * See the comment above dsl_pool_hold() for details.  In summary, a long
 643  * hold is used to prevent destruction of a dataset while the pool hold
 644  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
 645  *
 646  * The dataset and pool must be held when this function is called.  After it
 647  * is called, the pool hold may be released while the dataset is still held
 648  * and accessed.
 649  */
 650 void
 651 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
 652 {
 653         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 654         (void) refcount_add(&ds->ds_longholds, tag);
 655 }
 656
 657 void
 658 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
 659 {
 660         (void) refcount_remove(&ds->ds_longholds, tag);
 661 }
 662
 663 /* Return B_TRUE if there are any long holds on this dataset. */
 664 boolean_t
 665 dsl_dataset_long_held(dsl_dataset_t *ds)
 666 {
 667         return (!refcount_is_zero(&ds->ds_longholds));
 668 }
 669
 670 void
 671 dsl_dataset_name(dsl_dataset_t *ds, char *name)
 672 {
 673         if (ds == NULL) {
 674                 (void) strcpy(name, "mos");
 675         } else {
 676                 dsl_dir_name(ds->ds_dir, name);
 677                 VERIFY0(dsl_dataset_get_snapname(ds));
 678                 if (ds->ds_snapname[0]) {
 679                         VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
 680                             <, ZFS_MAX_DATASET_NAME_LEN);
 681                         /*
 682                          * We use a "recursive" mutex so that we
 683                          * can call dprintf_ds() with ds_lock held.
 684                          */
 685                         if (!MUTEX_HELD(&ds->ds_lock)) {
 686                                 mutex_enter(&ds->ds_lock);
 687                                 VERIFY3U(strlcat(name, ds->ds_snapname,
 688                                     ZFS_MAX_DATASET_NAME_LEN), <,
 689                                     ZFS_MAX_DATASET_NAME_LEN);
 690                                 mutex_exit(&ds->ds_lock);
 691                         } else {
 692                                 VERIFY3U(strlcat(name, ds->ds_snapname,
 693                                     ZFS_MAX_DATASET_NAME_LEN), <,
 694                                     ZFS_MAX_DATASET_NAME_LEN);
 695                         }
 696                 }
 697         }
 698 }
 699
 700 int
 701 dsl_dataset_namelen(dsl_dataset_t *ds)
 702 {
 703         VERIFY0(dsl_dataset_get_snapname(ds));
 704         mutex_enter(&ds->ds_lock);
 705         int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
 706         mutex_exit(&ds->ds_lock);
 707         return (len);
 708 }
 709
 710 void
 711 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
 712 {
 713         dmu_buf_rele(ds->ds_dbuf, tag);
 714 }
 715
 716 void
 717 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
 718 {
 719         ASSERT3P(ds->ds_owner, ==, tag);
 720         ASSERT(ds->ds_dbuf != NULL);
 721
 722         mutex_enter(&ds->ds_lock);
 723         ds->ds_owner = NULL;
 724         mutex_exit(&ds->ds_lock);
 725         dsl_dataset_long_rele(ds, tag);
 726         dsl_dataset_rele(ds, tag);
 727 }
 728
 729 boolean_t
 730 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
 731 {
 732         boolean_t gotit = FALSE;
 733
 734         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 735         mutex_enter(&ds->ds_lock);
 736         if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
 737                 ds->ds_owner = tag;
 738                 dsl_dataset_long_hold(ds, tag);
 739                 gotit = TRUE;
 740         }
 741         mutex_exit(&ds->ds_lock);
 742         return (gotit);
 743 }
 744
 745 boolean_t
 746 dsl_dataset_has_owner(dsl_dataset_t *ds)
 747 {
 748         boolean_t rv;
 749         mutex_enter(&ds->ds_lock);
 750         rv = (ds->ds_owner != NULL);
 751         mutex_exit(&ds->ds_lock);
 752         return (rv);
 753 }
 754
 755 static void
 756 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
 757 {
 758         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 759         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
 760         uint64_t zero = 0;
 761
 762         VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
 763
 764         spa_feature_incr(spa, f, tx);
 765         dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
 766
 767         VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
 768             sizeof (zero), 1, &zero, tx));
 769 }
 770
 771 void
 772 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
 773 {
 774         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 775         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
 776
 777         VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
 778
 779         VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
 780         spa_feature_decr(spa, f, tx);
 781 }
 782
 783 uint64_t
 784 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
 785     uint64_t flags, dmu_tx_t *tx)
 786 {
 787         dsl_pool_t *dp = dd->dd_pool;
 788         dmu_buf_t *dbuf;
 789         dsl_dataset_phys_t *dsphys;
 790         uint64_t dsobj;
 791         objset_t *mos = dp->dp_meta_objset;
 792
 793         if (origin == NULL)
 794                 origin = dp->dp_origin_snap;
 795
 796         ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
 797         ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
 798         ASSERT(dmu_tx_is_syncing(tx));
 799         ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
 800
 801         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
 802             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
 803         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
 804         dmu_buf_will_dirty(dbuf, tx);
 805         dsphys = dbuf->db_data;
 806         bzero(dsphys, sizeof (dsl_dataset_phys_t));
 807         dsphys->ds_dir_obj = dd->dd_object;
 808         dsphys->ds_flags = flags;
 809         dsphys->ds_fsid_guid = unique_create();
 810         do {
 811                 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
 812                     sizeof (dsphys->ds_guid));
 813         } while (dsphys->ds_guid == 0);
 814         dsphys->ds_snapnames_zapobj =
 815             zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
 816             DMU_OT_NONE, 0, tx);
 817         dsphys->ds_creation_time = gethrestime_sec();
 818         dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
 819
 820         if (origin == NULL) {
 821                 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
 822         } else {
 823                 dsl_dataset_t *ohds; /* head of the origin snapshot */
 824
 825                 dsphys->ds_prev_snap_obj = origin->ds_object;
 826                 dsphys->ds_prev_snap_txg =
 827                     dsl_dataset_phys(origin)->ds_creation_txg;
 828                 dsphys->ds_referenced_bytes =
 829                     dsl_dataset_phys(origin)->ds_referenced_bytes;
 830                 dsphys->ds_compressed_bytes =
 831                     dsl_dataset_phys(origin)->ds_compressed_bytes;
 832                 dsphys->ds_uncompressed_bytes =
 833                     dsl_dataset_phys(origin)->ds_uncompressed_bytes;
 834                 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
 835
 836                 /*
 837                  * Inherit flags that describe the dataset's contents
 838                  * (INCONSISTENT) or properties (Case Insensitive).
 839                  */
 840                 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
 841                     (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
 842
 843                 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
 844                         if (origin->ds_feature_inuse[f])
 845                                 dsl_dataset_activate_feature(dsobj, f, tx);
 846                 }
 847
 848                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
 849                 dsl_dataset_phys(origin)->ds_num_children++;
 850
 851                 VERIFY0(dsl_dataset_hold_obj(dp,
 852                     dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
 853                     FTAG, &ohds));
 854                 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
 855                     dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
 856                 dsl_dataset_rele(ohds, FTAG);
 857
 858                 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
 859                         if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
 860                                 dsl_dataset_phys(origin)->ds_next_clones_obj =
 861                                     zap_create(mos,
 862                                     DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
 863                         }
 864                         VERIFY0(zap_add_int(mos,
 865                             dsl_dataset_phys(origin)->ds_next_clones_obj,
 866                             dsobj, tx));
 867                 }
 868
 869                 dmu_buf_will_dirty(dd->dd_dbuf, tx);
 870                 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
 871                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 872                         if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
 873                                 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
 874                                 dsl_dir_phys(origin->ds_dir)->dd_clones =
 875                                     zap_create(mos,
 876                                     DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
 877                         }
 878                         VERIFY0(zap_add_int(mos,
 879                             dsl_dir_phys(origin->ds_dir)->dd_clones,
 880                             dsobj, tx));
 881                 }
 882         }
 883
 884         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
 885                 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 886
 887         dmu_buf_rele(dbuf, FTAG);
 888
 889         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 890         dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
 891
 892         return (dsobj);
 893 }
 894
 895 static void
 896 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
 897 {
 898         objset_t *os;
 899
 900         VERIFY0(dmu_objset_from_ds(ds, &os));
 901         bzero(&os->os_zil_header, sizeof (os->os_zil_header));
 902         dsl_dataset_dirty(ds, tx);
 903 }
 904
 905 uint64_t
 906 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
 907     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
 908 {
 909         dsl_pool_t *dp = pdd->dd_pool;
 910         uint64_t dsobj, ddobj;
 911         dsl_dir_t *dd;
 912
 913         ASSERT(dmu_tx_is_syncing(tx));
 914         ASSERT(lastname[0] != '@');
 915
 916         ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
 917         VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
 918
 919         dsobj = dsl_dataset_create_sync_dd(dd, origin,
 920             flags & ~DS_CREATE_FLAG_NODIRTY, tx);
 921
 922         dsl_deleg_set_create_perms(dd, tx, cr);
 923
 924         /*
 925          * Since we're creating a new node we know it's a leaf, so we can
 926          * initialize the counts if the limit feature is active.
 927          */
 928         if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
 929                 uint64_t cnt = 0;
 930                 objset_t *os = dd->dd_pool->dp_meta_objset;
 931
 932                 dsl_dir_zapify(dd, tx);
 933                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
 934                     sizeof (cnt), 1, &cnt, tx));
 935                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
 936                     sizeof (cnt), 1, &cnt, tx));
 937         }
 938
 939         dsl_dir_rele(dd, FTAG);
 940
 941         /*
 942          * If we are creating a clone, make sure we zero out any stale
 943          * data from the origin snapshots zil header.
 944          */
 945         if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
 946                 dsl_dataset_t *ds;
 947
 948                 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
 949                 dsl_dataset_zero_zil(ds, tx);
 950                 dsl_dataset_rele(ds, FTAG);
 951         }
 952
 953         return (dsobj);
 954 }
 955
 956 #ifdef __FreeBSD__
 957 /* FreeBSD ioctl compat begin */
 958 struct destroyarg {
 959         nvlist_t *nvl;
 960         const char *snapname;
 961 };
 962
 963 static int
 964 dsl_check_snap_cb(const char *name, void *arg)
 965 {
 966         struct destroyarg *da = arg;
 967         dsl_dataset_t *ds;
 968         char *dsname;
 969
 970         dsname = kmem_asprintf("%s@%s", name, da->snapname);
 971         fnvlist_add_boolean(da->nvl, dsname);
 972         kmem_free(dsname, strlen(dsname) + 1);
 973
 974         return (0);
 975 }
 976
 977 int
 978 dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
 979     nvlist_t *snaps)
 980 {
 981         struct destroyarg *da;
 982         int err;
 983
 984         da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP);
 985         da->nvl = snaps;
 986         da->snapname = snapname;
 987         err = dmu_objset_find(fsname, dsl_check_snap_cb, da,
 988             DS_FIND_CHILDREN);
 989         kmem_free(da, sizeof (struct destroyarg));
 990
 991         return (err);
 992 }
 993 /* FreeBSD ioctl compat end */
 994 #endif /* __FreeBSD__ */
 995
 996 /*
 997  * The unique space in the head dataset can be calculated by subtracting
 998  * the space used in the most recent snapshot, that is still being used
 999  * in this file system, from the space currently in use.  To figure out
1000  * the space in the most recent snapshot still in use, we need to take
1001  * the total space used in the snapshot and subtract out the space that
1002  * has been freed up since the snapshot was taken.
