module/zfs/dsl_dataset.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
  24  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  25  * Copyright (c) 2014 RackTop Systems.
  26  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  27  * Copyright (c) 2016 Actifio, Inc. All rights reserved.
  28  * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
  29  */
  30
  31 #include <sys/dmu_objset.h>
  32 #include <sys/dsl_dataset.h>
  33 #include <sys/dsl_dir.h>
  34 #include <sys/dsl_prop.h>
  35 #include <sys/dsl_synctask.h>
  36 #include <sys/dmu_traverse.h>
  37 #include <sys/dmu_impl.h>
  38 #include <sys/dmu_tx.h>
  39 #include <sys/arc.h>
  40 #include <sys/zio.h>
  41 #include <sys/zap.h>
  42 #include <sys/zfeature.h>
  43 #include <sys/unique.h>
  44 #include <sys/zfs_context.h>
  45 #include <sys/zfs_ioctl.h>
  46 #include <sys/spa.h>
  47 #include <sys/zfs_znode.h>
  48 #include <sys/zfs_onexit.h>
  49 #include <sys/zvol.h>
  50 #include <sys/dsl_scan.h>
  51 #include <sys/dsl_deadlist.h>
  52 #include <sys/dsl_destroy.h>
  53 #include <sys/dsl_userhold.h>
  54 #include <sys/dsl_bookmark.h>
  55 #include <sys/policy.h>
  56 #include <sys/dmu_send.h>
  57 #include <sys/zio_compress.h>
  58 #include <zfs_fletcher.h>
  59
  60 /*
  61  * The SPA supports block sizes up to 16MB.  However, very large blocks
  62  * can have an impact on i/o latency (e.g. tying up a spinning disk for
  63  * ~300ms), and also potentially on the memory allocator.  Therefore,
  64  * we do not allow the recordsize to be set larger than zfs_max_recordsize
  65  * (default 1MB).  Larger blocks can be created by changing this tunable,
  66  * and pools with larger blocks can always be imported and used, regardless
  67  * of this setting.
  68  */
  69 int zfs_max_recordsize = 1 * 1024 * 1024;
  70
  71 #define SWITCH64(x, y) \
  72         { \
  73                 uint64_t __tmp = (x); \
  74                 (x) = (y); \
  75                 (y) = __tmp; \
  76         }
  77
  78 #define DS_REF_MAX      (1ULL << 62)
  79
  80 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
  81
  82 extern int spa_asize_inflation;
  83
  84 /*
  85  * Figure out how much of this delta should be propogated to the dsl_dir
  86  * layer.  If there's a refreservation, that space has already been
  87  * partially accounted for in our ancestors.
  88  */
  89 static int64_t
  90 parent_delta(dsl_dataset_t *ds, int64_t delta)
  91 {
  92         dsl_dataset_phys_t *ds_phys;
  93         uint64_t old_bytes, new_bytes;
  94
  95         if (ds->ds_reserved == 0)
  96                 return (delta);
  97
  98         ds_phys = dsl_dataset_phys(ds);
  99         old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
 100         new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
 101
 102         ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
 103         return (new_bytes - old_bytes);
 104 }
 105
 106 void
 107 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
 108 {
 109         int used, compressed, uncompressed;
 110         int64_t delta;
 111
 112         used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 113         compressed = BP_GET_PSIZE(bp);
 114         uncompressed = BP_GET_UCSIZE(bp);
 115
 116         dprintf_bp(bp, "ds=%p", ds);
 117
 118         ASSERT(dmu_tx_is_syncing(tx));
 119         /* It could have been compressed away to nothing */
 120         if (BP_IS_HOLE(bp))
 121                 return;
 122         ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
 123         ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
 124         if (ds == NULL) {
 125                 dsl_pool_mos_diduse_space(tx->tx_pool,
 126                     used, compressed, uncompressed);
 127                 return;
 128         }
 129
 130         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 131         mutex_enter(&ds->ds_lock);
 132         delta = parent_delta(ds, used);
 133         dsl_dataset_phys(ds)->ds_referenced_bytes += used;
 134         dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
 135         dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
 136         dsl_dataset_phys(ds)->ds_unique_bytes += used;
 137         if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
 138                 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
 139                     B_TRUE;
 140         }
 141         mutex_exit(&ds->ds_lock);
 142         dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
 143             compressed, uncompressed, tx);
 144         dsl_dir_transfer_space(ds->ds_dir, used - delta,
 145             DD_USED_REFRSRV, DD_USED_HEAD, tx);
 146 }
 147
 148 int
 149 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
 150     boolean_t async)
 151 {
 152         int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
 153         int compressed = BP_GET_PSIZE(bp);
 154         int uncompressed = BP_GET_UCSIZE(bp);
 155
 156         if (BP_IS_HOLE(bp))
 157                 return (0);
 158
 159         ASSERT(dmu_tx_is_syncing(tx));
 160         ASSERT(bp->blk_birth <= tx->tx_txg);
 161
 162         if (ds == NULL) {
 163                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 164                 dsl_pool_mos_diduse_space(tx->tx_pool,
 165                     -used, -compressed, -uncompressed);
 166                 return (used);
 167         }
 168         ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
 169
 170         ASSERT(!ds->ds_is_snapshot);
 171         dmu_buf_will_dirty(ds->ds_dbuf, tx);
 172
 173         if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
 174                 int64_t delta;
 175
 176                 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
 177                 dsl_free(tx->tx_pool, tx->tx_txg, bp);
 178
 179                 mutex_enter(&ds->ds_lock);
 180                 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
 181                     !DS_UNIQUE_IS_ACCURATE(ds));
 182                 delta = parent_delta(ds, -used);
 183                 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
 184                 mutex_exit(&ds->ds_lock);
 185                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
 186                     delta, -compressed, -uncompressed, tx);
 187                 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
 188                     DD_USED_REFRSRV, DD_USED_HEAD, tx);
 189         } else {
 190                 dprintf_bp(bp, "putting on dead list: %s", "");
 191                 if (async) {
 192                         /*
 193                          * We are here as part of zio's write done callback,
 194                          * which means we're a zio interrupt thread.  We can't
 195                          * call dsl_deadlist_insert() now because it may block
 196                          * waiting for I/O.  Instead, put bp on the deferred
 197                          * queue and let dsl_pool_sync() finish the job.
 198                          */
 199                         bplist_append(&ds->ds_pending_deadlist, bp);
 200                 } else {
 201                         dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
 202                 }
 203                 ASSERT3U(ds->ds_prev->ds_object, ==,
 204                     dsl_dataset_phys(ds)->ds_prev_snap_obj);
 205                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
 206                 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
 207                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
 208                     ds->ds_object && bp->blk_birth >
 209                     dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
 210                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
 211                         mutex_enter(&ds->ds_prev->ds_lock);
 212                         dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
 213                         mutex_exit(&ds->ds_prev->ds_lock);
 214                 }
 215                 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
 216                         dsl_dir_transfer_space(ds->ds_dir, used,
 217                             DD_USED_HEAD, DD_USED_SNAP, tx);
 218                 }
 219         }
 220         mutex_enter(&ds->ds_lock);
 221         ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
 222         dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
 223         ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
 224         dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
 225         ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
 226         dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
 227         mutex_exit(&ds->ds_lock);
 228
 229         return (used);
 230 }
 231
 232 uint64_t
 233 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
 234 {
 235         uint64_t trysnap = 0;
 236
 237         if (ds == NULL)
 238                 return (0);
 239         /*
 240          * The snapshot creation could fail, but that would cause an
 241          * incorrect FALSE return, which would only result in an
 242          * overestimation of the amount of space that an operation would
 243          * consume, which is OK.
 244          *
 245          * There's also a small window where we could miss a pending
 246          * snapshot, because we could set the sync task in the quiescing
 247          * phase.  So this should only be used as a guess.
 248          */
 249         if (ds->ds_trysnap_txg >
 250             spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
 251                 trysnap = ds->ds_trysnap_txg;
 252         return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
 253 }
 254
 255 boolean_t
 256 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
 257     uint64_t blk_birth)
 258 {
 259         if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
 260             (bp != NULL && BP_IS_HOLE(bp)))
 261                 return (B_FALSE);
 262
 263         ddt_prefetch(dsl_dataset_get_spa(ds), bp);
 264
 265         return (B_TRUE);
 266 }
 267
 268 static void
 269 dsl_dataset_evict(void *dbu)
 270 {
 271         dsl_dataset_t *ds = dbu;
 272
 273         ASSERT(ds->ds_owner == NULL);
 274
 275         ds->ds_dbuf = NULL;
 276
 277         unique_remove(ds->ds_fsid_guid);
 278
 279         if (ds->ds_objset != NULL)
 280                 dmu_objset_evict(ds->ds_objset);
 281
 282         if (ds->ds_prev) {
 283                 dsl_dataset_rele(ds->ds_prev, ds);
 284                 ds->ds_prev = NULL;
 285         }
 286
 287         bplist_destroy(&ds->ds_pending_deadlist);
 288         if (ds->ds_deadlist.dl_os != NULL)
 289                 dsl_deadlist_close(&ds->ds_deadlist);
 290         if (ds->ds_dir)
 291                 dsl_dir_async_rele(ds->ds_dir, ds);
 292
 293         ASSERT(!list_link_active(&ds->ds_synced_link));
 294
 295         list_destroy(&ds->ds_prop_cbs);
 296         mutex_destroy(&ds->ds_lock);
 297         mutex_destroy(&ds->ds_opening_lock);
 298         mutex_destroy(&ds->ds_sendstream_lock);
 299         refcount_destroy(&ds->ds_longholds);
 300
 301         kmem_free(ds, sizeof (dsl_dataset_t));
 302 }
 303
 304 int
 305 dsl_dataset_get_snapname(dsl_dataset_t *ds)
 306 {
 307         dsl_dataset_phys_t *headphys;
 308         int err;
 309         dmu_buf_t *headdbuf;
 310         dsl_pool_t *dp = ds->ds_dir->dd_pool;
 311         objset_t *mos = dp->dp_meta_objset;
 312
 313         if (ds->ds_snapname[0])
 314                 return (0);
 315         if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
 316                 return (0);
 317
 318         err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
 319             FTAG, &headdbuf);
 320         if (err != 0)
 321                 return (err);
 322         headphys = headdbuf->db_data;
 323         err = zap_value_search(dp->dp_meta_objset,
 324             headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
 325         if (err != 0 && zfs_recover == B_TRUE) {
 326                 err = 0;
 327                 (void) snprintf(ds->ds_snapname, sizeof (ds->ds_snapname),
 328                     "SNAPOBJ=%llu-ERR=%d",
 329                     (unsigned long long)ds->ds_object, err);
 330         }
 331         dmu_buf_rele(headdbuf, FTAG);
 332         return (err);
 333 }
 334
 335 int
 336 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
 337 {
 338         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 339         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 340         matchtype_t mt;
 341         int err;
 342
 343         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 344                 mt = MT_FIRST;
 345         else
 346                 mt = MT_EXACT;
 347
 348         err = zap_lookup_norm(mos, snapobj, name, 8, 1,
 349             value, mt, NULL, 0, NULL);
 350         if (err == ENOTSUP && mt == MT_FIRST)
 351                 err = zap_lookup(mos, snapobj, name, 8, 1, value);
 352         return (err);
 353 }
 354
 355 int
 356 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
 357     boolean_t adj_cnt)
 358 {
 359         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 360         uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
 361         matchtype_t mt;
 362         int err;
 363
 364         dsl_dir_snap_cmtime_update(ds->ds_dir);
 365
 366         if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
 367                 mt = MT_FIRST;
 368         else
 369                 mt = MT_EXACT;
 370
 371         err = zap_remove_norm(mos, snapobj, name, mt, tx);
 372         if (err == ENOTSUP && mt == MT_FIRST)
 373                 err = zap_remove(mos, snapobj, name, tx);
 374
 375         if (err == 0 && adj_cnt)
 376                 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
 377                     DD_FIELD_SNAPSHOT_COUNT, tx);
 378
 379         return (err);
 380 }
 381
 382 boolean_t
 383 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
 384 {
 385         dmu_buf_t *dbuf = ds->ds_dbuf;
 386         boolean_t result = B_FALSE;
 387
 388         if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
 389             ds->ds_object, DMU_BONUS_BLKID, tag)) {
 390
 391                 if (ds == dmu_buf_get_user(dbuf))
 392                         result = B_TRUE;
 393                 else
 394                         dmu_buf_rele(dbuf, tag);
 395         }
 396
 397         return (result);
 398 }
 399
 400 int
 401 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 402     dsl_dataset_t **dsp)
 403 {
 404         objset_t *mos = dp->dp_meta_objset;
 405         dmu_buf_t *dbuf;
 406         dsl_dataset_t *ds;
 407         int err;
 408         dmu_object_info_t doi;
 409
 410         ASSERT(dsl_pool_config_held(dp));
 411
 412         err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
 413         if (err != 0)
 414                 return (err);
 415
 416         /* Make sure dsobj has the correct object type. */
 417         dmu_object_info_from_db(dbuf, &doi);
 418         if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
 419                 dmu_buf_rele(dbuf, tag);
 420                 return (SET_ERROR(EINVAL));
 421         }
 422
 423         ds = dmu_buf_get_user(dbuf);
 424         if (ds == NULL) {
 425                 dsl_dataset_t *winner = NULL;
 426
 427                 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
 428                 ds->ds_dbuf = dbuf;
 429                 ds->ds_object = dsobj;
 430                 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
 431                 list_link_init(&ds->ds_synced_link);
 432
 433                 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
 434                 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 435                 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
 436                 refcount_create(&ds->ds_longholds);
 437
 438                 bplist_create(&ds->ds_pending_deadlist);
 439                 dsl_deadlist_open(&ds->ds_deadlist,
 440                     mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
 441
 442                 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
 443                     offsetof(dmu_sendarg_t, dsa_link));
 444
 445                 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
 446                     offsetof(dsl_prop_cb_record_t, cbr_ds_node));
 447
 448                 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 449                         spa_feature_t f;
 450
 451                         for (f = 0; f < SPA_FEATURES; f++) {
 452                                 if (!(spa_feature_table[f].fi_flags &
 453                                     ZFEATURE_FLAG_PER_DATASET))
 454                                         continue;
 455                                 err = zap_contains(mos, dsobj,
 456                                     spa_feature_table[f].fi_guid);
 457                                 if (err == 0) {
 458                                         ds->ds_feature_inuse[f] = B_TRUE;
 459                                 } else {
 460                                         ASSERT3U(err, ==, ENOENT);
 461                                         err = 0;
 462                                 }
 463                         }
 464                 }
 465
 466                 err = dsl_dir_hold_obj(dp,
 467                     dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
 468                 if (err != 0) {
 469                         mutex_destroy(&ds->ds_lock);
 470                         mutex_destroy(&ds->ds_opening_lock);
 471                         mutex_destroy(&ds->ds_sendstream_lock);
 472                         refcount_destroy(&ds->ds_longholds);
 473                         bplist_destroy(&ds->ds_pending_deadlist);
 474                         dsl_deadlist_close(&ds->ds_deadlist);
 475                         kmem_free(ds, sizeof (dsl_dataset_t));
 476                         dmu_buf_rele(dbuf, tag);
 477                         return (err);
 478                 }
 479
 480                 if (!ds->ds_is_snapshot) {
 481                         ds->ds_snapname[0] = '\0';
