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