1003  */
1004 void
1005 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1006 {
1007         uint64_t mrs_used;
1008         uint64_t dlused, dlcomp, dluncomp;
1009
1010         ASSERT(!ds->ds_is_snapshot);
1011
1012         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1013                 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1014         else
1015                 mrs_used = 0;
1016
1017         dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1018
1019         ASSERT3U(dlused, <=, mrs_used);
1020         dsl_dataset_phys(ds)->ds_unique_bytes =
1021             dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1022
1023         if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1024             SPA_VERSION_UNIQUE_ACCURATE)
1025                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1026 }
1027
1028 void
1029 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1030     dmu_tx_t *tx)
1031 {
1032         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1033         uint64_t count;
1034         int err;
1035
1036         ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1037         err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1038             obj, tx);
1039         /*
1040          * The err should not be ENOENT, but a bug in a previous version
1041          * of the code could cause upgrade_clones_cb() to not set
1042          * ds_next_snap_obj when it should, leading to a missing entry.
1043          * If we knew that the pool was created after
1044          * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1045          * ENOENT.  However, at least we can check that we don't have
1046          * too many entries in the next_clones_obj even after failing to
1047          * remove this one.
1048          */
1049         if (err != ENOENT)
1050                 VERIFY0(err);
1051         ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1052             &count));
1053         ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1054 }
1055
1056
1057 blkptr_t *
1058 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1059 {
1060         return (&dsl_dataset_phys(ds)->ds_bp);
1061 }
1062
1063 spa_t *
1064 dsl_dataset_get_spa(dsl_dataset_t *ds)
1065 {
1066         return (ds->ds_dir->dd_pool->dp_spa);
1067 }
1068
1069 void
1070 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1071 {
1072         dsl_pool_t *dp;
1073
1074         if (ds == NULL) /* this is the meta-objset */
1075                 return;
1076
1077         ASSERT(ds->ds_objset != NULL);
1078
1079         if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1080                 panic("dirtying snapshot!");
1081
1082         dp = ds->ds_dir->dd_pool;
1083
1084         if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1085                 /* up the hold count until we can be written out */
1086                 dmu_buf_add_ref(ds->ds_dbuf, ds);
1087         }
1088 }
1089
1090 boolean_t
1091 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1092 {
1093         for (int t = 0; t < TXG_SIZE; t++) {
1094                 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1095                     ds, t))
1096                         return (B_TRUE);
1097         }
1098         return (B_FALSE);
1099 }
1100
1101 static int
1102 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1103 {
1104         uint64_t asize;
1105
1106         if (!dmu_tx_is_syncing(tx))
1107                 return (0);
1108
1109         /*
1110          * If there's an fs-only reservation, any blocks that might become
1111          * owned by the snapshot dataset must be accommodated by space
1112          * outside of the reservation.
1113          */
1114         ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1115         asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1116         if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1117                 return (SET_ERROR(ENOSPC));
1118
1119         /*
1120          * Propagate any reserved space for this snapshot to other
1121          * snapshot checks in this sync group.
1122          */
1123         if (asize > 0)
1124                 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1125
1126         return (0);
1127 }
1128
1129 typedef struct dsl_dataset_snapshot_arg {
1130         nvlist_t *ddsa_snaps;
1131         nvlist_t *ddsa_props;
1132         nvlist_t *ddsa_errors;
1133         cred_t *ddsa_cr;
1134 } dsl_dataset_snapshot_arg_t;
1135
1136 int
1137 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1138     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1139 {
1140         int error;
1141         uint64_t value;
1142
1143         ds->ds_trysnap_txg = tx->tx_txg;
1144
1145         if (!dmu_tx_is_syncing(tx))
1146                 return (0);
1147
1148         /*
1149          * We don't allow multiple snapshots of the same txg.  If there
1150          * is already one, try again.
1151          */
1152         if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1153                 return (SET_ERROR(EAGAIN));
1154
1155         /*
1156          * Check for conflicting snapshot name.
1157          */
1158         error = dsl_dataset_snap_lookup(ds, snapname, &value);
1159         if (error == 0)
1160                 return (SET_ERROR(EEXIST));
1161         if (error != ENOENT)
1162                 return (error);
1163
1164         /*
1165          * We don't allow taking snapshots of inconsistent datasets, such as
1166          * those into which we are currently receiving.  However, if we are
1167          * creating this snapshot as part of a receive, this check will be
1168          * executed atomically with respect to the completion of the receive
1169          * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1170          * case we ignore this, knowing it will be fixed up for us shortly in
1171          * dmu_recv_end_sync().
1172          */
1173         if (!recv && DS_IS_INCONSISTENT(ds))
1174                 return (SET_ERROR(EBUSY));
1175
1176         /*
1177          * Skip the check for temporary snapshots or if we have already checked
1178          * the counts in dsl_dataset_snapshot_check. This means we really only
1179          * check the count here when we're receiving a stream.
1180          */
1181         if (cnt != 0 && cr != NULL) {
1182                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1183                     ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1184                 if (error != 0)
1185                         return (error);
1186         }
1187
1188         error = dsl_dataset_snapshot_reserve_space(ds, tx);
1189         if (error != 0)
1190                 return (error);
1191
1192         return (0);
1193 }
1194
1195 static int
1196 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1197 {
1198         dsl_dataset_snapshot_arg_t *ddsa = arg;
1199         dsl_pool_t *dp = dmu_tx_pool(tx);
1200         nvpair_t *pair;
1201         int rv = 0;
1202
1203         /*
1204          * Pre-compute how many total new snapshots will be created for each
1205          * level in the tree and below. This is needed for validating the
1206          * snapshot limit when either taking a recursive snapshot or when
1207          * taking multiple snapshots.
1208          *
1209          * The problem is that the counts are not actually adjusted when
1210          * we are checking, only when we finally sync. For a single snapshot,
1211          * this is easy, the count will increase by 1 at each node up the tree,
1212          * but its more complicated for the recursive/multiple snapshot case.
1213          *
1214          * The dsl_fs_ss_limit_check function does recursively check the count
1215          * at each level up the tree but since it is validating each snapshot
1216          * independently we need to be sure that we are validating the complete
1217          * count for the entire set of snapshots. We do this by rolling up the
1218          * counts for each component of the name into an nvlist and then
1219          * checking each of those cases with the aggregated count.
1220          *
1221          * This approach properly handles not only the recursive snapshot
1222          * case (where we get all of those on the ddsa_snaps list) but also
1223          * the sibling case (e.g. snapshot a/b and a/c so that we will also
1224          * validate the limit on 'a' using a count of 2).
1225          *
1226          * We validate the snapshot names in the third loop and only report
1227          * name errors once.
1228          */
1229         if (dmu_tx_is_syncing(tx)) {
1230                 nvlist_t *cnt_track = NULL;
1231                 cnt_track = fnvlist_alloc();
1232
1233                 /* Rollup aggregated counts into the cnt_track list */
1234                 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1235                     pair != NULL;
1236                     pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1237                         char *pdelim;
1238                         uint64_t val;
1239                         char nm[MAXPATHLEN];
1240
1241                         (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1242                         pdelim = strchr(nm, '@');
1243                         if (pdelim == NULL)
1244                                 continue;
1245                         *pdelim = '\0';
1246
1247                         do {
1248                                 if (nvlist_lookup_uint64(cnt_track, nm,
1249                                     &val) == 0) {
1250                                         /* update existing entry */
1251                                         fnvlist_add_uint64(cnt_track, nm,
1252                                             val + 1);
1253                                 } else {
1254                                         /* add to list */
1255                                         fnvlist_add_uint64(cnt_track, nm, 1);
1256                                 }
1257
1258                                 pdelim = strrchr(nm, '/');
1259                                 if (pdelim != NULL)
1260                                         *pdelim = '\0';
1261                         } while (pdelim != NULL);
1262                 }
1263
1264                 /* Check aggregated counts at each level */
1265                 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1266                     pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1267                         int error = 0;
1268                         char *name;
1269                         uint64_t cnt = 0;
1270                         dsl_dataset_t *ds;
1271
1272                         name = nvpair_name(pair);
1273                         cnt = fnvpair_value_uint64(pair);
1274                         ASSERT(cnt > 0);
1275
1276                         error = dsl_dataset_hold(dp, name, FTAG, &ds);
1277                         if (error == 0) {
1278                                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1279                                     ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1280                                     ddsa->ddsa_cr);
1281                                 dsl_dataset_rele(ds, FTAG);
1282                         }
1283
1284                         if (error != 0) {
1285                                 if (ddsa->ddsa_errors != NULL)
1286                                         fnvlist_add_int32(ddsa->ddsa_errors,
1287                                             name, error);
1288                                 rv = error;
1289                                 /* only report one error for this check */
1290                                 break;
1291                         }
1292                 }
1293                 nvlist_free(cnt_track);
1294         }
1295
1296         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1297             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1298                 int error = 0;
1299                 dsl_dataset_t *ds;
1300                 char *name, *atp;
1301                 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1302
1303                 name = nvpair_name(pair);
1304                 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1305                         error = SET_ERROR(ENAMETOOLONG);
1306                 if (error == 0) {
1307                         atp = strchr(name, '@');
1308                         if (atp == NULL)
1309                                 error = SET_ERROR(EINVAL);
1310                         if (error == 0)
1311                                 (void) strlcpy(dsname, name, atp - name + 1);
1312                 }
1313                 if (error == 0)
1314                         error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1315                 if (error == 0) {
1316                         /* passing 0/NULL skips dsl_fs_ss_limit_check */
1317                         error = dsl_dataset_snapshot_check_impl(ds,
1318                             atp + 1, tx, B_FALSE, 0, NULL);
1319                         dsl_dataset_rele(ds, FTAG);
1320                 }
1321
1322                 if (error != 0) {
1323                         if (ddsa->ddsa_errors != NULL) {
1324                                 fnvlist_add_int32(ddsa->ddsa_errors,
1325                                     name, error);
1326                         }
1327                         rv = error;
1328                 }
1329         }
1330
1331         return (rv);
1332 }
1333
1334 void
1335 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1336     dmu_tx_t *tx)
1337 {
1338         static zil_header_t zero_zil;
1339
1340         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1341         dmu_buf_t *dbuf;
1342         dsl_dataset_phys_t *dsphys;
1343         uint64_t dsobj, crtxg;
1344         objset_t *mos = dp->dp_meta_objset;
1345         objset_t *os;
1346
1347         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1348
1349         /*
1350          * If we are on an old pool, the zil must not be active, in which
1351          * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1352          */
1353         ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1354             dmu_objset_from_ds(ds, &os) != 0 ||
1355             bcmp(&os->os_phys->os_zil_header, &zero_zil,
1356             sizeof (zero_zil)) == 0);
1357
1358         dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1359
1360         /*
1361          * The origin's ds_creation_txg has to be < TXG_INITIAL
1362          */
1363         if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1364                 crtxg = 1;
1365         else
1366                 crtxg = tx->tx_txg;
1367
1368         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1369             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1370         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1371         dmu_buf_will_dirty(dbuf, tx);
1372         dsphys = dbuf->db_data;
1373         bzero(dsphys, sizeof (dsl_dataset_phys_t));
1374         dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1375         dsphys->ds_fsid_guid = unique_create();
1376         do {
1377                 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1378                     sizeof (dsphys->ds_guid));
1379         } while (dsphys->ds_guid == 0);
1380         dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1381         dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1382         dsphys->ds_next_snap_obj = ds->ds_object;
1383         dsphys->ds_num_children = 1;
1384         dsphys->ds_creation_time = gethrestime_sec();
1385         dsphys->ds_creation_txg = crtxg;
1386         dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1387         dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1388         dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1389         dsphys->ds_uncompressed_bytes =
1390             dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1391         dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1392         dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1393         dmu_buf_rele(dbuf, FTAG);
1394
1395         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1396                 if (ds->ds_feature_inuse[f])
1397                         dsl_dataset_activate_feature(dsobj, f, tx);
1398         }
1399
1400         ASSERT3U(ds->ds_prev != 0, ==,
1401             dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1402         if (ds->ds_prev) {
1403                 uint64_t next_clones_obj =
1404                     dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1405                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1406                     ds->ds_object ||
1407                     dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1408                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1409                     ds->ds_object) {
1410                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1411                         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1412                             dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1413                         dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1414                 } else if (next_clones_obj != 0) {
1415                         dsl_dataset_remove_from_next_clones(ds->ds_prev,
1416                             dsphys->ds_next_snap_obj, tx);
1417                         VERIFY0(zap_add_int(mos,
1418                             next_clones_obj, dsobj, tx));
1419                 }
1420         }
1421
1422         /*
1423          * If we have a reference-reservation on this dataset, we will
1424          * need to increase the amount of refreservation being charged
1425          * since our unique space is going to zero.
