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 https://opensource.org/licenses/CDDL-1.0. 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 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2020 by Delphix. All rights reserved. 25 * Copyright (c) 2014, Joyent, Inc. All rights reserved. 26 * Copyright (c) 2014 RackTop Systems. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 * Copyright (c) 2016 Actifio, Inc. All rights reserved. 29 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved. 30 * Copyright 2017 Nexenta Systems, Inc. 31 * Copyright (c) 2019, Klara Inc. 32 * Copyright (c) 2019, Allan Jude 33 * Copyright (c) 2020 The FreeBSD Foundation [1] 34 * 35 * [1] Portions of this software were developed by Allan Jude 36 * under sponsorship from the FreeBSD Foundation. 37 */ 38 39 #include <sys/dmu_objset.h> 40 #include <sys/dsl_dataset.h> 41 #include <sys/dsl_dir.h> 42 #include <sys/dsl_prop.h> 43 #include <sys/dsl_synctask.h> 44 #include <sys/dmu_traverse.h> 45 #include <sys/dmu_impl.h> 46 #include <sys/dmu_tx.h> 47 #include <sys/arc.h> 48 #include <sys/zio.h> 49 #include <sys/zap.h> 50 #include <sys/zfeature.h> 51 #include <sys/unique.h> 52 #include <sys/zfs_context.h> 53 #include <sys/zfs_ioctl.h> 54 #include <sys/spa.h> 55 #include <sys/spa_impl.h> 56 #include <sys/vdev.h> 57 #include <sys/zfs_znode.h> 58 #include <sys/zfs_onexit.h> 59 #include <sys/zvol.h> 60 #include <sys/dsl_scan.h> 61 #include <sys/dsl_deadlist.h> 62 #include <sys/dsl_destroy.h> 63 #include <sys/dsl_userhold.h> 64 #include <sys/dsl_bookmark.h> 65 #include <sys/policy.h> 66 #include <sys/dmu_send.h> 67 #include <sys/dmu_recv.h> 68 #include <sys/zio_compress.h> 69 #include <zfs_fletcher.h> 70 #include <sys/zio_checksum.h> 71 72 /* 73 * The SPA supports block sizes up to 16MB. However, very large blocks 74 * can have an impact on i/o latency (e.g. tying up a spinning disk for 75 * ~300ms), and also potentially on the memory allocator. Therefore, 76 * we did not allow the recordsize to be set larger than zfs_max_recordsize 77 * (former default: 1MB). Larger blocks could be created by changing this 78 * tunable, and pools with larger blocks could always be imported and used, 79 * regardless of this setting. 80 * 81 * We do, however, still limit it by default to 1M on x86_32, because Linux's 82 * 3/1 memory split doesn't leave much room for 16M chunks. 83 */ 84 #ifdef _ILP32 85 uint_t zfs_max_recordsize = 1 * 1024 * 1024; 86 #else 87 uint_t zfs_max_recordsize = 16 * 1024 * 1024; 88 #endif 89 static int zfs_allow_redacted_dataset_mount = 0; 90 91 int zfs_snapshot_history_enabled = 1; 92 93 #define SWITCH64(x, y) \ 94 { \ 95 uint64_t __tmp = (x); \ 96 (x) = (y); \ 97 (y) = __tmp; \ 98 } 99 100 #define DS_REF_MAX (1ULL << 62) 101 102 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, 103 uint64_t obj, dmu_tx_t *tx); 104 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, 105 dmu_tx_t *tx); 106 107 static void unload_zfeature(dsl_dataset_t *ds, spa_feature_t f); 108 109 extern uint_t spa_asize_inflation; 110 111 static zil_header_t zero_zil; 112 113 /* 114 * Figure out how much of this delta should be propagated to the dsl_dir 115 * layer. If there's a refreservation, that space has already been 116 * partially accounted for in our ancestors. 117 */ 118 static int64_t 119 parent_delta(dsl_dataset_t *ds, int64_t delta) 120 { 121 dsl_dataset_phys_t *ds_phys; 122 uint64_t old_bytes, new_bytes; 123 124 if (ds->ds_reserved == 0) 125 return (delta); 126 127 ds_phys = dsl_dataset_phys(ds); 128 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved); 129 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 130 131 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 132 return (new_bytes - old_bytes); 133 } 134 135 void 136 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 137 { 138 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 139 int used = bp_get_dsize_sync(spa, bp); 140 int compressed = BP_GET_PSIZE(bp); 141 int uncompressed = BP_GET_UCSIZE(bp); 142 int64_t delta; 143 spa_feature_t f; 144 145 dprintf_bp(bp, "ds=%p", ds); 146 147 ASSERT(dmu_tx_is_syncing(tx)); 148 /* It could have been compressed away to nothing */ 149 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp)) 150 return; 151 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 152 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 153 if (ds == NULL) { 154 dsl_pool_mos_diduse_space(tx->tx_pool, 155 used, compressed, uncompressed); 156 return; 157 } 158 159 ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 160 dmu_buf_will_dirty(ds->ds_dbuf, tx); 161 mutex_enter(&ds->ds_lock); 162 delta = parent_delta(ds, used); 163 dsl_dataset_phys(ds)->ds_referenced_bytes += used; 164 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed; 165 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed; 166 dsl_dataset_phys(ds)->ds_unique_bytes += used; 167 168 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) { 169 ds->ds_feature_activation[SPA_FEATURE_LARGE_BLOCKS] = 170 (void *)B_TRUE; 171 } 172 173 174 f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp)); 175 if (f != SPA_FEATURE_NONE) { 176 ASSERT3S(spa_feature_table[f].fi_type, ==, 177 ZFEATURE_TYPE_BOOLEAN); 178 ds->ds_feature_activation[f] = (void *)B_TRUE; 179 } 180 181 f = zio_compress_to_feature(BP_GET_COMPRESS(bp)); 182 if (f != SPA_FEATURE_NONE) { 183 ASSERT3S(spa_feature_table[f].fi_type, ==, 184 ZFEATURE_TYPE_BOOLEAN); 185 ds->ds_feature_activation[f] = (void *)B_TRUE; 186 } 187 188 /* 189 * Track block for livelist, but ignore embedded blocks because 190 * they do not need to be freed. 191 */ 192 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) && 193 bp->blk_birth > ds->ds_dir->dd_origin_txg && 194 !(BP_IS_EMBEDDED(bp))) { 195 ASSERT(dsl_dir_is_clone(ds->ds_dir)); 196 ASSERT(spa_feature_is_enabled(spa, 197 SPA_FEATURE_LIVELIST)); 198 bplist_append(&ds->ds_dir->dd_pending_allocs, bp); 199 } 200 201 mutex_exit(&ds->ds_lock); 202 dsl_dir_diduse_transfer_space(ds->ds_dir, delta, 203 compressed, uncompressed, used, 204 DD_USED_REFRSRV, DD_USED_HEAD, tx); 205 } 206 207 /* 208 * Called when the specified segment has been remapped, and is thus no 209 * longer referenced in the head dataset. The vdev must be indirect. 210 * 211 * If the segment is referenced by a snapshot, put it on the remap deadlist. 212 * Otherwise, add this segment to the obsolete spacemap. 213 */ 214 void 215 dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, uint64_t offset, 216 uint64_t size, uint64_t birth, dmu_tx_t *tx) 217 { 218 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 219 220 ASSERT(dmu_tx_is_syncing(tx)); 221 ASSERT(birth <= tx->tx_txg); 222 ASSERT(!ds->ds_is_snapshot); 223 224 if (birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 225 spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx); 226 } else { 227 blkptr_t fakebp; 228 dva_t *dva = &fakebp.blk_dva[0]; 229 230 ASSERT(ds != NULL); 231 232 mutex_enter(&ds->ds_remap_deadlist_lock); 233 if (!dsl_dataset_remap_deadlist_exists(ds)) { 234 dsl_dataset_create_remap_deadlist(ds, tx); 235 } 236 mutex_exit(&ds->ds_remap_deadlist_lock); 237 238 BP_ZERO(&fakebp); 239 fakebp.blk_birth = birth; 240 DVA_SET_VDEV(dva, vdev); 241 DVA_SET_OFFSET(dva, offset); 242 DVA_SET_ASIZE(dva, size); 243 dsl_deadlist_insert(&ds->ds_remap_deadlist, &fakebp, B_FALSE, 244 tx); 245 } 246 } 247 248 int 249 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 250 boolean_t async) 251 { 252 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 253 254 int used = bp_get_dsize_sync(spa, bp); 255 int compressed = BP_GET_PSIZE(bp); 256 int uncompressed = BP_GET_UCSIZE(bp); 257 258 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp)) 259 return (0); 260 261 ASSERT(dmu_tx_is_syncing(tx)); 262 ASSERT(bp->blk_birth <= tx->tx_txg); 263 264 if (ds == NULL) { 265 dsl_free(tx->tx_pool, tx->tx_txg, bp); 266 dsl_pool_mos_diduse_space(tx->tx_pool, 267 -used, -compressed, -uncompressed); 268 return (used); 269 } 270 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 271 272 ASSERT(!ds->ds_is_snapshot); 273 dmu_buf_will_dirty(ds->ds_dbuf, tx); 274 275 /* 276 * Track block for livelist, but ignore embedded blocks because 277 * they do not need to be freed. 278 */ 279 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) && 280 bp->blk_birth > ds->ds_dir->dd_origin_txg && 281 !(BP_IS_EMBEDDED(bp))) { 282 ASSERT(dsl_dir_is_clone(ds->ds_dir)); 283 ASSERT(spa_feature_is_enabled(spa, 284 SPA_FEATURE_LIVELIST)); 285 bplist_append(&ds->ds_dir->dd_pending_frees, bp); 286 } 287 288 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 289 int64_t delta; 290 291 dprintf_bp(bp, "freeing ds=%llu", (u_longlong_t)ds->ds_object); 292 dsl_free(tx->tx_pool, tx->tx_txg, bp); 293 294 mutex_enter(&ds->ds_lock); 295 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used || 296 !DS_UNIQUE_IS_ACCURATE(ds)); 297 delta = parent_delta(ds, -used); 298 dsl_dataset_phys(ds)->ds_unique_bytes -= used; 299 mutex_exit(&ds->ds_lock); 300 dsl_dir_diduse_transfer_space(ds->ds_dir, 301 delta, -compressed, -uncompressed, -used, 302 DD_USED_REFRSRV, DD_USED_HEAD, tx); 303 } else { 304 dprintf_bp(bp, "putting on dead list: %s", ""); 305 if (async) { 306 /* 307 * We are here as part of zio's write done callback, 308 * which means we're a zio interrupt thread. We can't 309 * call dsl_deadlist_insert() now because it may block 310 * waiting for I/O. Instead, put bp on the deferred 311 * queue and let dsl_pool_sync() finish the job. 312 */ 313 bplist_append(&ds->ds_pending_deadlist, bp); 314 } else { 315 dsl_deadlist_insert(&ds->ds_deadlist, bp, B_FALSE, tx); 316 } 317 ASSERT3U(ds->ds_prev->ds_object, ==, 318 dsl_dataset_phys(ds)->ds_prev_snap_obj); 319 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0); 320 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 321 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 322 ds->ds_object && bp->blk_birth > 323 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) { 324 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 325 mutex_enter(&ds->ds_prev->ds_lock); 326 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used; 327 mutex_exit(&ds->ds_prev->ds_lock); 328 } 329 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 330 dsl_dir_transfer_space(ds->ds_dir, used, 331 DD_USED_HEAD, DD_USED_SNAP, tx); 332 } 333 } 334 335 dsl_bookmark_block_killed(ds, bp, tx); 336 337 mutex_enter(&ds->ds_lock); 338 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used); 339 dsl_dataset_phys(ds)->ds_referenced_bytes -= used; 340 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed); 341 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed; 342 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed); 343 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed; 344 mutex_exit(&ds->ds_lock); 345 346 return (used); 347 } 348 349 struct feature_type_uint64_array_arg { 350 uint64_t length; 351 uint64_t *array; 352 }; 353 354 static void 355 unload_zfeature(dsl_dataset_t *ds, spa_feature_t f) 356 { 357 switch (spa_feature_table[f].fi_type) { 358 case ZFEATURE_TYPE_BOOLEAN: 359 break; 360 case ZFEATURE_TYPE_UINT64_ARRAY: 361 { 362 struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f]; 363 kmem_free(ftuaa->array, ftuaa->length * sizeof (uint64_t)); 364 kmem_free(ftuaa, sizeof (*ftuaa)); 365 break; 366 } 367 default: 368 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); 369 } 370 } 371 372 static int 373 load_zfeature(objset_t *mos, dsl_dataset_t *ds, spa_feature_t f) 374 { 375 int err = 0; 376 switch (spa_feature_table[f].fi_type) { 377 case ZFEATURE_TYPE_BOOLEAN: 378 err = zap_contains(mos, ds->ds_object, 379 spa_feature_table[f].fi_guid); 380 if (err == 0) { 381 ds->ds_feature[f] = (void *)B_TRUE; 382 } else { 383 ASSERT3U(err, ==, ENOENT); 384 err = 0; 385 } 386 break; 387 case ZFEATURE_TYPE_UINT64_ARRAY: 388 { 389 uint64_t int_size, num_int; 390 uint64_t *data; 391 err = zap_length(mos, ds->ds_object, 392 spa_feature_table[f].fi_guid, &int_size, &num_int); 393 if (err != 0) { 394 ASSERT3U(err, ==, ENOENT); 395 err = 0; 396 break; 397 } 398 ASSERT3U(int_size, ==, sizeof (uint64_t)); 399 data = kmem_alloc(int_size * num_int, KM_SLEEP); 400 VERIFY0(zap_lookup(mos, ds->ds_object, 401 spa_feature_table[f].fi_guid, int_size, num_int, data)); 402 struct feature_type_uint64_array_arg *ftuaa = 403 kmem_alloc(sizeof (*ftuaa), KM_SLEEP); 404 ftuaa->length = num_int; 405 ftuaa->array = data; 406 ds->ds_feature[f] = ftuaa; 407 break; 408 } 409 default: 410 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); 411 } 412 return (err); 413 } 414 415 /* 416 * We have to release the fsid synchronously or we risk that a subsequent 417 * mount of the same dataset will fail to unique_insert the fsid. This 418 * failure would manifest itself as the fsid of this dataset changing 419 * between mounts which makes NFS clients quite unhappy. 420 */ 421 static void 422 dsl_dataset_evict_sync(void *dbu) 423 { 424 dsl_dataset_t *ds = dbu; 425 426 ASSERT(ds->ds_owner == NULL); 427 428 unique_remove(ds->ds_fsid_guid); 429 } 430 431 static void 432 dsl_dataset_evict_async(void *dbu) 433 { 434 dsl_dataset_t *ds = dbu; 435 436 ASSERT(ds->ds_owner == NULL); 437 438 ds->ds_dbuf = NULL; 439 440 if (ds->ds_objset != NULL) 441 dmu_objset_evict(ds->ds_objset); 442 443 if (ds->ds_prev) { 444 dsl_dataset_rele(ds->ds_prev, ds); 445 ds->ds_prev = NULL; 446 } 447 448 dsl_bookmark_fini_ds(ds); 449 450 bplist_destroy(&ds->ds_pending_deadlist); 451 if (dsl_deadlist_is_open(&ds->ds_deadlist)) 452 dsl_deadlist_close(&ds->ds_deadlist); 453 if (dsl_deadlist_is_open(&ds->ds_remap_deadlist)) 454 dsl_deadlist_close(&ds->ds_remap_deadlist); 455 if (ds->ds_dir) 456 dsl_dir_async_rele(ds->ds_dir, ds); 457 458 ASSERT(!list_link_active(&ds->ds_synced_link)); 459 460 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 461 if (dsl_dataset_feature_is_active(ds, f)) 462 unload_zfeature(ds, f); 463 } 464 465 list_destroy(&ds->ds_prop_cbs); 466 mutex_destroy(&ds->ds_lock); 467 mutex_destroy(&ds->ds_opening_lock); 468 mutex_destroy(&ds->ds_sendstream_lock); 469 mutex_destroy(&ds->ds_remap_deadlist_lock); 470 zfs_refcount_destroy(&ds->ds_longholds); 471 rrw_destroy(&ds->ds_bp_rwlock); 472 473 kmem_free(ds, sizeof (dsl_dataset_t)); 474 } 475 476 int 477 dsl_dataset_get_snapname(dsl_dataset_t *ds) 478 { 479 dsl_dataset_phys_t *headphys; 480 int err; 481 dmu_buf_t *headdbuf; 482 dsl_pool_t *dp = ds->ds_dir->dd_pool; 483 objset_t *mos = dp->dp_meta_objset; 484 485 if (ds->ds_snapname[0]) 486 return (0); 487 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) 488 return (0); 489 490 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, 491 FTAG, &headdbuf); 492 if (err != 0) 493 return (err); 494 headphys = headdbuf->db_data; 495 err = zap_value_search(dp->dp_meta_objset, 496 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 497 if (err != 0 && zfs_recover == B_TRUE) { 498 err = 0; 499 (void) snprintf(ds->ds_snapname, sizeof (ds->ds_snapname), 500 "SNAPOBJ=%llu-ERR=%d", 501 (unsigned long long)ds->ds_object, err); 502 } 503 dmu_buf_rele(headdbuf, FTAG); 504 return (err); 505 } 506 507 int 508 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 509 { 510 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 511 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 512 matchtype_t mt = 0; 513 int err; 514 515 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 516 mt = MT_NORMALIZE; 517 518 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 519 value, mt, NULL, 0, NULL); 520 if (err == ENOTSUP && (mt & MT_NORMALIZE)) 521 err = zap_lookup(mos, snapobj, name, 8, 1, value); 522 return (err); 523 } 524 525 int 526 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 527 boolean_t adj_cnt) 528 { 529 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 530 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 531 matchtype_t mt = 0; 532 int err; 533 534 dsl_dir_snap_cmtime_update(ds->ds_dir, tx); 535 536 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 537 mt = MT_NORMALIZE; 538 539 err = zap_remove_norm(mos, snapobj, name, mt, tx); 540 if (err == ENOTSUP && (mt & MT_NORMALIZE)) 541 err = zap_remove(mos, snapobj, name, tx); 542 543 if (err == 0 && adj_cnt) 544 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 545 DD_FIELD_SNAPSHOT_COUNT, tx); 546 547 return (err); 548 } 549 550 boolean_t 551 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, const void *tag) 552 { 553 dmu_buf_t *dbuf = ds->ds_dbuf; 554 boolean_t result = B_FALSE; 555 556 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset, 557 ds->ds_object, DMU_BONUS_BLKID, tag)) { 558 559 if (ds == dmu_buf_get_user(dbuf)) 560 result = B_TRUE; 561 else 562 dmu_buf_rele(dbuf, tag); 563 } 564 565 return (result); 566 } 567 568 int 569 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, const void *tag, 570 dsl_dataset_t **dsp) 571 { 572 objset_t *mos = dp->dp_meta_objset; 573 dmu_buf_t *dbuf; 574 dsl_dataset_t *ds; 575 int err; 576 dmu_object_info_t doi; 577 578 ASSERT(dsl_pool_config_held(dp)); 579 580 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 581 if (err != 0) 582 return (err); 583 584 /* Make sure dsobj has the correct object type. */ 585 dmu_object_info_from_db(dbuf, &doi); 586 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 587 dmu_buf_rele(dbuf, tag); 588 return (SET_ERROR(EINVAL)); 589 } 590 591 ds = dmu_buf_get_user(dbuf); 592 if (ds == NULL) { 593 dsl_dataset_t *winner = NULL; 594 595 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 596 ds->ds_dbuf = dbuf; 597 ds->ds_object = dsobj; 598 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0; 599 list_link_init(&ds->ds_synced_link); 600 601 err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj, 602 NULL, ds, &ds->ds_dir); 603 if (err != 0) { 604 kmem_free(ds, sizeof (dsl_dataset_t)); 605 dmu_buf_rele(dbuf, tag); 606 return (err); 607 } 608 609 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 610 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 611 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 612 mutex_init(&ds->ds_remap_deadlist_lock, 613 NULL, MUTEX_DEFAULT, NULL); 614 rrw_init(&ds->ds_bp_rwlock, B_FALSE); 615 zfs_refcount_create(&ds->ds_longholds); 616 617 bplist_create(&ds->ds_pending_deadlist); 618 619 list_create(&ds->ds_sendstreams, sizeof (dmu_sendstatus_t), 620 offsetof(dmu_sendstatus_t, dss_link)); 621 622 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t), 623 offsetof(dsl_prop_cb_record_t, cbr_ds_node)); 624 625 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 626 spa_feature_t f; 627 628 for (f = 0; f < SPA_FEATURES; f++) { 629 if (!