 482                         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
 483                                 err = dsl_dataset_hold_obj(dp,
 484                                     dsl_dataset_phys(ds)->ds_prev_snap_obj,
 485                                     ds, &ds->ds_prev);
 486                         }
 487                         if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
 488                                 int zaperr = zap_lookup(mos, ds->ds_object,
 489                                     DS_FIELD_BOOKMARK_NAMES,
 490                                     sizeof (ds->ds_bookmarks), 1,
 491                                     &ds->ds_bookmarks);
 492                                 if (zaperr != ENOENT)
 493                                         VERIFY0(zaperr);
 494                         }
 495                 } else {
 496                         if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
 497                                 err = dsl_dataset_get_snapname(ds);
 498                         if (err == 0 &&
 499                             dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
 500                                 err = zap_count(
 501                                     ds->ds_dir->dd_pool->dp_meta_objset,
 502                                     dsl_dataset_phys(ds)->ds_userrefs_obj,
 503                                     &ds->ds_userrefs);
 504                         }
 505                 }
 506
 507                 if (err == 0 && !ds->ds_is_snapshot) {
 508                         err = dsl_prop_get_int_ds(ds,
 509                             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
 510                             &ds->ds_reserved);
 511                         if (err == 0) {
 512                                 err = dsl_prop_get_int_ds(ds,
 513                                     zfs_prop_to_name(ZFS_PROP_REFQUOTA),
 514                                     &ds->ds_quota);
 515                         }
 516                 } else {
 517                         ds->ds_reserved = ds->ds_quota = 0;
 518                 }
 519
 520                 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
 521                 if (err == 0)
 522                         winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
 523
 524                 if (err != 0 || winner != NULL) {
 525                         bplist_destroy(&ds->ds_pending_deadlist);
 526                         dsl_deadlist_close(&ds->ds_deadlist);
 527                         if (ds->ds_prev)
 528                                 dsl_dataset_rele(ds->ds_prev, ds);
 529                         dsl_dir_rele(ds->ds_dir, ds);
 530                         mutex_destroy(&ds->ds_lock);
 531                         mutex_destroy(&ds->ds_opening_lock);
 532                         mutex_destroy(&ds->ds_sendstream_lock);
 533                         refcount_destroy(&ds->ds_longholds);
 534                         kmem_free(ds, sizeof (dsl_dataset_t));
 535                         if (err != 0) {
 536                                 dmu_buf_rele(dbuf, tag);
 537                                 return (err);
 538                         }
 539                         ds = winner;
 540                 } else {
 541                         ds->ds_fsid_guid =
 542                             unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
 543                 }
 544         }
 545         ASSERT3P(ds->ds_dbuf, ==, dbuf);
 546         ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
 547         ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
 548             spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
 549             dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
 550         *dsp = ds;
 551         return (0);
 552 }
 553
 554 int
 555 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
 556     void *tag, dsl_dataset_t **dsp)
 557 {
 558         dsl_dir_t *dd;
 559         const char *snapname;
 560         uint64_t obj;
 561         int err = 0;
 562         dsl_dataset_t *ds;
 563
 564         err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
 565         if (err != 0)
 566                 return (err);
 567
 568         ASSERT(dsl_pool_config_held(dp));
 569         obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
 570         if (obj != 0)
 571                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
 572         else
 573                 err = SET_ERROR(ENOENT);
 574
 575         /* we may be looking for a snapshot */
 576         if (err == 0 && snapname != NULL) {
 577                 dsl_dataset_t *snap_ds;
 578
 579                 if (*snapname++ != '@') {
 580                         dsl_dataset_rele(ds, tag);
 581                         dsl_dir_rele(dd, FTAG);
 582                         return (SET_ERROR(ENOENT));
 583                 }
 584
 585                 dprintf("looking for snapshot '%s'\n", snapname);
 586                 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
 587                 if (err == 0)
 588                         err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
 589                 dsl_dataset_rele(ds, tag);
 590
 591                 if (err == 0) {
 592                         mutex_enter(&snap_ds->ds_lock);
 593                         if (snap_ds->ds_snapname[0] == 0)
 594                                 (void) strlcpy(snap_ds->ds_snapname, snapname,
 595                                     sizeof (snap_ds->ds_snapname));
 596                         mutex_exit(&snap_ds->ds_lock);
 597                         ds = snap_ds;
 598                 }
 599         }
 600         if (err == 0)
 601                 *dsp = ds;
 602         dsl_dir_rele(dd, FTAG);
 603         return (err);
 604 }
 605
 606 int
 607 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
 608     void *tag, dsl_dataset_t **dsp)
 609 {
 610         int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
 611         if (err != 0)
 612                 return (err);
 613         if (!dsl_dataset_tryown(*dsp, tag)) {
 614                 dsl_dataset_rele(*dsp, tag);
 615                 *dsp = NULL;
 616                 return (SET_ERROR(EBUSY));
 617         }
 618         return (0);
 619 }
 620
 621 int
 622 dsl_dataset_own(dsl_pool_t *dp, const char *name,
 623     void *tag, dsl_dataset_t **dsp)
 624 {
 625         int err = dsl_dataset_hold(dp, name, tag, dsp);
 626         if (err != 0)
 627                 return (err);
 628         if (!dsl_dataset_tryown(*dsp, tag)) {
 629                 dsl_dataset_rele(*dsp, tag);
 630                 return (SET_ERROR(EBUSY));
 631         }
 632         return (0);
 633 }
 634
 635 /*
 636  * See the comment above dsl_pool_hold() for details.  In summary, a long
 637  * hold is used to prevent destruction of a dataset while the pool hold
 638  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
 639  *
 640  * The dataset and pool must be held when this function is called.  After it
 641  * is called, the pool hold may be released while the dataset is still held
 642  * and accessed.
 643  */
 644 void
 645 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
 646 {
 647         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 648         (void) refcount_add(&ds->ds_longholds, tag);
 649 }
 650
 651 void
 652 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
 653 {
 654         (void) refcount_remove(&ds->ds_longholds, tag);
 655 }
 656
 657 /* Return B_TRUE if there are any long holds on this dataset. */
 658 boolean_t
 659 dsl_dataset_long_held(dsl_dataset_t *ds)
 660 {
 661         return (!refcount_is_zero(&ds->ds_longholds));
 662 }
 663
 664 void
 665 dsl_dataset_name(dsl_dataset_t *ds, char *name)
 666 {
 667         if (ds == NULL) {
 668                 (void) strcpy(name, "mos");
 669         } else {
 670                 dsl_dir_name(ds->ds_dir, name);
 671                 VERIFY0(dsl_dataset_get_snapname(ds));
 672                 if (ds->ds_snapname[0]) {
 673                         VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
 674                             <, ZFS_MAX_DATASET_NAME_LEN);
 675                         /*
 676                          * We use a "recursive" mutex so that we
 677                          * can call dprintf_ds() with ds_lock held.
 678                          */
 679                         if (!MUTEX_HELD(&ds->ds_lock)) {
 680                                 mutex_enter(&ds->ds_lock);
 681                                 VERIFY3U(strlcat(name, ds->ds_snapname,
 682                                     ZFS_MAX_DATASET_NAME_LEN), <,
 683                                     ZFS_MAX_DATASET_NAME_LEN);
 684                                 mutex_exit(&ds->ds_lock);
 685                         } else {
 686                                 VERIFY3U(strlcat(name, ds->ds_snapname,
 687                                     ZFS_MAX_DATASET_NAME_LEN), <,
 688                                     ZFS_MAX_DATASET_NAME_LEN);
 689                         }
 690                 }
 691         }
 692 }
 693
 694 int
 695 dsl_dataset_namelen(dsl_dataset_t *ds)
 696 {
 697         int len;
 698         VERIFY0(dsl_dataset_get_snapname(ds));
 699         mutex_enter(&ds->ds_lock);
 700         len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
 701         mutex_exit(&ds->ds_lock);
 702         return (len);
 703 }
 704
 705 void
 706 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
 707 {
 708         dmu_buf_rele(ds->ds_dbuf, tag);
 709 }
 710
 711 void
 712 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
 713 {
 714         ASSERT3P(ds->ds_owner, ==, tag);
 715         ASSERT(ds->ds_dbuf != NULL);
 716
 717         mutex_enter(&ds->ds_lock);
 718         ds->ds_owner = NULL;
 719         mutex_exit(&ds->ds_lock);
 720         dsl_dataset_long_rele(ds, tag);
 721         dsl_dataset_rele(ds, tag);
 722 }
 723
 724 boolean_t
 725 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
 726 {
 727         boolean_t gotit = FALSE;
 728
 729         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 730         mutex_enter(&ds->ds_lock);
 731         if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
 732                 ds->ds_owner = tag;
 733                 dsl_dataset_long_hold(ds, tag);
 734                 gotit = TRUE;
 735         }
 736         mutex_exit(&ds->ds_lock);
 737         return (gotit);
 738 }
 739
 740 boolean_t
 741 dsl_dataset_has_owner(dsl_dataset_t *ds)
 742 {
 743         boolean_t rv;
 744         mutex_enter(&ds->ds_lock);
 745         rv = (ds->ds_owner != NULL);
 746         mutex_exit(&ds->ds_lock);
 747         return (rv);
 748 }
 749
 750 static void
 751 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
 752 {
 753         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 754         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
 755         uint64_t zero = 0;
 756
 757         VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
 758
 759         spa_feature_incr(spa, f, tx);
 760         dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
 761
 762         VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
 763             sizeof (zero), 1, &zero, tx));
 764 }
 765
 766 void
 767 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
 768 {
 769         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 770         objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
 771
 772         VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
 773
 774         VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
 775         spa_feature_decr(spa, f, tx);
 776 }
 777
 778 uint64_t
 779 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
 780     uint64_t flags, dmu_tx_t *tx)
 781 {
 782         dsl_pool_t *dp = dd->dd_pool;
 783         dmu_buf_t *dbuf;
 784         dsl_dataset_phys_t *dsphys;
 785         uint64_t dsobj;
 786         objset_t *mos = dp->dp_meta_objset;
 787
 788         if (origin == NULL)
 789                 origin = dp->dp_origin_snap;
 790
 791         ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
 792         ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
 793         ASSERT(dmu_tx_is_syncing(tx));
 794         ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
 795
 796         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
 797             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
 798         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
 799         dmu_buf_will_dirty(dbuf, tx);
 800         dsphys = dbuf->db_data;
 801         bzero(dsphys, sizeof (dsl_dataset_phys_t));
 802         dsphys->ds_dir_obj = dd->dd_object;
 803         dsphys->ds_flags = flags;
 804         dsphys->ds_fsid_guid = unique_create();
 805         (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
 806             sizeof (dsphys->ds_guid));
 807         dsphys->ds_snapnames_zapobj =
 808             zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
 809             DMU_OT_NONE, 0, tx);
 810         dsphys->ds_creation_time = gethrestime_sec();
 811         dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
 812
 813         if (origin == NULL) {
 814                 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
 815         } else {
 816                 spa_feature_t f;
 817                 dsl_dataset_t *ohds; /* head of the origin snapshot */
 818
 819                 dsphys->ds_prev_snap_obj = origin->ds_object;
 820                 dsphys->ds_prev_snap_txg =
 821                     dsl_dataset_phys(origin)->ds_creation_txg;
 822                 dsphys->ds_referenced_bytes =
 823                     dsl_dataset_phys(origin)->ds_referenced_bytes;
 824                 dsphys->ds_compressed_bytes =
 825                     dsl_dataset_phys(origin)->ds_compressed_bytes;
 826                 dsphys->ds_uncompressed_bytes =
 827                     dsl_dataset_phys(origin)->ds_uncompressed_bytes;
 828                 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
 829
 830                 /*
 831                  * Inherit flags that describe the dataset's contents
 832                  * (INCONSISTENT) or properties (Case Insensitive).
 833                  */
 834                 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
 835                     (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
 836
 837                 for (f = 0; f < SPA_FEATURES; f++) {
 838                         if (origin->ds_feature_inuse[f])
 839                                 dsl_dataset_activate_feature(dsobj, f, tx);
 840                 }
 841
 842                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
 843                 dsl_dataset_phys(origin)->ds_num_children++;
 844
 845                 VERIFY0(dsl_dataset_hold_obj(dp,
 846                     dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
 847                     FTAG, &ohds));
 848                 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
 849                     dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
 850                 dsl_dataset_rele(ohds, FTAG);
 851
 852                 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
 853                         if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
 854                                 dsl_dataset_phys(origin)->ds_next_clones_obj =
 855                                     zap_create(mos,
 856                                     DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
 857                         }
 858                         VERIFY0(zap_add_int(mos,
 859                             dsl_dataset_phys(origin)->ds_next_clones_obj,
 860                             dsobj, tx));
 861                 }
 862
 863                 dmu_buf_will_dirty(dd->dd_dbuf, tx);
 864                 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
 865                 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
 866                         if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
 867                                 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
 868                                 dsl_dir_phys(origin->ds_dir)->dd_clones =
 869                                     zap_create(mos,
 870                                     DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
 871                         }
 872                         VERIFY0(zap_add_int(mos,
 873                             dsl_dir_phys(origin->ds_dir)->dd_clones,
 874                             dsobj, tx));
 875                 }
 876         }
 877
 878         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
 879                 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 880
 881         dmu_buf_rele(dbuf, FTAG);
 882
 883         dmu_buf_will_dirty(dd->dd_dbuf, tx);
 884         dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
 885
 886         return (dsobj);
 887 }
 888
 889 static void
 890 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
 891 {
 892         objset_t *os;
 893
 894         VERIFY0(dmu_objset_from_ds(ds, &os));
 895         bzero(&os->os_zil_header, sizeof (os->os_zil_header));
 896         dsl_dataset_dirty(ds, tx);
 897 }
 898
 899 uint64_t
 900 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
 901     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
 902 {
 903         dsl_pool_t *dp = pdd->dd_pool;
 904         uint64_t dsobj, ddobj;
 905         dsl_dir_t *dd;
 906
 907         ASSERT(dmu_tx_is_syncing(tx));
 908         ASSERT(lastname[0] != '@');
 909
 910         ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
 911         VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
 912
 913         dsobj = dsl_dataset_create_sync_dd(dd, origin,
 914             flags & ~DS_CREATE_FLAG_NODIRTY, tx);
 915
 916         dsl_deleg_set_create_perms(dd, tx, cr);
 917
 918         /*
 919          * Since we're creating a new node we know it's a leaf, so we can
 920          * initialize the counts if the limit feature is active.
