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