(spa_feature_table[f].fi_flags & 630 ZFEATURE_FLAG_PER_DATASET)) 631 continue; 632 err = load_zfeature(mos, ds, f); 633 } 634 } 635 636 if (!ds->ds_is_snapshot) { 637 ds->ds_snapname[0] = '\0'; 638 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 639 err = dsl_dataset_hold_obj(dp, 640 dsl_dataset_phys(ds)->ds_prev_snap_obj, 641 ds, &ds->ds_prev); 642 } 643 if (err != 0) 644 goto after_dsl_bookmark_fini; 645 err = dsl_bookmark_init_ds(ds); 646 } else { 647 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 648 err = dsl_dataset_get_snapname(ds); 649 if (err == 0 && 650 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { 651 err = zap_count( 652 ds->ds_dir->dd_pool->dp_meta_objset, 653 dsl_dataset_phys(ds)->ds_userrefs_obj, 654 &ds->ds_userrefs); 655 } 656 } 657 658 if (err == 0 && !ds->ds_is_snapshot) { 659 err = dsl_prop_get_int_ds(ds, 660 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 661 &ds->ds_reserved); 662 if (err == 0) { 663 err = dsl_prop_get_int_ds(ds, 664 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 665 &ds->ds_quota); 666 } 667 } else { 668 ds->ds_reserved = ds->ds_quota = 0; 669 } 670 671 if (err == 0 && ds->ds_dir->dd_crypto_obj != 0 && 672 ds->ds_is_snapshot && 673 zap_contains(mos, dsobj, DS_FIELD_IVSET_GUID) != 0) { 674 dp->dp_spa->spa_errata = 675 ZPOOL_ERRATA_ZOL_8308_ENCRYPTION; 676 } 677 678 dsl_deadlist_open(&ds->ds_deadlist, 679 mos, dsl_dataset_phys(ds)->ds_deadlist_obj); 680 uint64_t remap_deadlist_obj = 681 dsl_dataset_get_remap_deadlist_object(ds); 682 if (remap_deadlist_obj != 0) { 683 dsl_deadlist_open(&ds->ds_remap_deadlist, mos, 684 remap_deadlist_obj); 685 } 686 687 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync, 688 dsl_dataset_evict_async, &ds->ds_dbuf); 689 if (err == 0) 690 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu); 691 692 if (err != 0 || winner != NULL) { 693 dsl_deadlist_close(&ds->ds_deadlist); 694 if (dsl_deadlist_is_open(&ds->ds_remap_deadlist)) 695 dsl_deadlist_close(&ds->ds_remap_deadlist); 696 dsl_bookmark_fini_ds(ds); 697 after_dsl_bookmark_fini: 698 if (ds->ds_prev) 699 dsl_dataset_rele(ds->ds_prev, ds); 700 dsl_dir_rele(ds->ds_dir, ds); 701 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 702 if (dsl_dataset_feature_is_active(ds, f)) 703 unload_zfeature(ds, f); 704 } 705 706 list_destroy(&ds->ds_prop_cbs); 707 list_destroy(&ds->ds_sendstreams); 708 bplist_destroy(&ds->ds_pending_deadlist); 709 mutex_destroy(&ds->ds_lock); 710 mutex_destroy(&ds->ds_opening_lock); 711 mutex_destroy(&ds->ds_sendstream_lock); 712 mutex_destroy(&ds->ds_remap_deadlist_lock); 713 zfs_refcount_destroy(&ds->ds_longholds); 714 rrw_destroy(&ds->ds_bp_rwlock); 715 kmem_free(ds, sizeof (dsl_dataset_t)); 716 if (err != 0) { 717 dmu_buf_rele(dbuf, tag); 718 return (err); 719 } 720 ds = winner; 721 } else { 722 ds->ds_fsid_guid = 723 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); 724 if (ds->ds_fsid_guid != 725 dsl_dataset_phys(ds)->ds_fsid_guid) { 726 zfs_dbgmsg("ds_fsid_guid changed from " 727 "%llx to %llx for pool %s dataset id %llu", 728 (long long) 729 dsl_dataset_phys(ds)->ds_fsid_guid, 730 (long long)ds->ds_fsid_guid, 731 spa_name(dp->dp_spa), 732 (u_longlong_t)dsobj); 733 } 734 } 735 } 736 737 ASSERT3P(ds->ds_dbuf, ==, dbuf); 738 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); 739 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || 740 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 741 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 742 *dsp = ds; 743 744 return (0); 745 } 746 747 int 748 dsl_dataset_create_key_mapping(dsl_dataset_t *ds) 749 { 750 dsl_dir_t *dd = ds->ds_dir; 751 752 if (dd->dd_crypto_obj == 0) 753 return (0); 754 755 return (spa_keystore_create_mapping(dd->dd_pool->dp_spa, 756 ds, ds, &ds->ds_key_mapping)); 757 } 758 759 int 760 dsl_dataset_hold_obj_flags(dsl_pool_t *dp, uint64_t dsobj, 761 ds_hold_flags_t flags, const void *tag, dsl_dataset_t **dsp) 762 { 763 int err; 764 765 err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 766 if (err != 0) 767 return (err); 768 769 ASSERT3P(*dsp, !=, NULL); 770 771 if (flags & DS_HOLD_FLAG_DECRYPT) { 772 err = dsl_dataset_create_key_mapping(*dsp); 773 if (err != 0) 774 dsl_dataset_rele(*dsp, tag); 775 } 776 777 return (err); 778 } 779 780 int 781 dsl_dataset_hold_flags(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, 782 const void *tag, dsl_dataset_t **dsp) 783 { 784 dsl_dir_t *dd; 785 const char *snapname; 786 uint64_t obj; 787 int err = 0; 788 dsl_dataset_t *ds; 789 790 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 791 if (err != 0) 792 return (err); 793 794 ASSERT(dsl_pool_config_held(dp)); 795 obj = dsl_dir_phys(dd)->dd_head_dataset_obj; 796 if (obj != 0) 797 err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, &ds); 798 else 799 err = SET_ERROR(ENOENT); 800 801 /* we may be looking for a snapshot */ 802 if (err == 0 && snapname != NULL) { 803 dsl_dataset_t *snap_ds; 804 805 if (*snapname++ != '@') { 806 dsl_dataset_rele_flags(ds, flags, tag); 807 dsl_dir_rele(dd, FTAG); 808 return (SET_ERROR(ENOENT)); 809 } 810 811 dprintf("looking for snapshot '%s'\n", snapname); 812 err = dsl_dataset_snap_lookup(ds, snapname, &obj); 813 if (err == 0) { 814 err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, 815 &snap_ds); 816 } 817 dsl_dataset_rele_flags(ds, flags, tag); 818 819 if (err == 0) { 820 mutex_enter(&snap_ds->ds_lock); 821 if (snap_ds->ds_snapname[0] == 0) 822 (void) strlcpy(snap_ds->ds_snapname, snapname, 823 sizeof (snap_ds->ds_snapname)); 824 mutex_exit(&snap_ds->ds_lock); 825 ds = snap_ds; 826 } 827 } 828 if (err == 0) 829 *dsp = ds; 830 dsl_dir_rele(dd, FTAG); 831 return (err); 832 } 833 834 int 835 dsl_dataset_hold(dsl_pool_t *dp, const char *name, const void *tag, 836 dsl_dataset_t **dsp) 837 { 838 return (dsl_dataset_hold_flags(dp, name, 0, tag, dsp)); 839 } 840 841 static int 842 dsl_dataset_own_obj_impl(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags, 843 const void *tag, boolean_t override, dsl_dataset_t **dsp) 844 { 845 int err = dsl_dataset_hold_obj_flags(dp, dsobj, flags, tag, dsp); 846 if (err != 0) 847 return (err); 848 if (!dsl_dataset_tryown(*dsp, tag, override)) { 849 dsl_dataset_rele_flags(*dsp, flags, tag); 850 *dsp = NULL; 851 return (SET_ERROR(EBUSY)); 852 } 853 return (0); 854 } 855 856 857 int 858 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags, 859 const void *tag, dsl_dataset_t **dsp) 860 { 861 return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_FALSE, dsp)); 862 } 863 864 int 865 dsl_dataset_own_obj_force(dsl_pool_t *dp, uint64_t dsobj, 866 ds_hold_flags_t flags, const void *tag, dsl_dataset_t **dsp) 867 { 868 return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_TRUE, dsp)); 869 } 870 871 static int 872 dsl_dataset_own_impl(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, 873 const void *tag, boolean_t override, dsl_dataset_t **dsp) 874 { 875 int err = dsl_dataset_hold_flags(dp, name, flags, tag, dsp); 876 if (err != 0) 877 return (err); 878 if (!dsl_dataset_tryown(*dsp, tag, override)) { 879 dsl_dataset_rele_flags(*dsp, flags, tag); 880 return (SET_ERROR(EBUSY)); 881 } 882 return (0); 883 } 884 885 int 886 dsl_dataset_own_force(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, 887 const void *tag, dsl_dataset_t **dsp) 888 { 889 return (dsl_dataset_own_impl(dp, name, flags, tag, B_TRUE, dsp)); 890 } 891 892 int 893 dsl_dataset_own(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, 894 const void *tag, dsl_dataset_t **dsp) 895 { 896 return (dsl_dataset_own_impl(dp, name, flags, tag, B_FALSE, dsp)); 897 } 898 899 /* 900 * See the comment above dsl_pool_hold() for details. In summary, a long 901 * hold is used to prevent destruction of a dataset while the pool hold 902 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 903 * 904 * The dataset and pool must be held when this function is called. After it 905 * is called, the pool hold may be released while the dataset is still held 906 * and accessed. 907 */ 908 void 909 dsl_dataset_long_hold(dsl_dataset_t *ds, const void *tag) 910 { 911 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 912 (void) zfs_refcount_add(&ds->ds_longholds, tag); 913 } 914 915 void 916 dsl_dataset_long_rele(dsl_dataset_t *ds, const void *tag) 917 { 918 (void) zfs_refcount_remove(&ds->ds_longholds, tag); 919 } 920 921 /* Return B_TRUE if there are any long holds on this dataset. */ 922 boolean_t 923 dsl_dataset_long_held(dsl_dataset_t *ds) 924 { 925 return (!zfs_refcount_is_zero(&ds->ds_longholds)); 926 } 927 928 void 929 dsl_dataset_name(dsl_dataset_t *ds, char *name) 930 { 931 if (ds == NULL) { 932 (void) strlcpy(name, "mos", ZFS_MAX_DATASET_NAME_LEN); 933 } else { 934 dsl_dir_name(ds->ds_dir, name); 935 VERIFY0(dsl_dataset_get_snapname(ds)); 936 if (ds->ds_snapname[0]) { 937 VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN), 938 <, ZFS_MAX_DATASET_NAME_LEN); 939 /* 940 * We use a "recursive" mutex so that we 941 * can call dprintf_ds() with ds_lock held. 942 */ 943 if (!MUTEX_HELD(&ds->ds_lock)) { 944 mutex_enter(&ds->ds_lock); 945 VERIFY3U(strlcat(name, ds->ds_snapname, 946 ZFS_MAX_DATASET_NAME_LEN), <, 947 ZFS_MAX_DATASET_NAME_LEN); 948 mutex_exit(&ds->ds_lock); 949 } else { 950 VERIFY3U(strlcat(name, ds->ds_snapname, 951 ZFS_MAX_DATASET_NAME_LEN), <, 952 ZFS_MAX_DATASET_NAME_LEN); 953 } 954 } 955 } 956 } 957 958 int 959 dsl_dataset_namelen(dsl_dataset_t *ds) 960 { 961 VERIFY0(dsl_dataset_get_snapname(ds)); 962 mutex_enter(&ds->ds_lock); 963 int len = strlen(ds->ds_snapname); 964 mutex_exit(&ds->ds_lock); 965 /* add '@' if ds is a snap */ 966 if (len > 0) 967 len++; 968 len += dsl_dir_namelen(ds->ds_dir); 969 return (len); 970 } 971 972 void 973 dsl_dataset_rele(dsl_dataset_t *ds, const void *tag) 974 { 975 dmu_buf_rele(ds->ds_dbuf, tag); 976 } 977 978 void 979 dsl_dataset_remove_key_mapping(dsl_dataset_t *ds) 980 { 981 dsl_dir_t *dd = ds->ds_dir; 982 983 if (dd == NULL || dd->dd_crypto_obj == 0) 984 return; 985 986 (void) spa_keystore_remove_mapping(dd->dd_pool->dp_spa, 987 ds->ds_object, ds); 988 } 989 990 void 991 dsl_dataset_rele_flags(dsl_dataset_t *ds, ds_hold_flags_t flags, 992 const void *tag) 993 { 994 if (flags & DS_HOLD_FLAG_DECRYPT) 995 dsl_dataset_remove_key_mapping(ds); 996 997 dsl_dataset_rele(ds, tag); 998 } 999 1000 void 1001 dsl_dataset_disown(dsl_dataset_t *ds, ds_hold_flags_t flags, const void *tag) 1002 { 1003 ASSERT3P(ds->ds_owner, ==, tag); 1004 ASSERT(ds->ds_dbuf != NULL); 1005 1006 mutex_enter(&ds->ds_lock); 1007 ds->ds_owner = NULL; 1008 mutex_exit(&ds->ds_lock); 1009 dsl_dataset_long_rele(ds, tag); 1010 dsl_dataset_rele_flags(ds, flags, tag); 1011 } 1012 1013 boolean_t 1014 dsl_dataset_tryown(dsl_dataset_t *ds, const void *tag, boolean_t override) 1015 { 1016 boolean_t gotit = FALSE; 1017 1018 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1019 mutex_enter(&ds->ds_lock); 1020 if (ds->ds_owner == NULL && (override || !(DS_IS_INCONSISTENT(ds) || 1021 (dsl_dataset_feature_is_active(ds, 1022 SPA_FEATURE_REDACTED_DATASETS) && 1023 !zfs_allow_redacted_dataset_mount)))) { 1024 ds->ds_owner = tag; 1025 dsl_dataset_long_hold(ds, tag); 1026 gotit = TRUE; 1027 } 1028 mutex_exit(&ds->ds_lock); 1029 return (gotit); 1030 } 1031 1032 boolean_t 1033 dsl_dataset_has_owner(dsl_dataset_t *ds) 1034 { 1035 boolean_t rv; 1036 mutex_enter(&ds->ds_lock); 1037 rv = (ds->ds_owner != NULL); 1038 mutex_exit(&ds->ds_lock); 1039 return (rv); 1040 } 1041 1042 static boolean_t 1043 zfeature_active(spa_feature_t f, void *arg) 1044 { 1045 switch (spa_feature_table[f].fi_type) { 1046 case ZFEATURE_TYPE_BOOLEAN: { 1047 boolean_t val = (boolean_t)(uintptr_t)arg; 1048 ASSERT(val == B_FALSE || val == B_TRUE); 1049 return (val); 1050 } 1051 case ZFEATURE_TYPE_UINT64_ARRAY: 1052 /* 1053 * In this case, arg is a uint64_t array. The feature is active 1054 * if the array is non-null. 1055 */ 1056 return (arg != NULL); 1057 default: 1058 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); 1059 return (B_FALSE); 1060 } 1061 } 1062 1063 boolean_t 1064 dsl_dataset_feature_is_active(dsl_dataset_t *ds, spa_feature_t f) 1065 { 1066 return (zfeature_active(f, ds->ds_feature[f])); 1067 } 1068 1069 /* 1070 * The buffers passed out by this function are references to internal buffers; 1071 * they should not be freed by callers of this function, and they should not be 1072 * used after the dataset has been released. 1073 */ 1074 boolean_t 1075 dsl_dataset_get_uint64_array_feature(dsl_dataset_t *ds, spa_feature_t f, 1076 uint64_t *outlength, uint64_t **outp) 1077 { 1078 VERIFY(spa_feature_table[f].fi_type & ZFEATURE_TYPE_UINT64_ARRAY); 1079 if (!dsl_dataset_feature_is_active(ds, f)) { 1080 return (B_FALSE); 1081 } 1082 struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f]; 1083 *outp = ftuaa->array; 1084 *outlength = ftuaa->length; 1085 return (B_TRUE); 1086 } 1087 1088 void 1089 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, void *arg, 1090 dmu_tx_t *tx) 1091 { 1092 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1093 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 1094 uint64_t zero = 0; 1095 1096 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 1097 1098 spa_feature_incr(spa, f, tx); 1099 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 1100 1101 switch (spa_feature_table[f].fi_type) { 1102 case ZFEATURE_TYPE_BOOLEAN: 1103 ASSERT3S((boolean_t)(uintptr_t)arg, ==, B_TRUE); 1104 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, 1105 sizeof (zero), 1, &zero, tx)); 1106 break; 1107 case ZFEATURE_TYPE_UINT64_ARRAY: 1108 { 1109 struct feature_type_uint64_array_arg *ftuaa = arg; 1110 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, 1111 sizeof (uint64_t), ftuaa->length, ftuaa->array, tx)); 1112 break; 1113 } 1114 default: 1115 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); 1116 } 1117 } 1118 1119 static void 1120 dsl_dataset_deactivate_feature_impl(dsl_dataset_t *ds, spa_feature_t f, 1121 dmu_tx_t *tx) 1122 { 1123 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1124 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 1125 uint64_t dsobj = ds->ds_object; 1126 1127 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 1128 1129 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx)); 1130 spa_feature_decr(spa, f, tx); 1131 ds->ds_feature[f] = NULL; 1132 } 1133 1134 void 1135 dsl_dataset_deactivate_feature(dsl_dataset_t *ds, spa_feature_t f, dmu_tx_t *tx) 1136 { 1137 unload_zfeature(ds, f); 1138 dsl_dataset_deactivate_feature_impl(ds, f, tx); 1139 } 1140 1141 uint64_t 1142 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 1143 dsl_crypto_params_t *dcp, uint64_t flags, dmu_tx_t *tx) 1144 { 1145 dsl_pool_t *dp = dd->dd_pool; 1146 dmu_buf_t *dbuf; 1147 dsl_dataset_phys_t *dsphys; 1148 uint64_t dsobj; 1149 objset_t *mos = dp->dp_meta_objset; 1150 1151 if (origin == NULL) 1152 origin = dp->dp_origin_snap; 1153 1154 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 1155 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); 1156 ASSERT(dmu_tx_is_syncing(tx)); 1157 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 1158 1159 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1160 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1161 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1162 dmu_buf_will_dirty(dbuf, tx); 1163 dsphys = dbuf->db_data; 1164 memset(dsphys, 0, sizeof (dsl_dataset_phys_t)); 1165 dsphys->ds_dir_obj = dd->dd_object; 1166 dsphys->ds_flags = flags; 1167 dsphys->ds_fsid_guid = unique_create(); 1168 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1169 sizeof (dsphys->ds_guid)); 1170 dsphys->ds_snapnames_zapobj = 1171 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 1172 DMU_OT_NONE, 0, tx); 1173 dsphys->ds_creation_time = gethrestime_sec(); 1174 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 1175 1176 if (origin == NULL) { 1177 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 1178 } else { 1179 dsl_dataset_t *ohds; /* head of the origin snapshot */ 1180 1181 dsphys->ds_prev_snap_obj = origin->ds_object; 1182 dsphys->ds_prev_snap_txg = 1183 dsl_dataset_phys(origin)->ds_creation_txg; 1184 dsphys->ds_referenced_bytes = 1185 dsl_dataset_phys(origin)->ds_referenced_bytes; 1186 dsphys->ds_compressed_bytes = 1187 dsl_dataset_phys(origin)->ds_compressed_bytes; 1188 dsphys->ds_uncompressed_bytes = 1189 dsl_dataset_phys(origin)->ds_uncompressed_bytes; 1190 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG); 1191 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; 1192 rrw_exit(&origin->ds_bp_rwlock, FTAG); 1193 1194 /* 1195 * Inherit flags that describe the dataset's contents 1196 * (INCONSISTENT) or properties (Case Insensitive). 1197 */ 1198 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & 1199 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 1200 1201 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1202 if (zfeature_active(f, origin->ds_feature[f])) { 1203 dsl_dataset_activate_feature(dsobj, f, 1204 origin->ds_feature[f], tx); 1205 } 1206 } 1207 1208 dmu_buf_will_dirty(origin->ds_dbuf, tx); 1209 dsl_dataset_phys(origin)->ds_num_children++; 1210 1211 VERIFY0(dsl_dataset_hold_obj(dp, 1212 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, 1213 FTAG, &ohds)); 1214 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 1215 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 1216 dsl_dataset_rele(ohds, FTAG); 1217 1218 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 1219 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { 1220 dsl_dataset_phys(origin)->ds_next_clones_obj = 1221 zap_create(mos, 1222 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 1223 } 1224 VERIFY0(zap_add_int(mos, 1225 dsl_dataset_phys(origin)->ds_next_clones_obj, 1226 dsobj, tx)); 1227 } 1228 1229 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1230 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; 1231 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 1232 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 1233 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 1234 dsl_dir_phys(origin->ds_dir)->dd_clones = 1235 zap_create(mos, 1236 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 1237 } 1238 VERIFY0(zap_add_int(mos, 1239 dsl_dir_phys(origin->ds_dir)->dd_clones, 1240 dsobj, tx)); 1241 } 1242 } 1243 1244 /* handle encryption */ 1245 dsl_dataset_create_crypt_sync(dsobj, dd, origin, dcp, tx); 1246 1247 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1248 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1249 1250 dmu_buf_rele(dbuf, FTAG); 1251 1252 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1253 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; 1254 1255 return (dsobj); 1256 } 1257 1258 static void 1259 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 1260 { 1261 objset_t *os; 1262 1263 VERIFY0(dmu_objset_from_ds(ds, &os)); 1264 if (memcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) { 1265 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1266 zio_t *zio; 1267 1268 memset(&os->os_zil_header, 0, sizeof (os->os_zil_header)); 1269 if (os->os_encrypted) 1270 os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_TRUE; 1271 1272 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1273 dsl_dataset_sync(ds, zio, tx); 1274 VERIFY0(zio_wait(zio)); 1275 1276 /* dsl_dataset_sync_done will drop this reference. */ 1277 dmu_buf_add_ref(ds->ds_dbuf, ds); 1278 dsl_dataset_sync_done(ds, tx); 1279 } 1280 } 1281 1282 uint64_t 1283 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 1284 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, 1285 dsl_crypto_params_t *dcp, dmu_tx_t *tx) 1286 { 1287 dsl_pool_t *dp = pdd->dd_pool; 1288 uint64_t dsobj, ddobj; 1289 dsl_dir_t *dd; 1290 1291 ASSERT(dmu_tx_is_syncing(tx)); 1292 ASSERT(lastname[0] != '@'); 1293 /* 1294 * Filesystems will eventually have their origin set to dp_origin_snap, 1295 * but that's taken care of in dsl_dataset_create_sync_dd. When 1296 * creating a filesystem, this function is called with origin equal to 1297 * NULL. 1298 */ 1299 if (origin != NULL) 1300 ASSERT3P(origin, !=, dp->dp_origin_snap); 1301 1302 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 1303 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 1304 1305 dsobj = dsl_dataset_create_sync_dd(dd, origin, dcp, 1306 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 1307 1308 dsl_deleg_set_create_perms(dd, tx, cr); 1309 1310 /* 1311 * If we are creating a clone and the livelist feature is enabled, 1312 * add the entry DD_FIELD_LIVELIST to ZAP. 1313 */ 1314 if (origin != NULL && 1315 spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LIVELIST)) { 1316 objset_t *mos = dd->dd_pool->dp_meta_objset; 1317 dsl_dir_zapify(dd, tx); 1318 uint64_t obj = dsl_deadlist_alloc(mos, tx); 1319 VERIFY0(zap_add(mos, dd->dd_object, DD_FIELD_LIVELIST, 1320 sizeof (uint64_t), 1, &obj, tx)); 1321 spa_feature_incr(dp->dp_spa, SPA_FEATURE_LIVELIST, tx); 1322 } 1323 1324 /* 1325 * Since we're creating a new node we know it's a leaf, so we can 1326 * initialize the counts if the limit feature is active. 