 921          */
 922         if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
 923                 uint64_t cnt = 0;
 924                 objset_t *os = dd->dd_pool->dp_meta_objset;
 925
 926                 dsl_dir_zapify(dd, tx);
 927                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
 928                     sizeof (cnt), 1, &cnt, tx));
 929                 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
 930                     sizeof (cnt), 1, &cnt, tx));
 931         }
 932
 933         dsl_dir_rele(dd, FTAG);
 934
 935         /*
 936          * If we are creating a clone, make sure we zero out any stale
 937          * data from the origin snapshots zil header.
 938          */
 939         if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
 940                 dsl_dataset_t *ds;
 941
 942                 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
 943                 dsl_dataset_zero_zil(ds, tx);
 944                 dsl_dataset_rele(ds, FTAG);
 945         }
 946
 947         return (dsobj);
 948 }
 949
 950 /*
 951  * The unique space in the head dataset can be calculated by subtracting
 952  * the space used in the most recent snapshot, that is still being used
 953  * in this file system, from the space currently in use.  To figure out
 954  * the space in the most recent snapshot still in use, we need to take
 955  * the total space used in the snapshot and subtract out the space that
 956  * has been freed up since the snapshot was taken.
 957  */
 958 void
 959 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
 960 {
 961         uint64_t mrs_used;
 962         uint64_t dlused, dlcomp, dluncomp;
 963
 964         ASSERT(!ds->ds_is_snapshot);
 965
 966         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
 967                 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
 968         else
 969                 mrs_used = 0;
 970
 971         dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
 972
 973         ASSERT3U(dlused, <=, mrs_used);
 974         dsl_dataset_phys(ds)->ds_unique_bytes =
 975             dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
 976
 977         if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
 978             SPA_VERSION_UNIQUE_ACCURATE)
 979                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
 980 }
 981
 982 void
 983 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
 984     dmu_tx_t *tx)
 985 {
 986         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 987         int err;
 988         ASSERTV(uint64_t count);
 989
 990         ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
 991         err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
 992             obj, tx);
 993         /*
 994          * The err should not be ENOENT, but a bug in a previous version
 995          * of the code could cause upgrade_clones_cb() to not set
 996          * ds_next_snap_obj when it should, leading to a missing entry.
 997          * If we knew that the pool was created after
 998          * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
 999          * ENOENT.  However, at least we can check that we don't have
1000          * too many entries in the next_clones_obj even after failing to
1001          * remove this one.
1002          */
1003         if (err != ENOENT)
1004                 VERIFY0(err);
1005         ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1006             &count));
1007         ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1008 }
1009
1010
1011 blkptr_t *
1012 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1013 {
1014         return (&dsl_dataset_phys(ds)->ds_bp);
1015 }
1016
1017 void
1018 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1019 {
1020         ASSERT(dmu_tx_is_syncing(tx));
1021         /* If it's the meta-objset, set dp_meta_rootbp */
1022         if (ds == NULL) {
1023                 tx->tx_pool->dp_meta_rootbp = *bp;
1024         } else {
1025                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1026                 dsl_dataset_phys(ds)->ds_bp = *bp;
1027         }
1028 }
1029
1030 spa_t *
1031 dsl_dataset_get_spa(dsl_dataset_t *ds)
1032 {
1033         return (ds->ds_dir->dd_pool->dp_spa);
1034 }
1035
1036 void
1037 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1038 {
1039         dsl_pool_t *dp;
1040
1041         if (ds == NULL) /* this is the meta-objset */
1042                 return;
1043
1044         ASSERT(ds->ds_objset != NULL);
1045
1046         if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1047                 panic("dirtying snapshot!");
1048
1049         dp = ds->ds_dir->dd_pool;
1050
1051         if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1052                 /* up the hold count until we can be written out */
1053                 dmu_buf_add_ref(ds->ds_dbuf, ds);
1054         }
1055 }
1056
1057 boolean_t
1058 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1059 {
1060         int t;
1061
1062         for (t = 0; t < TXG_SIZE; t++) {
1063                 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1064                     ds, t))
1065                         return (B_TRUE);
1066         }
1067         return (B_FALSE);
1068 }
1069
1070 static int
1071 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1072 {
1073         uint64_t asize;
1074
1075         if (!dmu_tx_is_syncing(tx))
1076                 return (0);
1077
1078         /*
1079          * If there's an fs-only reservation, any blocks that might become
1080          * owned by the snapshot dataset must be accommodated by space
1081          * outside of the reservation.
1082          */
1083         ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1084         asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1085         if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1086                 return (SET_ERROR(ENOSPC));
1087
1088         /*
1089          * Propagate any reserved space for this snapshot to other
1090          * snapshot checks in this sync group.
1091          */
1092         if (asize > 0)
1093                 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1094
1095         return (0);
1096 }
1097
1098 typedef struct dsl_dataset_snapshot_arg {
1099         nvlist_t *ddsa_snaps;
1100         nvlist_t *ddsa_props;
1101         nvlist_t *ddsa_errors;
1102         cred_t *ddsa_cr;
1103 } dsl_dataset_snapshot_arg_t;
1104
1105 int
1106 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1107     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1108 {
1109         int error;
1110         uint64_t value;
1111
1112         ds->ds_trysnap_txg = tx->tx_txg;
1113
1114         if (!dmu_tx_is_syncing(tx))
1115                 return (0);
1116
1117         /*
1118          * We don't allow multiple snapshots of the same txg.  If there
1119          * is already one, try again.
1120          */
1121         if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1122                 return (SET_ERROR(EAGAIN));
1123
1124         /*
1125          * Check for conflicting snapshot name.
1126          */
1127         error = dsl_dataset_snap_lookup(ds, snapname, &value);
1128         if (error == 0)
1129                 return (SET_ERROR(EEXIST));
1130         if (error != ENOENT)
1131                 return (error);
1132
1133         /*
1134          * We don't allow taking snapshots of inconsistent datasets, such as
1135          * those into which we are currently receiving.  However, if we are
1136          * creating this snapshot as part of a receive, this check will be
1137          * executed atomically with respect to the completion of the receive
1138          * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1139          * case we ignore this, knowing it will be fixed up for us shortly in
1140          * dmu_recv_end_sync().
1141          */
1142         if (!recv && DS_IS_INCONSISTENT(ds))
1143                 return (SET_ERROR(EBUSY));
1144
1145         /*
1146          * Skip the check for temporary snapshots or if we have already checked
1147          * the counts in dsl_dataset_snapshot_check. This means we really only
1148          * check the count here when we're receiving a stream.
1149          */
1150         if (cnt != 0 && cr != NULL) {
1151                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1152                     ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1153                 if (error != 0)
1154                         return (error);
1155         }
1156
1157         error = dsl_dataset_snapshot_reserve_space(ds, tx);
1158         if (error != 0)
1159                 return (error);
1160
1161         return (0);
1162 }
1163
1164 static int
1165 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1166 {
1167         dsl_dataset_snapshot_arg_t *ddsa = arg;
1168         dsl_pool_t *dp = dmu_tx_pool(tx);
1169         nvpair_t *pair;
1170         int rv = 0;
1171
1172         /*
1173          * Pre-compute how many total new snapshots will be created for each
1174          * level in the tree and below. This is needed for validating the
1175          * snapshot limit when either taking a recursive snapshot or when
1176          * taking multiple snapshots.
1177          *
1178          * The problem is that the counts are not actually adjusted when
1179          * we are checking, only when we finally sync. For a single snapshot,
1180          * this is easy, the count will increase by 1 at each node up the tree,
1181          * but its more complicated for the recursive/multiple snapshot case.
1182          *
1183          * The dsl_fs_ss_limit_check function does recursively check the count
1184          * at each level up the tree but since it is validating each snapshot
1185          * independently we need to be sure that we are validating the complete
1186          * count for the entire set of snapshots. We do this by rolling up the
1187          * counts for each component of the name into an nvlist and then
1188          * checking each of those cases with the aggregated count.
1189          *
1190          * This approach properly handles not only the recursive snapshot
1191          * case (where we get all of those on the ddsa_snaps list) but also
1192          * the sibling case (e.g. snapshot a/b and a/c so that we will also
1193          * validate the limit on 'a' using a count of 2).
1194          *
1195          * We validate the snapshot names in the third loop and only report
1196          * name errors once.
1197          */
1198         if (dmu_tx_is_syncing(tx)) {
1199                 char *nm;
1200                 nvlist_t *cnt_track = NULL;
1201                 cnt_track = fnvlist_alloc();
1202
1203                 nm = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1204
1205                 /* Rollup aggregated counts into the cnt_track list */
1206                 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1207                     pair != NULL;
1208                     pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1209                         char *pdelim;
1210                         uint64_t val;
1211
1212                         (void) strlcpy(nm, nvpair_name(pair), MAXPATHLEN);
1213                         pdelim = strchr(nm, '@');
1214                         if (pdelim == NULL)
1215                                 continue;
1216                         *pdelim = '\0';
1217
1218                         do {
1219                                 if (nvlist_lookup_uint64(cnt_track, nm,
1220                                     &val) == 0) {
1221                                         /* update existing entry */
1222                                         fnvlist_add_uint64(cnt_track, nm,
1223                                             val + 1);
1224                                 } else {
1225                                         /* add to list */
1226                                         fnvlist_add_uint64(cnt_track, nm, 1);
1227                                 }
1228
1229                                 pdelim = strrchr(nm, '/');
1230                                 if (pdelim != NULL)
1231                                         *pdelim = '\0';
1232                         } while (pdelim != NULL);
1233                 }
1234
1235                 kmem_free(nm, MAXPATHLEN);
1236
1237                 /* Check aggregated counts at each level */
1238                 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1239                     pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1240                         int error = 0;
1241                         char *name;
1242                         uint64_t cnt = 0;
1243                         dsl_dataset_t *ds;
1244
1245                         name = nvpair_name(pair);
1246                         cnt = fnvpair_value_uint64(pair);
1247                         ASSERT(cnt > 0);
1248
1249                         error = dsl_dataset_hold(dp, name, FTAG, &ds);
1250                         if (error == 0) {
1251                                 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1252                                     ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1253                                     ddsa->ddsa_cr);
1254                                 dsl_dataset_rele(ds, FTAG);
1255                         }
1256
1257                         if (error != 0) {
1258                                 if (ddsa->ddsa_errors != NULL)
1259                                         fnvlist_add_int32(ddsa->ddsa_errors,
1260                                             name, error);
1261                                 rv = error;
1262                                 /* only report one error for this check */
1263                                 break;
1264                         }
1265                 }
1266                 nvlist_free(cnt_track);
1267         }
1268
1269         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1270             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1271                 int error = 0;
1272                 dsl_dataset_t *ds;
1273                 char *name, *atp;
1274                 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1275
1276                 name = nvpair_name(pair);
1277                 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1278                         error = SET_ERROR(ENAMETOOLONG);
1279                 if (error == 0) {
1280                         atp = strchr(name, '@');
1281                         if (atp == NULL)
1282                                 error = SET_ERROR(EINVAL);
1283                         if (error == 0)
1284                                 (void) strlcpy(dsname, name, atp - name + 1);
1285                 }
1286                 if (error == 0)
1287                         error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1288                 if (error == 0) {
1289                         /* passing 0/NULL skips dsl_fs_ss_limit_check */
1290                         error = dsl_dataset_snapshot_check_impl(ds,
1291                             atp + 1, tx, B_FALSE, 0, NULL);
1292                         dsl_dataset_rele(ds, FTAG);
1293                 }
1294
1295                 if (error != 0) {
1296                         if (ddsa->ddsa_errors != NULL) {
1297                                 fnvlist_add_int32(ddsa->ddsa_errors,
1298                                     name, error);
1299                         }
1300                         rv = error;
1301                 }
1302         }
1303
1304         return (rv);
1305 }
1306
1307 void
1308 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1309     dmu_tx_t *tx)
1310 {
1311         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1312         dmu_buf_t *dbuf;
1313         dsl_dataset_phys_t *dsphys;
1314         uint64_t dsobj, crtxg;
1315         objset_t *mos = dp->dp_meta_objset;
1316         spa_feature_t f;
1317         ASSERTV(static zil_header_t zero_zil);
1318         ASSERTV(objset_t *os);
1319
1320         ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1321
1322         /*
1323          * If we are on an old pool, the zil must not be active, in which
1324          * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1325          */
1326         ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1327             dmu_objset_from_ds(ds, &os) != 0 ||
1328             bcmp(&os->os_phys->os_zil_header, &zero_zil,
1329             sizeof (zero_zil)) == 0);
1330
1331         dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1332
1333         /*
1334          * The origin's ds_creation_txg has to be < TXG_INITIAL
1335          */
1336         if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1337                 crtxg = 1;
1338         else
1339                 crtxg = tx->tx_txg;
1340
1341         dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1342             DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1343         VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1344         dmu_buf_will_dirty(dbuf, tx);
1345         dsphys = dbuf->db_data;
1346         bzero(dsphys, sizeof (dsl_dataset_phys_t));
1347         dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1348         dsphys->ds_fsid_guid = unique_create();
1349         (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1350             sizeof (dsphys->ds_guid));
1351         dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1352         dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1353         dsphys->ds_next_snap_obj = ds->ds_object;
1354         dsphys->ds_num_children = 1;
1355         dsphys->ds_creation_time = gethrestime_sec();
1356         dsphys->ds_creation_txg = crtxg;
1357         dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1358         dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1359         dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1360         dsphys->ds_uncompressed_bytes =
1361             dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1362         dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1363         dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1364         dmu_buf_rele(dbuf, FTAG);
1365
1366         for (f = 0; f < SPA_FEATURES; f++) {
1367                 if (ds->ds_feature_inuse[f])
1368                         dsl_dataset_activate_feature(dsobj, f, tx);
1369         }
1370
1371         ASSERT3U(ds->ds_prev != 0, ==,
1372             dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1373         if (ds->ds_prev) {
1374                 uint64_t next_clones_obj =
1375                     dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1376                 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1377                     ds->ds_object ||
1378                     dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1379                 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1380                     ds->ds_object) {
1381                         dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1382                         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1383                             dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1384                         dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1385                 } else if (next_clones_obj != 0) {
1386                         dsl_dataset_remove_from_next_clones(ds->ds_prev,
1387                             dsphys->ds_next_snap_obj, tx);
1388                         VERIFY0(zap_add_int(mos,
1389                             next_clones_obj, dsobj, tx));
1390                 }
1391         }
1392
1393         /*
1394          * If we have a reference-reservation on this dataset, we will
1395          * need to increase the amount of refreservation being charged
1396          * since our unique space is going to zero.
