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