1327 */ 1328 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 1329 uint64_t cnt = 0; 1330 objset_t *os = dd->dd_pool->dp_meta_objset; 1331 1332 dsl_dir_zapify(dd, tx); 1333 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 1334 sizeof (cnt), 1, &cnt, tx)); 1335 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 1336 sizeof (cnt), 1, &cnt, tx)); 1337 } 1338 1339 dsl_dir_rele(dd, FTAG); 1340 1341 /* 1342 * If we are creating a clone, make sure we zero out any stale 1343 * data from the origin snapshots zil header. 1344 */ 1345 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 1346 dsl_dataset_t *ds; 1347 1348 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 1349 dsl_dataset_zero_zil(ds, tx); 1350 dsl_dataset_rele(ds, FTAG); 1351 } 1352 1353 return (dsobj); 1354 } 1355 1356 /* 1357 * The unique space in the head dataset can be calculated by subtracting 1358 * the space used in the most recent snapshot, that is still being used 1359 * in this file system, from the space currently in use. To figure out 1360 * the space in the most recent snapshot still in use, we need to take 1361 * the total space used in the snapshot and subtract out the space that 1362 * has been freed up since the snapshot was taken. 1363 */ 1364 void 1365 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 1366 { 1367 uint64_t mrs_used; 1368 uint64_t dlused, dlcomp, dluncomp; 1369 1370 ASSERT(!ds->ds_is_snapshot); 1371 1372 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) 1373 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; 1374 else 1375 mrs_used = 0; 1376 1377 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 1378 1379 ASSERT3U(dlused, <=, mrs_used); 1380 dsl_dataset_phys(ds)->ds_unique_bytes = 1381 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); 1382 1383 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1384 SPA_VERSION_UNIQUE_ACCURATE) 1385 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1386 } 1387 1388 void 1389 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 1390 dmu_tx_t *tx) 1391 { 1392 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1393 uint64_t count __maybe_unused; 1394 int err; 1395 1396 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); 1397 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1398 obj, tx); 1399 /* 1400 * The err should not be ENOENT, but a bug in a previous version 1401 * of the code could cause upgrade_clones_cb() to not set 1402 * ds_next_snap_obj when it should, leading to a missing entry. 1403 * If we knew that the pool was created after 1404 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 1405 * ENOENT. However, at least we can check that we don't have 1406 * too many entries in the next_clones_obj even after failing to 1407 * remove this one. 1408 */ 1409 if (err != ENOENT) 1410 VERIFY0(err); 1411 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1412 &count)); 1413 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); 1414 } 1415 1416 1417 blkptr_t * 1418 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1419 { 1420 return (&dsl_dataset_phys(ds)->ds_bp); 1421 } 1422 1423 spa_t * 1424 dsl_dataset_get_spa(dsl_dataset_t *ds) 1425 { 1426 return (ds->ds_dir->dd_pool->dp_spa); 1427 } 1428 1429 void 1430 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1431 { 1432 dsl_pool_t *dp; 1433 1434 if (ds == NULL) /* this is the meta-objset */ 1435 return; 1436 1437 ASSERT(ds->ds_objset != NULL); 1438 1439 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) 1440 panic("dirtying snapshot!"); 1441 1442 /* Must not dirty a dataset in the same txg where it got snapshotted. */ 1443 ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 1444 1445 dp = ds->ds_dir->dd_pool; 1446 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1447 objset_t *os = ds->ds_objset; 1448 1449 /* up the hold count until we can be written out */ 1450 dmu_buf_add_ref(ds->ds_dbuf, ds); 1451 1452 /* if this dataset is encrypted, grab a reference to the DCK */ 1453 if (ds->ds_dir->dd_crypto_obj != 0 && 1454 !os->os_raw_receive && 1455 !os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { 1456 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1457 key_mapping_add_ref(ds->ds_key_mapping, ds); 1458 } 1459 } 1460 } 1461 1462 static int 1463 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1464 { 1465 uint64_t asize; 1466 1467 if (!dmu_tx_is_syncing(tx)) 1468 return (0); 1469 1470 /* 1471 * If there's an fs-only reservation, any blocks that might become 1472 * owned by the snapshot dataset must be accommodated by space 1473 * outside of the reservation. 1474 */ 1475 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1476 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); 1477 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1478 return (SET_ERROR(ENOSPC)); 1479 1480 /* 1481 * Propagate any reserved space for this snapshot to other 1482 * snapshot checks in this sync group. 1483 */ 1484 if (asize > 0) 1485 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1486 1487 return (0); 1488 } 1489 1490 int 1491 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1492 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr, proc_t *proc) 1493 { 1494 int error; 1495 uint64_t value; 1496 1497 ds->ds_trysnap_txg = tx->tx_txg; 1498 1499 if (!dmu_tx_is_syncing(tx)) 1500 return (0); 1501 1502 /* 1503 * We don't allow multiple snapshots of the same txg. If there 1504 * is already one, try again. 1505 */ 1506 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) 1507 return (SET_ERROR(EAGAIN)); 1508 1509 /* 1510 * Check for conflicting snapshot name. 1511 */ 1512 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1513 if (error == 0) 1514 return (SET_ERROR(EEXIST)); 1515 if (error != ENOENT) 1516 return (error); 1517 1518 /* 1519 * We don't allow taking snapshots of inconsistent datasets, such as 1520 * those into which we are currently receiving. However, if we are 1521 * creating this snapshot as part of a receive, this check will be 1522 * executed atomically with respect to the completion of the receive 1523 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1524 * case we ignore this, knowing it will be fixed up for us shortly in 1525 * dmu_recv_end_sync(). 1526 */ 1527 if (!recv && DS_IS_INCONSISTENT(ds)) 1528 return (SET_ERROR(EBUSY)); 1529 1530 /* 1531 * Skip the check for temporary snapshots or if we have already checked 1532 * the counts in dsl_dataset_snapshot_check. This means we really only 1533 * check the count here when we're receiving a stream. 1534 */ 1535 if (cnt != 0 && cr != NULL) { 1536 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1537 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr, proc); 1538 if (error != 0) 1539 return (error); 1540 } 1541 1542 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1543 if (error != 0) 1544 return (error); 1545 1546 return (0); 1547 } 1548 1549 int 1550 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1551 { 1552 dsl_dataset_snapshot_arg_t *ddsa = arg; 1553 dsl_pool_t *dp = dmu_tx_pool(tx); 1554 nvpair_t *pair; 1555 int rv = 0; 1556 1557 /* 1558 * Pre-compute how many total new snapshots will be created for each 1559 * level in the tree and below. This is needed for validating the 1560 * snapshot limit when either taking a recursive snapshot or when 1561 * taking multiple snapshots. 1562 * 1563 * The problem is that the counts are not actually adjusted when 1564 * we are checking, only when we finally sync. For a single snapshot, 1565 * this is easy, the count will increase by 1 at each node up the tree, 1566 * but its more complicated for the recursive/multiple snapshot case. 1567 * 1568 * The dsl_fs_ss_limit_check function does recursively check the count 1569 * at each level up the tree but since it is validating each snapshot 1570 * independently we need to be sure that we are validating the complete 1571 * count for the entire set of snapshots. We do this by rolling up the 1572 * counts for each component of the name into an nvlist and then 1573 * checking each of those cases with the aggregated count. 1574 * 1575 * This approach properly handles not only the recursive snapshot 1576 * case (where we get all of those on the ddsa_snaps list) but also 1577 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1578 * validate the limit on 'a' using a count of 2). 1579 * 1580 * We validate the snapshot names in the third loop and only report 1581 * name errors once. 1582 */ 1583 if (dmu_tx_is_syncing(tx)) { 1584 char *nm; 1585 nvlist_t *cnt_track = NULL; 1586 cnt_track = fnvlist_alloc(); 1587 1588 nm = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1589 1590 /* Rollup aggregated counts into the cnt_track list */ 1591 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1592 pair != NULL; 1593 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1594 char *pdelim; 1595 uint64_t val; 1596 1597 (void) strlcpy(nm, nvpair_name(pair), MAXPATHLEN); 1598 pdelim = strchr(nm, '@'); 1599 if (pdelim == NULL) 1600 continue; 1601 *pdelim = '\0'; 1602 1603 do { 1604 if (nvlist_lookup_uint64(cnt_track, nm, 1605 &val) == 0) { 1606 /* update existing entry */ 1607 fnvlist_add_uint64(cnt_track, nm, 1608 val + 1); 1609 } else { 1610 /* add to list */ 1611 fnvlist_add_uint64(cnt_track, nm, 1); 1612 } 1613 1614 pdelim = strrchr(nm, '/'); 1615 if (pdelim != NULL) 1616 *pdelim = '\0'; 1617 } while (pdelim != NULL); 1618 } 1619 1620 kmem_free(nm, MAXPATHLEN); 1621 1622 /* Check aggregated counts at each level */ 1623 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1624 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1625 int error = 0; 1626 char *name; 1627 uint64_t cnt = 0; 1628 dsl_dataset_t *ds; 1629 1630 name = nvpair_name(pair); 1631 cnt = fnvpair_value_uint64(pair); 1632 ASSERT(cnt > 0); 1633 1634 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1635 if (error == 0) { 1636 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1637 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1638 ddsa->ddsa_cr, ddsa->ddsa_proc); 1639 dsl_dataset_rele(ds, FTAG); 1640 } 1641 1642 if (error != 0) { 1643 if (ddsa->ddsa_errors != NULL) 1644 fnvlist_add_int32(ddsa->ddsa_errors, 1645 name, error); 1646 rv = error; 1647 /* only report one error for this check */ 1648 break; 1649 } 1650 } 1651 nvlist_free(cnt_track); 1652 } 1653 1654 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1655 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1656 int error = 0; 1657 dsl_dataset_t *ds; 1658 char *name, *atp = NULL; 1659 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1660 1661 name = nvpair_name(pair); 1662 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN) 1663 error = SET_ERROR(ENAMETOOLONG); 1664 if (error == 0) { 1665 atp = strchr(name, '@'); 1666 if (atp == NULL) 1667 error = SET_ERROR(EINVAL); 1668 if (error == 0) 1669 (void) strlcpy(dsname, name, atp - name + 1); 1670 } 1671 if (error == 0) 1672 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1673 if (error == 0) { 1674 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1675 error = dsl_dataset_snapshot_check_impl(ds, 1676 atp + 1, tx, B_FALSE, 0, NULL, NULL); 1677 dsl_dataset_rele(ds, FTAG); 1678 } 1679 1680 if (error != 0) { 1681 if (ddsa->ddsa_errors != NULL) { 1682 fnvlist_add_int32(ddsa->ddsa_errors, 1683 name, error); 1684 } 1685 rv = error; 1686 } 1687 } 1688 1689 return (rv); 1690 } 1691 1692 void 1693 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1694 dmu_tx_t *tx) 1695 { 1696 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1697 dmu_buf_t *dbuf; 1698 dsl_dataset_phys_t *dsphys; 1699 uint64_t dsobj, crtxg; 1700 objset_t *mos = dp->dp_meta_objset; 1701 static zil_header_t zero_zil __maybe_unused; 1702 objset_t *os __maybe_unused; 1703 1704 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1705 1706 /* 1707 * If we are on an old pool, the zil must not be active, in which 1708 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1709 */ 1710 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1711 dmu_objset_from_ds(ds, &os) != 0 || 1712 memcmp(&os->os_phys->os_zil_header, &zero_zil, 1713 sizeof (zero_zil)) == 0); 1714 1715 /* Should not snapshot a dirty dataset. */ 1716 ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1717 ds, tx->tx_txg)); 1718 1719 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1720 1721 /* 1722 * The origin's ds_creation_txg has to be < TXG_INITIAL 1723 */ 1724 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1725 crtxg = 1; 1726 else 1727 crtxg = tx->tx_txg; 1728 1729 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1730 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1731 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1732 dmu_buf_will_dirty(dbuf, tx); 1733 dsphys = dbuf->db_data; 1734 memset(dsphys, 0, sizeof (dsl_dataset_phys_t)); 1735 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1736 dsphys->ds_fsid_guid = unique_create(); 1737 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1738 sizeof (dsphys->ds_guid)); 1739 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1740 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 1741 dsphys->ds_next_snap_obj = ds->ds_object; 1742 dsphys->ds_num_children = 1; 1743 dsphys->ds_creation_time = gethrestime_sec(); 1744 dsphys->ds_creation_txg = crtxg; 1745 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; 1746 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; 1747 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; 1748 dsphys->ds_uncompressed_bytes = 1749 dsl_dataset_phys(ds)->ds_uncompressed_bytes; 1750 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; 1751 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1752 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; 1753 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1754 dmu_buf_rele(dbuf, FTAG); 1755 1756 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1757 if (zfeature_active(f, ds->ds_feature[f])) { 1758 dsl_dataset_activate_feature(dsobj, f, 1759 ds->ds_feature[f], tx); 1760 } 1761 } 1762 1763 ASSERT3U(ds->ds_prev != 0, ==, 1764 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); 1765 if (ds->ds_prev) { 1766 uint64_t next_clones_obj = 1767 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; 1768 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1769 ds->ds_object || 1770 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); 1771 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1772 ds->ds_object) { 1773 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1774 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, 1775 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); 1776 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; 1777 } else if (next_clones_obj != 0) { 1778 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1779 dsphys->ds_next_snap_obj, tx); 1780 VERIFY0(zap_add_int(mos, 1781 next_clones_obj, dsobj, tx)); 1782 } 1783 } 1784 1785 /* 1786 * If we have a reference-reservation on this dataset, we will 1787 * need to increase the amount of refreservation being charged 1788 * since our unique space is going to zero. 1789 */ 1790 if (ds->ds_reserved) { 1791 int64_t delta; 1792 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1793 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, 1794 ds->ds_reserved); 1795 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1796 delta, 0, 0, tx); 1797 } 1798 1799 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1800 dsl_dataset_phys(ds)->ds_deadlist_obj = 1801 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, 1802 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 1803 dsl_deadlist_close(&ds->ds_deadlist); 1804 dsl_deadlist_open(&ds->ds_deadlist, mos, 1805 dsl_dataset_phys(ds)->ds_deadlist_obj); 1806 dsl_deadlist_add_key(&ds->ds_deadlist, 1807 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); 1808 dsl_bookmark_snapshotted(ds, tx); 1809 1810 if (dsl_dataset_remap_deadlist_exists(ds)) { 1811 uint64_t remap_deadlist_obj = 1812 dsl_dataset_get_remap_deadlist_object(ds); 1813 /* 1814 * Move the remap_deadlist to the snapshot. The head 1815 * will create a new remap deadlist on demand, from 1816 * dsl_dataset_block_remapped(). 1817 */ 1818 dsl_dataset_unset_remap_deadlist_object(ds, tx); 1819 dsl_deadlist_close(&ds->ds_remap_deadlist); 1820 1821 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 1822 VERIFY0(zap_add(mos, dsobj, DS_FIELD_REMAP_DEADLIST, 1823 sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj, tx)); 1824 } 1825 1826 /* 1827 * Create a ivset guid for this snapshot if the dataset is 1828 * encrypted. This may be overridden by a raw receive. A 1829 * previous implementation of this code did not have this 1830 * field as part of the on-disk format for ZFS encryption 1831 * (see errata #4). As part of the remediation for this 1832 * issue, we ask the user to enable the bookmark_v2 feature 1833 * which is now a dependency of the encryption feature. We 1834 * use this as a heuristic to determine when the user has 1835 * elected to correct any datasets created with the old code. 1836 * As a result, we only do this step if the bookmark_v2 1837 * feature is enabled, which limits the number of states a 1838 * given pool / dataset can be in with regards to terms of 1839 * correcting the issue. 1840 */ 1841 if (ds->ds_dir->dd_crypto_obj != 0 && 1842 spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARK_V2)) { 1843 uint64_t ivset_guid = unique_create(); 1844 1845 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 1846 VERIFY0(zap_add(mos, dsobj, DS_FIELD_IVSET_GUID, 1847 sizeof (ivset_guid), 1, &ivset_guid, tx)); 1848 } 1849 1850 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); 1851 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; 1852 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; 1853 dsl_dataset_phys(ds)->ds_unique_bytes = 0; 1854 1855 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1856 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1857 1858 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1859 snapname, 8, 1, &dsobj, tx)); 1860 1861 if (ds->ds_prev) 1862 dsl_dataset_rele(ds->ds_prev, ds); 1863 VERIFY0(dsl_dataset_hold_obj(dp, 1864 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); 1865 1866 dsl_scan_ds_snapshotted(ds, tx); 1867 1868 dsl_dir_snap_cmtime_update(ds->ds_dir, tx); 1869 1870 if (zfs_snapshot_history_enabled) 1871 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, " "); 1872 } 1873 1874 void 1875 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1876 { 1877 dsl_dataset_snapshot_arg_t *ddsa = arg; 1878 dsl_pool_t *dp = dmu_tx_pool(tx); 1879 nvpair_t *pair; 1880 1881 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1882 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1883 dsl_dataset_t *ds; 1884 char *name, *atp; 1885 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1886 1887 name = nvpair_name(pair); 1888 atp = strchr(name, '@'); 1889 (void) strlcpy(dsname, name, atp - name + 1); 1890 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1891 1892 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1893 if (ddsa->ddsa_props != NULL) { 1894 dsl_props_set_sync_impl(ds->ds_prev, 1895 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1896 } 1897 dsl_dataset_rele(ds, FTAG); 1898 } 1899 } 1900 1901 /* 1902 * The snapshots must all be in the same pool. 1903 * All-or-nothing: if there are any failures, nothing will be modified. 