1397          */
1398         if (ds->ds_reserved) {
1399                 int64_t delta;
1400                 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1401                 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1402                     ds->ds_reserved);
1403                 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1404                     delta, 0, 0, tx);
1405         }
1406
1407         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1408         dsl_dataset_phys(ds)->ds_deadlist_obj =
1409             dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1410             dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1411         dsl_deadlist_close(&ds->ds_deadlist);
1412         dsl_deadlist_open(&ds->ds_deadlist, mos,
1413             dsl_dataset_phys(ds)->ds_deadlist_obj);
1414         dsl_deadlist_add_key(&ds->ds_deadlist,
1415             dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1416
1417         ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1418         dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1419         dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1420         dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1421         if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1422                 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1423
1424         VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1425             snapname, 8, 1, &dsobj, tx));
1426
1427         if (ds->ds_prev)
1428                 dsl_dataset_rele(ds->ds_prev, ds);
1429         VERIFY0(dsl_dataset_hold_obj(dp,
1430             dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1431
1432         dsl_scan_ds_snapshotted(ds, tx);
1433
1434         dsl_dir_snap_cmtime_update(ds->ds_dir);
1435
1436         spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1437 }
1438
1439 static void
1440 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1441 {
1442         dsl_dataset_snapshot_arg_t *ddsa = arg;
1443         dsl_pool_t *dp = dmu_tx_pool(tx);
1444         nvpair_t *pair;
1445
1446         for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1447             pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1448                 dsl_dataset_t *ds;
1449                 char *name, *atp;
1450                 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1451
1452                 name = nvpair_name(pair);
1453                 atp = strchr(name, '@');
1454                 (void) strlcpy(dsname, name, atp - name + 1);
1455                 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1456
1457                 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1458                 if (ddsa->ddsa_props != NULL) {
1459                         dsl_props_set_sync_impl(ds->ds_prev,
1460                             ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1461                 }
1462                 zvol_create_minors(dp->dp_spa, nvpair_name(pair), B_TRUE);
1463                 dsl_dataset_rele(ds, FTAG);
1464         }
1465 }
1466
1467 /*
1468  * The snapshots must all be in the same pool.
1469  * All-or-nothing: if there are any failures, nothing will be modified.
1470  */
1471 int
1472 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1473 {
1474         dsl_dataset_snapshot_arg_t ddsa;
1475         nvpair_t *pair;
1476         boolean_t needsuspend;
1477         int error;
1478         spa_t *spa;
1479         char *firstname;
1480         nvlist_t *suspended = NULL;
1481
1482         pair = nvlist_next_nvpair(snaps, NULL);
1483         if (pair == NULL)
1484                 return (0);
1485         firstname = nvpair_name(pair);
1486
1487         error = spa_open(firstname, &spa, FTAG);
1488         if (error != 0)
1489                 return (error);
1490         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1491         spa_close(spa, FTAG);
1492
1493         if (needsuspend) {
1494                 suspended = fnvlist_alloc();
1495                 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1496                     pair = nvlist_next_nvpair(snaps, pair)) {
1497                         char fsname[ZFS_MAX_DATASET_NAME_LEN];
1498                         char *snapname = nvpair_name(pair);
1499                         char *atp;
1500                         void *cookie;
1501
1502                         atp = strchr(snapname, '@');
1503                         if (atp == NULL) {
1504                                 error = SET_ERROR(EINVAL);
1505                                 break;
1506                         }
1507                         (void) strlcpy(fsname, snapname, atp - snapname + 1);
1508
1509                         error = zil_suspend(fsname, &cookie);
1510                         if (error != 0)
1511                                 break;
1512                         fnvlist_add_uint64(suspended, fsname,
1513                             (uintptr_t)cookie);
1514                 }
1515         }
1516
1517         ddsa.ddsa_snaps = snaps;
1518         ddsa.ddsa_props = props;
1519         ddsa.ddsa_errors = errors;
1520         ddsa.ddsa_cr = CRED();
1521
1522         if (error == 0) {
1523                 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1524                     dsl_dataset_snapshot_sync, &ddsa,
1525                     fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1526         }
1527
1528         if (suspended != NULL) {
1529                 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1530                     pair = nvlist_next_nvpair(suspended, pair)) {
1531                         zil_resume((void *)(uintptr_t)
1532                             fnvpair_value_uint64(pair));
1533                 }
1534                 fnvlist_free(suspended);
1535         }
1536
1537         return (error);
1538 }
1539
1540 typedef struct dsl_dataset_snapshot_tmp_arg {
1541         const char *ddsta_fsname;
1542         const char *ddsta_snapname;
1543         minor_t ddsta_cleanup_minor;
1544         const char *ddsta_htag;
1545 } dsl_dataset_snapshot_tmp_arg_t;
1546
1547 static int
1548 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1549 {
1550         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1551         dsl_pool_t *dp = dmu_tx_pool(tx);
1552         dsl_dataset_t *ds;
1553         int error;
1554
1555         error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1556         if (error != 0)
1557                 return (error);
1558
1559         /* NULL cred means no limit check for tmp snapshot */
1560         error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1561             tx, B_FALSE, 0, NULL);
1562         if (error != 0) {
1563                 dsl_dataset_rele(ds, FTAG);
1564                 return (error);
1565         }
1566
1567         if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1568                 dsl_dataset_rele(ds, FTAG);
1569                 return (SET_ERROR(ENOTSUP));
1570         }
1571         error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1572             B_TRUE, tx);
1573         if (error != 0) {
1574                 dsl_dataset_rele(ds, FTAG);
1575                 return (error);
1576         }
1577
1578         dsl_dataset_rele(ds, FTAG);
1579         return (0);
1580 }
1581
1582 static void
1583 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1584 {
1585         dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1586         dsl_pool_t *dp = dmu_tx_pool(tx);
1587         dsl_dataset_t *ds;
1588
1589         VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1590
1591         dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1592         dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1593             ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1594         dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1595
1596         dsl_dataset_rele(ds, FTAG);
1597 }
1598
1599 int
1600 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1601     minor_t cleanup_minor, const char *htag)
1602 {
1603         dsl_dataset_snapshot_tmp_arg_t ddsta;
1604         int error;
1605         spa_t *spa;
1606         boolean_t needsuspend;
1607         void *cookie;
1608
1609         ddsta.ddsta_fsname = fsname;
1610         ddsta.ddsta_snapname = snapname;
1611         ddsta.ddsta_cleanup_minor = cleanup_minor;
1612         ddsta.ddsta_htag = htag;
1613
1614         error = spa_open(fsname, &spa, FTAG);
1615         if (error != 0)
1616                 return (error);
1617         needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1618         spa_close(spa, FTAG);
1619
1620         if (needsuspend) {
1621                 error = zil_suspend(fsname, &cookie);
1622                 if (error != 0)
1623                         return (error);
1624         }
1625
1626         error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1627             dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1628
1629         if (needsuspend)
1630                 zil_resume(cookie);
1631         return (error);
1632 }
1633
1634
1635 void
1636 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1637 {
1638         spa_feature_t f;
1639
1640         ASSERT(dmu_tx_is_syncing(tx));
1641         ASSERT(ds->ds_objset != NULL);
1642         ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1643
1644         /*
1645          * in case we had to change ds_fsid_guid when we opened it,
1646          * sync it out now.
1647          */
1648         dmu_buf_will_dirty(ds->ds_dbuf, tx);
1649         dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1650
1651         if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1652                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1653                     ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1654                     &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1655                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1656                     ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1657                     &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1658                 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1659                     ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1660                     &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1661                 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1662                 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1663                 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1664         }
1665
1666         dmu_objset_sync(ds->ds_objset, zio, tx);
1667
1668         for (f = 0; f < SPA_FEATURES; f++) {
1669                 if (ds->ds_feature_activation_needed[f]) {
1670                         if (ds->ds_feature_inuse[f])
1671                                 continue;
1672                         dsl_dataset_activate_feature(ds->ds_object, f, tx);
1673                         ds->ds_feature_inuse[f] = B_TRUE;
1674                 }
1675         }
1676 }
1677
1678 static void
1679 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1680 {
1681         uint64_t count = 0;
1682         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1683         zap_cursor_t zc;
1684         zap_attribute_t za;
1685         nvlist_t *propval = fnvlist_alloc();
1686         nvlist_t *val = fnvlist_alloc();
1687
1688         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1689
1690         /*
1691          * There may be missing entries in ds_next_clones_obj
1692          * due to a bug in a previous version of the code.
1693          * Only trust it if it has the right number of entries.
1694          */
1695         if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1696                 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1697                     &count));
1698         }
1699         if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1700                 goto fail;
1701         for (zap_cursor_init(&zc, mos,
1702             dsl_dataset_phys(ds)->ds_next_clones_obj);
1703             zap_cursor_retrieve(&zc, &za) == 0;
1704             zap_cursor_advance(&zc)) {
1705                 dsl_dataset_t *clone;
1706                 char buf[ZFS_MAX_DATASET_NAME_LEN];
1707                 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1708                     za.za_first_integer, FTAG, &clone));
1709                 dsl_dir_name(clone->ds_dir, buf);
1710                 fnvlist_add_boolean(val, buf);
1711                 dsl_dataset_rele(clone, FTAG);
1712         }
1713         zap_cursor_fini(&zc);
1714         fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1715         fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1716 fail:
1717         nvlist_free(val);
1718         nvlist_free(propval);
1719 }
1720
1721 static void
1722 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1723 {
1724         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1725
1726         if (dsl_dataset_has_resume_receive_state(ds)) {
1727                 char *str;
1728                 void *packed;
1729                 uint8_t *compressed;
1730                 uint64_t val;
1731                 nvlist_t *token_nv = fnvlist_alloc();
1732                 size_t packed_size, compressed_size;
1733                 zio_cksum_t cksum;
1734                 char *propval;
1735                 char buf[MAXNAMELEN];
1736                 int i;
1737
1738                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1739                     DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1740                         fnvlist_add_uint64(token_nv, "fromguid", val);
1741                 }
1742                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1743                     DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1744                         fnvlist_add_uint64(token_nv, "object", val);
1745                 }
1746                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1747                     DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1748                         fnvlist_add_uint64(token_nv, "offset", val);
1749                 }
1750                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1751                     DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1752                         fnvlist_add_uint64(token_nv, "bytes", val);
1753                 }
1754                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1755                     DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1756                         fnvlist_add_uint64(token_nv, "toguid", val);
1757                 }
1758                 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1759                     DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1760                         fnvlist_add_string(token_nv, "toname", buf);
1761                 }
1762                 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1763                     DS_FIELD_RESUME_EMBEDOK) == 0) {
1764                         fnvlist_add_boolean(token_nv, "embedok");
1765                 }
1766                 packed = fnvlist_pack(token_nv, &packed_size);
1767                 fnvlist_free(token_nv);
1768                 compressed = kmem_alloc(packed_size, KM_SLEEP);
1769
1770                 compressed_size = gzip_compress(packed, compressed,
1771                     packed_size, packed_size, 6);
1772
1773                 fletcher_4_native(compressed, compressed_size, &cksum);
1774
1775                 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1776                 for (i = 0; i < compressed_size; i++) {
1777                         (void) sprintf(str + i * 2, "%02x", compressed[i]);
1778                 }
1779                 str[compressed_size * 2] = '\0';
1780                 propval = kmem_asprintf("%u-%llx-%llx-%s",
1781                     ZFS_SEND_RESUME_TOKEN_VERSION,
1782                     (longlong_t)cksum.zc_word[0],
1783                     (longlong_t)packed_size, str);
1784                 dsl_prop_nvlist_add_string(nv,
1785                     ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1786                 kmem_free(packed, packed_size);
1787                 kmem_free(str, compressed_size * 2 + 1);
1788                 kmem_free(compressed, packed_size);
1789                 strfree(propval);
1790         }
1791 }
1792
1793 void
1794 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1795 {
1796         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1797         uint64_t refd, avail, uobjs, aobjs, ratio;
1798
1799         ASSERT(dsl_pool_config_held(dp));
1800
1801         ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1802             (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1803             dsl_dataset_phys(ds)->ds_compressed_bytes);
1804
1805         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1806         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1807             dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1808
1809         if (ds->ds_is_snapshot) {
1810                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1811                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1812                     dsl_dataset_phys(ds)->ds_unique_bytes);
1813                 get_clones_stat(ds, nv);
1814         } else {
1815                 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1816                         char buf[ZFS_MAX_DATASET_NAME_LEN];
1817                         dsl_dataset_name(ds->ds_prev, buf);
1818                         dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1819                 }
1820
1821                 dsl_dir_stats(ds->ds_dir, nv);
1822         }
1823
1824         dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1825         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1826         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1827
1828         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1829             dsl_dataset_phys(ds)->ds_creation_time);
1830         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1831             dsl_dataset_phys(ds)->ds_creation_txg);
1832         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1833             ds->ds_quota);
1834         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1835             ds->ds_reserved);
1836         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1837             dsl_dataset_phys(ds)->ds_guid);
1838         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1839             dsl_dataset_phys(ds)->ds_unique_bytes);
1840         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1841             ds->ds_object);
1842         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1843             ds->ds_userrefs);
1844         dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1845             DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1846
1847         if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1848                 uint64_t written, comp, uncomp;
1849                 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1850                 dsl_dataset_t *prev;
1851                 int err;
1852
1853                 err = dsl_dataset_hold_obj(dp,
1854                     dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1855                 if (err == 0) {
1856                         err = dsl_dataset_space_written(prev, ds, &written,
1857                             &comp, &uncomp);
1858                         dsl_dataset_rele(prev, FTAG);
1859                         if (err == 0) {
1860                                 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1861                                     written);
1862                         }
1863                 }
1864         }
1865
1866         if (!dsl_dataset_is_snapshot(ds)) {
1867                 /* 6 extra bytes for /%recv */
1868                 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1869                 dsl_dataset_t *recv_ds;
1870
1871                 /*
1872                  * A failed "newfs" (e.g. full) resumable receive leaves
1873                  * the stats set on this dataset.  Check here for the prop.
