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

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