1904 */ 1905 int 1906 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1907 { 1908 dsl_dataset_snapshot_arg_t ddsa; 1909 nvpair_t *pair; 1910 boolean_t needsuspend; 1911 int error; 1912 spa_t *spa; 1913 char *firstname; 1914 nvlist_t *suspended = NULL; 1915 1916 pair = nvlist_next_nvpair(snaps, NULL); 1917 if (pair == NULL) 1918 return (0); 1919 firstname = nvpair_name(pair); 1920 1921 error = spa_open(firstname, &spa, FTAG); 1922 if (error != 0) 1923 return (error); 1924 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1925 spa_close(spa, FTAG); 1926 1927 if (needsuspend) { 1928 suspended = fnvlist_alloc(); 1929 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1930 pair = nvlist_next_nvpair(snaps, pair)) { 1931 char fsname[ZFS_MAX_DATASET_NAME_LEN]; 1932 char *snapname = nvpair_name(pair); 1933 char *atp; 1934 void *cookie; 1935 1936 atp = strchr(snapname, '@'); 1937 if (atp == NULL) { 1938 error = SET_ERROR(EINVAL); 1939 break; 1940 } 1941 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1942 1943 error = zil_suspend(fsname, &cookie); 1944 if (error != 0) 1945 break; 1946 fnvlist_add_uint64(suspended, fsname, 1947 (uintptr_t)cookie); 1948 } 1949 } 1950 1951 ddsa.ddsa_snaps = snaps; 1952 ddsa.ddsa_props = props; 1953 ddsa.ddsa_errors = errors; 1954 ddsa.ddsa_cr = CRED(); 1955 ddsa.ddsa_proc = curproc; 1956 1957 if (error == 0) { 1958 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1959 dsl_dataset_snapshot_sync, &ddsa, 1960 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1961 } 1962 1963 if (suspended != NULL) { 1964 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1965 pair = nvlist_next_nvpair(suspended, pair)) { 1966 zil_resume((void *)(uintptr_t) 1967 fnvpair_value_uint64(pair)); 1968 } 1969 fnvlist_free(suspended); 1970 } 1971 1972 if (error == 0) { 1973 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1974 pair = nvlist_next_nvpair(snaps, pair)) { 1975 zvol_create_minor(nvpair_name(pair)); 1976 } 1977 } 1978 1979 return (error); 1980 } 1981 1982 typedef struct dsl_dataset_snapshot_tmp_arg { 1983 const char *ddsta_fsname; 1984 const char *ddsta_snapname; 1985 minor_t ddsta_cleanup_minor; 1986 const char *ddsta_htag; 1987 } dsl_dataset_snapshot_tmp_arg_t; 1988 1989 static int 1990 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1991 { 1992 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1993 dsl_pool_t *dp = dmu_tx_pool(tx); 1994 dsl_dataset_t *ds; 1995 int error; 1996 1997 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1998 if (error != 0) 1999 return (error); 2000 2001 /* NULL cred means no limit check for tmp snapshot */ 2002 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 2003 tx, B_FALSE, 0, NULL, NULL); 2004 if (error != 0) { 2005 dsl_dataset_rele(ds, FTAG); 2006 return (error); 2007 } 2008 2009 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 2010 dsl_dataset_rele(ds, FTAG); 2011 return (SET_ERROR(ENOTSUP)); 2012 } 2013 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 2014 B_TRUE, tx); 2015 if (error != 0) { 2016 dsl_dataset_rele(ds, FTAG); 2017 return (error); 2018 } 2019 2020 dsl_dataset_rele(ds, FTAG); 2021 return (0); 2022 } 2023 2024 static void 2025 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 2026 { 2027 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 2028 dsl_pool_t *dp = dmu_tx_pool(tx); 2029 dsl_dataset_t *ds = NULL; 2030 2031 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 2032 2033 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 2034 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 2035 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 2036 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 2037 2038 dsl_dataset_rele(ds, FTAG); 2039 } 2040 2041 int 2042 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 2043 minor_t cleanup_minor, const char *htag) 2044 { 2045 dsl_dataset_snapshot_tmp_arg_t ddsta; 2046 int error; 2047 spa_t *spa; 2048 boolean_t needsuspend; 2049 void *cookie; 2050 2051 ddsta.ddsta_fsname = fsname; 2052 ddsta.ddsta_snapname = snapname; 2053 ddsta.ddsta_cleanup_minor = cleanup_minor; 2054 ddsta.ddsta_htag = htag; 2055 2056 error = spa_open(fsname, &spa, FTAG); 2057 if (error != 0) 2058 return (error); 2059 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 2060 spa_close(spa, FTAG); 2061 2062 if (needsuspend) { 2063 error = zil_suspend(fsname, &cookie); 2064 if (error != 0) 2065 return (error); 2066 } 2067 2068 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 2069 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 2070 2071 if (needsuspend) 2072 zil_resume(cookie); 2073 return (error); 2074 } 2075 2076 void 2077 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 2078 { 2079 ASSERT(dmu_tx_is_syncing(tx)); 2080 ASSERT(ds->ds_objset != NULL); 2081 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); 2082 2083 /* 2084 * in case we had to change ds_fsid_guid when we opened it, 2085 * sync it out now. 2086 */ 2087 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2088 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; 2089 2090 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) { 2091 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 2092 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1, 2093 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx)); 2094 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 2095 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1, 2096 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx)); 2097 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 2098 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1, 2099 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx)); 2100 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0; 2101 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0; 2102 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0; 2103 } 2104 2105 dmu_objset_sync(ds->ds_objset, zio, tx); 2106 2107 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 2108 if (zfeature_active(f, ds->ds_feature_activation[f])) { 2109 if (zfeature_active(f, ds->ds_feature[f])) 2110 continue; 2111 dsl_dataset_activate_feature(ds->ds_object, f, 2112 ds->ds_feature_activation[f], tx); 2113 ds->ds_feature[f] = ds->ds_feature_activation[f]; 2114 } 2115 } 2116 } 2117 2118 /* 2119 * Check if the percentage of blocks shared between the clone and the 2120 * snapshot (as opposed to those that are clone only) is below a certain 2121 * threshold 2122 */ 2123 static boolean_t 2124 dsl_livelist_should_disable(dsl_dataset_t *ds) 2125 { 2126 uint64_t used, referenced; 2127 int percent_shared; 2128 2129 used = dsl_dir_get_usedds(ds->ds_dir); 2130 referenced = dsl_get_referenced(ds); 2131 if (referenced == 0) 2132 return (B_FALSE); 2133 percent_shared = (100 * (referenced - used)) / referenced; 2134 if (percent_shared <= zfs_livelist_min_percent_shared) 2135 return (B_TRUE); 2136 return (B_FALSE); 2137 } 2138 2139 /* 2140 * Check if it is possible to combine two livelist entries into one. 2141 * This is the case if the combined number of 'live' blkptrs (ALLOCs that 2142 * don't have a matching FREE) is under the maximum sublist size. 2143 * We check this by subtracting twice the total number of frees from the total 2144 * number of blkptrs. FREEs are counted twice because each FREE blkptr 2145 * will cancel out an ALLOC blkptr when the livelist is processed. 2146 */ 2147 static boolean_t 2148 dsl_livelist_should_condense(dsl_deadlist_entry_t *first, 2149 dsl_deadlist_entry_t *next) 2150 { 2151 uint64_t total_free = first->dle_bpobj.bpo_phys->bpo_num_freed + 2152 next->dle_bpobj.bpo_phys->bpo_num_freed; 2153 uint64_t total_entries = first->dle_bpobj.bpo_phys->bpo_num_blkptrs + 2154 next->dle_bpobj.bpo_phys->bpo_num_blkptrs; 2155 if ((total_entries - (2 * total_free)) < zfs_livelist_max_entries) 2156 return (B_TRUE); 2157 return (B_FALSE); 2158 } 2159 2160 typedef struct try_condense_arg { 2161 spa_t *spa; 2162 dsl_dataset_t *ds; 2163 } try_condense_arg_t; 2164 2165 /* 2166 * Iterate over the livelist entries, searching for a pair to condense. 2167 * A nonzero return value means stop, 0 means keep looking. 2168 */ 2169 static int 2170 dsl_livelist_try_condense(void *arg, dsl_deadlist_entry_t *first) 2171 { 2172 try_condense_arg_t *tca = arg; 2173 spa_t *spa = tca->spa; 2174 dsl_dataset_t *ds = tca->ds; 2175 dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist; 2176 dsl_deadlist_entry_t *next; 2177 2178 /* The condense thread has not yet been created at import */ 2179 if (spa->spa_livelist_condense_zthr == NULL) 2180 return (1); 2181 2182 /* A condense is already in progress */ 2183 if (spa->spa_to_condense.ds != NULL) 2184 return (1); 2185 2186 next = AVL_NEXT(&ll->dl_tree, &first->dle_node); 2187 /* The livelist has only one entry - don't condense it */ 2188 if (next == NULL) 2189 return (1); 2190 2191 /* Next is the newest entry - don't condense it */ 2192 if (AVL_NEXT(&ll->dl_tree, &next->dle_node) == NULL) 2193 return (1); 2194 2195 /* This pair is not ready to condense but keep looking */ 2196 if (!dsl_livelist_should_condense(first, next)) 2197 return (0); 2198 2199 /* 2200 * Add a ref to prevent the dataset from being evicted while 2201 * the condense zthr or synctask are running. Ref will be 2202 * released at the end of the condense synctask 2203 */ 2204 dmu_buf_add_ref(ds->ds_dbuf, spa); 2205 2206 spa->spa_to_condense.ds = ds; 2207 spa->spa_to_condense.first = first; 2208 spa->spa_to_condense.next = next; 2209 spa->spa_to_condense.syncing = B_FALSE; 2210 spa->spa_to_condense.cancelled = B_FALSE; 2211 2212 zthr_wakeup(spa->spa_livelist_condense_zthr); 2213 return (1); 2214 } 2215 2216 static void 2217 dsl_flush_pending_livelist(dsl_dataset_t *ds, dmu_tx_t *tx) 2218 { 2219 dsl_dir_t *dd = ds->ds_dir; 2220 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 2221 dsl_deadlist_entry_t *last = dsl_deadlist_last(&dd->dd_livelist); 2222 2223 /* Check if we need to add a new sub-livelist */ 2224 if (last == NULL) { 2225 /* The livelist is empty */ 2226 dsl_deadlist_add_key(&dd->dd_livelist, 2227 tx->tx_txg - 1, tx); 2228 } else if (spa_sync_pass(spa) == 1) { 2229 /* 2230 * Check if the newest entry is full. If it is, make a new one. 2231 * We only do this once per sync because we could overfill a 2232 * sublist in one sync pass and don't want to add another entry 2233 * for a txg that is already represented. This ensures that 2234 * blkptrs born in the same txg are stored in the same sublist. 2235 */ 2236 bpobj_t bpobj = last->dle_bpobj; 2237 uint64_t all = bpobj.bpo_phys->bpo_num_blkptrs; 2238 uint64_t free = bpobj.bpo_phys->bpo_num_freed; 2239 uint64_t alloc = all - free; 2240 if (alloc > zfs_livelist_max_entries) { 2241 dsl_deadlist_add_key(&dd->dd_livelist, 2242 tx->tx_txg - 1, tx); 2243 } 2244 } 2245 2246 /* Insert each entry into the on-disk livelist */ 2247 bplist_iterate(&dd->dd_pending_allocs, 2248 dsl_deadlist_insert_alloc_cb, &dd->dd_livelist, tx); 2249 bplist_iterate(&dd->dd_pending_frees, 2250 dsl_deadlist_insert_free_cb, &dd->dd_livelist, tx); 2251 2252 /* Attempt to condense every pair of adjacent entries */ 2253 try_condense_arg_t arg = { 2254 .spa = spa, 2255 .ds = ds 2256 }; 2257 dsl_deadlist_iterate(&dd->dd_livelist, dsl_livelist_try_condense, 2258 &arg); 2259 } 2260 2261 void 2262 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx) 2263 { 2264 objset_t *os = ds->ds_objset; 2265 2266 bplist_iterate(&ds->ds_pending_deadlist, 2267 dsl_deadlist_insert_alloc_cb, &ds->ds_deadlist, tx); 2268 2269 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist)) { 2270 dsl_flush_pending_livelist(ds, tx); 2271 if (dsl_livelist_should_disable(ds)) { 2272 dsl_dir_remove_livelist(ds->ds_dir, tx, B_TRUE); 2273 } 2274 } 2275 2276 dsl_bookmark_sync_done(ds, tx); 2277 2278 multilist_destroy(&os->os_synced_dnodes); 2279 2280 if (os->os_encrypted) 2281 os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_FALSE; 2282 else 2283 ASSERT0(os->os_next_write_raw[tx->tx_txg & TXG_MASK]); 2284 2285 ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx))); 2286 2287 dmu_buf_rele(ds->ds_dbuf, ds); 2288 } 2289 2290 int 2291 get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val) 2292 { 2293 uint64_t count = 0; 2294 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 2295 zap_cursor_t zc; 2296 zap_attribute_t za; 2297 2298 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 2299 2300 /* 2301 * There may be missing entries in ds_next_clones_obj 2302 * due to a bug in a previous version of the code. 2303 * Only trust it if it has the right number of entries. 2304 */ 2305 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { 2306 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 2307 &count)); 2308 } 2309 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) { 2310 return (SET_ERROR(ENOENT)); 2311 } 2312 for (zap_cursor_init(&zc, mos, 2313 dsl_dataset_phys(ds)->ds_next_clones_obj); 2314 zap_cursor_retrieve(&zc, &za) == 0; 2315 zap_cursor_advance(&zc)) { 2316 dsl_dataset_t *clone; 2317 char buf[ZFS_MAX_DATASET_NAME_LEN]; 2318 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 2319 za.za_first_integer, FTAG, &clone)); 2320 dsl_dir_name(clone->ds_dir, buf); 2321 fnvlist_add_boolean(val, buf); 2322 dsl_dataset_rele(clone, FTAG); 2323 } 2324 zap_cursor_fini(&zc); 2325 return (0); 2326 } 2327 2328 void 2329 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 2330 { 2331 nvlist_t *propval = fnvlist_alloc(); 2332 nvlist_t *val = fnvlist_alloc(); 2333 2334 if (get_clones_stat_impl(ds, val) == 0) { 2335 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 2336 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), 2337 propval); 2338 } 2339 2340 nvlist_free(val); 2341 nvlist_free(propval); 2342 } 2343 2344 static char * 2345 get_receive_resume_token_impl(dsl_dataset_t *ds) 2346 { 2347 if (!dsl_dataset_has_resume_receive_state(ds)) 2348 return (NULL); 2349 2350 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2351 char *str; 2352 void *packed; 2353 uint8_t *compressed; 2354 uint64_t val; 2355 nvlist_t *token_nv = fnvlist_alloc(); 2356 size_t packed_size, compressed_size; 2357 2358 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2359 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) { 2360 fnvlist_add_uint64(token_nv, "fromguid", val); 2361 } 2362 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2363 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) { 2364 fnvlist_add_uint64(token_nv, "object", val); 2365 } 2366 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2367 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) { 2368 fnvlist_add_uint64(token_nv, "offset", val); 2369 } 2370 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2371 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) { 2372 fnvlist_add_uint64(token_nv, "bytes", val); 2373 } 2374 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2375 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) { 2376 fnvlist_add_uint64(token_nv, "toguid", val); 2377 } 2378 char buf[MAXNAMELEN]; 2379 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2380 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) { 2381 fnvlist_add_string(token_nv, "toname", buf); 2382 } 2383 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2384 DS_FIELD_RESUME_LARGEBLOCK) == 0) { 2385 fnvlist_add_boolean(token_nv, "largeblockok"); 2386 } 2387 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2388 DS_FIELD_RESUME_EMBEDOK) == 0) { 2389 fnvlist_add_boolean(token_nv, "embedok"); 2390 } 2391 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2392 DS_FIELD_RESUME_COMPRESSOK) == 0) { 2393 fnvlist_add_boolean(token_nv, "compressok"); 2394 } 2395 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2396 DS_FIELD_RESUME_RAWOK) == 0) { 2397 fnvlist_add_boolean(token_nv, "rawok"); 2398 } 2399 if (dsl_dataset_feature_is_active(ds, 2400 SPA_FEATURE_REDACTED_DATASETS)) { 2401 uint64_t num_redact_snaps = 0; 2402 uint64_t *redact_snaps = NULL; 2403 VERIFY3B(dsl_dataset_get_uint64_array_feature(ds, 2404 SPA_FEATURE_REDACTED_DATASETS, &num_redact_snaps, 2405 &redact_snaps), ==, B_TRUE); 2406 fnvlist_add_uint64_array(token_nv, "redact_snaps", 2407 redact_snaps, num_redact_snaps); 2408 } 2409 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2410 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS) == 0) { 2411 uint64_t num_redact_snaps = 0, int_size = 0; 2412 uint64_t *redact_snaps = NULL; 2413 VERIFY0(zap_length(dp->dp_meta_objset, ds->ds_object, 2414 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, &int_size, 2415 &num_redact_snaps)); 2416 ASSERT3U(int_size, ==, sizeof (uint64_t)); 2417 2418 redact_snaps = kmem_alloc(int_size * num_redact_snaps, 2419 KM_SLEEP); 2420 VERIFY0(zap_lookup(dp->dp_meta_objset, ds->ds_object, 2421 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, int_size, 2422 num_redact_snaps, redact_snaps)); 2423 fnvlist_add_uint64_array(token_nv, "book_redact_snaps", 2424 redact_snaps, num_redact_snaps); 2425 kmem_free(redact_snaps, int_size * num_redact_snaps); 2426 } 2427 packed = fnvlist_pack(token_nv, &packed_size); 2428 fnvlist_free(token_nv); 2429 compressed = kmem_alloc(packed_size, KM_SLEEP); 2430 2431 compressed_size = gzip_compress(packed, compressed, 2432 packed_size, packed_size, 6); 2433 2434 zio_cksum_t cksum; 2435 fletcher_4_native_varsize(compressed, compressed_size, &cksum); 2436 2437 size_t alloc_size = compressed_size * 2 + 1; 2438 str = kmem_alloc(alloc_size, KM_SLEEP); 2439 for (int i = 0; i < compressed_size; i++) { 2440 size_t offset = i * 2; 2441 (void) snprintf(str + offset, alloc_size - offset, 2442 "%02x", compressed[i]); 2443 } 2444 str[compressed_size * 2] = '\0'; 2445 char *propval = kmem_asprintf("%u-%llx-%llx-%s", 2446 ZFS_SEND_RESUME_TOKEN_VERSION, 2447 (longlong_t)cksum.zc_word[0], 2448 (longlong_t)packed_size, str); 2449 kmem_free(packed, packed_size); 2450 kmem_free(str, alloc_size); 2451 kmem_free(compressed, packed_size); 2452 return (propval); 2453 } 2454 2455 /* 2456 * Returns a string that represents the receive resume state token. It should 2457 * be freed with strfree(). NULL is returned if no resume state is present. 2458 */ 2459 char * 2460 get_receive_resume_token(dsl_dataset_t *ds) 2461 { 2462 /* 2463 * A failed "newfs" (e.g. full) resumable receive leaves 2464 * the stats set on this dataset. Check here for the prop. 2465 */ 2466 char *token = get_receive_resume_token_impl(ds); 2467 if (token != NULL) 2468 return (token); 2469 /* 2470 * A failed incremental resumable receive leaves the 2471 * stats set on our child named "%recv". Check the child 2472 * for the prop. 2473 */ 2474 /* 6 extra bytes for /%recv */ 2475 char name[ZFS_MAX_DATASET_NAME_LEN + 6]; 2476 dsl_dataset_t *recv_ds; 2477 dsl_dataset_name(ds, name); 2478 if (strlcat(name, "/", sizeof (name)) < sizeof (name) && 2479 strlcat(name, recv_clone_name, sizeof (name)) < sizeof (name) && 2480 dsl_dataset_hold(ds->ds_dir->dd_pool, name, FTAG, &recv_ds) == 0) { 2481 token = get_receive_resume_token_impl(recv_ds); 2482 dsl_dataset_rele(recv_ds, FTAG); 2483 } 2484 return (token); 2485 } 2486 2487 uint64_t 2488 dsl_get_refratio(dsl_dataset_t *ds) 2489 { 2490 uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : 2491 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / 2492 dsl_dataset_phys(ds)->ds_compressed_bytes); 2493 return (ratio); 2494 } 2495 2496 uint64_t 2497 dsl_get_logicalreferenced(dsl_dataset_t *ds) 2498 { 2499 return (dsl_dataset_phys(ds)->ds_uncompressed_bytes); 2500 } 2501 2502 uint64_t 2503 dsl_get_compressratio(dsl_dataset_t *ds) 2504 { 2505 if (ds->ds_is_snapshot) { 2506 return (dsl_get_refratio(ds)); 2507 } else { 2508 dsl_dir_t *dd = ds->ds_dir; 2509 mutex_enter(&dd->dd_lock); 2510 uint64_t val = dsl_dir_get_compressratio(dd); 2511 mutex_exit(&dd->dd_lock); 2512 return (val); 2513 } 2514 } 2515 2516 uint64_t 2517 dsl_get_used(dsl_dataset_t *ds) 2518 { 2519 if (ds->ds_is_snapshot) { 2520 return (dsl_dataset_phys(ds)->ds_unique_bytes); 2521 } else { 2522 dsl_dir_t *dd = ds->ds_dir; 2523 mutex_enter(&dd->dd_lock); 2524 uint64_t val = dsl_dir_get_used(dd); 2525 mutex_exit(&dd->dd_lock); 2526 return (val); 2527 } 2528 } 2529 2530 uint64_t 2531 dsl_get_creation(dsl_dataset_t *ds) 2532 { 2533 return (dsl_dataset_phys(ds)->ds_creation_time); 2534 } 2535 2536 uint64_t 2537 dsl_get_creationtxg(dsl_dataset_t *ds) 2538 { 2539 return (dsl_dataset_phys(ds)->ds_creation_txg); 2540 } 2541 2542 uint64_t 2543 dsl_get_refquota(dsl_dataset_t *ds) 2544 { 2545 return (ds->ds_quota); 2546 } 2547 2548 uint64_t 2549 dsl_get_refreservation(dsl_dataset_t *ds) 2550 { 2551 return (ds->ds_reserved); 2552 } 2553 2554 uint64_t 2555 dsl_get_guid(dsl_dataset_t *ds) 2556 { 2557 return (dsl_dataset_phys(ds)->ds_guid); 2558 } 2559 2560 uint64_t 2561 dsl_get_unique(dsl_dataset_t *ds) 2562 { 2563 return (dsl_dataset_phys(ds)->ds_unique_bytes); 2564 } 2565 2566 uint64_t 2567 dsl_get_objsetid(dsl_dataset_t *ds) 2568 { 2569 return (ds->ds_object); 2570 } 2571 2572 uint64_t 2573 dsl_get_userrefs(dsl_dataset_t *ds) 2574 { 2575 return (ds->ds_userrefs); 2576 } 2577 2578 uint64_t 2579 dsl_get_defer_destroy(dsl_dataset_t *ds) 2580 { 2581 return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 2582 } 2583 2584 uint64_t 2585 dsl_get_referenced(dsl_dataset_t *ds) 2586 { 2587 return (dsl_dataset_phys(ds)->ds_referenced_bytes); 2588 } 2589 2590 uint64_t 2591 dsl_get_numclones(dsl_dataset_t *ds) 2592 { 2593 ASSERT(ds->ds_is_snapshot); 2594 return (dsl_dataset_phys(ds)->ds_num_children - 1); 2595 } 2596 2597 uint64_t 2598 dsl_get_inconsistent(dsl_dataset_t *ds) 2599 { 2600 return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ? 