1426          */
1427         if (ds->ds_reserved) {
1428                 int64_t delta;
1429                 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1430                 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1431                     ds->ds_reserved);
1432                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1433                     delta, 0, 0, tx);
1434         }
1435
1436         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1437         dsl_dataset_phys(ds)->ds_deadlist_obj =
1438             dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1439             dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1440         dsl_deadlist_close(&ds->ds_deadlist);
1441         dsl_deadlist_open(&ds->ds_deadlist, mos,
1442             dsl_dataset_phys(ds)->ds_deadlist_obj);
1443         dsl_deadlist_add_key(&ds->ds_deadlist,
1444             dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1445
1446         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1447         dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1448         dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1449         dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1450         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1451                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1452
1453         VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1454             snapname, 8, 1, &dsobj, tx));
1455
1456         if (ds->ds_prev)
1457                 dsl_dataset_rele(ds->ds_prev, ds);
1458         VERIFY0(dsl_dataset_hold_obj(dp,
1459             dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1460
1461         dsl_scan_ds_snapshotted(ds, tx);
1462
1463         dsl_dir_snap_cmtime_update(ds->ds_dir);
1464
1465         spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1466 }
1467
1468 static void
1469 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1470 {
1471         dsl_dataset_snapshot_arg_t *ddsa = arg;
1472         dsl_pool_t *dp = dmu_tx_pool(tx);
1473         nvpair_t *pair;
1474
1475         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1476             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1477                 dsl_dataset_t *ds;
1478                 char *name, *atp;
1479                 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1480
1481                 name = nvpair_name(pair);
1482                 atp = strchr(name, '@');
1483                 (void) strlcpy(dsname, name, atp - name + 1);
1484                 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1485
1486                 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1487                 if (ddsa->ddsa_props != NULL) {
1488                         dsl_props_set_sync_impl(ds->ds_prev,
1489                             ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1490                 }
1491                 dsl_dataset_rele(ds, FTAG);
1492         }
1493 }
1494
1495 /*
1496  * The snapshots must all be in the same pool.
1497  * All-or-nothing: if there are any failures, nothing will be modified.
1498  */
1499 int
1500 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1501 {
1502         dsl_dataset_snapshot_arg_t ddsa;
1503         nvpair_t *pair;
1504         boolean_t needsuspend;
1505         int error;
1506         spa_t *spa;
1507         char *firstname;
1508         nvlist_t *suspended = NULL;
1509
1510         pair = nvlist_next_nvpair(snaps, NULL);
1511         if (pair == NULL)
1512                 return (0);
1513         firstname = nvpair_name(pair);
1514
1515         error = spa_open(firstname, &spa, FTAG);
1516         if (error != 0)
1517                 return (error);
1518         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1519         spa_close(spa, FTAG);
1520
1521         if (needsuspend) {
1522                 suspended = fnvlist_alloc();
1523                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1524                     pair = nvlist_next_nvpair(snaps, pair)) {
1525                         char fsname[ZFS_MAX_DATASET_NAME_LEN];
1526                         char *snapname = nvpair_name(pair);
1527                         char *atp;
1528                         void *cookie;
1529
1530                         atp = strchr(snapname, '@');
1531                         if (atp == NULL) {
1532                                 error = SET_ERROR(EINVAL);
1533                                 break;
1534                         }
1535                         (void) strlcpy(fsname, snapname, atp - snapname + 1);
1536
1537                         error = zil_suspend(fsname, &cookie);
1538                         if (error != 0)
1539                                 break;
1540                         fnvlist_add_uint64(suspended, fsname,
1541                             (uintptr_t)cookie);
1542                 }
1543         }
1544
1545         ddsa.ddsa_snaps = snaps;
1546         ddsa.ddsa_props = props;
1547         ddsa.ddsa_errors = errors;
1548         ddsa.ddsa_cr = CRED();
1549
1550         if (error == 0) {
1551                 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1552                     dsl_dataset_snapshot_sync, &ddsa,
1553                     fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1554         }
1555
1556         if (suspended != NULL) {
1557                 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1558                     pair = nvlist_next_nvpair(suspended, pair)) {
1559                         zil_resume((void *)(uintptr_t)
1560                             fnvpair_value_uint64(pair));
1561                 }
1562                 fnvlist_free(suspended);
1563         }
1564
1565 #ifdef __FreeBSD__
1566 #ifdef _KERNEL
1567         if (error == 0) {
1568                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1569                     pair = nvlist_next_nvpair(snaps, pair)) {
1570                         char *snapname = nvpair_name(pair);
1571                         zvol_create_minors(snapname);
1572                 }
1573         }
1574 #endif
1575 #endif
1576         return (error);
1577 }
1578
1579 typedef struct dsl_dataset_snapshot_tmp_arg {
1580         const char *ddsta_fsname;
1581         const char *ddsta_snapname;
1582         minor_t ddsta_cleanup_minor;
1583         const char *ddsta_htag;
1584 } dsl_dataset_snapshot_tmp_arg_t;
1585
1586 static int
1587 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1588 {
1589         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1590         dsl_pool_t *dp = dmu_tx_pool(tx);
1591         dsl_dataset_t *ds;
1592         int error;
1593
1594         error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1595         if (error != 0)
1596                 return (error);
1597
1598         /* NULL cred means no limit check for tmp snapshot */
1599         error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1600             tx, B_FALSE, 0, NULL);
1601         if (error != 0) {
1602                 dsl_dataset_rele(ds, FTAG);
1603                 return (error);
1604         }
1605
1606         if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1607                 dsl_dataset_rele(ds, FTAG);
1608                 return (SET_ERROR(ENOTSUP));
1609         }
1610         error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1611             B_TRUE, tx);
1612         if (error != 0) {
1613                 dsl_dataset_rele(ds, FTAG);
1614                 return (error);
1615         }
1616
1617         dsl_dataset_rele(ds, FTAG);
1618         return (0);
1619 }
1620
1621 static void
1622 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1623 {
1624         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1625         dsl_pool_t *dp = dmu_tx_pool(tx);
1626         dsl_dataset_t *ds;
1627
1628         VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1629
1630         dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1631         dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1632             ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1633         dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1634
1635         dsl_dataset_rele(ds, FTAG);
1636 }
1637
1638 int
1639 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1640     minor_t cleanup_minor, const char *htag)
1641 {
1642         dsl_dataset_snapshot_tmp_arg_t ddsta;
1643         int error;
1644         spa_t *spa;
1645         boolean_t needsuspend;
1646         void *cookie;
1647
1648         ddsta.ddsta_fsname = fsname;
1649         ddsta.ddsta_snapname = snapname;
1650         ddsta.ddsta_cleanup_minor = cleanup_minor;
1651         ddsta.ddsta_htag = htag;
1652
1653         error = spa_open(fsname, &spa, FTAG);
1654         if (error != 0)
1655                 return (error);
1656         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1657         spa_close(spa, FTAG);
1658
1659         if (needsuspend) {
1660                 error = zil_suspend(fsname, &cookie);
1661                 if (error != 0)
1662                         return (error);
1663         }
1664
1665         error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1666             dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1667
1668         if (needsuspend)
1669                 zil_resume(cookie);
1670         return (error);
1671 }
1672
1673
1674 void
1675 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1676 {
1677         ASSERT(dmu_tx_is_syncing(tx));
1678         ASSERT(ds->ds_objset != NULL);
1679         ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1680
1681         /*
1682          * in case we had to change ds_fsid_guid when we opened it,
1683          * sync it out now.
1684          */
1685         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1686         dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1687
1688         if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1689                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1690                     ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1691                     &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1692                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1693                     ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1694                     &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1695                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1696                     ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1697                     &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1698                 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1699                 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1700                 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1701         }
1702
1703         dmu_objset_sync(ds->ds_objset, zio, tx);
1704
1705         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1706                 if (ds->ds_feature_activation_needed[f]) {
1707                         if (ds->ds_feature_inuse[f])
1708                                 continue;
1709                         dsl_dataset_activate_feature(ds->ds_object, f, tx);
1710                         ds->ds_feature_inuse[f] = B_TRUE;
1711                 }
1712         }
1713 }
1714
1715 static void
1716 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1717 {
1718         uint64_t count = 0;
1719         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1720         zap_cursor_t zc;
1721         zap_attribute_t za;
1722         nvlist_t *propval = fnvlist_alloc();
1723         nvlist_t *val = fnvlist_alloc();
1724
1725         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1726
1727         /*
1728          * There may be missing entries in ds_next_clones_obj
1729          * due to a bug in a previous version of the code.
1730          * Only trust it if it has the right number of entries.