1874                  */
1875                 get_receive_resume_stats(ds, nv);
1876
1877                 /*
1878                  * A failed incremental resumable receive leaves the
1879                  * stats set on our child named "%recv".  Check the child
1880                  * for the prop.
1881                  */
1882                 dsl_dataset_name(ds, recvname);
1883                 if (strlcat(recvname, "/", sizeof (recvname)) <
1884                     sizeof (recvname) &&
1885                     strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1886                     sizeof (recvname) &&
1887                     dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1888                         get_receive_resume_stats(recv_ds, nv);
1889                         dsl_dataset_rele(recv_ds, FTAG);
1890                 }
1891         }
1892 }
1893
1894 void
1895 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1896 {
1897         dsl_pool_t *dp = ds->ds_dir->dd_pool;
1898         ASSERT(dsl_pool_config_held(dp));
1899
1900         stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1901         stat->dds_inconsistent =
1902             dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1903         stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1904         stat->dds_origin[0] = '\0';
1905         if (ds->ds_is_snapshot) {
1906                 stat->dds_is_snapshot = B_TRUE;
1907                 stat->dds_num_clones =
1908                     dsl_dataset_phys(ds)->ds_num_children - 1;
1909         } else {
1910                 stat->dds_is_snapshot = B_FALSE;
1911                 stat->dds_num_clones = 0;
1912
1913                 if (dsl_dir_is_clone(ds->ds_dir)) {
1914                         dsl_dataset_t *ods;
1915
1916                         VERIFY0(dsl_dataset_hold_obj(dp,
1917                             dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1918                             FTAG, &ods));
1919                         dsl_dataset_name(ods, stat->dds_origin);
1920                         dsl_dataset_rele(ods, FTAG);
1921                 }
1922         }
1923 }
1924
1925 uint64_t
1926 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1927 {
1928         return (ds->ds_fsid_guid);
1929 }
1930
1931 void
1932 dsl_dataset_space(dsl_dataset_t *ds,
1933     uint64_t *refdbytesp, uint64_t *availbytesp,
1934     uint64_t *usedobjsp, uint64_t *availobjsp)
1935 {
1936         *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1937         *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1938         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1939                 *availbytesp +=
1940                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1941         if (ds->ds_quota != 0) {
1942                 /*
1943                  * Adjust available bytes according to refquota
1944                  */
1945                 if (*refdbytesp < ds->ds_quota)
1946                         *availbytesp = MIN(*availbytesp,
1947                             ds->ds_quota - *refdbytesp);
1948                 else
1949                         *availbytesp = 0;
1950         }
1951         *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1952         *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1953 }
1954
1955 boolean_t
1956 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1957 {
1958         ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1959         if (snap == NULL)
1960                 return (B_FALSE);
1961         if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1962             dsl_dataset_phys(snap)->ds_creation_txg) {
1963                 objset_t *os, *os_snap;
1964                 /*
1965                  * It may be that only the ZIL differs, because it was
1966                  * reset in the head.  Don't count that as being
1967                  * modified.
1968                  */
1969                 if (dmu_objset_from_ds(ds, &os) != 0)
1970                         return (B_TRUE);
1971                 if (dmu_objset_from_ds(snap, &os_snap) != 0)
1972                         return (B_TRUE);
1973                 return (bcmp(&os->os_phys->os_meta_dnode,
1974                     &os_snap->os_phys->os_meta_dnode,
1975                     sizeof (os->os_phys->os_meta_dnode)) != 0);
1976         }
1977         return (B_FALSE);
1978 }
1979
1980 typedef struct dsl_dataset_rename_snapshot_arg {
1981         const char *ddrsa_fsname;
1982         const char *ddrsa_oldsnapname;
1983         const char *ddrsa_newsnapname;
1984         boolean_t ddrsa_recursive;
1985         dmu_tx_t *ddrsa_tx;
1986 } dsl_dataset_rename_snapshot_arg_t;
1987
1988 /* ARGSUSED */
1989 static int
1990 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1991     dsl_dataset_t *hds, void *arg)
1992 {
1993         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1994         int error;
1995         uint64_t val;
1996
1997         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1998         if (error != 0) {
1999                 /* ignore nonexistent snapshots */
2000                 return (error == ENOENT ? 0 : error);
2001         }
2002
2003         /* new name should not exist */
2004         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2005         if (error == 0)
2006                 error = SET_ERROR(EEXIST);
2007         else if (error == ENOENT)
2008                 error = 0;
2009
2010         /* dataset name + 1 for the "@" + the new snapshot name must fit */
2011         if (dsl_dir_namelen(hds->ds_dir) + 1 +
2012             strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2013                 error = SET_ERROR(ENAMETOOLONG);
2014
2015         return (error);
2016 }
2017
2018 static int
2019 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2020 {
2021         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2022         dsl_pool_t *dp = dmu_tx_pool(tx);
2023         dsl_dataset_t *hds;
2024         int error;
2025
2026         error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2027         if (error != 0)
2028                 return (error);
2029
2030         if (ddrsa->ddrsa_recursive) {
2031                 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2032                     dsl_dataset_rename_snapshot_check_impl, ddrsa,
2033                     DS_FIND_CHILDREN);
2034         } else {
2035                 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2036         }
2037         dsl_dataset_rele(hds, FTAG);
2038         return (error);
2039 }
2040
2041 static int
2042 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2043     dsl_dataset_t *hds, void *arg)
2044 {
2045         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2046         dsl_dataset_t *ds;
2047         uint64_t val;
2048         dmu_tx_t *tx = ddrsa->ddrsa_tx;
2049         int error;
2050
2051         error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2052         ASSERT(error == 0 || error == ENOENT);
2053         if (error == ENOENT) {
2054                 /* ignore nonexistent snapshots */
2055                 return (0);
2056         }
2057
2058         VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2059
2060         /* log before we change the name */
2061         spa_history_log_internal_ds(ds, "rename", tx,
2062             "-> @%s", ddrsa->ddrsa_newsnapname);
2063
2064         VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2065             B_FALSE));
2066         mutex_enter(&ds->ds_lock);
2067         (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2068         mutex_exit(&ds->ds_lock);
2069         VERIFY0(zap_add(dp->dp_meta_objset,
2070             dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2071             ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2072         zvol_rename_minors(dp->dp_spa, ddrsa->ddrsa_oldsnapname,
2073             ddrsa->ddrsa_newsnapname, B_TRUE);
2074
2075         dsl_dataset_rele(ds, FTAG);
2076         return (0);
2077 }
2078
2079 static void
2080 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2081 {
2082         dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2083         dsl_pool_t *dp = dmu_tx_pool(tx);
2084         dsl_dataset_t *hds;
2085
2086         VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2087         ddrsa->ddrsa_tx = tx;
2088         if (ddrsa->ddrsa_recursive) {
2089                 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2090                     dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2091                     DS_FIND_CHILDREN));
2092         } else {
2093                 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2094         }
2095         dsl_dataset_rele(hds, FTAG);
2096 }
2097
2098 int
2099 dsl_dataset_rename_snapshot(const char *fsname,
2100     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2101 {
2102         dsl_dataset_rename_snapshot_arg_t ddrsa;
2103
2104         ddrsa.ddrsa_fsname = fsname;
2105         ddrsa.ddrsa_oldsnapname = oldsnapname;
2106         ddrsa.ddrsa_newsnapname = newsnapname;
2107         ddrsa.ddrsa_recursive = recursive;
2108
2109         return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2110             dsl_dataset_rename_snapshot_sync, &ddrsa,
2111             1, ZFS_SPACE_CHECK_RESERVED));
2112 }
2113
2114 /*
2115  * If we're doing an ownership handoff, we need to make sure that there is
2116  * only one long hold on the dataset.  We're not allowed to change anything here
2117  * so we don't permanently release the long hold or regular hold here.  We want
2118  * to do this only when syncing to avoid the dataset unexpectedly going away
2119  * when we release the long hold.  Allow a long hold to exist for volumes, this
2120  * may occur when asynchronously registering the minor with the kernel.
2121  */
2122 static int
2123 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2124 {
2125         boolean_t held;
2126
2127         if (!dmu_tx_is_syncing(tx))
2128                 return (0);
2129
2130         if (owner != NULL) {
2131                 VERIFY3P(ds->ds_owner, ==, owner);
2132                 dsl_dataset_long_rele(ds, owner);
2133         }
2134
2135         held = (dsl_dataset_long_held(ds) && (ds->ds_owner != zvol_tag));
2136
2137         if (owner != NULL)
2138                 dsl_dataset_long_hold(ds, owner);
2139
2140         if (held)
2141                 return (SET_ERROR(EBUSY));
2142
2143         return (0);
2144 }
2145
2146 typedef struct dsl_dataset_rollback_arg {
2147         const char *ddra_fsname;
2148         void *ddra_owner;
2149         nvlist_t *ddra_result;
2150 } dsl_dataset_rollback_arg_t;
2151
2152 static int
2153 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2154 {
2155         dsl_dataset_rollback_arg_t *ddra = arg;
2156         dsl_pool_t *dp = dmu_tx_pool(tx);
2157         dsl_dataset_t *ds;
2158         int64_t unused_refres_delta;
2159         int error;
2160         nvpair_t *pair;
2161         nvlist_t *proprequest, *bookmarks;
2162
2163         error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2164         if (error != 0)
2165                 return (error);
2166
2167         /* must not be a snapshot */
2168         if (ds->ds_is_snapshot) {
2169                 dsl_dataset_rele(ds, FTAG);
2170                 return (SET_ERROR(EINVAL));
2171         }
2172
2173         /* must have a most recent snapshot */
2174         if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2175                 dsl_dataset_rele(ds, FTAG);
2176                 return (SET_ERROR(EINVAL));
2177         }
2178
2179         /* must not have any bookmarks after the most recent snapshot */
2180         proprequest = fnvlist_alloc();
2181         fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2182         bookmarks = fnvlist_alloc();
2183         error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2184         fnvlist_free(proprequest);
2185         if (error != 0)
2186                 return (error);
2187         for (pair = nvlist_next_nvpair(bookmarks, NULL);
2188             pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2189                 nvlist_t *valuenv =
2190                     fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2191                     zfs_prop_to_name(ZFS_PROP_CREATETXG));
2192                 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2193                 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2194                         fnvlist_free(bookmarks);
2195                         dsl_dataset_rele(ds, FTAG);
2196                         return (SET_ERROR(EEXIST));
2197                 }
2198         }
2199         fnvlist_free(bookmarks);
2200
2201         error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2202         if (error != 0) {
2203                 dsl_dataset_rele(ds, FTAG);
2204                 return (error);
2205         }
2206
2207         /*
2208          * Check if the snap we are rolling back to uses more than
2209          * the refquota.
2210          */
2211         if (ds->ds_quota != 0 &&
2212             dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2213                 dsl_dataset_rele(ds, FTAG);
2214                 return (SET_ERROR(EDQUOT));
2215         }
2216
2217         /*
2218          * When we do the clone swap, we will temporarily use more space
2219          * due to the refreservation (the head will no longer have any
2220          * unique space, so the entire amount of the refreservation will need
2221          * to be free).  We will immediately destroy the clone, freeing
2222          * this space, but the freeing happens over many txg's.
2223          */
2224         unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2225             dsl_dataset_phys(ds)->ds_unique_bytes);
2226
2227         if (unused_refres_delta > 0 &&
2228             unused_refres_delta >
2229             dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2230                 dsl_dataset_rele(ds, FTAG);
2231                 return (SET_ERROR(ENOSPC));
2232         }
2233
2234         dsl_dataset_rele(ds, FTAG);
2235         return (0);
2236 }
2237
2238 static void
2239 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2240 {
2241         dsl_dataset_rollback_arg_t *ddra = arg;
2242         dsl_pool_t *dp = dmu_tx_pool(tx);
2243         dsl_dataset_t *ds, *clone;
2244         uint64_t cloneobj;
2245         char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2246
2247         VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2248
2249         dsl_dataset_name(ds->ds_prev, namebuf);
2250         fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2251
2252         cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2253             ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2254
2255         VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2256
2257         dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2258         dsl_dataset_zero_zil(ds, tx);
2259
2260         dsl_destroy_head_sync_impl(clone, tx);
2261
2262         dsl_dataset_rele(clone, FTAG);
2263         dsl_dataset_rele(ds, FTAG);
2264 }
2265
2266 /*
2267  * Rolls back the given filesystem or volume to the most recent snapshot.
2268  * The name of the most recent snapshot will be returned under key "target"
2269  * in the result nvlist.
2270  *
2271  * If owner != NULL:
2272  * - The existing dataset MUST be owned by the specified owner at entry
2273  * - Upon return, dataset will still be held by the same owner, whether we
2274  *   succeed or not.
2275  *
2276  * This mode is required any time the existing filesystem is mounted.  See
2277  * notes above zfs_suspend_fs() for further details.