2601 1 : 0); 2602 } 2603 2604 uint64_t 2605 dsl_get_redacted(dsl_dataset_t *ds) 2606 { 2607 return (dsl_dataset_feature_is_active(ds, 2608 SPA_FEATURE_REDACTED_DATASETS)); 2609 } 2610 2611 uint64_t 2612 dsl_get_available(dsl_dataset_t *ds) 2613 { 2614 uint64_t refdbytes = dsl_get_referenced(ds); 2615 uint64_t availbytes = dsl_dir_space_available(ds->ds_dir, 2616 NULL, 0, TRUE); 2617 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 2618 availbytes += 2619 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 2620 } 2621 if (ds->ds_quota != 0) { 2622 /* 2623 * Adjust available bytes according to refquota 2624 */ 2625 if (refdbytes < ds->ds_quota) { 2626 availbytes = MIN(availbytes, 2627 ds->ds_quota - refdbytes); 2628 } else { 2629 availbytes = 0; 2630 } 2631 } 2632 return (availbytes); 2633 } 2634 2635 int 2636 dsl_get_written(dsl_dataset_t *ds, uint64_t *written) 2637 { 2638 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2639 dsl_dataset_t *prev; 2640 int err = dsl_dataset_hold_obj(dp, 2641 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 2642 if (err == 0) { 2643 uint64_t comp, uncomp; 2644 err = dsl_dataset_space_written(prev, ds, written, 2645 &comp, &uncomp); 2646 dsl_dataset_rele(prev, FTAG); 2647 } 2648 return (err); 2649 } 2650 2651 /* 2652 * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN. 2653 */ 2654 int 2655 dsl_get_prev_snap(dsl_dataset_t *ds, char *snap) 2656 { 2657 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2658 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 2659 dsl_dataset_name(ds->ds_prev, snap); 2660 return (0); 2661 } else { 2662 return (SET_ERROR(ENOENT)); 2663 } 2664 } 2665 2666 void 2667 dsl_get_redact_snaps(dsl_dataset_t *ds, nvlist_t *propval) 2668 { 2669 uint64_t nsnaps; 2670 uint64_t *snaps; 2671 if (dsl_dataset_get_uint64_array_feature(ds, 2672 SPA_FEATURE_REDACTED_DATASETS, &nsnaps, &snaps)) { 2673 fnvlist_add_uint64_array(propval, ZPROP_VALUE, snaps, 2674 nsnaps); 2675 } 2676 } 2677 2678 /* 2679 * Returns the mountpoint property and source for the given dataset in the value 2680 * and source buffers. The value buffer must be at least as large as MAXPATHLEN 2681 * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN. 2682 * Returns 0 on success and an error on failure. 2683 */ 2684 int 2685 dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value, 2686 char *source) 2687 { 2688 int error; 2689 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2690 2691 /* Retrieve the mountpoint value stored in the zap object */ 2692 error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1, 2693 ZAP_MAXVALUELEN, value, source); 2694 if (error != 0) { 2695 return (error); 2696 } 2697 2698 /* 2699 * Process the dsname and source to find the full mountpoint string. 2700 * Can be skipped for 'legacy' or 'none'. 2701 */ 2702 if (value[0] == '/') { 2703 char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP); 2704 char *root = buf; 2705 const char *relpath; 2706 2707 /* 2708 * If we inherit the mountpoint, even from a dataset 2709 * with a received value, the source will be the path of 2710 * the dataset we inherit from. If source is 2711 * ZPROP_SOURCE_VAL_RECVD, the received value is not 2712 * inherited. 2713 */ 2714 if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) { 2715 relpath = ""; 2716 } else { 2717 ASSERT0(strncmp(dsname, source, strlen(source))); 2718 relpath = dsname + strlen(source); 2719 if (relpath[0] == '/') 2720 relpath++; 2721 } 2722 2723 spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN); 2724 2725 /* 2726 * Special case an alternate root of '/'. This will 2727 * avoid having multiple leading slashes in the 2728 * mountpoint path. 2729 */ 2730 if (strcmp(root, "/") == 0) 2731 root++; 2732 2733 /* 2734 * If the mountpoint is '/' then skip over this 2735 * if we are obtaining either an alternate root or 2736 * an inherited mountpoint. 2737 */ 2738 char *mnt = value; 2739 if (value[1] == '\0' && (root[0] != '\0' || 2740 relpath[0] != '\0')) 2741 mnt = value + 1; 2742 2743 mnt = kmem_strdup(mnt); 2744 2745 if (relpath[0] == '\0') { 2746 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s", 2747 root, mnt); 2748 } else { 2749 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s", 2750 root, mnt, relpath[0] == '@' ? "" : "/", 2751 relpath); 2752 } 2753 kmem_free(buf, ZAP_MAXVALUELEN); 2754 kmem_strfree(mnt); 2755 } 2756 2757 return (0); 2758 } 2759 2760 void 2761 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 2762 { 2763 dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool; 2764 2765 ASSERT(dsl_pool_config_held(dp)); 2766 2767 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, 2768 dsl_get_refratio(ds)); 2769 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 2770 dsl_get_logicalreferenced(ds)); 2771 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 2772 dsl_get_compressratio(ds)); 2773 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 2774 dsl_get_used(ds)); 2775 2776 if (ds->ds_is_snapshot) { 2777 get_clones_stat(ds, nv); 2778 } else { 2779 char buf[ZFS_MAX_DATASET_NAME_LEN]; 2780 if (dsl_get_prev_snap(ds, buf) == 0) 2781 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, 2782 buf); 2783 dsl_dir_stats(ds->ds_dir, nv); 2784 } 2785 2786 nvlist_t *propval = fnvlist_alloc(); 2787 dsl_get_redact_snaps(ds, propval); 2788 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_REDACT_SNAPS), 2789 propval); 2790 nvlist_free(propval); 2791 2792 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, 2793 dsl_get_available(ds)); 2794 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, 2795 dsl_get_referenced(ds)); 2796 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 2797 dsl_get_creation(ds)); 2798 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 2799 dsl_get_creationtxg(ds)); 2800 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 2801 dsl_get_refquota(ds)); 2802 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 2803 dsl_get_refreservation(ds)); 2804 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 2805 dsl_get_guid(ds)); 2806 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 2807 dsl_get_unique(ds)); 2808 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 2809 dsl_get_objsetid(ds)); 2810 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 2811 dsl_get_userrefs(ds)); 2812 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 2813 dsl_get_defer_destroy(ds)); 2814 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOTS_CHANGED, 2815 dsl_dir_snap_cmtime(ds->ds_dir).tv_sec); 2816 dsl_dataset_crypt_stats(ds, nv); 2817 2818 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 2819 uint64_t written; 2820 if (dsl_get_written(ds, &written) == 0) { 2821 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 2822 written); 2823 } 2824 } 2825 2826 if (!dsl_dataset_is_snapshot(ds)) { 2827 char *token = get_receive_resume_token(ds); 2828 if (token != NULL) { 2829 dsl_prop_nvlist_add_string(nv, 2830 ZFS_PROP_RECEIVE_RESUME_TOKEN, token); 2831 kmem_strfree(token); 2832 } 2833 } 2834 } 2835 2836 void 2837 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 2838 { 2839 dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool; 2840 ASSERT(dsl_pool_config_held(dp)); 2841 2842 stat->dds_creation_txg = dsl_get_creationtxg(ds); 2843 stat->dds_inconsistent = dsl_get_inconsistent(ds); 2844 stat->dds_guid = dsl_get_guid(ds); 2845 stat->dds_redacted = dsl_get_redacted(ds); 2846 stat->dds_origin[0] = '\0'; 2847 if (ds->ds_is_snapshot) { 2848 stat->dds_is_snapshot = B_TRUE; 2849 stat->dds_num_clones = dsl_get_numclones(ds); 2850 } else { 2851 stat->dds_is_snapshot = B_FALSE; 2852 stat->dds_num_clones = 0; 2853 2854 if (dsl_dir_is_clone(ds->ds_dir)) { 2855 dsl_dir_get_origin(ds->ds_dir, stat->dds_origin); 2856 } 2857 } 2858 } 2859 2860 uint64_t 2861 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 2862 { 2863 return (ds->ds_fsid_guid); 2864 } 2865 2866 void 2867 dsl_dataset_space(dsl_dataset_t *ds, 2868 uint64_t *refdbytesp, uint64_t *availbytesp, 2869 uint64_t *usedobjsp, uint64_t *availobjsp) 2870 { 2871 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; 2872 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 2873 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) 2874 *availbytesp += 2875 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 2876 if (ds->ds_quota != 0) { 2877 /* 2878 * Adjust available bytes according to refquota 2879 */ 2880 if (*refdbytesp < ds->ds_quota) 2881 *availbytesp = MIN(*availbytesp, 2882 ds->ds_quota - *refdbytesp); 2883 else 2884 *availbytesp = 0; 2885 } 2886 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2887 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); 2888 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2889 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 2890 } 2891 2892 boolean_t 2893 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 2894 { 2895 dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool; 2896 uint64_t birth; 2897 2898 ASSERT(dsl_pool_config_held(dp)); 2899 if (snap == NULL) 2900 return (B_FALSE); 2901 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2902 birth = dsl_dataset_get_blkptr(ds)->blk_birth; 2903 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2904 if (birth > dsl_dataset_phys(snap)->ds_creation_txg) { 2905 objset_t *os, *os_snap; 2906 /* 2907 * It may be that only the ZIL differs, because it was 2908 * reset in the head. Don't count that as being 2909 * modified. 2910 */ 2911 if (dmu_objset_from_ds(ds, &os) != 0) 2912 return (B_TRUE); 2913 if (dmu_objset_from_ds(snap, &os_snap) != 0) 2914 return (B_TRUE); 2915 return (memcmp(&os->os_phys->os_meta_dnode, 2916 &os_snap->os_phys->os_meta_dnode, 2917 sizeof (os->os_phys->os_meta_dnode)) != 0); 2918 } 2919 return (B_FALSE); 2920 } 2921 2922 static int 2923 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 2924 dsl_dataset_t *hds, void *arg) 2925 { 2926 (void) dp; 2927 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2928 int error; 2929 uint64_t val; 2930 2931 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2932 if (error != 0) { 2933 /* ignore nonexistent snapshots */ 2934 return (error == ENOENT ? 0 : error); 2935 } 2936 2937 /* new name should not exist */ 2938 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 2939 if (error == 0) 2940 error = SET_ERROR(EEXIST); 2941 else if (error == ENOENT) 2942 error = 0; 2943 2944 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2945 if (dsl_dir_namelen(hds->ds_dir) + 1 + 2946 strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN) 2947 error = SET_ERROR(ENAMETOOLONG); 2948 2949 return (error); 2950 } 2951 2952 int 2953 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 2954 { 2955 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2956 dsl_pool_t *dp = dmu_tx_pool(tx); 2957 dsl_dataset_t *hds; 2958 int error; 2959 2960 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 2961 if (error != 0) 2962 return (error); 2963 2964 if (ddrsa->ddrsa_recursive) { 2965 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 2966 dsl_dataset_rename_snapshot_check_impl, ddrsa, 2967 DS_FIND_CHILDREN); 2968 } else { 2969 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 2970 } 2971 dsl_dataset_rele(hds, FTAG); 2972 return (error); 2973 } 2974 2975 static int 2976 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 2977 dsl_dataset_t *hds, void *arg) 2978 { 2979 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2980 dsl_dataset_t *ds; 2981 uint64_t val; 2982 dmu_tx_t *tx = ddrsa->ddrsa_tx; 2983 int error; 2984 2985 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2986 ASSERT(error == 0 || error == ENOENT); 2987 if (error == ENOENT) { 2988 /* ignore nonexistent snapshots */ 2989 return (0); 2990 } 2991 2992 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 2993 2994 /* log before we change the name */ 2995 spa_history_log_internal_ds(ds, "rename", tx, 2996 "-> @%s", ddrsa->ddrsa_newsnapname); 2997 2998 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 2999 B_FALSE)); 3000 mutex_enter(&ds->ds_lock); 3001 (void) strlcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname, 3002 sizeof (ds->ds_snapname)); 3003 mutex_exit(&ds->ds_lock); 3004 VERIFY0(zap_add(dp->dp_meta_objset, 3005 dsl_dataset_phys(hds)->ds_snapnames_zapobj, 3006 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 3007 zvol_rename_minors(dp->dp_spa, ddrsa->ddrsa_oldsnapname, 3008 ddrsa->ddrsa_newsnapname, B_TRUE); 3009 3010 dsl_dataset_rele(ds, FTAG); 3011 return (0); 3012 } 3013 3014 void 3015 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 3016 { 3017 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 3018 dsl_pool_t *dp = dmu_tx_pool(tx); 3019 dsl_dataset_t *hds = NULL; 3020 3021 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 3022 ddrsa->ddrsa_tx = tx; 3023 if (ddrsa->ddrsa_recursive) { 3024 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 3025 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 3026 DS_FIND_CHILDREN)); 3027 } else { 3028 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 3029 } 3030 dsl_dataset_rele(hds, FTAG); 3031 } 3032 3033 int 3034 dsl_dataset_rename_snapshot(const char *fsname, 3035 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 3036 { 3037 dsl_dataset_rename_snapshot_arg_t ddrsa; 3038 3039 ddrsa.ddrsa_fsname = fsname; 3040 ddrsa.ddrsa_oldsnapname = oldsnapname; 3041 ddrsa.ddrsa_newsnapname = newsnapname; 3042 ddrsa.ddrsa_recursive = recursive; 3043 3044 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 3045 dsl_dataset_rename_snapshot_sync, &ddrsa, 3046 1, ZFS_SPACE_CHECK_RESERVED)); 3047 } 3048 3049 /* 3050 * If we're doing an ownership handoff, we need to make sure that there is 3051 * only one long hold on the dataset. We're not allowed to change anything here 3052 * so we don't permanently release the long hold or regular hold here. We want 3053 * to do this only when syncing to avoid the dataset unexpectedly going away 3054 * when we release the long hold. 3055 */ 3056 static int 3057 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 3058 { 3059 boolean_t held = B_FALSE; 3060 3061 if (!dmu_tx_is_syncing(tx)) 3062 return (0); 3063 3064 dsl_dir_t *dd = ds->ds_dir; 3065 mutex_enter(&dd->dd_activity_lock); 3066 uint64_t holds = zfs_refcount_count(&ds->ds_longholds) - 3067 (owner != NULL ? 1 : 0); 3068 /* 3069 * The value of dd_activity_waiters can chance as soon as we drop the 3070 * lock, but we're fine with that; new waiters coming in or old 3071 * waiters leaving doesn't cause problems, since we're going to cancel 3072 * waiters later anyway. The goal of this check is to verify that no 3073 * non-waiters have long-holds, and all new long-holds will be 3074 * prevented because we're holding the pool config as writer. 3075 */ 3076 if (holds != dd->dd_activity_waiters) 3077 held = B_TRUE; 3078 mutex_exit(&dd->dd_activity_lock); 3079 3080 if (held) 3081 return (SET_ERROR(EBUSY)); 3082 3083 return (0); 3084 } 3085 3086 int 3087 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 3088 { 3089 dsl_dataset_rollback_arg_t *ddra = arg; 3090 dsl_pool_t *dp = dmu_tx_pool(tx); 3091 dsl_dataset_t *ds; 3092 int64_t unused_refres_delta; 3093 int error; 3094 3095 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 3096 if (error != 0) 3097 return (error); 3098 3099 /* must not be a snapshot */ 3100 if (ds->ds_is_snapshot) { 3101 dsl_dataset_rele(ds, FTAG); 3102 return (SET_ERROR(EINVAL)); 3103 } 3104 3105 /* must have a most recent snapshot */ 3106 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { 3107 dsl_dataset_rele(ds, FTAG); 3108 return (SET_ERROR(ESRCH)); 3109 } 3110 3111 /* 3112 * No rollback to a snapshot created in the current txg, because 3113 * the rollback may dirty the dataset and create blocks that are 3114 * not reachable from the rootbp while having a birth txg that 3115 * falls into the snapshot's range. 3116 */ 3117 if (dmu_tx_is_syncing(tx) && 3118 dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) { 3119 dsl_dataset_rele(ds, FTAG); 3120 return (SET_ERROR(EAGAIN)); 3121 } 3122 3123 /* 3124 * If the expected target snapshot is specified, then check that 3125 * the latest snapshot is it. 3126 */ 3127 if (ddra->ddra_tosnap != NULL) { 3128 dsl_dataset_t *snapds; 3129 3130 /* Check if the target snapshot exists at all. */ 3131 error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds); 3132 if (error != 0) { 3133 /* 3134 * ESRCH is used to signal that the target snapshot does 3135 * not exist, while ENOENT is used to report that 3136 * the rolled back dataset does not exist. 3137 * ESRCH is also used to cover other cases where the 3138 * target snapshot is not related to the dataset being 3139 * rolled back such as being in a different pool. 3140 */ 3141 if (error == ENOENT || error == EXDEV) 3142 error = SET_ERROR(ESRCH); 3143 dsl_dataset_rele(ds, FTAG); 3144 return (error); 3145 } 3146 ASSERT(snapds->ds_is_snapshot); 3147 3148 /* Check if the snapshot is the latest snapshot indeed. */ 3149 if (snapds != ds->ds_prev) { 3150 /* 3151 * Distinguish between the case where the only problem 3152 * is intervening snapshots (EEXIST) vs the snapshot 3153 * not being a valid target for rollback (ESRCH). 3154 */ 3155 if (snapds->ds_dir == ds->ds_dir || 3156 (dsl_dir_is_clone(ds->ds_dir) && 3157 dsl_dir_phys(ds->ds_dir)->dd_origin_obj == 3158 snapds->ds_object)) { 3159 error = SET_ERROR(EEXIST); 3160 } else { 3161 error = SET_ERROR(ESRCH); 3162 } 3163 dsl_dataset_rele(snapds, FTAG); 3164 dsl_dataset_rele(ds, FTAG); 3165 return (error); 3166 } 3167 dsl_dataset_rele(snapds, FTAG); 3168 } 3169 3170 /* must not have any bookmarks after the most recent snapshot */ 3171 if (dsl_bookmark_latest_txg(ds) > 3172 dsl_dataset_phys(ds)->ds_prev_snap_txg) { 3173 dsl_dataset_rele(ds, FTAG); 3174 return (SET_ERROR(EEXIST)); 3175 } 3176 3177 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 3178 if (error != 0) { 3179 dsl_dataset_rele(ds, FTAG); 3180 return (error); 3181 } 3182 3183 /* 3184 * Check if the snap we are rolling back to uses more than 3185 * the refquota. 