1731          */
1732         if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1733                 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1734                     &count));
1735         }
1736         if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1737                 goto fail;
1738         for (zap_cursor_init(&zc, mos,
1739             dsl_dataset_phys(ds)->ds_next_clones_obj);
1740             zap_cursor_retrieve(&zc, &za) == 0;
1741             zap_cursor_advance(&zc)) {
1742                 dsl_dataset_t *clone;
1743                 char buf[ZFS_MAX_DATASET_NAME_LEN];
1744                 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1745                     za.za_first_integer, FTAG, &clone));
1746                 dsl_dir_name(clone->ds_dir, buf);
1747                 fnvlist_add_boolean(val, buf);
1748                 dsl_dataset_rele(clone, FTAG);
1749         }
1750         zap_cursor_fini(&zc);
1751         fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1752         fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1753 fail:
1754         nvlist_free(val);
1755         nvlist_free(propval);
1756 }
1757
1758 static void
1759 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1760 {
1761         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1762
1763         if (dsl_dataset_has_resume_receive_state(ds)) {
1764                 char *str;
1765                 void *packed;
1766                 uint8_t *compressed;
1767                 uint64_t val;
1768                 nvlist_t *token_nv = fnvlist_alloc();
1769                 size_t packed_size, compressed_size;
1770
1771                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1772                     DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1773                         fnvlist_add_uint64(token_nv, "fromguid", val);
1774                 }
1775                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1776                     DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1777                         fnvlist_add_uint64(token_nv, "object", val);
1778                 }
1779                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1780                     DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1781                         fnvlist_add_uint64(token_nv, "offset", val);
1782                 }
1783                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1784                     DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1785                         fnvlist_add_uint64(token_nv, "bytes", val);
1786                 }
1787                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1788                     DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1789                         fnvlist_add_uint64(token_nv, "toguid", val);
1790                 }
1791                 char buf[256];
1792                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1793                     DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1794                         fnvlist_add_string(token_nv, "toname", buf);
1795                 }
1796                 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1797                     DS_FIELD_RESUME_EMBEDOK) == 0) {
1798                         fnvlist_add_boolean(token_nv, "embedok");
1799                 }
1800                 packed = fnvlist_pack(token_nv, &packed_size);
1801                 fnvlist_free(token_nv);
1802                 compressed = kmem_alloc(packed_size, KM_SLEEP);
1803
1804                 compressed_size = gzip_compress(packed, compressed,
1805                     packed_size, packed_size, 6);
1806
1807                 zio_cksum_t cksum;
1808                 fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1809
1810                 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1811                 for (int i = 0; i < compressed_size; i++) {
1812                         (void) sprintf(str + i * 2, "%02x", compressed[i]);
1813                 }
1814                 str[compressed_size * 2] = '\0';
1815                 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1816                     ZFS_SEND_RESUME_TOKEN_VERSION,
1817                     (longlong_t)cksum.zc_word[0],
1818                     (longlong_t)packed_size, str);
1819                 dsl_prop_nvlist_add_string(nv,
1820                     ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1821                 kmem_free(packed, packed_size);
1822                 kmem_free(str, compressed_size * 2 + 1);
1823                 kmem_free(compressed, packed_size);
1824                 strfree(propval);
1825         }
1826 }
1827
1828 void
1829 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1830 {
1831         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1832         uint64_t refd, avail, uobjs, aobjs, ratio;
1833
1834         ASSERT(dsl_pool_config_held(dp));
1835
1836         ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1837             (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1838             dsl_dataset_phys(ds)->ds_compressed_bytes);
1839
1840         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1841         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1842             dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1843
1844         if (ds->ds_is_snapshot) {
1845                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1846                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1847                     dsl_dataset_phys(ds)->ds_unique_bytes);
1848                 get_clones_stat(ds, nv);
1849         } else {
1850                 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1851                         char buf[ZFS_MAX_DATASET_NAME_LEN];
1852                         dsl_dataset_name(ds->ds_prev, buf);
1853                         dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1854                 }
1855
1856                 dsl_dir_stats(ds->ds_dir, nv);
1857         }
1858
1859         dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1860         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1861         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1862
1863         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1864             dsl_dataset_phys(ds)->ds_creation_time);
1865         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1866             dsl_dataset_phys(ds)->ds_creation_txg);
1867         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1868             ds->ds_quota);
1869         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1870             ds->ds_reserved);
1871         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1872             dsl_dataset_phys(ds)->ds_guid);
1873         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1874             dsl_dataset_phys(ds)->ds_unique_bytes);
1875         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1876             ds->ds_object);
1877         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1878             ds->ds_userrefs);
1879         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1880             DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1881
1882         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1883                 uint64_t written, comp, uncomp;
1884                 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1885                 dsl_dataset_t *prev;
1886
1887                 int err = dsl_dataset_hold_obj(dp,
1888                     dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1889                 if (err == 0) {
1890                         err = dsl_dataset_space_written(prev, ds, &written,
1891                             &comp, &uncomp);
1892                         dsl_dataset_rele(prev, FTAG);
1893                         if (err == 0) {
1894                                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1895                                     written);
1896                         }
1897                 }
1898         }
1899
1900         if (!dsl_dataset_is_snapshot(ds)) {
1901                 /*
1902                  * A failed "newfs" (e.g. full) resumable receive leaves
1903                  * the stats set on this dataset.  Check here for the prop.
1904                  */
1905                 get_receive_resume_stats(ds, nv);
1906
1907                 /*
1908                  * A failed incremental resumable receive leaves the
1909                  * stats set on our child named "%recv".  Check the child
1910                  * for the prop.
1911                  */
1912                 /* 6 extra bytes for /%recv */
1913                 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1914                 dsl_dataset_t *recv_ds;
1915                 dsl_dataset_name(ds, recvname);
1916                 if (strlcat(recvname, "/", sizeof (recvname)) <
1917                     sizeof (recvname) &&
1918                     strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1919                     sizeof (recvname) &&
1920                     dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1921                         get_receive_resume_stats(recv_ds, nv);
1922                         dsl_dataset_rele(recv_ds, FTAG);
1923                 }
1924         }
1925 }
1926
1927 void
1928 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1929 {
1930         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1931         ASSERT(dsl_pool_config_held(dp));
1932
1933         stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1934         stat->dds_inconsistent =
1935             dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1936         stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1937         stat->dds_origin[0] = '\0';
1938         if (ds->ds_is_snapshot) {
1939                 stat->dds_is_snapshot = B_TRUE;
1940                 stat->dds_num_clones =
1941                     dsl_dataset_phys(ds)->ds_num_children - 1;
1942         } else {
1943                 stat->dds_is_snapshot = B_FALSE;
1944                 stat->dds_num_clones = 0;
1945
1946                 if (dsl_dir_is_clone(ds->ds_dir)) {
1947                         dsl_dataset_t *ods;
1948
1949                         VERIFY0(dsl_dataset_hold_obj(dp,
1950                             dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1951                             FTAG, &ods));
1952                         dsl_dataset_name(ods, stat->dds_origin);
1953                         dsl_dataset_rele(ods, FTAG);
1954                 }
1955         }
1956 }
1957
1958 uint64_t
1959 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1960 {
1961         return (ds->ds_fsid_guid);
1962 }
1963
1964 void
1965 dsl_dataset_space(dsl_dataset_t *ds,
1966     uint64_t *refdbytesp, uint64_t *availbytesp,
1967     uint64_t *usedobjsp, uint64_t *availobjsp)
1968 {
1969         *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1970         *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1971         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1972                 *availbytesp +=
1973                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1974         if (ds->ds_quota != 0) {
1975                 /*
1976                  * Adjust available bytes according to refquota
1977                  */
1978                 if (*refdbytesp < ds->ds_quota)
1979                         *availbytesp = MIN(*availbytesp,
1980                             ds->ds_quota - *refdbytesp);
1981                 else
1982                         *availbytesp = 0;
1983         }
1984         *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1985         *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1986 }
1987
1988 boolean_t
1989 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1990 {
1991         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1992
1993         ASSERT(dsl_pool_config_held(dp));
1994         if (snap == NULL)
1995                 return (B_FALSE);
1996         if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1997             dsl_dataset_phys(snap)->ds_creation_txg) {
1998                 objset_t *os, *os_snap;
1999                 /*
2000                  * It may be that only the ZIL differs, because it was
2001                  * reset in the head.  Don't count that as being
2002                  * modified.
2003                  */
2004                 if (dmu_objset_from_ds(ds, &os) != 0)
2005                         return (B_TRUE);
2006                 if (dmu_objset_from_ds(snap, &os_snap) != 0)
2007                         return (B_TRUE);
2008                 return (bcmp(&os->os_phys->os_meta_dnode,
2009                     &os_snap->os_phys->os_meta_dnode,
2010                     sizeof (os->os_phys->os_meta_dnode)) != 0);
2011         }
2012         return (B_FALSE);
2013 }
2014
2015 typedef struct dsl_dataset_rename_snapshot_arg {
2016         const char *ddrsa_fsname;
2017         const char *ddrsa_oldsnapname;
2018         const char *ddrsa_newsnapname;
2019         boolean_t ddrsa_recursive;
2020         dmu_tx_t *ddrsa_tx;
2021 } dsl_dataset_rename_snapshot_arg_t;
2022
2023 /* ARGSUSED */
2024 static int
2025 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2026     dsl_dataset_t *hds, void *arg)
2027 {
2028         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2029         int error;
2030         uint64_t val;
2031
2032         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2033         if (error != 0) {
2034                 /* ignore nonexistent snapshots */
2035                 return (error == ENOENT ? 0 : error);
2036         }
2037
2038         /* new name should not exist */
2039         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2040         if (error == 0)
2041                 error = SET_ERROR(EEXIST);
2042         else if (error == ENOENT)
2043                 error = 0;
2044
2045         /* dataset name + 1 for the "@" + the new snapshot name must fit */
2046         if (dsl_dir_namelen(hds->ds_dir) + 1 +
2047             strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2048                 error = SET_ERROR(ENAMETOOLONG);
2049
2050         return (error);
2051 }
2052
2053 static int
2054 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2055 {
2056         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2057         dsl_pool_t *dp = dmu_tx_pool(tx);
2058         dsl_dataset_t *hds;
2059         int error;
2060
2061         error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2062         if (error != 0)
2063                 return (error);
2064
2065         if (ddrsa->ddrsa_recursive) {
2066                 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2067                     dsl_dataset_rename_snapshot_check_impl, ddrsa,
2068                     DS_FIND_CHILDREN);
2069         } else {
2070                 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2071         }
2072         dsl_dataset_rele(hds, FTAG);
2073         return (error);
2074 }
2075
2076 static int
2077 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2078     dsl_dataset_t *hds, void *arg)
2079 {
2080 #ifdef __FreeBSD__
2081 #ifdef _KERNEL
2082         char *oldname, *newname;
2083 #endif
2084 #endif
2085         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2086         dsl_dataset_t *ds;
2087         uint64_t val;
2088         dmu_tx_t *tx = ddrsa->ddrsa_tx;
2089         int error;
2090
2091         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2092         ASSERT(error == 0 || error == ENOENT);
2093         if (error == ENOENT) {
2094                 /* ignore nonexistent snapshots */
2095                 return (0);
2096         }
2097
2098         VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2099
2100         /* log before we change the name */
2101         spa_history_log_internal_ds(ds, "rename", tx,
2102             "-> @%s", ddrsa->ddrsa_newsnapname);
2103
2104         VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2105             B_FALSE));
2106         mutex_enter(&ds->ds_lock);
2107         (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2108         mutex_exit(&ds->ds_lock);
2109         VERIFY0(zap_add(dp->dp_meta_objset,
2110             dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2111             ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2112
2113 #ifdef __FreeBSD__
2114 #ifdef _KERNEL
2115         oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2116         newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2117         snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
2118             ddrsa->ddrsa_oldsnapname);
2119         snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
2120             ddrsa->ddrsa_newsnapname);
2121         zfsvfs_update_fromname(oldname, newname);
2122         zvol_rename_minors(oldname, newname);
2123         kmem_free(newname, MAXPATHLEN);
2124         kmem_free(oldname, MAXPATHLEN);
2125 #endif
2126 #endif
2127         dsl_dataset_rele(ds, FTAG);
2128
2129         return (0);
2130 }
2131
2132 static void
2133 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2134 {
2135         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2136         dsl_pool_t *dp = dmu_tx_pool(tx);
2137         dsl_dataset_t *hds;
2138
2139         VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2140         ddrsa->ddrsa_tx = tx;
2141         if (ddrsa->ddrsa_recursive) {
2142                 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2143                     dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2144                     DS_FIND_CHILDREN));
2145         } else {
2146                 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2147         }
2148         dsl_dataset_rele(hds, FTAG);
2149 }
2150
2151 int
2152 dsl_dataset_rename_snapshot(const char *fsname,
2153     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2154 {
2155         dsl_dataset_rename_snapshot_arg_t ddrsa;
2156
2157         ddrsa.ddrsa_fsname = fsname;
2158         ddrsa.ddrsa_oldsnapname = oldsnapname;
2159         ddrsa.ddrsa_newsnapname = newsnapname;
2160         ddrsa.ddrsa_recursive = recursive;
2161
2162         return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2163             dsl_dataset_rename_snapshot_sync, &ddrsa,
2164             1, ZFS_SPACE_CHECK_RESERVED));
2165 }
2166
2167 /*
2168  * If we're doing an ownership handoff, we need to make sure that there is
2169  * only one long hold on the dataset.  We're not allowed to change anything here
2170  * so we don't permanently release the long hold or regular hold here.  We want
2171  * to do this only when syncing to avoid the dataset unexpectedly going away
2172  * when we release the long hold.