2278  */
2279 int
2280 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2281 {
2282         dsl_dataset_rollback_arg_t ddra;
2283
2284         ddra.ddra_fsname = fsname;
2285         ddra.ddra_owner = owner;
2286         ddra.ddra_result = result;
2287
2288         return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2289             dsl_dataset_rollback_sync, &ddra,
2290             1, ZFS_SPACE_CHECK_RESERVED));
2291 }
2292
2293 struct promotenode {
2294         list_node_t link;
2295         dsl_dataset_t *ds;
2296 };
2297
2298 typedef struct dsl_dataset_promote_arg {
2299         const char *ddpa_clonename;
2300         dsl_dataset_t *ddpa_clone;
2301         list_t shared_snaps, origin_snaps, clone_snaps;
2302         dsl_dataset_t *origin_origin; /* origin of the origin */
2303         uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2304         char *err_ds;
2305         cred_t *cr;
2306 } dsl_dataset_promote_arg_t;
2307
2308 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2309 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2310     void *tag);
2311 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2312
2313 static int
2314 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2315 {
2316         dsl_dataset_promote_arg_t *ddpa = arg;
2317         dsl_pool_t *dp = dmu_tx_pool(tx);
2318         dsl_dataset_t *hds;
2319         struct promotenode *snap;
2320         dsl_dataset_t *origin_ds;
2321         int err;
2322         uint64_t unused;
2323         uint64_t ss_mv_cnt;
2324         size_t max_snap_len;
2325
2326         err = promote_hold(ddpa, dp, FTAG);
2327         if (err != 0)
2328                 return (err);
2329
2330         hds = ddpa->ddpa_clone;
2331         max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2332
2333         if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2334                 promote_rele(ddpa, FTAG);
2335                 return (SET_ERROR(EXDEV));
2336         }
2337
2338         /*
2339          * Compute and check the amount of space to transfer.  Since this is
2340          * so expensive, don't do the preliminary check.
2341          */
2342         if (!dmu_tx_is_syncing(tx)) {
2343                 promote_rele(ddpa, FTAG);
2344                 return (0);
2345         }
2346
2347         snap = list_head(&ddpa->shared_snaps);
2348         origin_ds = snap->ds;
2349
2350         /* compute origin's new unique space */
2351         snap = list_tail(&ddpa->clone_snaps);
2352         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2353             origin_ds->ds_object);
2354         dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2355             dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2356             &ddpa->unique, &unused, &unused);
2357
2358         /*
2359          * Walk the snapshots that we are moving
2360          *
2361          * Compute space to transfer.  Consider the incremental changes
2362          * to used by each snapshot:
2363          * (my used) = (prev's used) + (blocks born) - (blocks killed)
2364          * So each snapshot gave birth to:
2365          * (blocks born) = (my used) - (prev's used) + (blocks killed)
2366          * So a sequence would look like:
2367          * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2368          * Which simplifies to:
2369          * uN + kN + kN-1 + ... + k1 + k0
2370          * Note however, if we stop before we reach the ORIGIN we get:
2371          * uN + kN + kN-1 + ... + kM - uM-1
2372          */
2373         ss_mv_cnt = 0;
2374         ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2375         ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2376         ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2377         for (snap = list_head(&ddpa->shared_snaps); snap;
2378             snap = list_next(&ddpa->shared_snaps, snap)) {
2379                 uint64_t val, dlused, dlcomp, dluncomp;
2380                 dsl_dataset_t *ds = snap->ds;
2381
2382                 ss_mv_cnt++;
2383
2384                 /*
2385                  * If there are long holds, we won't be able to evict
2386                  * the objset.
2387                  */
2388                 if (dsl_dataset_long_held(ds)) {
2389                         err = SET_ERROR(EBUSY);
2390                         goto out;
2391                 }
2392
2393                 /* Check that the snapshot name does not conflict */
2394                 VERIFY0(dsl_dataset_get_snapname(ds));
2395                 if (strlen(ds->ds_snapname) >= max_snap_len) {
2396                         err = SET_ERROR(ENAMETOOLONG);
2397                         goto out;
2398                 }
2399                 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2400                 if (err == 0) {
2401                         (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2402                         err = SET_ERROR(EEXIST);
2403                         goto out;
2404                 }
2405                 if (err != ENOENT)
2406                         goto out;
2407
2408                 /* The very first snapshot does not have a deadlist */
2409                 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2410                         continue;
2411
2412                 dsl_deadlist_space(&ds->ds_deadlist,
2413                     &dlused, &dlcomp, &dluncomp);
2414                 ddpa->used += dlused;
2415                 ddpa->comp += dlcomp;
2416                 ddpa->uncomp += dluncomp;
2417         }
2418
2419         /*
2420          * If we are a clone of a clone then we never reached ORIGIN,
2421          * so we need to subtract out the clone origin's used space.
2422          */
2423         if (ddpa->origin_origin) {
2424                 ddpa->used -=
2425                     dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2426                 ddpa->comp -=
2427                     dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2428                 ddpa->uncomp -=
2429                     dsl_dataset_phys(ddpa->origin_origin)->
2430                     ds_uncompressed_bytes;
2431         }
2432
2433         /* Check that there is enough space and limit headroom here */
2434         err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2435             0, ss_mv_cnt, ddpa->used, ddpa->cr);
2436         if (err != 0)
2437                 goto out;
2438
2439         /*
2440          * Compute the amounts of space that will be used by snapshots
2441          * after the promotion (for both origin and clone).  For each,
2442          * it is the amount of space that will be on all of their
2443          * deadlists (that was not born before their new origin).
2444          */
2445         if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2446                 uint64_t space;
2447
2448                 /*
2449                  * Note, typically this will not be a clone of a clone,
2450                  * so dd_origin_txg will be < TXG_INITIAL, so
2451                  * these snaplist_space() -> dsl_deadlist_space_range()
2452                  * calls will be fast because they do not have to
2453                  * iterate over all bps.
2454                  */
2455                 snap = list_head(&ddpa->origin_snaps);
2456                 err = snaplist_space(&ddpa->shared_snaps,
2457                     snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2458                 if (err != 0)
2459                         goto out;
2460
2461                 err = snaplist_space(&ddpa->clone_snaps,
2462                     snap->ds->ds_dir->dd_origin_txg, &space);
2463                 if (err != 0)
2464                         goto out;
2465                 ddpa->cloneusedsnap += space;
2466         }
2467         if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2468             DD_FLAG_USED_BREAKDOWN) {
2469                 err = snaplist_space(&ddpa->origin_snaps,
2470                     dsl_dataset_phys(origin_ds)->ds_creation_txg,
2471                     &ddpa->originusedsnap);
2472                 if (err != 0)
2473                         goto out;
2474         }
2475
2476 out:
2477         promote_rele(ddpa, FTAG);
2478         return (err);
2479 }
2480
2481 static void
2482 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2483 {
2484         dsl_dataset_promote_arg_t *ddpa = arg;
2485         dsl_pool_t *dp = dmu_tx_pool(tx);
2486         dsl_dataset_t *hds;
2487         struct promotenode *snap;
2488         dsl_dataset_t *origin_ds;
2489         dsl_dataset_t *origin_head;
2490         dsl_dir_t *dd;
2491         dsl_dir_t *odd = NULL;
2492         uint64_t oldnext_obj;
2493         int64_t delta;
2494
2495         VERIFY0(promote_hold(ddpa, dp, FTAG));
2496         hds = ddpa->ddpa_clone;
2497
2498         ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2499
2500         snap = list_head(&ddpa->shared_snaps);
2501         origin_ds = snap->ds;
2502         dd = hds->ds_dir;
2503
2504         snap = list_head(&ddpa->origin_snaps);
2505         origin_head = snap->ds;
2506
2507         /*
2508          * We need to explicitly open odd, since origin_ds's dd will be
2509          * changing.
2510          */
2511         VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2512             NULL, FTAG, &odd));
2513
2514         /* change origin's next snap */
2515         dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2516         oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2517         snap = list_tail(&ddpa->clone_snaps);
2518         ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2519             origin_ds->ds_object);
2520         dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2521
2522         /* change the origin's next clone */
2523         if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2524                 dsl_dataset_remove_from_next_clones(origin_ds,
2525                     snap->ds->ds_object, tx);
2526                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2527                     dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2528                     oldnext_obj, tx));
2529         }
2530
2531         /* change origin */
2532         dmu_buf_will_dirty(dd->dd_dbuf, tx);
2533         ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2534         dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2535         dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2536         dmu_buf_will_dirty(odd->dd_dbuf, tx);
2537         dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2538         origin_head->ds_dir->dd_origin_txg =
2539             dsl_dataset_phys(origin_ds)->ds_creation_txg;
2540
2541         /* change dd_clone entries */
2542         if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2543                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2544                     dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2545                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2546                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2547                     hds->ds_object, tx));
2548
2549                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2550                     dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2551                     origin_head->ds_object, tx));
2552                 if (dsl_dir_phys(dd)->dd_clones == 0) {
2553                         dsl_dir_phys(dd)->dd_clones =
2554                             zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2555                             DMU_OT_NONE, 0, tx);
2556                 }
2557                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2558                     dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2559         }
2560
2561         /* move snapshots to this dir */
2562         for (snap = list_head(&ddpa->shared_snaps); snap;
2563             snap = list_next(&ddpa->shared_snaps, snap)) {
2564                 dsl_dataset_t *ds = snap->ds;
2565
2566                 /*
2567                  * Property callbacks are registered to a particular
2568                  * dsl_dir.  Since ours is changing, evict the objset
2569                  * so that they will be unregistered from the old dsl_dir.
2570                  */
2571                 if (ds->ds_objset) {
2572                         dmu_objset_evict(ds->ds_objset);
2573                         ds->ds_objset = NULL;
2574                 }
2575
2576                 /* move snap name entry */
2577                 VERIFY0(dsl_dataset_get_snapname(ds));
2578                 VERIFY0(dsl_dataset_snap_remove(origin_head,
2579                     ds->ds_snapname, tx, B_TRUE));
2580                 VERIFY0(zap_add(dp->dp_meta_objset,
2581                     dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2582                     8, 1, &ds->ds_object, tx));
2583                 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2584                     DD_FIELD_SNAPSHOT_COUNT, tx);
2585
2586                 /* change containing dsl_dir */
2587                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2588                 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2589                 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2590                 ASSERT3P(ds->ds_dir, ==, odd);
2591                 dsl_dir_rele(ds->ds_dir, ds);
2592                 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2593                     NULL, ds, &ds->ds_dir));
2594
2595                 /* move any clone references */
2596                 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2597                     spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2598                         zap_cursor_t zc;
2599                         zap_attribute_t za;
2600
2601                         for (zap_cursor_init(&zc, dp->dp_meta_objset,
2602                             dsl_dataset_phys(ds)->ds_next_clones_obj);
2603                             zap_cursor_retrieve(&zc, &za) == 0;
2604                             zap_cursor_advance(&zc)) {
2605                                 dsl_dataset_t *cnds;
2606                                 uint64_t o;
2607
2608                                 if (za.za_first_integer == oldnext_obj) {
2609                                         /*
2610                                          * We've already moved the
2611                                          * origin's reference.
2612                                          */
2613                                         continue;
2614                                 }
2615
2616                                 VERIFY0(dsl_dataset_hold_obj(dp,
2617                                     za.za_first_integer, FTAG, &cnds));
2618                                 o = dsl_dir_phys(cnds->ds_dir)->
2619                                     dd_head_dataset_obj;
2620
2621                                 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2622                                     dsl_dir_phys(odd)->dd_clones, o, tx));
2623                                 VERIFY0(zap_add_int(dp->dp_meta_objset,
2624                                     dsl_dir_phys(dd)->dd_clones, o, tx));
2625                                 dsl_dataset_rele(cnds, FTAG);
2626                         }
2627                         zap_cursor_fini(&zc);
2628                 }
2629
2630                 ASSERT(!dsl_prop_hascb(ds));
2631         }
2632
2633         /*
2634          * Change space accounting.
2635          * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2636          * both be valid, or both be 0 (resulting in delta == 0).  This
2637          * is true for each of {clone,origin} independently.
2638          */
2639
2640         delta = ddpa->cloneusedsnap -
2641             dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2642         ASSERT3S(delta, >=, 0);
2643         ASSERT3U(ddpa->used, >=, delta);
2644         dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2645         dsl_dir_diduse_space(dd, DD_USED_HEAD,
2646             ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2647
2648         delta = ddpa->originusedsnap -
2649             dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2650         ASSERT3S(delta, <=, 0);
2651         ASSERT3U(ddpa->used, >=, -delta);
2652         dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2653         dsl_dir_diduse_space(odd, DD_USED_HEAD,
2654             -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2655
2656         dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2657
2658         /* log history record */
2659         spa_history_log_internal_ds(hds, "promote", tx, "");
2660
2661         dsl_dir_rele(odd, FTAG);
2662         promote_rele(ddpa, FTAG);
2663 }
2664
2665 /*
2666  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2667  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2668  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2669  * snapshots back to this dataset's origin.
2670  */
2671 static int
2672 snaplist_make(dsl_pool_t *dp,
2673     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2674 {
2675         uint64_t obj = last_obj;
2676
2677         list_create(l, sizeof (struct promotenode),
2678             offsetof(struct promotenode, link));
2679
2680         while (obj != first_obj) {
2681                 dsl_dataset_t *ds;
2682                 struct promotenode *snap;
2683                 int err;
2684
2685                 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2686                 ASSERT(err != ENOENT);
2687                 if (err != 0)
2688                         return (err);
2689
2690                 if (first_obj == 0)
2691                         first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2692
2693                 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2694                 snap->ds = ds;
2695                 list_insert_tail(l, snap);
2696                 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2697         }
2698
2699         return (0);
2700 }
2701
2702 static int
2703 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2704 {
2705         struct promotenode *snap;
2706
2707         *spacep = 0;
2708         for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2709                 uint64_t used, comp, uncomp;
2710                 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2711                     mintxg, UINT64_MAX, &used, &comp, &uncomp);
2712                 *spacep += used;
2713         }
2714         return (0);
2715 }
2716
2717 static void
2718 snaplist_destroy(list_t *l, void *tag)
2719 {
2720         struct promotenode *snap;
2721
2722         if (l == NULL || !list_link_active(&l->list_head))
2723                 return;
2724
2725         while ((snap = list_tail(l)) != NULL) {
2726                 list_remove(l, snap);
2727                 dsl_dataset_rele(snap->ds, tag);
2728                 kmem_free(snap, sizeof (*snap));
2729         }
2730         list_destroy(l);
2731 }
2732
2733 static int
2734 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2735 {
2736         int error;
2737         dsl_dir_t *dd;
2738         struct promotenode *snap;
2739
2740         error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2741             &ddpa->ddpa_clone);
2742         if (error != 0)
2743                 return (error);
2744         dd = ddpa->ddpa_clone->ds_dir;
2745
2746         if (ddpa->ddpa_clone->ds_is_snapshot ||
2747             !dsl_dir_is_clone(dd)) {
2748                 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2749                 return (SET_ERROR(EINVAL));
2750         }
2751
2752         error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2753             &ddpa->shared_snaps, tag);
2754         if (error != 0)
2755                 goto out;
2756
2757         error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2758             &ddpa->clone_snaps, tag);
2759         if (error != 0)
2760                 goto out;
2761
2762         snap = list_head(&ddpa->shared_snaps);
2763         ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2764         error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2765             dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2766             &ddpa->origin_snaps, tag);
2767         if (error != 0)
2768                 goto out;
2769
2770         if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2771                 error = dsl_dataset_hold_obj(dp,
2772                     dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2773                     tag, &ddpa->origin_origin);
2774                 if (error != 0)
2775                         goto out;
2776         }
2777 out:
2778         if (error != 0)
2779                 promote_rele(ddpa, tag);
2780         return (error);
2781 }
2782
2783 static void
2784 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2785 {
2786         snaplist_destroy(&ddpa->shared_snaps, tag);
2787         snaplist_destroy(&ddpa->clone_snaps, tag);
2788         snaplist_destroy(&ddpa->origin_snaps, tag);
2789         if (ddpa->origin_origin != NULL)
2790                 dsl_dataset_rele(ddpa->origin_origin, tag);
2791         dsl_dataset_rele(ddpa->ddpa_clone, tag);
2792 }
2793
2794 /*
2795  * Promote a clone.