3186 */ 3187 if (ds->ds_quota != 0 && 3188 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { 3189 dsl_dataset_rele(ds, FTAG); 3190 return (SET_ERROR(EDQUOT)); 3191 } 3192 3193 /* 3194 * When we do the clone swap, we will temporarily use more space 3195 * due to the refreservation (the head will no longer have any 3196 * unique space, so the entire amount of the refreservation will need 3197 * to be free). We will immediately destroy the clone, freeing 3198 * this space, but the freeing happens over many txg's. 3199 */ 3200 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 3201 dsl_dataset_phys(ds)->ds_unique_bytes); 3202 3203 if (unused_refres_delta > 0 && 3204 unused_refres_delta > 3205 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 3206 dsl_dataset_rele(ds, FTAG); 3207 return (SET_ERROR(ENOSPC)); 3208 } 3209 3210 dsl_dataset_rele(ds, FTAG); 3211 return (0); 3212 } 3213 3214 void 3215 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 3216 { 3217 dsl_dataset_rollback_arg_t *ddra = arg; 3218 dsl_pool_t *dp = dmu_tx_pool(tx); 3219 dsl_dataset_t *ds, *clone; 3220 uint64_t cloneobj; 3221 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 3222 3223 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 3224 3225 dsl_dataset_name(ds->ds_prev, namebuf); 3226 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 3227 3228 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 3229 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, NULL, tx); 3230 3231 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 3232 3233 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 3234 dsl_dataset_zero_zil(ds, tx); 3235 3236 dsl_destroy_head_sync_impl(clone, tx); 3237 3238 dsl_dataset_rele(clone, FTAG); 3239 dsl_dataset_rele(ds, FTAG); 3240 } 3241 3242 /* 3243 * Rolls back the given filesystem or volume to the most recent snapshot. 3244 * The name of the most recent snapshot will be returned under key "target" 3245 * in the result nvlist. 3246 * 3247 * If owner != NULL: 3248 * - The existing dataset MUST be owned by the specified owner at entry 3249 * - Upon return, dataset will still be held by the same owner, whether we 3250 * succeed or not. 3251 * 3252 * This mode is required any time the existing filesystem is mounted. See 3253 * notes above zfs_suspend_fs() for further details. 3254 */ 3255 int 3256 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner, 3257 nvlist_t *result) 3258 { 3259 dsl_dataset_rollback_arg_t ddra; 3260 3261 ddra.ddra_fsname = fsname; 3262 ddra.ddra_tosnap = tosnap; 3263 ddra.ddra_owner = owner; 3264 ddra.ddra_result = result; 3265 3266 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 3267 dsl_dataset_rollback_sync, &ddra, 3268 1, ZFS_SPACE_CHECK_RESERVED)); 3269 } 3270 3271 struct promotenode { 3272 list_node_t link; 3273 dsl_dataset_t *ds; 3274 }; 3275 3276 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 3277 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 3278 const void *tag); 3279 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, const void *tag); 3280 3281 int 3282 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 3283 { 3284 dsl_dataset_promote_arg_t *ddpa = arg; 3285 dsl_pool_t *dp = dmu_tx_pool(tx); 3286 dsl_dataset_t *hds; 3287 struct promotenode *snap; 3288 int err; 3289 uint64_t unused; 3290 uint64_t ss_mv_cnt; 3291 size_t max_snap_len; 3292 boolean_t conflicting_snaps; 3293 3294 err = promote_hold(ddpa, dp, FTAG); 3295 if (err != 0) 3296 return (err); 3297 3298 hds = ddpa->ddpa_clone; 3299 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; 3300 3301 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { 3302 promote_rele(ddpa, FTAG); 3303 return (SET_ERROR(EXDEV)); 3304 } 3305 3306 snap = list_head(&ddpa->shared_snaps); 3307 if (snap == NULL) { 3308 err = SET_ERROR(ENOENT); 3309 goto out; 3310 } 3311 dsl_dataset_t *const origin_ds = snap->ds; 3312 3313 /* 3314 * Encrypted clones share a DSL Crypto Key with their origin's dsl dir. 3315 * When doing a promote we must make sure the encryption root for 3316 * both the target and the target's origin does not change to avoid 3317 * needing to rewrap encryption keys 3318 */ 3319 err = dsl_dataset_promote_crypt_check(hds->ds_dir, origin_ds->ds_dir); 3320 if (err != 0) 3321 goto out; 3322 3323 /* 3324 * Compute and check the amount of space to transfer. Since this is 3325 * so expensive, don't do the preliminary check. 3326 */ 3327 if (!dmu_tx_is_syncing(tx)) { 3328 promote_rele(ddpa, FTAG); 3329 return (0); 3330 } 3331 3332 /* compute origin's new unique space */ 3333 snap = list_tail(&ddpa->clone_snaps); 3334 ASSERT(snap != NULL); 3335 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 3336 origin_ds->ds_object); 3337 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 3338 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, 3339 &ddpa->unique, &unused, &unused); 3340 3341 /* 3342 * Walk the snapshots that we are moving 3343 * 3344 * Compute space to transfer. Consider the incremental changes 3345 * to used by each snapshot: 3346 * (my used) = (prev's used) + (blocks born) - (blocks killed) 3347 * So each snapshot gave birth to: 3348 * (blocks born) = (my used) - (prev's used) + (blocks killed) 3349 * So a sequence would look like: 3350 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 3351 * Which simplifies to: 3352 * uN + kN + kN-1 + ... + k1 + k0 3353 * Note however, if we stop before we reach the ORIGIN we get: 3354 * uN + kN + kN-1 + ... + kM - uM-1 3355 */ 3356 conflicting_snaps = B_FALSE; 3357 ss_mv_cnt = 0; 3358 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; 3359 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; 3360 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; 3361 for (snap = list_head(&ddpa->shared_snaps); snap; 3362 snap = list_next(&ddpa->shared_snaps, snap)) { 3363 uint64_t val, dlused, dlcomp, dluncomp; 3364 dsl_dataset_t *ds = snap->ds; 3365 3366 ss_mv_cnt++; 3367 3368 /* 3369 * If there are long holds, we won't be able to evict 3370 * the objset. 3371 */ 3372 if (dsl_dataset_long_held(ds)) { 3373 err = SET_ERROR(EBUSY); 3374 goto out; 3375 } 3376 3377 /* Check that the snapshot name does not conflict */ 3378 VERIFY0(dsl_dataset_get_snapname(ds)); 3379 if (strlen(ds->ds_snapname) >= max_snap_len) { 3380 err = SET_ERROR(ENAMETOOLONG); 3381 goto out; 3382 } 3383 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 3384 if (err == 0) { 3385 fnvlist_add_boolean(ddpa->err_ds, 3386 snap->ds->ds_snapname); 3387 conflicting_snaps = B_TRUE; 3388 } else if (err != ENOENT) { 3389 goto out; 3390 } 3391 3392 /* The very first snapshot does not have a deadlist */ 3393 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) 3394 continue; 3395 3396 dsl_deadlist_space(&ds->ds_deadlist, 3397 &dlused, &dlcomp, &dluncomp); 3398 ddpa->used += dlused; 3399 ddpa->comp += dlcomp; 3400 ddpa->uncomp += dluncomp; 3401 } 3402 3403 /* 3404 * Check that bookmarks that are being transferred don't have 3405 * name conflicts. 3406 */ 3407 for (dsl_bookmark_node_t *dbn = avl_first(&origin_ds->ds_bookmarks); 3408 dbn != NULL && dbn->dbn_phys.zbm_creation_txg <= 3409 dsl_dataset_phys(origin_ds)->ds_creation_txg; 3410 dbn = AVL_NEXT(&origin_ds->ds_bookmarks, dbn)) { 3411 if (strlen(dbn->dbn_name) >= max_snap_len) { 3412 err = SET_ERROR(ENAMETOOLONG); 3413 goto out; 3414 } 3415 zfs_bookmark_phys_t bm; 3416 err = dsl_bookmark_lookup_impl(ddpa->ddpa_clone, 3417 dbn->dbn_name, &bm); 3418 3419 if (err == 0) { 3420 fnvlist_add_boolean(ddpa->err_ds, dbn->dbn_name); 3421 conflicting_snaps = B_TRUE; 3422 } else if (err == ESRCH) { 3423 err = 0; 3424 } else if (err != 0) { 3425 goto out; 3426 } 3427 } 3428 3429 /* 3430 * In order to return the full list of conflicting snapshots, we check 3431 * whether there was a conflict after traversing all of them. 3432 */ 3433 if (conflicting_snaps) { 3434 err = SET_ERROR(EEXIST); 3435 goto out; 3436 } 3437 3438 /* 3439 * If we are a clone of a clone then we never reached ORIGIN, 3440 * so we need to subtract out the clone origin's used space. 3441 */ 3442 if (ddpa->origin_origin) { 3443 ddpa->used -= 3444 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; 3445 ddpa->comp -= 3446 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; 3447 ddpa->uncomp -= 3448 dsl_dataset_phys(ddpa->origin_origin)-> 3449 ds_uncompressed_bytes; 3450 } 3451 3452 /* Check that there is enough space and limit headroom here */ 3453 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 3454 0, ss_mv_cnt, ddpa->used, ddpa->cr, ddpa->proc); 3455 if (err != 0) 3456 goto out; 3457 3458 /* 3459 * Compute the amounts of space that will be used by snapshots 3460 * after the promotion (for both origin and clone). For each, 3461 * it is the amount of space that will be on all of their 3462 * deadlists (that was not born before their new origin). 3463 */ 3464 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 3465 uint64_t space; 3466 3467 /* 3468 * Note, typically this will not be a clone of a clone, 3469 * so dd_origin_txg will be < TXG_INITIAL, so 3470 * these snaplist_space() -> dsl_deadlist_space_range() 3471 * calls will be fast because they do not have to 3472 * iterate over all bps. 3473 */ 3474 snap = list_head(&ddpa->origin_snaps); 3475 if (snap == NULL) { 3476 err = SET_ERROR(ENOENT); 3477 goto out; 3478 } 3479 err = snaplist_space(&ddpa->shared_snaps, 3480 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 3481 if (err != 0) 3482 goto out; 3483 3484 err = snaplist_space(&ddpa->clone_snaps, 3485 snap->ds->ds_dir->dd_origin_txg, &space); 3486 if (err != 0) 3487 goto out; 3488 ddpa->cloneusedsnap += space; 3489 } 3490 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & 3491 DD_FLAG_USED_BREAKDOWN) { 3492 err = snaplist_space(&ddpa->origin_snaps, 3493 dsl_dataset_phys(origin_ds)->ds_creation_txg, 3494 &ddpa->originusedsnap); 3495 if (err != 0) 3496 goto out; 3497 } 3498 3499 out: 3500 promote_rele(ddpa, FTAG); 3501 return (err); 3502 } 3503 3504 void 3505 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 3506 { 3507 dsl_dataset_promote_arg_t *ddpa = arg; 3508 dsl_pool_t *dp = dmu_tx_pool(tx); 3509 dsl_dataset_t *hds; 3510 struct promotenode *snap; 3511 dsl_dataset_t *origin_ds; 3512 dsl_dataset_t *origin_head; 3513 dsl_dir_t *dd; 3514 dsl_dir_t *odd = NULL; 3515 uint64_t oldnext_obj; 3516 int64_t delta; 3517 3518 ASSERT(nvlist_empty(ddpa->err_ds)); 3519 3520 VERIFY0(promote_hold(ddpa, dp, FTAG)); 3521 hds = ddpa->ddpa_clone; 3522 3523 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); 3524 3525 snap = list_head(&ddpa->shared_snaps); 3526 origin_ds = snap->ds; 3527 dd = hds->ds_dir; 3528 3529 snap = list_head(&ddpa->origin_snaps); 3530 origin_head = snap->ds; 3531 3532 /* 3533 * We need to explicitly open odd, since origin_ds's dd will be 3534 * changing. 3535 */ 3536 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 3537 NULL, FTAG, &odd)); 3538 3539 dsl_dataset_promote_crypt_sync(hds->ds_dir, odd, tx); 3540 3541 /* change origin's next snap */ 3542 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 3543 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; 3544 snap = list_tail(&ddpa->clone_snaps); 3545 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 3546 origin_ds->ds_object); 3547 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; 3548 3549 /* change the origin's next clone */ 3550 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { 3551 dsl_dataset_remove_from_next_clones(origin_ds, 3552 snap->ds->ds_object, tx); 3553 VERIFY0(zap_add_int(dp->dp_meta_objset, 3554 dsl_dataset_phys(origin_ds)->ds_next_clones_obj, 3555 oldnext_obj, tx)); 3556 } 3557 3558 /* change origin */ 3559 dmu_buf_will_dirty(dd->dd_dbuf, tx); 3560 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); 3561 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; 3562 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 3563 dmu_buf_will_dirty(odd->dd_dbuf, tx); 3564 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; 3565 origin_head->ds_dir->dd_origin_txg = 3566 dsl_dataset_phys(origin_ds)->ds_creation_txg; 3567 3568 /* change dd_clone entries */ 3569 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 3570 VERIFY0(zap_remove_int(dp->dp_meta_objset, 3571 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); 3572 VERIFY0(zap_add_int(dp->dp_meta_objset, 3573 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 3574 hds->ds_object, tx)); 3575 3576 VERIFY0(zap_remove_int(dp->dp_meta_objset, 3577 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 3578 origin_head->ds_object, tx)); 3579 if (dsl_dir_phys(dd)->dd_clones == 0) { 3580 dsl_dir_phys(dd)->dd_clones = 3581 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, 3582 DMU_OT_NONE, 0, tx); 3583 } 3584 VERIFY0(zap_add_int(dp->dp_meta_objset, 3585 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); 3586 } 3587 3588 /* 3589 * Move bookmarks to this dir. 3590 */ 3591 dsl_bookmark_node_t *dbn_next; 3592 for (dsl_bookmark_node_t *dbn = avl_first(&origin_head->ds_bookmarks); 3593 dbn != NULL && dbn->dbn_phys.zbm_creation_txg <= 3594 dsl_dataset_phys(origin_ds)->ds_creation_txg; 3595 dbn = dbn_next) { 3596 dbn_next = AVL_NEXT(&origin_head->ds_bookmarks, dbn); 3597 3598 avl_remove(&origin_head->ds_bookmarks, dbn); 3599 VERIFY0(zap_remove(dp->dp_meta_objset, 3600 origin_head->ds_bookmarks_obj, dbn->dbn_name, tx)); 3601 3602 dsl_bookmark_node_add(hds, dbn, tx); 3603 } 3604 3605 dsl_bookmark_next_changed(hds, origin_ds, tx); 3606 3607 /* move snapshots to this dir */ 3608 for (snap = list_head(&ddpa->shared_snaps); snap; 3609 snap = list_next(&ddpa->shared_snaps, snap)) { 3610 dsl_dataset_t *ds = snap->ds; 3611 3612 /* 3613 * Property callbacks are registered to a particular 3614 * dsl_dir. Since ours is changing, evict the objset 3615 * so that they will be unregistered from the old dsl_dir. 3616 */ 3617 if (ds->ds_objset) { 3618 dmu_objset_evict(ds->ds_objset); 3619 ds->ds_objset = NULL; 3620 } 3621 3622 /* move snap name entry */ 3623 VERIFY0(dsl_dataset_get_snapname(ds)); 3624 VERIFY0(dsl_dataset_snap_remove(origin_head, 3625 ds->ds_snapname, tx, B_TRUE)); 3626 VERIFY0(zap_add(dp->dp_meta_objset, 3627 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 3628 8, 1, &ds->ds_object, tx)); 3629 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 3630 DD_FIELD_SNAPSHOT_COUNT, tx); 3631 3632 /* change containing dsl_dir */ 3633 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3634 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); 3635 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; 3636 ASSERT3P(ds->ds_dir, ==, odd); 3637 dsl_dir_rele(ds->ds_dir, ds); 3638 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 3639 NULL, ds, &ds->ds_dir)); 3640 3641 /* move any clone references */ 3642 if (dsl_dataset_phys(ds)->ds_next_clones_obj && 3643 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 3644 zap_cursor_t zc; 3645 zap_attribute_t za; 3646 3647 for (zap_cursor_init(&zc, dp->dp_meta_objset, 3648 dsl_dataset_phys(ds)->ds_next_clones_obj); 3649 zap_cursor_retrieve(&zc, &za) == 0; 3650 zap_cursor_advance(&zc)) { 3651 dsl_dataset_t *cnds; 3652 uint64_t o; 3653 3654 if (za.za_first_integer == oldnext_obj) { 3655 /* 3656 * We've already moved the 3657 * origin's reference. 3658 */ 3659 continue; 3660 } 3661 3662 VERIFY0(dsl_dataset_hold_obj(dp, 3663 za.za_first_integer, FTAG, &cnds)); 3664 o = dsl_dir_phys(cnds->ds_dir)-> 3665 dd_head_dataset_obj; 3666 3667 VERIFY0(zap_remove_int(dp->dp_meta_objset, 3668 dsl_dir_phys(odd)->dd_clones, o, tx)); 3669 VERIFY0(zap_add_int(dp->dp_meta_objset, 3670 dsl_dir_phys(dd)->dd_clones, o, tx)); 3671 dsl_dataset_rele(cnds, FTAG); 3672 } 3673 zap_cursor_fini(&zc); 3674 } 3675 3676 ASSERT(!dsl_prop_hascb(ds)); 3677 } 3678 3679 /* 3680 * Change space accounting. 3681 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 3682 * both be valid, or both be 0 (resulting in delta == 0). This 3683 * is true for each of {clone,origin} independently. 3684 */ 3685 3686 delta = ddpa->cloneusedsnap - 3687 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; 3688 ASSERT3S(delta, >=, 0); 3689 ASSERT3U(ddpa->used, >=, delta); 3690 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 3691 dsl_dir_diduse_space(dd, DD_USED_HEAD, 3692 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 3693 3694 delta = ddpa->originusedsnap - 3695 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; 3696 ASSERT3S(delta, <=, 0); 3697 ASSERT3U(ddpa->used, >=, -delta); 3698 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 3699 dsl_dir_diduse_space(odd, DD_USED_HEAD, 3700 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 3701 3702 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; 3703 3704 /* 3705 * Since livelists are specific to a clone's origin txg, they 3706 * are no longer accurate. Destroy the livelist from the clone being 3707 * promoted. If the origin dataset is a clone, destroy its livelist 3708 * as well. 3709 */ 3710 dsl_dir_remove_livelist(dd, tx, B_TRUE); 3711 dsl_dir_remove_livelist(odd, tx, B_TRUE); 3712 3713 /* log history record */ 3714 spa_history_log_internal_ds(hds, "promote", tx, " "); 3715 3716 dsl_dir_rele(odd, FTAG); 3717 promote_rele(ddpa, FTAG); 3718 3719 /* 3720 * Transfer common error blocks from old head to new head. 3721 */ 3722 if (spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_HEAD_ERRLOG)) { 3723 uint64_t old_head = origin_head->ds_object; 3724 uint64_t new_head = hds->ds_object; 3725 spa_swap_errlog(dp->dp_spa, new_head, old_head, tx); 3726 } 3727 } 3728 3729 /* 3730 * Make a list of dsl_dataset_t's for the snapshots between first_obj 3731 * (exclusive) and last_obj (inclusive). The list will be in reverse 3732 * order (last_obj will be the list_head()). If first_obj == 0, do all 3733 * snapshots back to this dataset's origin. 3734 */ 3735 static int 3736 snaplist_make(dsl_pool_t *dp, 3737 uint64_t first_obj, uint64_t last_obj, list_t *l, const void *tag) 3738 { 3739 uint64_t obj = last_obj; 3740 3741 list_create(l, sizeof (struct promotenode), 3742 offsetof(struct promotenode, link)); 3743 3744 while (obj != first_obj) { 3745 dsl_dataset_t *ds; 3746 struct promotenode *snap; 3747 int err; 3748 3749 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 3750 ASSERT(err != ENOENT); 3751 if (err != 0) 3752 return (err); 3753 3754 if (first_obj == 0) 3755 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; 3756 3757 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 3758 snap->ds = ds; 3759 list_insert_tail(l, snap); 3760 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3761 } 3762 3763 return (0); 3764 } 3765 3766 static int 3767 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 3768 { 3769 struct promotenode *snap; 3770 3771 *spacep = 0; 3772 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 3773 uint64_t used, comp, uncomp; 3774 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 3775 mintxg, UINT64_MAX, &used, &comp, &uncomp); 3776 *spacep += used; 3777 } 3778 return (0); 3779 } 3780 3781 static void 3782 snaplist_destroy(list_t *l, const void *tag) 3783 { 3784 struct promotenode *snap; 3785 3786 if (l == NULL || !list_link_active(&l->list_head)) 3787 return; 3788 3789 while ((snap = list_tail(l)) != NULL) { 3790 list_remove(l, snap); 3791 dsl_dataset_rele(snap->ds, tag); 3792 kmem_free(snap, sizeof (*snap)); 3793 } 3794 list_destroy(l); 3795 } 3796 3797 static int 3798 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, const void *tag) 3799 { 3800 int error; 3801 dsl_dir_t *dd; 3802 struct promotenode *snap; 3803 3804 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 3805 &ddpa->ddpa_clone); 3806 if (error != 0) 3807 return (error); 3808 dd = ddpa->ddpa_clone->ds_dir; 3809 3810 if (ddpa->ddpa_clone->ds_is_snapshot || 3811 !dsl_dir_is_clone(dd)) { 3812 dsl_dataset_rele(ddpa->ddpa_clone, tag); 3813 return (SET_ERROR(EINVAL)); 3814 } 3815 3816 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, 3817 &ddpa->shared_snaps, tag); 3818 if (error != 0) 3819 goto out; 3820 3821 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 3822 &ddpa->clone_snaps, tag); 3823 if (error != 0) 3824 goto out; 3825 3826 snap = list_head(&ddpa->shared_snaps); 3827 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); 3828 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, 3829 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, 3830 &ddpa->origin_snaps, tag); 3831 if (error != 0) 3832 goto out; 3833 3834 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { 3835 error = dsl_dataset_hold_obj(dp, 3836 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, 3837 tag, &ddpa->origin_origin); 3838 if (error != 0) 3839 goto out; 3840 } 3841 out: 3842 if (error != 0) 3843 promote_rele(ddpa, tag); 3844 return (error); 3845 } 3846 3847 static void 3848 promote_rele(dsl_dataset_promote_arg_t *ddpa, const void *tag) 3849 { 3850 snaplist_destroy(&ddpa->shared_snaps, tag); 3851 snaplist_destroy(&ddpa->clone_snaps, tag); 3852 snaplist_destroy(&ddpa->origin_snaps, tag); 3853 if (ddpa->origin_origin != NULL) 3854 dsl_dataset_rele(ddpa->origin_origin, tag); 3855 dsl_dataset_rele(ddpa->ddpa_clone, tag); 3856 } 3857 3858 /* 3859 * Promote a clone. 