2173  */
2174 static int
2175 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2176 {
2177         boolean_t held;
2178
2179         if (!dmu_tx_is_syncing(tx))
2180                 return (0);
2181
2182         if (owner != NULL) {
2183                 VERIFY3P(ds->ds_owner, ==, owner);
2184                 dsl_dataset_long_rele(ds, owner);
2185         }
2186
2187         held = dsl_dataset_long_held(ds);
2188
2189         if (owner != NULL)
2190                 dsl_dataset_long_hold(ds, owner);
2191
2192         if (held)
2193                 return (SET_ERROR(EBUSY));
2194
2195         return (0);
2196 }
2197
2198 typedef struct dsl_dataset_rollback_arg {
2199         const char *ddra_fsname;
2200         void *ddra_owner;
2201         nvlist_t *ddra_result;
2202 } dsl_dataset_rollback_arg_t;
2203
2204 static int
2205 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2206 {
2207         dsl_dataset_rollback_arg_t *ddra = arg;
2208         dsl_pool_t *dp = dmu_tx_pool(tx);
2209         dsl_dataset_t *ds;
2210         int64_t unused_refres_delta;
2211         int error;
2212
2213         error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2214         if (error != 0)
2215                 return (error);
2216
2217         /* must not be a snapshot */
2218         if (ds->ds_is_snapshot) {
2219                 dsl_dataset_rele(ds, FTAG);
2220                 return (SET_ERROR(EINVAL));
2221         }
2222
2223         /* must have a most recent snapshot */
2224         if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2225                 dsl_dataset_rele(ds, FTAG);
2226                 return (SET_ERROR(EINVAL));
2227         }
2228
2229         /* must not have any bookmarks after the most recent snapshot */
2230         nvlist_t *proprequest = fnvlist_alloc();
2231         fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2232         nvlist_t *bookmarks = fnvlist_alloc();
2233         error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2234         fnvlist_free(proprequest);
2235         if (error != 0)
2236                 return (error);
2237         for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2238             pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2239                 nvlist_t *valuenv =
2240                     fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2241                     zfs_prop_to_name(ZFS_PROP_CREATETXG));
2242                 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2243                 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2244                         fnvlist_free(bookmarks);
2245                         dsl_dataset_rele(ds, FTAG);
2246                         return (SET_ERROR(EEXIST));
2247                 }
2248         }
2249         fnvlist_free(bookmarks);
2250
2251         error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2252         if (error != 0) {
2253                 dsl_dataset_rele(ds, FTAG);
2254                 return (error);
2255         }
2256
2257         /*
2258          * Check if the snap we are rolling back to uses more than
2259          * the refquota.
2260          */
2261         if (ds->ds_quota != 0 &&
2262             dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2263                 dsl_dataset_rele(ds, FTAG);
2264                 return (SET_ERROR(EDQUOT));
2265         }
2266
2267         /*
2268          * When we do the clone swap, we will temporarily use more space
2269          * due to the refreservation (the head will no longer have any
2270          * unique space, so the entire amount of the refreservation will need
2271          * to be free).  We will immediately destroy the clone, freeing
2272          * this space, but the freeing happens over many txg's.
2273          */
2274         unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2275             dsl_dataset_phys(ds)->ds_unique_bytes);
2276
2277         if (unused_refres_delta > 0 &&
2278             unused_refres_delta >
2279             dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2280                 dsl_dataset_rele(ds, FTAG);
2281                 return (SET_ERROR(ENOSPC));
2282         }
2283
2284         dsl_dataset_rele(ds, FTAG);
2285         return (0);
2286 }
2287
2288 static void
2289 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2290 {
2291         dsl_dataset_rollback_arg_t *ddra = arg;
2292         dsl_pool_t *dp = dmu_tx_pool(tx);
2293         dsl_dataset_t *ds, *clone;
2294         uint64_t cloneobj;
2295         char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2296
2297         VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2298
2299         dsl_dataset_name(ds->ds_prev, namebuf);
2300         fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2301
2302         cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2303             ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2304
2305         VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2306
2307         dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2308         dsl_dataset_zero_zil(ds, tx);
2309
2310         dsl_destroy_head_sync_impl(clone, tx);
2311
2312         dsl_dataset_rele(clone, FTAG);
2313         dsl_dataset_rele(ds, FTAG);
2314 }
2315
2316 /*
2317  * Rolls back the given filesystem or volume to the most recent snapshot.
2318  * The name of the most recent snapshot will be returned under key "target"
2319  * in the result nvlist.
2320  *
2321  * If owner != NULL:
2322  * - The existing dataset MUST be owned by the specified owner at entry
2323  * - Upon return, dataset will still be held by the same owner, whether we
2324  *   succeed or not.
2325  *
2326  * This mode is required any time the existing filesystem is mounted.  See
2327  * notes above zfs_suspend_fs() for further details.
2328  */
2329 int
2330 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2331 {
2332         dsl_dataset_rollback_arg_t ddra;
2333
2334         ddra.ddra_fsname = fsname;
2335         ddra.ddra_owner = owner;
2336         ddra.ddra_result = result;
2337
2338         return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2339             dsl_dataset_rollback_sync, &ddra,
2340             1, ZFS_SPACE_CHECK_RESERVED));
2341 }
2342
2343 struct promotenode {
2344         list_node_t link;
2345         dsl_dataset_t *ds;
2346 };
2347
2348 typedef struct dsl_dataset_promote_arg {
2349         const char *ddpa_clonename;
2350         dsl_dataset_t *ddpa_clone;
2351         list_t shared_snaps, origin_snaps, clone_snaps;
2352         dsl_dataset_t *origin_origin; /* origin of the origin */
2353         uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2354         char *err_ds;
2355         cred_t *cr;
2356 } dsl_dataset_promote_arg_t;
2357
2358 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2359 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2360     void *tag);
2361 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2362
2363 static int
2364 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2365 {
2366         dsl_dataset_promote_arg_t *ddpa = arg;
2367         dsl_pool_t *dp = dmu_tx_pool(tx);
2368         dsl_dataset_t *hds;
2369         struct promotenode *snap;
2370         dsl_dataset_t *origin_ds;
2371         int err;
2372         uint64_t unused;
2373         uint64_t ss_mv_cnt;
2374         size_t max_snap_len;
2375
2376         err = promote_hold(ddpa, dp, FTAG);
2377         if (err != 0)
2378                 return (err);
2379
2380         hds = ddpa->ddpa_clone;
2381         max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2382
2383         if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2384                 promote_rele(ddpa, FTAG);
2385                 return (SET_ERROR(EXDEV));
2386         }
2387
2388         /*
2389          * Compute and check the amount of space to transfer.  Since this is
2390          * so expensive, don't do the preliminary check.
2391          */
2392         if (!dmu_tx_is_syncing(tx)) {
2393                 promote_rele(ddpa, FTAG);
2394                 return (0);
2395         }
2396
2397         snap = list_head(&ddpa->shared_snaps);
2398         origin_ds = snap->ds;
2399
2400         /* compute origin's new unique space */
2401         snap = list_tail(&ddpa->clone_snaps);
2402         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2403             origin_ds->ds_object);
2404         dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2405             dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2406             &ddpa->unique, &unused, &unused);
2407
2408         /*
2409          * Walk the snapshots that we are moving
2410          *
2411          * Compute space to transfer.  Consider the incremental changes
2412          * to used by each snapshot:
2413          * (my used) = (prev's used) + (blocks born) - (blocks killed)
2414          * So each snapshot gave birth to:
2415          * (blocks born) = (my used) - (prev's used) + (blocks killed)
2416          * So a sequence would look like:
2417          * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2418          * Which simplifies to:
2419          * uN + kN + kN-1 + ... + k1 + k0
2420          * Note however, if we stop before we reach the ORIGIN we get:
2421          * uN + kN + kN-1 + ... + kM - uM-1
2422          */
2423         ss_mv_cnt = 0;
2424         ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2425         ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2426         ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2427         for (snap = list_head(&ddpa->shared_snaps); snap;
2428             snap = list_next(&ddpa->shared_snaps, snap)) {
2429                 uint64_t val, dlused, dlcomp, dluncomp;
2430                 dsl_dataset_t *ds = snap->ds;
2431
2432                 ss_mv_cnt++;
2433
2434                 /*
2435                  * If there are long holds, we won't be able to evict
2436                  * the objset.
2437                  */
2438                 if (dsl_dataset_long_held(ds)) {
2439                         err = SET_ERROR(EBUSY);
2440                         goto out;
2441                 }
2442
2443                 /* Check that the snapshot name does not conflict */
2444                 VERIFY0(dsl_dataset_get_snapname(ds));
2445                 if (strlen(ds->ds_snapname) >= max_snap_len) {
2446                         err = SET_ERROR(ENAMETOOLONG);
2447                         goto out;
2448                 }
2449                 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2450                 if (err == 0) {
2451                         (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2452                         err = SET_ERROR(EEXIST);
2453                         goto out;
2454                 }
2455                 if (err != ENOENT)
2456                         goto out;
2457
2458                 /* The very first snapshot does not have a deadlist */
2459                 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2460                         continue;
2461
2462                 dsl_deadlist_space(&ds->ds_deadlist,
2463                     &dlused, &dlcomp, &dluncomp);
2464                 ddpa->used += dlused;
2465                 ddpa->comp += dlcomp;
2466                 ddpa->uncomp += dluncomp;
2467         }
2468
2469         /*
2470          * If we are a clone of a clone then we never reached ORIGIN,
2471          * so we need to subtract out the clone origin's used space.
2472          */
2473         if (ddpa->origin_origin) {
2474                 ddpa->used -=
2475                     dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2476                 ddpa->comp -=
2477                     dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2478                 ddpa->uncomp -=
2479                     dsl_dataset_phys(ddpa->origin_origin)->
2480                     ds_uncompressed_bytes;
2481         }
2482
2483         /* Check that there is enough space and limit headroom here */
2484         err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2485             0, ss_mv_cnt, ddpa->used, ddpa->cr);
2486         if (err != 0)
2487                 goto out;
2488
2489         /*
2490          * Compute the amounts of space that will be used by snapshots
2491          * after the promotion (for both origin and clone).  For each,
2492          * it is the amount of space that will be on all of their
2493          * deadlists (that was not born before their new origin).
2494          */
2495         if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2496                 uint64_t space;
2497
2498                 /*
2499                  * Note, typically this will not be a clone of a clone,
2500                  * so dd_origin_txg will be < TXG_INITIAL, so
2501                  * these snaplist_space() -> dsl_deadlist_space_range()
2502                  * calls will be fast because they do not have to
2503                  * iterate over all bps.