2796  *
2797  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2798  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
2799  */
2800 int
2801 dsl_dataset_promote(const char *name, char *conflsnap)
2802 {
2803         dsl_dataset_promote_arg_t ddpa = { 0 };
2804         uint64_t numsnaps;
2805         int error;
2806         objset_t *os;
2807
2808         /*
2809          * We will modify space proportional to the number of
2810          * snapshots.  Compute numsnaps.
2811          */
2812         error = dmu_objset_hold(name, FTAG, &os);
2813         if (error != 0)
2814                 return (error);
2815         error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2816             dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2817             &numsnaps);
2818         dmu_objset_rele(os, FTAG);
2819         if (error != 0)
2820                 return (error);
2821
2822         ddpa.ddpa_clonename = name;
2823         ddpa.err_ds = conflsnap;
2824         ddpa.cr = CRED();
2825
2826         return (dsl_sync_task(name, dsl_dataset_promote_check,
2827             dsl_dataset_promote_sync, &ddpa,
2828             2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2829 }
2830
2831 int
2832 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2833     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2834 {
2835         /*
2836          * "slack" factor for received datasets with refquota set on them.
2837          * See the bottom of this function for details on its use.
2838          */
2839         uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
2840         int64_t unused_refres_delta;
2841
2842         /* they should both be heads */
2843         if (clone->ds_is_snapshot ||
2844             origin_head->ds_is_snapshot)
2845                 return (SET_ERROR(EINVAL));
2846
2847         /* if we are not forcing, the branch point should be just before them */
2848         if (!force && clone->ds_prev != origin_head->ds_prev)
2849                 return (SET_ERROR(EINVAL));
2850
2851         /* clone should be the clone (unless they are unrelated) */
2852         if (clone->ds_prev != NULL &&
2853             clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2854             origin_head->ds_dir != clone->ds_prev->ds_dir)
2855                 return (SET_ERROR(EINVAL));
2856
2857         /* the clone should be a child of the origin */
2858         if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2859                 return (SET_ERROR(EINVAL));
2860
2861         /* origin_head shouldn't be modified unless 'force' */
2862         if (!force &&
2863             dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2864                 return (SET_ERROR(ETXTBSY));
2865
2866         /* origin_head should have no long holds (e.g. is not mounted) */
2867         if (dsl_dataset_handoff_check(origin_head, owner, tx))
2868                 return (SET_ERROR(EBUSY));
2869
2870         /* check amount of any unconsumed refreservation */
2871         unused_refres_delta =
2872             (int64_t)MIN(origin_head->ds_reserved,
2873             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2874             (int64_t)MIN(origin_head->ds_reserved,
2875             dsl_dataset_phys(clone)->ds_unique_bytes);
2876
2877         if (unused_refres_delta > 0 &&
2878             unused_refres_delta >
2879             dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2880                 return (SET_ERROR(ENOSPC));
2881
2882         /*
2883          * The clone can't be too much over the head's refquota.
2884          *
2885          * To ensure that the entire refquota can be used, we allow one
2886          * transaction to exceed the the refquota.  Therefore, this check
2887          * needs to also allow for the space referenced to be more than the
2888          * refquota.  The maximum amount of space that one transaction can use
2889          * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
2890          * overage ensures that we are able to receive a filesystem that
2891          * exceeds the refquota on the source system.
2892          *
2893          * So that overage is the refquota_slack we use below.
2894          */
2895         if (origin_head->ds_quota != 0 &&
2896             dsl_dataset_phys(clone)->ds_referenced_bytes >
2897             origin_head->ds_quota + refquota_slack)
2898                 return (SET_ERROR(EDQUOT));
2899
2900         return (0);
2901 }
2902
2903 void
2904 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2905     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2906 {
2907         spa_feature_t f;
2908         dsl_pool_t *dp = dmu_tx_pool(tx);
2909         int64_t unused_refres_delta;
2910
2911         ASSERT(clone->ds_reserved == 0);
2912         /*
2913          * NOTE: On DEBUG kernels there could be a race between this and
2914          * the check function if spa_asize_inflation is adjusted...
2915          */
2916         ASSERT(origin_head->ds_quota == 0 ||
2917             dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
2918             DMU_MAX_ACCESS * spa_asize_inflation);
2919         ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2920
2921         /*
2922          * Swap per-dataset feature flags.
2923          */
2924         for (f = 0; f < SPA_FEATURES; f++) {
2925                 boolean_t clone_inuse;
2926                 boolean_t origin_head_inuse;
2927
2928                 if (!(spa_feature_table[f].fi_flags &
2929                     ZFEATURE_FLAG_PER_DATASET)) {
2930                         ASSERT(!clone->ds_feature_inuse[f]);
2931                         ASSERT(!origin_head->ds_feature_inuse[f]);
2932                         continue;
2933                 }
2934
2935                 clone_inuse = clone->ds_feature_inuse[f];
2936                 origin_head_inuse = origin_head->ds_feature_inuse[f];
2937
2938                 if (clone_inuse) {
2939                         dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2940                         clone->ds_feature_inuse[f] = B_FALSE;
2941                 }
2942                 if (origin_head_inuse) {
2943                         dsl_dataset_deactivate_feature(origin_head->ds_object,
2944                             f, tx);
2945                         origin_head->ds_feature_inuse[f] = B_FALSE;
2946                 }
2947                 if (clone_inuse) {
2948                         dsl_dataset_activate_feature(origin_head->ds_object,
2949                             f, tx);
2950                         origin_head->ds_feature_inuse[f] = B_TRUE;
2951                 }
2952                 if (origin_head_inuse) {
2953                         dsl_dataset_activate_feature(clone->ds_object, f, tx);
2954                         clone->ds_feature_inuse[f] = B_TRUE;
2955                 }
2956         }
2957
2958         dmu_buf_will_dirty(clone->ds_dbuf, tx);
2959         dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2960
2961         if (clone->ds_objset != NULL) {
2962                 dmu_objset_evict(clone->ds_objset);
2963                 clone->ds_objset = NULL;
2964         }
2965
2966         if (origin_head->ds_objset != NULL) {
2967                 dmu_objset_evict(origin_head->ds_objset);
2968                 origin_head->ds_objset = NULL;
2969         }
2970
2971         unused_refres_delta =
2972             (int64_t)MIN(origin_head->ds_reserved,
2973             dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2974             (int64_t)MIN(origin_head->ds_reserved,
2975             dsl_dataset_phys(clone)->ds_unique_bytes);
2976
2977         /*
2978          * Reset origin's unique bytes, if it exists.
2979          */
2980         if (clone->ds_prev) {
2981                 dsl_dataset_t *origin = clone->ds_prev;
2982                 uint64_t comp, uncomp;
2983
2984                 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2985                 dsl_deadlist_space_range(&clone->ds_deadlist,
2986                     dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2987                     &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2988         }
2989
2990         /* swap blkptrs */
2991         {
2992                 blkptr_t tmp;
2993                 tmp = dsl_dataset_phys(origin_head)->ds_bp;
2994                 dsl_dataset_phys(origin_head)->ds_bp =
2995                     dsl_dataset_phys(clone)->ds_bp;
2996                 dsl_dataset_phys(clone)->ds_bp = tmp;
2997         }
2998
2999         /* set dd_*_bytes */
3000         {
3001                 int64_t dused, dcomp, duncomp;
3002                 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3003                 uint64_t odl_used, odl_comp, odl_uncomp;
3004
3005                 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3006                     dd_used_breakdown[DD_USED_SNAP], ==, 0);
3007
3008                 dsl_deadlist_space(&clone->ds_deadlist,
3009                     &cdl_used, &cdl_comp, &cdl_uncomp);
3010                 dsl_deadlist_space(&origin_head->ds_deadlist,
3011                     &odl_used, &odl_comp, &odl_uncomp);
3012
3013                 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3014                     cdl_used -
3015                     (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3016                     odl_used);
3017                 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3018                     cdl_comp -
3019                     (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3020                     odl_comp);
3021                 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3022                     cdl_uncomp -
3023                     (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3024                     odl_uncomp);
3025
3026                 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3027                     dused, dcomp, duncomp, tx);
3028                 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3029                     -dused, -dcomp, -duncomp, tx);
3030
3031                 /*
3032                  * The difference in the space used by snapshots is the
3033                  * difference in snapshot space due to the head's
3034                  * deadlist (since that's the only thing that's
3035                  * changing that affects the snapused).
3036                  */
3037                 dsl_deadlist_space_range(&clone->ds_deadlist,
3038                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3039                     &cdl_used, &cdl_comp, &cdl_uncomp);
3040                 dsl_deadlist_space_range(&origin_head->ds_deadlist,
3041                     origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3042                     &odl_used, &odl_comp, &odl_uncomp);
3043                 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3044                     DD_USED_HEAD, DD_USED_SNAP, tx);
3045         }
3046
3047         /* swap ds_*_bytes */
3048         SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3049             dsl_dataset_phys(clone)->ds_referenced_bytes);
3050         SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3051             dsl_dataset_phys(clone)->ds_compressed_bytes);
3052         SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3053             dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3054         SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3055             dsl_dataset_phys(clone)->ds_unique_bytes);
3056
3057         /* apply any parent delta for change in unconsumed refreservation */
3058         dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3059             unused_refres_delta, 0, 0, tx);
3060
3061         /*
3062          * Swap deadlists.
3063          */
3064         dsl_deadlist_close(&clone->ds_deadlist);
3065         dsl_deadlist_close(&origin_head->ds_deadlist);
3066         SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3067             dsl_dataset_phys(clone)->ds_deadlist_obj);
3068         dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3069             dsl_dataset_phys(clone)->ds_deadlist_obj);
3070         dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3071             dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3072
3073         dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3074
3075         spa_history_log_internal_ds(clone, "clone swap", tx,
3076             "parent=%s", origin_head->ds_dir->dd_myname);
3077 }
3078
3079 /*
3080  * Given a pool name and a dataset object number in that pool,
3081  * return the name of that dataset.
3082  */
3083 int
3084 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3085 {
3086         dsl_pool_t *dp;
3087         dsl_dataset_t *ds;
3088         int error;
3089
3090         error = dsl_pool_hold(pname, FTAG, &dp);
3091         if (error != 0)
3092                 return (error);
3093
3094         error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3095         if (error == 0) {
3096                 dsl_dataset_name(ds, buf);
3097                 dsl_dataset_rele(ds, FTAG);
3098         }
3099         dsl_pool_rele(dp, FTAG);
3100
3101         return (error);
3102 }
3103
3104 int
3105 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3106     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3107 {
3108         int error = 0;
3109
3110         ASSERT3S(asize, >, 0);
3111
3112         /*
3113          * *ref_rsrv is the portion of asize that will come from any
3114          * unconsumed refreservation space.
3115          */
3116         *ref_rsrv = 0;
3117
3118         mutex_enter(&ds->ds_lock);
3119         /*
3120          * Make a space adjustment for reserved bytes.
3121          */
3122         if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3123                 ASSERT3U(*used, >=,
3124                     ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3125                 *used -=
3126                     (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3127                 *ref_rsrv =
3128                     asize - MIN(asize, parent_delta(ds, asize + inflight));
3129         }
3130
3131         if (!check_quota || ds->ds_quota == 0) {
3132                 mutex_exit(&ds->ds_lock);
3133                 return (0);
3134         }
3135         /*
3136          * If they are requesting more space, and our current estimate
3137          * is over quota, they get to try again unless the actual
3138          * on-disk is over quota and there are no pending changes (which
3139          * may free up space for us).