3860 * 3861 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 3862 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.) 3863 */ 3864 int 3865 dsl_dataset_promote(const char *name, char *conflsnap) 3866 { 3867 dsl_dataset_promote_arg_t ddpa = { 0 }; 3868 uint64_t numsnaps; 3869 int error; 3870 nvpair_t *snap_pair; 3871 objset_t *os; 3872 3873 /* 3874 * We will modify space proportional to the number of 3875 * snapshots. Compute numsnaps. 3876 */ 3877 error = dmu_objset_hold(name, FTAG, &os); 3878 if (error != 0) 3879 return (error); 3880 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 3881 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, 3882 &numsnaps); 3883 dmu_objset_rele(os, FTAG); 3884 if (error != 0) 3885 return (error); 3886 3887 ddpa.ddpa_clonename = name; 3888 ddpa.err_ds = fnvlist_alloc(); 3889 ddpa.cr = CRED(); 3890 ddpa.proc = curproc; 3891 3892 error = dsl_sync_task(name, dsl_dataset_promote_check, 3893 dsl_dataset_promote_sync, &ddpa, 3894 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED); 3895 3896 /* 3897 * Return the first conflicting snapshot found. 3898 */ 3899 snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL); 3900 if (snap_pair != NULL && conflsnap != NULL) 3901 (void) strlcpy(conflsnap, nvpair_name(snap_pair), 3902 ZFS_MAX_DATASET_NAME_LEN); 3903 3904 fnvlist_free(ddpa.err_ds); 3905 return (error); 3906 } 3907 3908 int 3909 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 3910 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 3911 { 3912 /* 3913 * "slack" factor for received datasets with refquota set on them. 3914 * See the bottom of this function for details on its use. 3915 */ 3916 uint64_t refquota_slack = (uint64_t)DMU_MAX_ACCESS * 3917 spa_asize_inflation; 3918 int64_t unused_refres_delta; 3919 3920 /* they should both be heads */ 3921 if (clone->ds_is_snapshot || 3922 origin_head->ds_is_snapshot) 3923 return (SET_ERROR(EINVAL)); 3924 3925 /* if we are not forcing, the branch point should be just before them */ 3926 if (!force && clone->ds_prev != origin_head->ds_prev) 3927 return (SET_ERROR(EINVAL)); 3928 3929 /* clone should be the clone (unless they are unrelated) */ 3930 if (clone->ds_prev != NULL && 3931 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 3932 origin_head->ds_dir != clone->ds_prev->ds_dir) 3933 return (SET_ERROR(EINVAL)); 3934 3935 /* the clone should be a child of the origin */ 3936 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 3937 return (SET_ERROR(EINVAL)); 3938 3939 /* origin_head shouldn't be modified unless 'force' */ 3940 if (!force && 3941 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 3942 return (SET_ERROR(ETXTBSY)); 3943 3944 /* origin_head should have no long holds (e.g. is not mounted) */ 3945 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 3946 return (SET_ERROR(EBUSY)); 3947 3948 /* check amount of any unconsumed refreservation */ 3949 unused_refres_delta = 3950 (int64_t)MIN(origin_head->ds_reserved, 3951 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 3952 (int64_t)MIN(origin_head->ds_reserved, 3953 dsl_dataset_phys(clone)->ds_unique_bytes); 3954 3955 if (unused_refres_delta > 0 && 3956 unused_refres_delta > 3957 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 3958 return (SET_ERROR(ENOSPC)); 3959 3960 /* 3961 * The clone can't be too much over the head's refquota. 3962 * 3963 * To ensure that the entire refquota can be used, we allow one 3964 * transaction to exceed the refquota. Therefore, this check 3965 * needs to also allow for the space referenced to be more than the 3966 * refquota. The maximum amount of space that one transaction can use 3967 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this 3968 * overage ensures that we are able to receive a filesystem that 3969 * exceeds the refquota on the source system. 3970 * 3971 * So that overage is the refquota_slack we use below. 3972 */ 3973 if (origin_head->ds_quota != 0 && 3974 dsl_dataset_phys(clone)->ds_referenced_bytes > 3975 origin_head->ds_quota + refquota_slack) 3976 return (SET_ERROR(EDQUOT)); 3977 3978 return (0); 3979 } 3980 3981 static void 3982 dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone, 3983 dsl_dataset_t *origin, dmu_tx_t *tx) 3984 { 3985 uint64_t clone_remap_dl_obj, origin_remap_dl_obj; 3986 dsl_pool_t *dp = dmu_tx_pool(tx); 3987 3988 ASSERT(dsl_pool_sync_context(dp)); 3989 3990 clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone); 3991 origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin); 3992 3993 if (clone_remap_dl_obj != 0) { 3994 dsl_deadlist_close(&clone->ds_remap_deadlist); 3995 dsl_dataset_unset_remap_deadlist_object(clone, tx); 3996 } 3997 if (origin_remap_dl_obj != 0) { 3998 dsl_deadlist_close(&origin->ds_remap_deadlist); 3999 dsl_dataset_unset_remap_deadlist_object(origin, tx); 4000 } 4001 4002 if (clone_remap_dl_obj != 0) { 4003 dsl_dataset_set_remap_deadlist_object(origin, 4004 clone_remap_dl_obj, tx); 4005 dsl_deadlist_open(&origin->ds_remap_deadlist, 4006 dp->dp_meta_objset, clone_remap_dl_obj); 4007 } 4008 if (origin_remap_dl_obj != 0) { 4009 dsl_dataset_set_remap_deadlist_object(clone, 4010 origin_remap_dl_obj, tx); 4011 dsl_deadlist_open(&clone->ds_remap_deadlist, 4012 dp->dp_meta_objset, origin_remap_dl_obj); 4013 } 4014 } 4015 4016 void 4017 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 4018 dsl_dataset_t *origin_head, dmu_tx_t *tx) 4019 { 4020 dsl_pool_t *dp = dmu_tx_pool(tx); 4021 int64_t unused_refres_delta; 4022 4023 ASSERT(clone->ds_reserved == 0); 4024 /* 4025 * NOTE: On DEBUG kernels there could be a race between this and 4026 * the check function if spa_asize_inflation is adjusted... 4027 */ 4028 ASSERT(origin_head->ds_quota == 0 || 4029 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota + 4030 DMU_MAX_ACCESS * spa_asize_inflation); 4031 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 4032 4033 dsl_dir_cancel_waiters(origin_head->ds_dir); 4034 4035 /* 4036 * Swap per-dataset feature flags. 4037 */ 4038 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 4039 if (!(spa_feature_table[f].fi_flags & 4040 ZFEATURE_FLAG_PER_DATASET)) { 4041 ASSERT(!dsl_dataset_feature_is_active(clone, f)); 4042 ASSERT(!dsl_dataset_feature_is_active(origin_head, f)); 4043 continue; 4044 } 4045 4046 boolean_t clone_inuse = dsl_dataset_feature_is_active(clone, f); 4047 void *clone_feature = clone->ds_feature[f]; 4048 boolean_t origin_head_inuse = 4049 dsl_dataset_feature_is_active(origin_head, f); 4050 void *origin_head_feature = origin_head->ds_feature[f]; 4051 4052 if (clone_inuse) 4053 dsl_dataset_deactivate_feature_impl(clone, f, tx); 4054 if (origin_head_inuse) 4055 dsl_dataset_deactivate_feature_impl(origin_head, f, tx); 4056 4057 if (clone_inuse) { 4058 dsl_dataset_activate_feature(origin_head->ds_object, f, 4059 clone_feature, tx); 4060 origin_head->ds_feature[f] = clone_feature; 4061 } 4062 if (origin_head_inuse) { 4063 dsl_dataset_activate_feature(clone->ds_object, f, 4064 origin_head_feature, tx); 4065 clone->ds_feature[f] = origin_head_feature; 4066 } 4067 } 4068 4069 dmu_buf_will_dirty(clone->ds_dbuf, tx); 4070 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 4071 4072 if (clone->ds_objset != NULL) { 4073 dmu_objset_evict(clone->ds_objset); 4074 clone->ds_objset = NULL; 4075 } 4076 4077 if (origin_head->ds_objset != NULL) { 4078 dmu_objset_evict(origin_head->ds_objset); 4079 origin_head->ds_objset = NULL; 4080 } 4081 4082 unused_refres_delta = 4083 (int64_t)MIN(origin_head->ds_reserved, 4084 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 4085 (int64_t)MIN(origin_head->ds_reserved, 4086 dsl_dataset_phys(clone)->ds_unique_bytes); 4087 4088 /* 4089 * Reset origin's unique bytes. 4090 */ 4091 { 4092 dsl_dataset_t *origin = clone->ds_prev; 4093 uint64_t comp, uncomp; 4094 4095 dmu_buf_will_dirty(origin->ds_dbuf, tx); 4096 dsl_deadlist_space_range(&clone->ds_deadlist, 4097 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, 4098 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); 4099 } 4100 4101 /* swap blkptrs */ 4102 { 4103 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG); 4104 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG); 4105 blkptr_t tmp; 4106 tmp = dsl_dataset_phys(origin_head)->ds_bp; 4107 dsl_dataset_phys(origin_head)->ds_bp = 4108 dsl_dataset_phys(clone)->ds_bp; 4109 dsl_dataset_phys(clone)->ds_bp = tmp; 4110 rrw_exit(&origin_head->ds_bp_rwlock, FTAG); 4111 rrw_exit(&clone->ds_bp_rwlock, FTAG); 4112 } 4113 4114 /* set dd_*_bytes */ 4115 { 4116 int64_t dused, dcomp, duncomp; 4117 uint64_t cdl_used, cdl_comp, cdl_uncomp; 4118 uint64_t odl_used, odl_comp, odl_uncomp; 4119 4120 ASSERT3U(dsl_dir_phys(clone->ds_dir)-> 4121 dd_used_breakdown[DD_USED_SNAP], ==, 0); 4122 4123 dsl_deadlist_space(&clone->ds_deadlist, 4124 &cdl_used, &cdl_comp, &cdl_uncomp); 4125 dsl_deadlist_space(&origin_head->ds_deadlist, 4126 &odl_used, &odl_comp, &odl_uncomp); 4127 4128 dused = dsl_dataset_phys(clone)->ds_referenced_bytes + 4129 cdl_used - 4130 (dsl_dataset_phys(origin_head)->ds_referenced_bytes + 4131 odl_used); 4132 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + 4133 cdl_comp - 4134 (dsl_dataset_phys(origin_head)->ds_compressed_bytes + 4135 odl_comp); 4136 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + 4137 cdl_uncomp - 4138 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + 4139 odl_uncomp); 4140 4141 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 4142 dused, dcomp, duncomp, tx); 4143 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 4144 -dused, -dcomp, -duncomp, tx); 4145 4146 /* 4147 * The difference in the space used by snapshots is the 4148 * difference in snapshot space due to the head's 4149 * deadlist (since that's the only thing that's 4150 * changing that affects the snapused). 4151 */ 4152 dsl_deadlist_space_range(&clone->ds_deadlist, 4153 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 4154 &cdl_used, &cdl_comp, &cdl_uncomp); 4155 dsl_deadlist_space_range(&origin_head->ds_deadlist, 4156 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 4157 &odl_used, &odl_comp, &odl_uncomp); 4158 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 4159 DD_USED_HEAD, DD_USED_SNAP, tx); 4160 } 4161 4162 /* swap ds_*_bytes */ 4163 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, 4164 dsl_dataset_phys(clone)->ds_referenced_bytes); 4165 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, 4166 dsl_dataset_phys(clone)->ds_compressed_bytes); 4167 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, 4168 dsl_dataset_phys(clone)->ds_uncompressed_bytes); 4169 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, 4170 dsl_dataset_phys(clone)->ds_unique_bytes); 4171 4172 /* apply any parent delta for change in unconsumed refreservation */ 4173 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 4174 unused_refres_delta, 0, 0, tx); 4175 4176 /* 4177 * Swap deadlists. 4178 */ 4179 dsl_deadlist_close(&clone->ds_deadlist); 4180 dsl_deadlist_close(&origin_head->ds_deadlist); 4181 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, 4182 dsl_dataset_phys(clone)->ds_deadlist_obj); 4183 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 4184 dsl_dataset_phys(clone)->ds_deadlist_obj); 4185 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 4186 dsl_dataset_phys(origin_head)->ds_deadlist_obj); 4187 dsl_dataset_swap_remap_deadlists(clone, origin_head, tx); 4188 4189 /* 4190 * If there is a bookmark at the origin, its "next dataset" is 4191 * changing, so we need to reset its FBN. 4192 */ 4193 dsl_bookmark_next_changed(origin_head, origin_head->ds_prev, tx); 4194 4195 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 4196 4197 /* 4198 * Destroy any livelists associated with the clone or the origin, 4199 * since after the swap the corresponding livelists are no longer 4200 * valid. 4201 */ 4202 dsl_dir_remove_livelist(clone->ds_dir, tx, B_TRUE); 4203 dsl_dir_remove_livelist(origin_head->ds_dir, tx, B_TRUE); 4204 4205 spa_history_log_internal_ds(clone, "clone swap", tx, 4206 "parent=%s", origin_head->ds_dir->dd_myname); 4207 } 4208 4209 /* 4210 * Given a pool name and a dataset object number in that pool, 4211 * return the name of that dataset. 4212 */ 4213 int 4214 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 4215 { 4216 dsl_pool_t *dp; 4217 dsl_dataset_t *ds; 4218 int error; 4219 4220 error = dsl_pool_hold(pname, FTAG, &dp); 4221 if (error != 0) 4222 return (error); 4223 4224 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 4225 if (error == 0) { 4226 dsl_dataset_name(ds, buf); 4227 dsl_dataset_rele(ds, FTAG); 4228 } 4229 dsl_pool_rele(dp, FTAG); 4230 4231 return (error); 4232 } 4233 4234 int 4235 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 4236 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 4237 { 4238 int error = 0; 4239 4240 ASSERT3S(asize, >, 0); 4241 4242 /* 4243 * *ref_rsrv is the portion of asize that will come from any 4244 * unconsumed refreservation space. 4245 */ 4246 *ref_rsrv = 0; 4247 4248 mutex_enter(&ds->ds_lock); 4249 /* 4250 * Make a space adjustment for reserved bytes. 4251 */ 4252 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 4253 ASSERT3U(*used, >=, 4254 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 4255 *used -= 4256 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 4257 *ref_rsrv = 4258 asize - MIN(asize, parent_delta(ds, asize + inflight)); 4259 } 4260 4261 if (!check_quota || ds->ds_quota == 0) { 4262 mutex_exit(&ds->ds_lock); 4263 return (0); 4264 } 4265 /* 4266 * If they are requesting more space, and our current estimate 4267 * is over quota, they get to try again unless the actual 4268 * on-disk is over quota and there are no pending changes (which 4269 * may free up space for us). 4270 */ 4271 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= 4272 ds->ds_quota) { 4273 if (inflight > 0 || 4274 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) 4275 error = SET_ERROR(ERESTART); 4276 else 4277 error = SET_ERROR(EDQUOT); 4278 } 4279 mutex_exit(&ds->ds_lock); 4280 4281 return (error); 4282 } 4283 4284 typedef struct dsl_dataset_set_qr_arg { 4285 const char *ddsqra_name; 4286 zprop_source_t ddsqra_source; 4287 uint64_t ddsqra_value; 4288 } dsl_dataset_set_qr_arg_t; 4289 4290 4291 static int 4292 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 4293 { 4294 dsl_dataset_set_qr_arg_t *ddsqra = arg; 4295 dsl_pool_t *dp = dmu_tx_pool(tx); 4296 dsl_dataset_t *ds; 4297 int error; 4298 uint64_t newval; 4299 4300 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 4301 return (SET_ERROR(ENOTSUP)); 4302 4303 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 4304 if (error != 0) 4305 return (error); 4306 4307 if (ds->ds_is_snapshot) { 4308 dsl_dataset_rele(ds, FTAG); 4309 return (SET_ERROR(EINVAL)); 4310 } 4311 4312 error = dsl_prop_predict(ds->ds_dir, 4313 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 4314 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 4315 if (error != 0) { 4316 dsl_dataset_rele(ds, FTAG); 4317 return (error); 4318 } 4319 4320 if (newval == 0) { 4321 dsl_dataset_rele(ds, FTAG); 4322 return (0); 4323 } 4324 4325 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || 4326 newval < ds->ds_reserved) { 4327 dsl_dataset_rele(ds, FTAG); 4328 return (SET_ERROR(ENOSPC)); 4329 } 4330 4331 dsl_dataset_rele(ds, FTAG); 4332 return (0); 4333 } 4334 4335 static void 4336 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 4337 { 4338 dsl_dataset_set_qr_arg_t *ddsqra = arg; 4339 dsl_pool_t *dp = dmu_tx_pool(tx); 4340 dsl_dataset_t *ds = NULL; 4341 uint64_t newval; 4342 4343 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 4344 4345 dsl_prop_set_sync_impl(ds, 4346 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 4347 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 4348 &ddsqra->ddsqra_value, tx); 4349 4350 VERIFY0(dsl_prop_get_int_ds(ds, 4351 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 4352 4353 if (ds->ds_quota != newval) { 4354 dmu_buf_will_dirty(ds->ds_dbuf, tx); 4355 ds->ds_quota = newval; 4356 } 4357 dsl_dataset_rele(ds, FTAG); 4358 } 4359 4360 int 4361 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 4362 uint64_t refquota) 4363 { 4364 dsl_dataset_set_qr_arg_t ddsqra; 4365 4366 ddsqra.ddsqra_name = dsname; 4367 ddsqra.ddsqra_source = source; 4368 ddsqra.ddsqra_value = refquota; 4369 4370 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 4371 dsl_dataset_set_refquota_sync, &ddsqra, 0, 4372 ZFS_SPACE_CHECK_EXTRA_RESERVED)); 4373 } 4374 4375 static int 4376 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 4377 { 4378 dsl_dataset_set_qr_arg_t *ddsqra = arg; 4379 dsl_pool_t *dp = dmu_tx_pool(tx); 4380 dsl_dataset_t *ds; 4381 int error; 4382 uint64_t newval, unique; 4383 4384 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 4385 return (SET_ERROR(ENOTSUP)); 4386 4387 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 4388 if (error != 0) 4389 return (error); 4390 4391 if (ds->ds_is_snapshot) { 4392 dsl_dataset_rele(ds, FTAG); 4393 return (SET_ERROR(EINVAL)); 4394 } 4395 4396 error = dsl_prop_predict(ds->ds_dir, 4397 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 4398 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 4399 if (error != 0) { 4400 dsl_dataset_rele(ds, FTAG); 4401 return (error); 4402 } 4403 4404 /* 4405 * If we are doing the preliminary check in open context, the 4406 * space estimates may be inaccurate. 