2504                  */
2505                 snap = list_head(&ddpa->origin_snaps);
2506                 err = snaplist_space(&ddpa->shared_snaps,
2507                     snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2508                 if (err != 0)
2509                         goto out;
2510
2511                 err = snaplist_space(&ddpa->clone_snaps,
2512                     snap->ds->ds_dir->dd_origin_txg, &space);
2513                 if (err != 0)
2514                         goto out;
2515                 ddpa->cloneusedsnap += space;
2516         }
2517         if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2518             DD_FLAG_USED_BREAKDOWN) {
2519                 err = snaplist_space(&ddpa->origin_snaps,
2520                     dsl_dataset_phys(origin_ds)->ds_creation_txg,
2521                     &ddpa->originusedsnap);
2522                 if (err != 0)
2523                         goto out;
2524         }
2525
2526 out:
2527         promote_rele(ddpa, FTAG);
2528         return (err);
2529 }
2530
2531 static void
2532 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2533 {
2534         dsl_dataset_promote_arg_t *ddpa = arg;
2535         dsl_pool_t *dp = dmu_tx_pool(tx);
2536         dsl_dataset_t *hds;
2537         struct promotenode *snap;
2538         dsl_dataset_t *origin_ds;
2539         dsl_dataset_t *origin_head;
2540         dsl_dir_t *dd;
2541         dsl_dir_t *odd = NULL;
2542         uint64_t oldnext_obj;
2543         int64_t delta;
2544 #if defined(__FreeBSD__) && defined(_KERNEL)
2545         char *oldname, *newname;
2546 #endif
2547
2548         VERIFY0(promote_hold(ddpa, dp, FTAG));
2549         hds = ddpa->ddpa_clone;
2550
2551         ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2552
2553         snap = list_head(&ddpa->shared_snaps);
2554         origin_ds = snap->ds;
2555         dd = hds->ds_dir;
2556
2557         snap = list_head(&ddpa->origin_snaps);
2558         origin_head = snap->ds;
2559
2560         /*
2561          * We need to explicitly open odd, since origin_ds's dd will be
2562          * changing.
2563          */
2564         VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2565             NULL, FTAG, &odd));
2566
2567         /* change origin's next snap */
2568         dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2569         oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2570         snap = list_tail(&ddpa->clone_snaps);
2571         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2572             origin_ds->ds_object);
2573         dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2574
2575         /* change the origin's next clone */
2576         if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2577                 dsl_dataset_remove_from_next_clones(origin_ds,
2578                     snap->ds->ds_object, tx);
2579                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2580                     dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2581                     oldnext_obj, tx));
2582         }
2583
2584         /* change origin */
2585         dmu_buf_will_dirty(dd->dd_dbuf, tx);
2586         ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2587         dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2588         dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2589         dmu_buf_will_dirty(odd->dd_dbuf, tx);
2590         dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2591         origin_head->ds_dir->dd_origin_txg =
2592             dsl_dataset_phys(origin_ds)->ds_creation_txg;
2593
2594         /* change dd_clone entries */
2595         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2596                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2597                     dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2598                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2599                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2600                     hds->ds_object, tx));
2601
2602                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2603                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2604                     origin_head->ds_object, tx));
2605                 if (dsl_dir_phys(dd)->dd_clones == 0) {
2606                         dsl_dir_phys(dd)->dd_clones =
2607                             zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2608                             DMU_OT_NONE, 0, tx);
2609                 }
2610                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2611                     dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2612         }
2613
2614 #if defined(__FreeBSD__) && defined(_KERNEL)
2615         /* Take the spa_namespace_lock early so zvol renames don't deadlock. */
2616         mutex_enter(&spa_namespace_lock);
2617
2618         oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2619         newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2620 #endif
2621
2622         /* move snapshots to this dir */
2623         for (snap = list_head(&ddpa->shared_snaps); snap;
2624             snap = list_next(&ddpa->shared_snaps, snap)) {
2625                 dsl_dataset_t *ds = snap->ds;
2626
2627                 /*
2628                  * Property callbacks are registered to a particular
2629                  * dsl_dir.  Since ours is changing, evict the objset
2630                  * so that they will be unregistered from the old dsl_dir.
2631                  */
2632                 if (ds->ds_objset) {
2633                         dmu_objset_evict(ds->ds_objset);
2634                         ds->ds_objset = NULL;
2635                 }
2636
2637                 /* move snap name entry */
2638                 VERIFY0(dsl_dataset_get_snapname(ds));
2639                 VERIFY0(dsl_dataset_snap_remove(origin_head,
2640                     ds->ds_snapname, tx, B_TRUE));
2641                 VERIFY0(zap_add(dp->dp_meta_objset,
2642                     dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2643                     8, 1, &ds->ds_object, tx));
2644                 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2645                     DD_FIELD_SNAPSHOT_COUNT, tx);
2646
2647                 /* change containing dsl_dir */
2648                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2649                 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2650                 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2651                 ASSERT3P(ds->ds_dir, ==, odd);
2652                 dsl_dir_rele(ds->ds_dir, ds);
2653                 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2654                     NULL, ds, &ds->ds_dir));
2655
2656 #if defined(__FreeBSD__) && defined(_KERNEL)
2657                 dsl_dataset_name(ds, newname);
2658                 zfsvfs_update_fromname(oldname, newname);
2659                 zvol_rename_minors(oldname, newname);
2660 #endif
2661
2662                 /* move any clone references */
2663                 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2664                     spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2665                         zap_cursor_t zc;
2666                         zap_attribute_t za;
2667
2668                         for (zap_cursor_init(&zc, dp->dp_meta_objset,
2669                             dsl_dataset_phys(ds)->ds_next_clones_obj);
2670                             zap_cursor_retrieve(&zc, &za) == 0;
2671                             zap_cursor_advance(&zc)) {
2672                                 dsl_dataset_t *cnds;
2673                                 uint64_t o;
2674
2675                                 if (za.za_first_integer == oldnext_obj) {
2676                                         /*
2677                                          * We've already moved the
2678                                          * origin's reference.
2679                                          */
2680                                         continue;
2681                                 }
2682
2683                                 VERIFY0(dsl_dataset_hold_obj(dp,
2684                                     za.za_first_integer, FTAG, &cnds));
2685                                 o = dsl_dir_phys(cnds->ds_dir)->
2686                                     dd_head_dataset_obj;
2687
2688                                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2689                                     dsl_dir_phys(odd)->dd_clones, o, tx));
2690                                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2691                                     dsl_dir_phys(dd)->dd_clones, o, tx));
2692                                 dsl_dataset_rele(cnds, FTAG);
2693                         }
2694                         zap_cursor_fini(&zc);
2695                 }
2696
2697                 ASSERT(!dsl_prop_hascb(ds));
2698         }
2699
2700 #if defined(__FreeBSD__) && defined(_KERNEL)
2701         mutex_exit(&spa_namespace_lock);
2702
2703         kmem_free(newname, MAXPATHLEN);
2704         kmem_free(oldname, MAXPATHLEN);
2705 #endif
2706         /*
2707          * Change space accounting.
2708          * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2709          * both be valid, or both be 0 (resulting in delta == 0).  This
2710          * is true for each of {clone,origin} independently.
2711          */
2712
2713         delta = ddpa->cloneusedsnap -
2714             dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2715         ASSERT3S(delta, >=, 0);
2716         ASSERT3U(ddpa->used, >=, delta);
2717         dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2718         dsl_dir_diduse_space(dd, DD_USED_HEAD,
2719             ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2720
2721         delta = ddpa->originusedsnap -
2722             dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2723         ASSERT3S(delta, <=, 0);
2724         ASSERT3U(ddpa->used, >=, -delta);
2725         dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2726         dsl_dir_diduse_space(odd, DD_USED_HEAD,
2727             -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2728
2729         dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2730
2731         /* log history record */
2732         spa_history_log_internal_ds(hds, "promote", tx, "");
2733
2734         dsl_dir_rele(odd, FTAG);
2735         promote_rele(ddpa, FTAG);
2736 }
2737
2738 /*
2739  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2740  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2741  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2742  * snapshots back to this dataset's origin.
2743  */
2744 static int
2745 snaplist_make(dsl_pool_t *dp,
2746     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2747 {
2748         uint64_t obj = last_obj;
2749
2750         list_create(l, sizeof (struct promotenode),
2751             offsetof(struct promotenode, link));
2752
2753         while (obj != first_obj) {
2754                 dsl_dataset_t *ds;
2755                 struct promotenode *snap;
2756                 int err;
2757
2758                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2759                 ASSERT(err != ENOENT);
2760                 if (err != 0)
2761                         return (err);
2762
2763                 if (first_obj == 0)
2764                         first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2765
2766                 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2767                 snap->ds = ds;
2768                 list_insert_tail(l, snap);
2769                 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2770         }
2771
2772         return (0);
2773 }
2774
2775 static int
2776 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2777 {
2778         struct promotenode *snap;
2779
2780         *spacep = 0;
2781         for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2782                 uint64_t used, comp, uncomp;
2783                 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2784                     mintxg, UINT64_MAX, &used, &comp, &uncomp);
2785                 *spacep += used;
2786         }
2787         return (0);
2788 }
2789
2790 static void
2791 snaplist_destroy(list_t *l, void *tag)
2792 {
2793         struct promotenode *snap;
2794
2795         if (l == NULL || !list_link_active(&l->list_head))
2796                 return;
2797
2798         while ((snap = list_tail(l)) != NULL) {
2799                 list_remove(l, snap);
2800                 dsl_dataset_rele(snap->ds, tag);
2801                 kmem_free(snap, sizeof (*snap));
2802         }
2803         list_destroy(l);
2804 }
2805
2806 static int
2807 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2808 {
2809         int error;
2810         dsl_dir_t *dd;
2811         struct promotenode *snap;
2812
2813         error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2814             &ddpa->ddpa_clone);
2815         if (error != 0)
2816                 return (error);
2817         dd = ddpa->ddpa_clone->ds_dir;
2818
2819         if (ddpa->ddpa_clone->ds_is_snapshot ||
2820             !dsl_dir_is_clone(dd)) {
2821                 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2822                 return (SET_ERROR(EINVAL));
2823         }
2824
2825         error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2826             &ddpa->shared_snaps, tag);
2827         if (error != 0)
2828                 goto out;
2829
2830         error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2831             &ddpa->clone_snaps, tag);
2832         if (error != 0)
2833                 goto out;
2834
2835         snap = list_head(&ddpa->shared_snaps);
2836         ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2837         error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2838             dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2839             &ddpa->origin_snaps, tag);
2840         if (error != 0)
2841                 goto out;
2842
2843         if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2844                 error = dsl_dataset_hold_obj(dp,
2845                     dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2846                     tag, &ddpa->origin_origin);
2847                 if (error != 0)
2848                         goto out;
2849         }
2850 out:
2851         if (error != 0)
2852                 promote_rele(ddpa, tag);
2853         return (error);
2854 }
2855
2856 static void
2857 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2858 {
2859         snaplist_destroy(&ddpa->shared_snaps, tag);
2860         snaplist_destroy(&ddpa->clone_snaps, tag);
2861         snaplist_destroy(&ddpa->origin_snaps, tag);
2862         if (ddpa->origin_origin != NULL)
2863                 dsl_dataset_rele(ddpa->origin_origin, tag);
2864         dsl_dataset_rele(ddpa->ddpa_clone, tag);
2865 }
2866
2867 /*
2868  * Promote a clone.
2869  *
2870  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2871  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
2872  */
2873 int
2874 dsl_dataset_promote(const char *name, char *conflsnap)
2875 {
2876         dsl_dataset_promote_arg_t ddpa = { 0 };
2877         uint64_t numsnaps;
2878         int error;
2879         objset_t *os;
2880
2881         /*
2882          * We will modify space proportional to the number of
2883          * snapshots.  Compute numsnaps.