3140          */
3141         if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3142             ds->ds_quota) {
3143                 if (inflight > 0 ||
3144                     dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3145                         error = SET_ERROR(ERESTART);
3146                 else
3147                         error = SET_ERROR(EDQUOT);
3148         }
3149         mutex_exit(&ds->ds_lock);
3150
3151         return (error);
3152 }
3153
3154 typedef struct dsl_dataset_set_qr_arg {
3155         const char *ddsqra_name;
3156         zprop_source_t ddsqra_source;
3157         uint64_t ddsqra_value;
3158 } dsl_dataset_set_qr_arg_t;
3159
3160
3161 /* ARGSUSED */
3162 static int
3163 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3164 {
3165         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3166         dsl_pool_t *dp = dmu_tx_pool(tx);
3167         dsl_dataset_t *ds;
3168         int error;
3169         uint64_t newval;
3170
3171         if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3172                 return (SET_ERROR(ENOTSUP));
3173
3174         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3175         if (error != 0)
3176                 return (error);
3177
3178         if (ds->ds_is_snapshot) {
3179                 dsl_dataset_rele(ds, FTAG);
3180                 return (SET_ERROR(EINVAL));
3181         }
3182
3183         error = dsl_prop_predict(ds->ds_dir,
3184             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3185             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3186         if (error != 0) {
3187                 dsl_dataset_rele(ds, FTAG);
3188                 return (error);
3189         }
3190
3191         if (newval == 0) {
3192                 dsl_dataset_rele(ds, FTAG);
3193                 return (0);
3194         }
3195
3196         if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3197             newval < ds->ds_reserved) {
3198                 dsl_dataset_rele(ds, FTAG);
3199                 return (SET_ERROR(ENOSPC));
3200         }
3201
3202         dsl_dataset_rele(ds, FTAG);
3203         return (0);
3204 }
3205
3206 static void
3207 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3208 {
3209         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3210         dsl_pool_t *dp = dmu_tx_pool(tx);
3211         dsl_dataset_t *ds;
3212         uint64_t newval;
3213
3214         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3215
3216         dsl_prop_set_sync_impl(ds,
3217             zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3218             ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3219             &ddsqra->ddsqra_value, tx);
3220
3221         VERIFY0(dsl_prop_get_int_ds(ds,
3222             zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3223
3224         if (ds->ds_quota != newval) {
3225                 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3226                 ds->ds_quota = newval;
3227         }
3228         dsl_dataset_rele(ds, FTAG);
3229 }
3230
3231 int
3232 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3233     uint64_t refquota)
3234 {
3235         dsl_dataset_set_qr_arg_t ddsqra;
3236
3237         ddsqra.ddsqra_name = dsname;
3238         ddsqra.ddsqra_source = source;
3239         ddsqra.ddsqra_value = refquota;
3240
3241         return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3242             dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3243 }
3244
3245 static int
3246 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3247 {
3248         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3249         dsl_pool_t *dp = dmu_tx_pool(tx);
3250         dsl_dataset_t *ds;
3251         int error;
3252         uint64_t newval, unique;
3253
3254         if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3255                 return (SET_ERROR(ENOTSUP));
3256
3257         error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3258         if (error != 0)
3259                 return (error);
3260
3261         if (ds->ds_is_snapshot) {
3262                 dsl_dataset_rele(ds, FTAG);
3263                 return (SET_ERROR(EINVAL));
3264         }
3265
3266         error = dsl_prop_predict(ds->ds_dir,
3267             zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3268             ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3269         if (error != 0) {
3270                 dsl_dataset_rele(ds, FTAG);
3271                 return (error);
3272         }
3273
3274         /*
3275          * If we are doing the preliminary check in open context, the
3276          * space estimates may be inaccurate.
3277          */
3278         if (!dmu_tx_is_syncing(tx)) {
3279                 dsl_dataset_rele(ds, FTAG);
3280                 return (0);
3281         }
3282
3283         mutex_enter(&ds->ds_lock);
3284         if (!DS_UNIQUE_IS_ACCURATE(ds))
3285                 dsl_dataset_recalc_head_uniq(ds);
3286         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3287         mutex_exit(&ds->ds_lock);
3288
3289         if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3290                 uint64_t delta = MAX(unique, newval) -
3291                     MAX(unique, ds->ds_reserved);
3292
3293                 if (delta >
3294                     dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3295                     (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3296                         dsl_dataset_rele(ds, FTAG);
3297                         return (SET_ERROR(ENOSPC));
3298                 }
3299         }
3300
3301         dsl_dataset_rele(ds, FTAG);
3302         return (0);
3303 }
3304
3305 void
3306 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3307     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3308 {
3309         uint64_t newval;
3310         uint64_t unique;
3311         int64_t delta;
3312
3313         dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3314             source, sizeof (value), 1, &value, tx);
3315
3316         VERIFY0(dsl_prop_get_int_ds(ds,
3317             zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3318
3319         dmu_buf_will_dirty(ds->ds_dbuf, tx);
3320         mutex_enter(&ds->ds_dir->dd_lock);
3321         mutex_enter(&ds->ds_lock);
3322         ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3323         unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3324         delta = MAX(0, (int64_t)(newval - unique)) -
3325             MAX(0, (int64_t)(ds->ds_reserved - unique));
3326         ds->ds_reserved = newval;
3327         mutex_exit(&ds->ds_lock);
3328
3329         dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3330         mutex_exit(&ds->ds_dir->dd_lock);
3331 }
3332
3333 static void
3334 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3335 {
3336         dsl_dataset_set_qr_arg_t *ddsqra = arg;
3337         dsl_pool_t *dp = dmu_tx_pool(tx);
3338         dsl_dataset_t *ds;
3339
3340         VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3341         dsl_dataset_set_refreservation_sync_impl(ds,
3342             ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3343         dsl_dataset_rele(ds, FTAG);
3344 }
3345
3346 int
3347 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3348     uint64_t refreservation)
3349 {
3350         dsl_dataset_set_qr_arg_t ddsqra;
3351
3352         ddsqra.ddsqra_name = dsname;
3353         ddsqra.ddsqra_source = source;
3354         ddsqra.ddsqra_value = refreservation;
3355
3356         return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3357             dsl_dataset_set_refreservation_sync, &ddsqra,
3358             0, ZFS_SPACE_CHECK_NONE));
3359 }
3360
3361 /*
3362  * Return (in *usedp) the amount of space written in new that is not
3363  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3364  * a snapshot before new, in new's filesystem (or its origin).  If not then
3365  * fail and return EINVAL.
3366  *
3367  * The written space is calculated by considering two components:  First, we
3368  * ignore any freed space, and calculate the written as new's used space
3369  * minus old's used space.  Next, we add in the amount of space that was freed
3370  * between the two snapshots, thus reducing new's used space relative to old's.
3371  * Specifically, this is the space that was born before old->ds_creation_txg,
3372  * and freed before new (ie. on new's deadlist or a previous deadlist).
3373  *
3374  * space freed                         [---------------------]
3375  * snapshots                       ---O-------O--------O-------O------
3376  *                                         oldsnap            new
3377  */
3378 int
3379 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3380     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3381 {
3382         int err = 0;
3383         uint64_t snapobj;
3384         dsl_pool_t *dp = new->ds_dir->dd_pool;
3385
3386         ASSERT(dsl_pool_config_held(dp));
3387
3388         *usedp = 0;
3389         *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3390         *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3391
3392         *compp = 0;
3393         *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3394         *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3395
3396         *uncompp = 0;
3397         *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3398         *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3399
3400         snapobj = new->ds_object;
3401         while (snapobj != oldsnap->ds_object) {
3402                 dsl_dataset_t *snap;
3403                 uint64_t used, comp, uncomp;
3404
3405                 if (snapobj == new->ds_object) {
3406                         snap = new;
3407                 } else {
3408                         err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3409                         if (err != 0)
3410                                 break;
3411                 }
3412
3413                 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3414                     dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3415                         /*
3416                          * The blocks in the deadlist can not be born after
3417                          * ds_prev_snap_txg, so get the whole deadlist space,
3418                          * which is more efficient (especially for old-format
3419                          * deadlists).  Unfortunately the deadlist code
3420                          * doesn't have enough information to make this
3421                          * optimization itself.
3422                          */
3423                         dsl_deadlist_space(&snap->ds_deadlist,
3424                             &used, &comp, &uncomp);
3425                 } else {
3426                         dsl_deadlist_space_range(&snap->ds_deadlist,
3427                             0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3428                             &used, &comp, &uncomp);
3429                 }
3430                 *usedp += used;
3431                 *compp += comp;
3432                 *uncompp += uncomp;
3433
3434                 /*
3435                  * If we get to the beginning of the chain of snapshots
3436                  * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3437                  * was not a snapshot of/before new.
3438                  */
3439                 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3440                 if (snap != new)
3441                         dsl_dataset_rele(snap, FTAG);
3442                 if (snapobj == 0) {
3443                         err = SET_ERROR(EINVAL);
3444                         break;
3445                 }
3446
3447         }
3448         return (err);
3449 }
3450
3451 /*
3452  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3453  * lastsnap, and all snapshots in between are deleted.
3454  *
3455  * blocks that would be freed            [---------------------------]
3456  * snapshots                       ---O-------O--------O-------O--------O
3457  *                                        firstsnap        lastsnap
3458  *
3459  * This is the set of blocks that were born after the snap before firstsnap,
3460  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3461  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3462  * We calculate this by iterating over the relevant deadlists (from the snap
3463  * after lastsnap, backward to the snap after firstsnap), summing up the
3464  * space on the deadlist that was born after the snap before firstsnap.
3465  */
3466 int
3467 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3468     dsl_dataset_t *lastsnap,
3469     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3470 {
3471         int err = 0;
3472         uint64_t snapobj;
3473         dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3474
3475         ASSERT(firstsnap->ds_is_snapshot);
3476         ASSERT(lastsnap->ds_is_snapshot);
3477
3478         /*
3479          * Check that the snapshots are in the same dsl_dir, and firstsnap
3480          * is before lastsnap.
3481          */
3482         if (firstsnap->ds_dir != lastsnap->ds_dir ||
3483             dsl_dataset_phys(firstsnap)->ds_creation_txg >
3484             dsl_dataset_phys(lastsnap)->ds_creation_txg)
3485                 return (SET_ERROR(EINVAL));
3486
3487         *usedp = *compp = *uncompp = 0;
3488
3489         snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3490         while (snapobj != firstsnap->ds_object) {
3491                 dsl_dataset_t *ds;
3492                 uint64_t used, comp, uncomp;
3493
3494                 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3495                 if (err != 0)
3496                         break;
3497
3498                 dsl_deadlist_space_range(&ds->ds_deadlist,
3499                     dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3500                     &used, &comp, &uncomp);
3501                 *usedp += used;
3502                 *compp += comp;
3503                 *uncompp += uncomp;
3504
3505                 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3506                 ASSERT3U(snapobj, !=, 0);
3507                 dsl_dataset_rele(ds, FTAG);
3508         }
3509         return (err);
3510 }
3511
3512 /*
3513  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3514  * For example, they could both be snapshots of the same filesystem, and
3515  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3516  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3517  * filesystem.  Or 'earlier' could be the origin's origin.
3518  *
3519  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3520  */
3521 boolean_t
3522 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3523         uint64_t earlier_txg)
3524 {
3525         dsl_pool_t *dp = later->ds_dir->dd_pool;
3526         int error;
3527         boolean_t ret;
3528         dsl_dataset_t *origin;
3529
3530         ASSERT(dsl_pool_config_held(dp));
3531         ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3532
3533         if (earlier_txg == 0)
3534                 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3535
3536         if (later->ds_is_snapshot &&
3537             earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3538                 return (B_FALSE);
3539
3540         if (later->ds_dir == earlier->ds_dir)
3541                 return (B_TRUE);
3542         if (!dsl_dir_is_clone(later->ds_dir))
3543                 return (B_FALSE);
3544
3545         if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3546                 return (B_TRUE);
3547         error = dsl_dataset_hold_obj(dp,
3548             dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3549         if (error != 0)
3550                 return (B_FALSE);
3551         ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3552         dsl_dataset_rele(origin, FTAG);
3553         return (ret);
3554 }
3555
3556 void
3557 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3558 {
3559         objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3560         dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3561 }
3562
3563 boolean_t
3564 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3565 {
3566         dmu_object_info_t doi;
3567
3568         dmu_object_info_from_db(ds->ds_dbuf, &doi);
3569         return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3570 }
3571
3572 boolean_t
3573 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3574 {
3575         return (dsl_dataset_is_zapified(ds) &&
3576             zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3577             ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3578 }
3579
3580 #if defined(_KERNEL) && defined(HAVE_SPL)
3581 #if defined(_LP64)
3582 module_param(zfs_max_recordsize, int, 0644);
3583 MODULE_PARM_DESC(zfs_max_recordsize, "Max allowed record size");
3584 #else
3585 /* Limited to 1M on 32-bit platforms due to lack of virtual address space */
3586 module_param(zfs_max_recordsize, int, 0444);
3587 MODULE_PARM_DESC(zfs_max_recordsize, "Max allowed record size");
3588 #endif
3589
3590 EXPORT_SYMBOL(dsl_dataset_hold);
3591 EXPORT_SYMBOL(dsl_dataset_hold_obj);
3592 EXPORT_SYMBOL(dsl_dataset_own);
3593 EXPORT_SYMBOL(dsl_dataset_own_obj);
3594 EXPORT_SYMBOL(dsl_dataset_name);
3595 EXPORT_SYMBOL(dsl_dataset_rele);
3596 EXPORT_SYMBOL(dsl_dataset_disown);
3597 EXPORT_SYMBOL(dsl_dataset_tryown);
3598 EXPORT_SYMBOL(dsl_dataset_create_sync);
3599 EXPORT_SYMBOL(dsl_dataset_create_sync_dd);
3600 EXPORT_SYMBOL(dsl_dataset_snapshot_check);
3601 EXPORT_SYMBOL(dsl_dataset_snapshot_sync);
3602 EXPORT_SYMBOL(dsl_dataset_promote);
3603 EXPORT_SYMBOL(dsl_dataset_user_hold);
3604 EXPORT_SYMBOL(dsl_dataset_user_release);
3605 EXPORT_SYMBOL(dsl_dataset_get_holds);
3606 EXPORT_SYMBOL(dsl_dataset_get_blkptr);
3607 EXPORT_SYMBOL(dsl_dataset_set_blkptr);
3608 EXPORT_SYMBOL(dsl_dataset_get_spa);
3609 EXPORT_SYMBOL(dsl_dataset_modified_since_snap);
3610 EXPORT_SYMBOL(dsl_dataset_space_written);
3611 EXPORT_SYMBOL(dsl_dataset_space_wouldfree);
3612 EXPORT_SYMBOL(dsl_dataset_sync);
3613 EXPORT_SYMBOL(dsl_dataset_block_born);
3614 EXPORT_SYMBOL(dsl_dataset_block_kill);
3615 EXPORT_SYMBOL(dsl_dataset_block_freeable);
3616 EXPORT_SYMBOL(dsl_dataset_prev_snap_txg);
3617 EXPORT_SYMBOL(dsl_dataset_dirty);
3618 EXPORT_SYMBOL(dsl_dataset_stats);
3619 EXPORT_SYMBOL(dsl_dataset_fast_stat);
3620 EXPORT_SYMBOL(dsl_dataset_space);
3621 EXPORT_SYMBOL(dsl_dataset_fsid_guid);
3622 EXPORT_SYMBOL(dsl_dsobj_to_dsname);
3623 EXPORT_SYMBOL(dsl_dataset_check_quota);
3624 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl);
3625 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl);
3626 #endif