4407 */ 4408 if (!dmu_tx_is_syncing(tx)) { 4409 dsl_dataset_rele(ds, FTAG); 4410 return (0); 4411 } 4412 4413 mutex_enter(&ds->ds_lock); 4414 if (!DS_UNIQUE_IS_ACCURATE(ds)) 4415 dsl_dataset_recalc_head_uniq(ds); 4416 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 4417 mutex_exit(&ds->ds_lock); 4418 4419 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 4420 uint64_t delta = MAX(unique, newval) - 4421 MAX(unique, ds->ds_reserved); 4422 4423 if (delta > 4424 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 4425 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 4426 dsl_dataset_rele(ds, FTAG); 4427 return (SET_ERROR(ENOSPC)); 4428 } 4429 } 4430 4431 dsl_dataset_rele(ds, FTAG); 4432 return (0); 4433 } 4434 4435 void 4436 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 4437 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 4438 { 4439 uint64_t newval; 4440 uint64_t unique; 4441 int64_t delta; 4442 4443 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 4444 source, sizeof (value), 1, &value, tx); 4445 4446 VERIFY0(dsl_prop_get_int_ds(ds, 4447 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 4448 4449 dmu_buf_will_dirty(ds->ds_dbuf, tx); 4450 mutex_enter(&ds->ds_dir->dd_lock); 4451 mutex_enter(&ds->ds_lock); 4452 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 4453 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 4454 delta = MAX(0, (int64_t)(newval - unique)) - 4455 MAX(0, (int64_t)(ds->ds_reserved - unique)); 4456 ds->ds_reserved = newval; 4457 mutex_exit(&ds->ds_lock); 4458 4459 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 4460 mutex_exit(&ds->ds_dir->dd_lock); 4461 } 4462 4463 static void 4464 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 4465 { 4466 dsl_dataset_set_qr_arg_t *ddsqra = arg; 4467 dsl_pool_t *dp = dmu_tx_pool(tx); 4468 dsl_dataset_t *ds = NULL; 4469 4470 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 4471 dsl_dataset_set_refreservation_sync_impl(ds, 4472 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 4473 dsl_dataset_rele(ds, FTAG); 4474 } 4475 4476 int 4477 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 4478 uint64_t refreservation) 4479 { 4480 dsl_dataset_set_qr_arg_t ddsqra; 4481 4482 ddsqra.ddsqra_name = dsname; 4483 ddsqra.ddsqra_source = source; 4484 ddsqra.ddsqra_value = refreservation; 4485 4486 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 4487 dsl_dataset_set_refreservation_sync, &ddsqra, 0, 4488 ZFS_SPACE_CHECK_EXTRA_RESERVED)); 4489 } 4490 4491 typedef struct dsl_dataset_set_compression_arg { 4492 const char *ddsca_name; 4493 zprop_source_t ddsca_source; 4494 uint64_t ddsca_value; 4495 } dsl_dataset_set_compression_arg_t; 4496 4497 static int 4498 dsl_dataset_set_compression_check(void *arg, dmu_tx_t *tx) 4499 { 4500 dsl_dataset_set_compression_arg_t *ddsca = arg; 4501 dsl_pool_t *dp = dmu_tx_pool(tx); 4502 4503 uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value); 4504 spa_feature_t f = zio_compress_to_feature(compval); 4505 4506 if (f == SPA_FEATURE_NONE) 4507 return (SET_ERROR(EINVAL)); 4508 4509 if (!spa_feature_is_enabled(dp->dp_spa, f)) 4510 return (SET_ERROR(ENOTSUP)); 4511 4512 return (0); 4513 } 4514 4515 static void 4516 dsl_dataset_set_compression_sync(void *arg, dmu_tx_t *tx) 4517 { 4518 dsl_dataset_set_compression_arg_t *ddsca = arg; 4519 dsl_pool_t *dp = dmu_tx_pool(tx); 4520 dsl_dataset_t *ds = NULL; 4521 4522 uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value); 4523 spa_feature_t f = zio_compress_to_feature(compval); 4524 ASSERT3S(f, !=, SPA_FEATURE_NONE); 4525 ASSERT3S(spa_feature_table[f].fi_type, ==, ZFEATURE_TYPE_BOOLEAN); 4526 4527 VERIFY0(dsl_dataset_hold(dp, ddsca->ddsca_name, FTAG, &ds)); 4528 if (zfeature_active(f, ds->ds_feature[f]) != B_TRUE) { 4529 ds->ds_feature_activation[f] = (void *)B_TRUE; 4530 dsl_dataset_activate_feature(ds->ds_object, f, 4531 ds->ds_feature_activation[f], tx); 4532 ds->ds_feature[f] = ds->ds_feature_activation[f]; 4533 } 4534 dsl_dataset_rele(ds, FTAG); 4535 } 4536 4537 int 4538 dsl_dataset_set_compression(const char *dsname, zprop_source_t source, 4539 uint64_t compression) 4540 { 4541 dsl_dataset_set_compression_arg_t ddsca; 4542 4543 /* 4544 * The sync task is only required for zstd in order to activate 4545 * the feature flag when the property is first set. 4546 */ 4547 if (ZIO_COMPRESS_ALGO(compression) != ZIO_COMPRESS_ZSTD) 4548 return (0); 4549 4550 ddsca.ddsca_name = dsname; 4551 ddsca.ddsca_source = source; 4552 ddsca.ddsca_value = compression; 4553 4554 return (dsl_sync_task(dsname, dsl_dataset_set_compression_check, 4555 dsl_dataset_set_compression_sync, &ddsca, 0, 4556 ZFS_SPACE_CHECK_EXTRA_RESERVED)); 4557 } 4558 4559 /* 4560 * Return (in *usedp) the amount of space referenced by "new" that was not 4561 * referenced at the time the bookmark corresponds to. "New" may be a 4562 * snapshot or a head. The bookmark must be before new, in 4563 * new's filesystem (or its origin) -- caller verifies this. 4564 * 4565 * The written space is calculated by considering two components: First, we 4566 * ignore any freed space, and calculate the written as new's used space 4567 * minus old's used space. Next, we add in the amount of space that was freed 4568 * between the two time points, thus reducing new's used space relative to 4569 * old's. Specifically, this is the space that was born before 4570 * zbm_creation_txg, and freed before new (ie. on new's deadlist or a 4571 * previous deadlist). 4572 * 4573 * space freed [---------------------] 4574 * snapshots ---O-------O--------O-------O------ 4575 * bookmark new 4576 * 4577 * Note, the bookmark's zbm_*_bytes_refd must be valid, but if the HAS_FBN 4578 * flag is not set, we will calculate the freed_before_next based on the 4579 * next snapshot's deadlist, rather than using zbm_*_freed_before_next_snap. 4580 */ 4581 static int 4582 dsl_dataset_space_written_impl(zfs_bookmark_phys_t *bmp, 4583 dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4584 { 4585 int err = 0; 4586 dsl_pool_t *dp = new->ds_dir->dd_pool; 4587 4588 ASSERT(dsl_pool_config_held(dp)); 4589 if (dsl_dataset_is_snapshot(new)) { 4590 ASSERT3U(bmp->zbm_creation_txg, <, 4591 dsl_dataset_phys(new)->ds_creation_txg); 4592 } 4593 4594 *usedp = 0; 4595 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; 4596 *usedp -= bmp->zbm_referenced_bytes_refd; 4597 4598 *compp = 0; 4599 *compp += dsl_dataset_phys(new)->ds_compressed_bytes; 4600 *compp -= bmp->zbm_compressed_bytes_refd; 4601 4602 *uncompp = 0; 4603 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; 4604 *uncompp -= bmp->zbm_uncompressed_bytes_refd; 4605 4606 dsl_dataset_t *snap = new; 4607 4608 while (dsl_dataset_phys(snap)->ds_prev_snap_txg > 4609 bmp->zbm_creation_txg) { 4610 uint64_t used, comp, uncomp; 4611 4612 dsl_deadlist_space_range(&snap->ds_deadlist, 4613 0, bmp->zbm_creation_txg, 4614 &used, &comp, &uncomp); 4615 *usedp += used; 4616 *compp += comp; 4617 *uncompp += uncomp; 4618 4619 uint64_t snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 4620 if (snap != new) 4621 dsl_dataset_rele(snap, FTAG); 4622 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 4623 if (err != 0) 4624 break; 4625 } 4626 4627 /* 4628 * We might not have the FBN if we are calculating written from 4629 * a snapshot (because we didn't know the correct "next" snapshot 4630 * until now). 4631 */ 4632 if (bmp->zbm_flags & ZBM_FLAG_HAS_FBN) { 4633 *usedp += bmp->zbm_referenced_freed_before_next_snap; 4634 *compp += bmp->zbm_compressed_freed_before_next_snap; 4635 *uncompp += bmp->zbm_uncompressed_freed_before_next_snap; 4636 } else { 4637 ASSERT3U(dsl_dataset_phys(snap)->ds_prev_snap_txg, ==, 4638 bmp->zbm_creation_txg); 4639 uint64_t used, comp, uncomp; 4640 dsl_deadlist_space(&snap->ds_deadlist, &used, &comp, &uncomp); 4641 *usedp += used; 4642 *compp += comp; 4643 *uncompp += uncomp; 4644 } 4645 if (snap != new) 4646 dsl_dataset_rele(snap, FTAG); 4647 return (err); 4648 } 4649 4650 /* 4651 * Return (in *usedp) the amount of space written in new that was not 4652 * present at the time the bookmark corresponds to. New may be a 4653 * snapshot or the head. Old must be a bookmark before new, in 4654 * new's filesystem (or its origin) -- caller verifies this. 4655 */ 4656 int 4657 dsl_dataset_space_written_bookmark(zfs_bookmark_phys_t *bmp, 4658 dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4659 { 4660 if (!(bmp->zbm_flags & ZBM_FLAG_HAS_FBN)) 4661 return (SET_ERROR(ENOTSUP)); 4662 return (dsl_dataset_space_written_impl(bmp, new, 4663 usedp, compp, uncompp)); 4664 } 4665 4666 /* 4667 * Return (in *usedp) the amount of space written in new that is not 4668 * present in oldsnap. New may be a snapshot or the head. Old must be 4669 * a snapshot before new, in new's filesystem (or its origin). If not then 4670 * fail and return EINVAL. 4671 */ 4672 int 4673 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 4674 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4675 { 4676 if (!dsl_dataset_is_before(new, oldsnap, 0)) 4677 return (SET_ERROR(EINVAL)); 4678 4679 zfs_bookmark_phys_t zbm = { 0 }; 4680 dsl_dataset_phys_t *dsp = dsl_dataset_phys(oldsnap); 4681 zbm.zbm_guid = dsp->ds_guid; 4682 zbm.zbm_creation_txg = dsp->ds_creation_txg; 4683 zbm.zbm_creation_time = dsp->ds_creation_time; 4684 zbm.zbm_referenced_bytes_refd = dsp->ds_referenced_bytes; 4685 zbm.zbm_compressed_bytes_refd = dsp->ds_compressed_bytes; 4686 zbm.zbm_uncompressed_bytes_refd = dsp->ds_uncompressed_bytes; 4687 4688 /* 4689 * If oldsnap is the origin (or origin's origin, ...) of new, 4690 * we can't easily calculate the effective FBN. Therefore, 4691 * we do not set ZBM_FLAG_HAS_FBN, so that the _impl will calculate 4692 * it relative to the correct "next": the next snapshot towards "new", 4693 * rather than the next snapshot in oldsnap's dsl_dir. 4694 */ 4695 return (dsl_dataset_space_written_impl(&zbm, new, 4696 usedp, compp, uncompp)); 4697 } 4698 4699 /* 4700 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 4701 * lastsnap, and all snapshots in between are deleted. 4702 * 4703 * blocks that would be freed [---------------------------] 4704 * snapshots ---O-------O--------O-------O--------O 4705 * firstsnap lastsnap 4706 * 4707 * This is the set of blocks that were born after the snap before firstsnap, 4708 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 4709 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 4710 * We calculate this by iterating over the relevant deadlists (from the snap 4711 * after lastsnap, backward to the snap after firstsnap), summing up the 4712 * space on the deadlist that was born after the snap before firstsnap. 4713 */ 4714 int 4715 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 4716 dsl_dataset_t *lastsnap, 4717 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4718 { 4719 int err = 0; 4720 uint64_t snapobj; 4721 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 4722 4723 ASSERT(firstsnap->ds_is_snapshot); 4724 ASSERT(lastsnap->ds_is_snapshot); 4725 4726 /* 4727 * Check that the snapshots are in the same dsl_dir, and firstsnap 4728 * is before lastsnap. 4729 */ 4730 if (firstsnap->ds_dir != lastsnap->ds_dir || 4731 dsl_dataset_phys(firstsnap)->ds_creation_txg > 4732 dsl_dataset_phys(lastsnap)->ds_creation_txg) 4733 return (SET_ERROR(EINVAL)); 4734 4735 *usedp = *compp = *uncompp = 0; 4736 4737 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; 4738 while (snapobj != firstsnap->ds_object) { 4739 dsl_dataset_t *ds; 4740 uint64_t used, comp, uncomp; 4741 4742 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 4743 if (err != 0) 4744 break; 4745 4746 dsl_deadlist_space_range(&ds->ds_deadlist, 4747 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, 4748 &used, &comp, &uncomp); 4749 *usedp += used; 4750 *compp += comp; 4751 *uncompp += uncomp; 4752 4753 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 4754 ASSERT3U(snapobj, !=, 0); 4755 dsl_dataset_rele(ds, FTAG); 4756 } 4757 return (err); 4758 } 4759 4760 /* 4761 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 4762 * For example, they could both be snapshots of the same filesystem, and 4763 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 4764 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 4765 * filesystem. Or 'earlier' could be the origin's origin. 4766 * 4767 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 4768 */ 4769 boolean_t 4770 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 4771 uint64_t earlier_txg) 4772 { 4773 dsl_pool_t *dp = later->ds_dir->dd_pool; 4774 int error; 4775 boolean_t ret; 4776 4777 ASSERT(dsl_pool_config_held(dp)); 4778 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); 4779 4780 if (earlier_txg == 0) 4781 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; 4782 4783 if (later->ds_is_snapshot && 4784 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) 4785 return (B_FALSE); 4786 4787 if (later->ds_dir == earlier->ds_dir) 4788 return (B_TRUE); 4789 4790 /* 4791 * We check dd_origin_obj explicitly here rather than using 4792 * dsl_dir_is_clone() so that we will return TRUE if "earlier" 4793 * is $ORIGIN@$ORIGIN. dsl_dataset_space_written() depends on 4794 * this behavior. 4795 */ 4796 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == 0) 4797 return (B_FALSE); 4798 4799 dsl_dataset_t *origin; 4800 error = dsl_dataset_hold_obj(dp, 4801 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); 4802 if (error != 0) 4803 return (B_FALSE); 4804 if (dsl_dataset_phys(origin)->ds_creation_txg == earlier_txg && 4805 origin->ds_dir == earlier->ds_dir) { 4806 dsl_dataset_rele(origin, FTAG); 4807 return (B_TRUE); 4808 } 4809 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 4810 dsl_dataset_rele(origin, FTAG); 4811 return (ret); 4812 } 4813 4814 void 4815 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 4816 { 4817 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 4818 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 4819 } 4820 4821 boolean_t 4822 dsl_dataset_is_zapified(dsl_dataset_t *ds) 4823 { 4824 dmu_object_info_t doi; 4825 4826 dmu_object_info_from_db(ds->ds_dbuf, &doi); 4827 return (doi.doi_type == DMU_OTN_ZAP_METADATA); 4828 } 4829 4830 boolean_t 4831 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds) 4832 { 4833 return (dsl_dataset_is_zapified(ds) && 4834 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset, 4835 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0); 4836 } 4837 4838 uint64_t 4839 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds) 4840 { 4841 uint64_t remap_deadlist_obj; 4842 int err; 4843 4844 if (!dsl_dataset_is_zapified(ds)) 4845 return (0); 4846 4847 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, 4848 DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1, 4849 &remap_deadlist_obj); 4850 4851 if (err != 0) { 4852 VERIFY3S(err, ==, ENOENT); 4853 return (0); 4854 } 4855 4856 ASSERT(remap_deadlist_obj != 0); 4857 return (remap_deadlist_obj); 4858 } 4859 4860 boolean_t 4861 dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds) 4862 { 4863 EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist), 4864 dsl_dataset_get_remap_deadlist_object(ds) != 0); 4865 return (dsl_deadlist_is_open(&ds->ds_remap_deadlist)); 4866 } 4867 4868 static void 4869 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj, 4870 dmu_tx_t *tx) 4871 { 4872 ASSERT(obj != 0); 4873 dsl_dataset_zapify(ds, tx); 4874 VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, 4875 DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx)); 4876 } 4877 4878 static void 4879 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx) 4880 { 4881 VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset, 4882 ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx)); 4883 } 4884 4885 void 4886 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx) 4887 { 4888 uint64_t remap_deadlist_object; 4889 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 4890 4891 ASSERT(dmu_tx_is_syncing(tx)); 4892 ASSERT(dsl_dataset_remap_deadlist_exists(ds)); 4893 4894 remap_deadlist_object = ds->ds_remap_deadlist.dl_object; 4895 dsl_deadlist_close(&ds->ds_remap_deadlist); 4896 dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx); 4897 dsl_dataset_unset_remap_deadlist_object(ds, tx); 4898 spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); 4899 } 4900 4901 void 4902 dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx) 4903 { 4904 uint64_t remap_deadlist_obj; 4905 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 4906 4907 ASSERT(dmu_tx_is_syncing(tx)); 4908 ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock)); 4909 /* 4910 * Currently we only create remap deadlists when there are indirect 4911 * vdevs with referenced mappings. 4912 */ 4913 ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL)); 4914 4915 remap_deadlist_obj = dsl_deadlist_clone( 4916 &ds->ds_deadlist, UINT64_MAX, 4917 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 4918 dsl_dataset_set_remap_deadlist_object(ds, 4919 remap_deadlist_obj, tx); 4920 dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa), 4921 remap_deadlist_obj); 4922 spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); 4923 } 4924 4925 void 4926 dsl_dataset_activate_redaction(dsl_dataset_t *ds, uint64_t *redact_snaps, 4927 uint64_t num_redact_snaps, dmu_tx_t *tx) 4928 { 4929 uint64_t dsobj = ds->ds_object; 4930 struct feature_type_uint64_array_arg *ftuaa = 4931 kmem_zalloc(sizeof (*ftuaa), KM_SLEEP); 4932 ftuaa->length = (int64_t)num_redact_snaps; 4933 if (num_redact_snaps > 0) { 4934 ftuaa->array = kmem_alloc(num_redact_snaps * sizeof (uint64_t), 4935 KM_SLEEP); 4936 memcpy(ftuaa->array, redact_snaps, num_redact_snaps * 4937 sizeof (uint64_t)); 4938 } 4939 dsl_dataset_activate_feature(dsobj, SPA_FEATURE_REDACTED_DATASETS, 4940 ftuaa, tx); 4941 ds->ds_feature[SPA_FEATURE_REDACTED_DATASETS] = ftuaa; 4942 } 4943 4944 /* 4945 * Find and return (in *oldest_dsobj) the oldest snapshot of the dsobj 4946 * dataset whose birth time is >= min_txg. 4947 */ 4948 int 4949 dsl_dataset_oldest_snapshot(spa_t *spa, uint64_t head_ds, uint64_t min_txg, 4950 uint64_t *oldest_dsobj) 4951 { 4952 dsl_dataset_t *ds; 4953 dsl_pool_t *dp = spa->spa_dsl_pool; 4954 4955 int error = dsl_dataset_hold_obj(dp, head_ds, FTAG, &ds); 4956 if (error != 0) 4957 return (error); 4958 4959 uint64_t prev_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 4960 uint64_t prev_obj_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 4961 4962 while (prev_obj != 0 && min_txg < prev_obj_txg) { 4963 dsl_dataset_rele(ds, FTAG); 4964 if ((error = dsl_dataset_hold_obj(dp, prev_obj, 4965 FTAG, &ds)) != 0) 4966 return (error); 4967 prev_obj_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 4968 prev_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 4969 } 4970 *oldest_dsobj = ds->ds_object; 4971 dsl_dataset_rele(ds, FTAG); 4972 return (0); 4973 } 4974 4975 ZFS_MODULE_PARAM(zfs, zfs_, max_recordsize, UINT, ZMOD_RW, 4976 "Max allowed record size"); 4977 4978 ZFS_MODULE_PARAM(zfs, zfs_, allow_redacted_dataset_mount, INT, ZMOD_RW, 4979 "Allow mounting of redacted datasets"); 4980 4981 ZFS_MODULE_PARAM(zfs, zfs_, snapshot_history_enabled, INT, ZMOD_RW, 4982 "Include snapshot events in pool history/events"); 4983 4984 EXPORT_SYMBOL(dsl_dataset_hold); 4985 EXPORT_SYMBOL(dsl_dataset_hold_flags); 4986 EXPORT_SYMBOL(dsl_dataset_hold_obj); 4987 EXPORT_SYMBOL(dsl_dataset_hold_obj_flags); 4988 EXPORT_SYMBOL(dsl_dataset_own); 4989 EXPORT_SYMBOL(dsl_dataset_own_obj); 4990 EXPORT_SYMBOL(dsl_dataset_name); 4991 EXPORT_SYMBOL(dsl_dataset_rele); 4992 EXPORT_SYMBOL(dsl_dataset_rele_flags); 4993 EXPORT_SYMBOL(dsl_dataset_disown); 4994 EXPORT_SYMBOL(dsl_dataset_tryown); 4995 EXPORT_SYMBOL(dsl_dataset_create_sync); 4996 EXPORT_SYMBOL(dsl_dataset_create_sync_dd); 4997 EXPORT_SYMBOL(dsl_dataset_snapshot_check); 4998 EXPORT_SYMBOL(dsl_dataset_snapshot_sync); 4999 EXPORT_SYMBOL(dsl_dataset_promote); 5000 EXPORT_SYMBOL(dsl_dataset_user_hold); 5001 EXPORT_SYMBOL(dsl_dataset_user_release); 5002 EXPORT_SYMBOL(dsl_dataset_get_holds); 5003 EXPORT_SYMBOL(dsl_dataset_get_blkptr); 5004 EXPORT_SYMBOL(dsl_dataset_get_spa); 5005 EXPORT_SYMBOL(dsl_dataset_modified_since_snap); 5006 EXPORT_SYMBOL(dsl_dataset_space_written); 5007 EXPORT_SYMBOL(dsl_dataset_space_wouldfree); 5008 EXPORT_SYMBOL(dsl_dataset_sync); 5009 EXPORT_SYMBOL(dsl_dataset_block_born); 5010 EXPORT_SYMBOL(dsl_dataset_block_kill); 5011 EXPORT_SYMBOL(dsl_dataset_dirty); 5012 EXPORT_SYMBOL(dsl_dataset_stats); 5013 EXPORT_SYMBOL(dsl_dataset_fast_stat); 5014 EXPORT_SYMBOL(dsl_dataset_space); 5015 EXPORT_SYMBOL(dsl_dataset_fsid_guid); 5016 EXPORT_SYMBOL(dsl_dsobj_to_dsname); 5017 EXPORT_SYMBOL(dsl_dataset_check_quota); 5018 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl); 5019 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl); 5020