2884          */
2885         error = dmu_objset_hold(name, FTAG, &os);
2886         if (error != 0)
2887                 return (error);
2888         error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2889             dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2890             &numsnaps);
2891         dmu_objset_rele(os, FTAG);
2892         if (error != 0)
2893                 return (error);
2894
2895         ddpa.ddpa_clonename = name;
2896         ddpa.err_ds = conflsnap;
2897         ddpa.cr = CRED();
2898
2899         return (dsl_sync_task(name, dsl_dataset_promote_check,
2900             dsl_dataset_promote_sync, &ddpa,
2901             2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2902 }
2903
2904 int
2905 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2906     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2907 {
2908         /*
2909          * "slack" factor for received datasets with refquota set on them.
2910          * See the bottom of this function for details on its use.
2911          */
2912         uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
2913         int64_t unused_refres_delta;
2914
2915         /* they should both be heads */
2916         if (clone->ds_is_snapshot ||
2917             origin_head->ds_is_snapshot)
2918                 return (SET_ERROR(EINVAL));
2919
2920         /* if we are not forcing, the branch point should be just before them */
2921         if (!force && clone->ds_prev != origin_head->ds_prev)
2922                 return (SET_ERROR(EINVAL));
2923
2924         /* clone should be the clone (unless they are unrelated) */
2925         if (clone->ds_prev != NULL &&
2926             clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2927             origin_head->ds_dir != clone->ds_prev->ds_dir)
2928                 return (SET_ERROR(EINVAL));
2929
2930         /* the clone should be a child of the origin */
2931         if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2932                 return (SET_ERROR(EINVAL));
2933
2934         /* origin_head shouldn't be modified unless 'force' */
2935         if (!force &&
2936             dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2937                 return (SET_ERROR(ETXTBSY));
2938
2939         /* origin_head should have no long holds (e.g. is not mounted) */
2940         if (dsl_dataset_handoff_check(origin_head, owner, tx))
2941                 return (SET_ERROR(EBUSY));
2942
2943         /* check amount of any unconsumed refreservation */
2944         unused_refres_delta =
2945             (int64_t)MIN(origin_head->ds_reserved,
2946             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2947             (int64_t)MIN(origin_head->ds_reserved,
2948             dsl_dataset_phys(clone)->ds_unique_bytes);
2949
2950         if (unused_refres_delta > 0 &&
2951             unused_refres_delta >
2952             dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2953                 return (SET_ERROR(ENOSPC));
2954
2955         /*
2956          * The clone can't be too much over the head's refquota.
2957          *
2958          * To ensure that the entire refquota can be used, we allow one
2959          * transaction to exceed the the refquota.  Therefore, this check
2960          * needs to also allow for the space referenced to be more than the
2961          * refquota.  The maximum amount of space that one transaction can use
2962          * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
2963          * overage ensures that we are able to receive a filesystem that
2964          * exceeds the refquota on the source system.
2965          *
2966          * So that overage is the refquota_slack we use below.
2967          */
2968         if (origin_head->ds_quota != 0 &&
2969             dsl_dataset_phys(clone)->ds_referenced_bytes >
2970             origin_head->ds_quota + refquota_slack)
2971                 return (SET_ERROR(EDQUOT));
2972
2973         return (0);
2974 }
2975
2976 void
2977 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2978     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2979 {
2980         dsl_pool_t *dp = dmu_tx_pool(tx);
2981         int64_t unused_refres_delta;
2982
2983         ASSERT(clone->ds_reserved == 0);
2984         /*
2985          * NOTE: On DEBUG kernels there could be a race between this and
2986          * the check function if spa_asize_inflation is adjusted...
2987          */
2988         ASSERT(origin_head->ds_quota == 0 ||
2989             dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
2990             DMU_MAX_ACCESS * spa_asize_inflation);
2991         ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2992
2993         /*
2994          * Swap per-dataset feature flags.
2995          */
2996         for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2997                 if (!(spa_feature_table[f].fi_flags &
2998                     ZFEATURE_FLAG_PER_DATASET)) {
2999                         ASSERT(!clone->ds_feature_inuse[f]);
3000                         ASSERT(!origin_head->ds_feature_inuse[f]);
3001                         continue;
3002                 }
3003
3004                 boolean_t clone_inuse = clone->ds_feature_inuse[f];
3005                 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
3006
3007                 if (clone_inuse) {
3008                         dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
3009                         clone->ds_feature_inuse[f] = B_FALSE;
3010                 }
3011                 if (origin_head_inuse) {
3012                         dsl_dataset_deactivate_feature(origin_head->ds_object,
3013                             f, tx);
3014                         origin_head->ds_feature_inuse[f] = B_FALSE;
3015                 }
3016                 if (clone_inuse) {
3017                         dsl_dataset_activate_feature(origin_head->ds_object,
3018                             f, tx);
3019                         origin_head->ds_feature_inuse[f] = B_TRUE;
3020                 }
3021                 if (origin_head_inuse) {
3022                         dsl_dataset_activate_feature(clone->ds_object, f, tx);
3023                         clone->ds_feature_inuse[f] = B_TRUE;
3024                 }
3025         }
3026
3027         dmu_buf_will_dirty(clone->ds_dbuf, tx);
3028         dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
3029
3030         if (clone->ds_objset != NULL) {
3031                 dmu_objset_evict(clone->ds_objset);
3032                 clone->ds_objset = NULL;
3033         }
3034
3035         if (origin_head->ds_objset != NULL) {
3036                 dmu_objset_evict(origin_head->ds_objset);
3037                 origin_head->ds_objset = NULL;
3038         }
3039
3040         unused_refres_delta =
3041             (int64_t)MIN(origin_head->ds_reserved,
3042             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3043             (int64_t)MIN(origin_head->ds_reserved,
3044             dsl_dataset_phys(clone)->ds_unique_bytes);
3045
3046         /*
3047          * Reset origin's unique bytes, if it exists.
3048          */
3049         if (clone->ds_prev) {
3050                 dsl_dataset_t *origin = clone->ds_prev;
3051                 uint64_t comp, uncomp;
3052
3053                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3054                 dsl_deadlist_space_range(&clone->ds_deadlist,
3055                     dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
3056                     &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
3057         }
3058
3059         /* swap blkptrs */
3060         {
3061                 blkptr_t tmp;
3062                 tmp = dsl_dataset_phys(origin_head)->ds_bp;
3063                 dsl_dataset_phys(origin_head)->ds_bp =
3064                     dsl_dataset_phys(clone)->ds_bp;
3065                 dsl_dataset_phys(clone)->ds_bp = tmp;
3066         }
3067
3068         /* set dd_*_bytes */
3069         {
3070                 int64_t dused, dcomp, duncomp;
3071                 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3072                 uint64_t odl_used, odl_comp, odl_uncomp;
3073
3074                 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3075                     dd_used_breakdown[DD_USED_SNAP], ==, 0);
3076
3077                 dsl_deadlist_space(&clone->ds_deadlist,
3078                     &cdl_used, &cdl_comp, &cdl_uncomp);
3079                 dsl_deadlist_space(&origin_head->ds_deadlist,
3080                     &odl_used, &odl_comp, &odl_uncomp);
3081
3082                 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3083                     cdl_used -
3084                     (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3085                     odl_used);
3086                 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3087                     cdl_comp -
3088                     (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3089                     odl_comp);
3090                 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3091                     cdl_uncomp -
3092                     (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3093                     odl_uncomp);
3094
3095                 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3096                     dused, dcomp, duncomp, tx);
3097                 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3098                     -dused, -dcomp, -duncomp, tx);
3099
3100                 /*
3101                  * The difference in the space used by snapshots is the
3102                  * difference in snapshot space due to the head's
3103                  * deadlist (since that's the only thing that's
3104                  * changing that affects the snapused).
3105                  */
3106                 dsl_deadlist_space_range(&clone->ds_deadlist,
3107                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3108                     &cdl_used, &cdl_comp, &cdl_uncomp);
3109                 dsl_deadlist_space_range(&origin_head->ds_deadlist,
3110                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3111                     &odl_used, &odl_comp, &odl_uncomp);
3112                 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3113                     DD_USED_HEAD, DD_USED_SNAP, NULL);
3114         }
3115
3116         /* swap ds_*_bytes */
3117         SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3118             dsl_dataset_phys(clone)->ds_referenced_bytes);
3119         SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3120             dsl_dataset_phys(clone)->ds_compressed_bytes);
3121         SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3122             dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3123         SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3124             dsl_dataset_phys(clone)->ds_unique_bytes);
3125
3126         /* apply any parent delta for change in unconsumed refreservation */
3127         dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3128             unused_refres_delta, 0, 0, tx);
3129
3130         /*
3131          * Swap deadlists.
3132          */
3133         dsl_deadlist_close(&clone->ds_deadlist);
3134         dsl_deadlist_close(&origin_head->ds_deadlist);
3135         SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3136             dsl_dataset_phys(clone)->ds_deadlist_obj);
3137         dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3138             dsl_dataset_phys(clone)->ds_deadlist_obj);
3139         dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3140             dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3141
3142         dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3143
3144         spa_history_log_internal_ds(clone, "clone swap", tx,
3145             "parent=%s", origin_head->ds_dir->dd_myname);
3146 }
3147
3148 /*
3149  * Given a pool name and a dataset object number in that pool,
3150  * return the name of that dataset.
3151  */
3152 int
3153 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3154 {
3155         dsl_pool_t *dp;
3156         dsl_dataset_t *ds;
3157         int error;
3158
3159         error = dsl_pool_hold(pname, FTAG, &dp);
3160         if (error != 0)
3161                 return (error);
3162
3163         error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3164         if (error == 0) {
3165                 dsl_dataset_name(ds, buf);
3166                 dsl_dataset_rele(ds, FTAG);
3167         }
3168         dsl_pool_rele(dp, FTAG);
3169
3170         return (error);
3171 }
3172
3173 int
3174 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3175     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3176 {
3177         int error = 0;
3178
3179         ASSERT3S(asize, >, 0);
3180
3181         /*
3182          * *ref_rsrv is the portion of asize that will come from any
3183          * unconsumed refreservation space.
3184          */
3185         *ref_rsrv = 0;
3186
3187         mutex_enter(&ds->ds_lock);
3188         /*
3189          * Make a space adjustment for reserved bytes.
3190          */
3191         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3192                 ASSERT3U(*used, >=,
3193                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3194                 *used -=
3195                     (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3196                 *ref_rsrv =
3197                     asize - MIN(asize, parent_delta(ds, asize + inflight));
3198         }
3199
3200         if (!check_quota || ds->ds_quota == 0) {
3201                 mutex_exit(&ds->ds_lock);
3202                 return (0);
3203         }
3204         /*
3205          * If they are requesting more space, and our current estimate
3206          * is over quota, they get to try again unless the actual
3207          * on-disk is over quota and there are no pending changes (which
3208          * may free up space for us).
3209          */
3210         if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3211             ds->ds_quota) {
3212                 if (inflight > 0 ||
3213                     dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3214                         error = SET_ERROR(ERESTART);
3215                 else
3216                         error = SET_ERROR(EDQUOT);
3217         }
3218         mutex_exit(&ds->ds_lock);
3219
3220         return (error);
3221 }
3222
3223 typedef struct dsl_dataset_set_qr_arg {
3224         const char *ddsqra_name;
3225         zprop_source_t ddsqra_source;
3226         uint64_t ddsqra_value;
3227 } dsl_dataset_set_qr_arg_t;
3228
3229
3230 /* ARGSUSED */
3231 static int
3232 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3233 {
3234         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3235         dsl_pool_t *dp = dmu_tx_pool(tx);
3236         dsl_dataset_t *ds;
3237         int error;
3238         uint64_t newval;
3239
3240         if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3241                 return (SET_ERROR(ENOTSUP));
3242
3243         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3244         if (error != 0)
3245                 return (error);
3246
3247         if (ds->ds_is_snapshot) {
3248                 dsl_dataset_rele(ds, FTAG);
3249                 return (SET_ERROR(EINVAL));
3250         }
3251
3252         error = dsl_prop_predict(ds->ds_dir,
3253             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3254             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3255         if (error != 0) {
3256                 dsl_dataset_rele(ds, FTAG);
3257                 return (error);
3258         }
3259
3260         if (newval == 0) {
3261                 dsl_dataset_rele(ds, FTAG);
3262                 return (0);
3263         }
3264
3265         if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3266             newval < ds->ds_reserved) {
3267                 dsl_dataset_rele(ds, FTAG);
3268                 return (SET_ERROR(ENOSPC));
3269         }
3270
3271         dsl_dataset_rele(ds, FTAG);
3272         return (0);
3273 }
3274
3275 static void
3276 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3277 {
3278         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3279         dsl_pool_t *dp = dmu_tx_pool(tx);
3280         dsl_dataset_t *ds;
3281         uint64_t newval;
3282
3283         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3284
3285         dsl_prop_set_sync_impl(ds,
3286             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3287             ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3288             &ddsqra->ddsqra_value, tx);
3289
3290         VERIFY0(dsl_prop_get_int_ds(ds,
3291             zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3292
3293         if (ds->ds_quota != newval) {
3294                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3295                 ds->ds_quota = newval;
3296         }
3297         dsl_dataset_rele(ds, FTAG);
3298 }
3299
3300 int
3301 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3302     uint64_t refquota)
3303 {
3304         dsl_dataset_set_qr_arg_t ddsqra;
3305
3306         ddsqra.ddsqra_name = dsname;
3307         ddsqra.ddsqra_source = source;
3308         ddsqra.ddsqra_value = refquota;
3309
3310         return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3311             dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3312 }
3313
3314 static int
3315 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3316 {
3317         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3318         dsl_pool_t *dp = dmu_tx_pool(tx);
3319         dsl_dataset_t *ds;
3320         int error;
3321         uint64_t newval, unique;
3322
3323         if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3324                 return (SET_ERROR(ENOTSUP));
3325
3326         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3327         if (error != 0)
3328                 return (error);
3329
3330         if (ds->ds_is_snapshot) {
3331                 dsl_dataset_rele(ds, FTAG);
3332                 return (SET_ERROR(EINVAL));
3333         }
3334
3335         error = dsl_prop_predict(ds->ds_dir,
3336             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3337             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3338         if (error != 0) {
3339                 dsl_dataset_rele(ds, FTAG);
3340                 return (error);
3341         }
3342
3343         /*
3344          * If we are doing the preliminary check in open context, the
3345          * space estimates may be inaccurate.
3346          */
3347         if (!dmu_tx_is_syncing(tx)) {
3348                 dsl_dataset_rele(ds, FTAG);
3349                 return (0);
3350         }
3351
3352         mutex_enter(&ds->ds_lock);
3353         if (!DS_UNIQUE_IS_ACCURATE(ds))
3354                 dsl_dataset_recalc_head_uniq(ds);
3355         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3356         mutex_exit(&ds->ds_lock);
3357
3358         if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3359                 uint64_t delta = MAX(unique, newval) -
3360                     MAX(unique, ds->ds_reserved);
3361
3362                 if (delta >
3363                     dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3364                     (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3365                         dsl_dataset_rele(ds, FTAG);
3366                         return (SET_ERROR(ENOSPC));
3367                 }
3368         }
3369
3370         dsl_dataset_rele(ds, FTAG);
3371         return (0);
3372 }
3373
3374 void
3375 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3376     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3377 {
3378         uint64_t newval;
3379         uint64_t unique;
3380         int64_t delta;
3381
3382         dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3383             source, sizeof (value), 1, &value, tx);
3384
3385         VERIFY0(dsl_prop_get_int_ds(ds,
3386             zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3387
3388         dmu_buf_will_dirty(ds->ds_dbuf, tx);
3389         mutex_enter(&ds->ds_dir->dd_lock);
3390         mutex_enter(&ds->ds_lock);
3391         ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3392         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3393         delta = MAX(0, (int64_t)(newval - unique)) -
3394             MAX(0, (int64_t)(ds->ds_reserved - unique));
3395         ds->ds_reserved = newval;
3396         mutex_exit(&ds->ds_lock);
3397
3398         dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3399         mutex_exit(&ds->ds_dir->dd_lock);
3400 }
3401
3402 static void
3403 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3404 {
3405         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3406         dsl_pool_t *dp = dmu_tx_pool(tx);
3407         dsl_dataset_t *ds;
3408
3409         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3410         dsl_dataset_set_refreservation_sync_impl(ds,
3411             ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3412         dsl_dataset_rele(ds, FTAG);
3413 }
3414
3415 int
3416 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3417     uint64_t refreservation)
3418 {
3419         dsl_dataset_set_qr_arg_t ddsqra;
3420
3421         ddsqra.ddsqra_name = dsname;
3422         ddsqra.ddsqra_source = source;
3423         ddsqra.ddsqra_value = refreservation;
3424
3425         return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3426             dsl_dataset_set_refreservation_sync, &ddsqra,
3427             0, ZFS_SPACE_CHECK_NONE));
3428 }
3429
3430 /*
3431  * Return (in *usedp) the amount of space written in new that is not
3432  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3433  * a snapshot before new, in new's filesystem (or its origin).  If not then
3434  * fail and return EINVAL.
3435  *
3436  * The written space is calculated by considering two components:  First, we
3437  * ignore any freed space, and calculate the written as new's used space
3438  * minus old's used space.  Next, we add in the amount of space that was freed
3439  * between the two snapshots, thus reducing new's used space relative to old's.
3440  * Specifically, this is the space that was born before old->ds_creation_txg,
3441  * and freed before new (ie. on new's deadlist or a previous deadlist).
3442  *
3443  * space freed                         [---------------------]
3444  * snapshots                       ---O-------O--------O-------O------
3445  *                                         oldsnap            new
3446  */
3447 int
3448 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3449     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3450 {
3451         int err = 0;
3452         uint64_t snapobj;
3453         dsl_pool_t *dp = new->ds_dir->dd_pool;
3454
3455         ASSERT(dsl_pool_config_held(dp));
3456
3457         *usedp = 0;
3458         *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3459         *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3460
3461         *compp = 0;
3462         *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3463         *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3464
3465         *uncompp = 0;
3466         *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3467         *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3468
3469         snapobj = new->ds_object;
3470         while (snapobj != oldsnap->ds_object) {
3471                 dsl_dataset_t *snap;
3472                 uint64_t used, comp, uncomp;
3473
3474                 if (snapobj == new->ds_object) {
3475                         snap = new;
3476                 } else {
3477                         err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3478                         if (err != 0)
3479                                 break;
3480                 }
3481
3482                 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3483                     dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3484                         /*
3485                          * The blocks in the deadlist can not be born after
3486                          * ds_prev_snap_txg, so get the whole deadlist space,
3487                          * which is more efficient (especially for old-format
3488                          * deadlists).  Unfortunately the deadlist code
3489                          * doesn't have enough information to make this
3490                          * optimization itself.
3491                          */
3492                         dsl_deadlist_space(&snap->ds_deadlist,
3493                             &used, &comp, &uncomp);
3494                 } else {
3495                         dsl_deadlist_space_range(&snap->ds_deadlist,
3496                             0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3497                             &used, &comp, &uncomp);
3498                 }
3499                 *usedp += used;
3500                 *compp += comp;
3501                 *uncompp += uncomp;
3502
3503                 /*
3504                  * If we get to the beginning of the chain of snapshots
3505                  * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3506                  * was not a snapshot of/before new.
3507                  */
3508                 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3509                 if (snap != new)
3510                         dsl_dataset_rele(snap, FTAG);
3511                 if (snapobj == 0) {
3512                         err = SET_ERROR(EINVAL);
3513                         break;
3514                 }
3515
3516         }
3517         return (err);
3518 }
3519
3520 /*
3521  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3522  * lastsnap, and all snapshots in between are deleted.
3523  *
3524  * blocks that would be freed            [---------------------------]
3525  * snapshots                       ---O-------O--------O-------O--------O
3526  *                                        firstsnap        lastsnap
3527  *
3528  * This is the set of blocks that were born after the snap before firstsnap,
3529  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3530  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3531  * We calculate this by iterating over the relevant deadlists (from the snap
3532  * after lastsnap, backward to the snap after firstsnap), summing up the
3533  * space on the deadlist that was born after the snap before firstsnap.
3534  */
3535 int
3536 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3537     dsl_dataset_t *lastsnap,
3538     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3539 {
3540         int err = 0;
3541         uint64_t snapobj;
3542         dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3543
3544         ASSERT(firstsnap->ds_is_snapshot);
3545         ASSERT(lastsnap->ds_is_snapshot);
3546
3547         /*
3548          * Check that the snapshots are in the same dsl_dir, and firstsnap
3549          * is before lastsnap.
3550          */
3551         if (firstsnap->ds_dir != lastsnap->ds_dir ||
3552             dsl_dataset_phys(firstsnap)->ds_creation_txg >
3553             dsl_dataset_phys(lastsnap)->ds_creation_txg)
3554                 return (SET_ERROR(EINVAL));
3555
3556         *usedp = *compp = *uncompp = 0;
3557
3558         snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3559         while (snapobj != firstsnap->ds_object) {
3560                 dsl_dataset_t *ds;
3561                 uint64_t used, comp, uncomp;
3562
3563                 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3564                 if (err != 0)
3565                         break;
3566
3567                 dsl_deadlist_space_range(&ds->ds_deadlist,
3568                     dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3569                     &used, &comp, &uncomp);
3570                 *usedp += used;
3571                 *compp += comp;
3572                 *uncompp += uncomp;
3573
3574                 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3575                 ASSERT3U(snapobj, !=, 0);
3576                 dsl_dataset_rele(ds, FTAG);
3577         }
3578         return (err);
3579 }
3580
3581 /*
3582  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3583  * For example, they could both be snapshots of the same filesystem, and
3584  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3585  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3586  * filesystem.  Or 'earlier' could be the origin's origin.
3587  *
3588  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3589  */
3590 boolean_t
3591 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3592     uint64_t earlier_txg)
3593 {
3594         dsl_pool_t *dp = later->ds_dir->dd_pool;
3595         int error;
3596         boolean_t ret;
3597
3598         ASSERT(dsl_pool_config_held(dp));
3599         ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3600
3601         if (earlier_txg == 0)
3602                 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3603
3604         if (later->ds_is_snapshot &&
3605             earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3606                 return (B_FALSE);
3607
3608         if (later->ds_dir == earlier->ds_dir)
3609                 return (B_TRUE);
3610         if (!dsl_dir_is_clone(later->ds_dir))
3611                 return (B_FALSE);
3612
3613         if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3614                 return (B_TRUE);
3615         dsl_dataset_t *origin;
3616         error = dsl_dataset_hold_obj(dp,
3617             dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3618         if (error != 0)
3619                 return (B_FALSE);
3620         ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3621         dsl_dataset_rele(origin, FTAG);
3622         return (ret);
3623 }
3624
3625 void
3626 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3627 {
3628         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3629         dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3630 }
3631
3632 boolean_t
3633 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3634 {
3635         dmu_object_info_t doi;
3636
3637         dmu_object_info_from_db(ds->ds_dbuf, &doi);
3638         return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3639 }
3640
3641 boolean_t
3642 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3643 {
3644         return (dsl_dataset_is_zapified(ds) &&
3645             zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3646             ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3647 }