1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 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 do not allow the recordsize to be set larger than zfs_max_recordsize 77 * (default 1MB). Larger blocks can be created by changing this tunable, 78 * and pools with larger blocks can always be imported and used, regardless 79 * of this setting. 80 */ 81 int zfs_max_recordsize = 1 * 1024 * 1024; 82 int zfs_allow_redacted_dataset_mount = 0; 83 84 #define SWITCH64(x, y) \ 85 { \ 86 uint64_t __tmp = (x); \ 87 (x) = (y); \ 88 (y) = __tmp; \ 89 } 90 91 #define DS_REF_MAX (1ULL << 62) 92 93 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds); 94 95 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, 96 uint64_t obj, dmu_tx_t *tx); 97 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, 98 dmu_tx_t *tx); 99 100 static void unload_zfeature(dsl_dataset_t *ds, spa_feature_t f); 101 102 extern int spa_asize_inflation; 103 104 static zil_header_t zero_zil; 105 106 /* 107 * Figure out how much of this delta should be propagated to the dsl_dir 108 * layer. If there's a refreservation, that space has already been 109 * partially accounted for in our ancestors. 110 */ 111 static int64_t 112 parent_delta(dsl_dataset_t *ds, int64_t delta) 113 { 114 dsl_dataset_phys_t *ds_phys; 115 uint64_t old_bytes, new_bytes; 116 117 if (ds->ds_reserved == 0) 118 return (delta); 119 120 ds_phys = dsl_dataset_phys(ds); 121 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved); 122 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 123 124 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 125 return (new_bytes - old_bytes); 126 } 127 128 void 129 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 130 { 131 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 132 int used = bp_get_dsize_sync(spa, bp); 133 int compressed = BP_GET_PSIZE(bp); 134 int uncompressed = BP_GET_UCSIZE(bp); 135 int64_t delta; 136 spa_feature_t f; 137 138 dprintf_bp(bp, "ds=%p", ds); 139 140 ASSERT(dmu_tx_is_syncing(tx)); 141 /* It could have been compressed away to nothing */ 142 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp)) 143 return; 144 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 145 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 146 if (ds == NULL) { 147 dsl_pool_mos_diduse_space(tx->tx_pool, 148 used, compressed, uncompressed); 149 return; 150 } 151 152 ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 153 dmu_buf_will_dirty(ds->ds_dbuf, tx); 154 mutex_enter(&ds->ds_lock); 155 delta = parent_delta(ds, used); 156 dsl_dataset_phys(ds)->ds_referenced_bytes += used; 157 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed; 158 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed; 159 dsl_dataset_phys(ds)->ds_unique_bytes += used; 160 161 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) { 162 ds->ds_feature_activation[SPA_FEATURE_LARGE_BLOCKS] = 163 (void *)B_TRUE; 164 } 165 166 167 f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp)); 168 if (f != SPA_FEATURE_NONE) { 169 ASSERT3S(spa_feature_table[f].fi_type, ==, 170 ZFEATURE_TYPE_BOOLEAN); 171 ds->ds_feature_activation[f] = (void *)B_TRUE; 172 } 173 174 f = zio_compress_to_feature(BP_GET_COMPRESS(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 /* 182 * Track block for livelist, but ignore embedded blocks because 183 * they do not need to be freed. 184 */ 185 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) && 186 bp->blk_birth > ds->ds_dir->dd_origin_txg && 187 !(BP_IS_EMBEDDED(bp))) { 188 ASSERT(dsl_dir_is_clone(ds->ds_dir)); 189 ASSERT(spa_feature_is_enabled(spa, 190 SPA_FEATURE_LIVELIST)); 191 bplist_append(&ds->ds_dir->dd_pending_allocs, bp); 192 } 193 194 mutex_exit(&ds->ds_lock); 195 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 196 compressed, uncompressed, tx); 197 dsl_dir_transfer_space(ds->ds_dir, used - delta, 198 DD_USED_REFRSRV, DD_USED_HEAD, tx); 199 } 200 201 /* 202 * Called when the specified segment has been remapped, and is thus no 203 * longer referenced in the head dataset. The vdev must be indirect. 204 * 205 * If the segment is referenced by a snapshot, put it on the remap deadlist. 206 * Otherwise, add this segment to the obsolete spacemap. 207 */ 208 void 209 dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, uint64_t offset, 210 uint64_t size, uint64_t birth, dmu_tx_t *tx) 211 { 212 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 213 214 ASSERT(dmu_tx_is_syncing(tx)); 215 ASSERT(birth <= tx->tx_txg); 216 ASSERT(!ds->ds_is_snapshot); 217 218 if (birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 219 spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx); 220 } else { 221 blkptr_t fakebp; 222 dva_t *dva = &fakebp.blk_dva[0]; 223 224 ASSERT(ds != NULL); 225 226 mutex_enter(&ds->ds_remap_deadlist_lock); 227 if (!dsl_dataset_remap_deadlist_exists(ds)) { 228 dsl_dataset_create_remap_deadlist(ds, tx); 229 } 230 mutex_exit(&ds->ds_remap_deadlist_lock); 231 232 BP_ZERO(&fakebp); 233 fakebp.blk_birth = birth; 234 DVA_SET_VDEV(dva, vdev); 235 DVA_SET_OFFSET(dva, offset); 236 DVA_SET_ASIZE(dva, size); 237 dsl_deadlist_insert(&ds->ds_remap_deadlist, &fakebp, B_FALSE, 238 tx); 239 } 240 } 241 242 int 243 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 244 boolean_t async) 245 { 246 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 247 248 int used = bp_get_dsize_sync(spa, bp); 249 int compressed = BP_GET_PSIZE(bp); 250 int uncompressed = BP_GET_UCSIZE(bp); 251 252 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp)) 253 return (0); 254 255 ASSERT(dmu_tx_is_syncing(tx)); 256 ASSERT(bp->blk_birth <= tx->tx_txg); 257 258 if (ds == NULL) { 259 dsl_free(tx->tx_pool, tx->tx_txg, bp); 260 dsl_pool_mos_diduse_space(tx->tx_pool, 261 -used, -compressed, -uncompressed); 262 return (used); 263 } 264 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 265 266 ASSERT(!ds->ds_is_snapshot); 267 dmu_buf_will_dirty(ds->ds_dbuf, tx); 268 269 /* 270 * Track block for livelist, but ignore embedded blocks because 271 * they do not need to be freed. 272 */ 273 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) && 274 bp->blk_birth > ds->ds_dir->dd_origin_txg && 275 !(BP_IS_EMBEDDED(bp))) { 276 ASSERT(dsl_dir_is_clone(ds->ds_dir)); 277 ASSERT(spa_feature_is_enabled(spa, 278 SPA_FEATURE_LIVELIST)); 279 bplist_append(&ds->ds_dir->dd_pending_frees, bp); 280 } 281 282 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 283 int64_t delta; 284 285 dprintf_bp(bp, "freeing ds=%llu", (u_longlong_t)ds->ds_object); 286 dsl_free(tx->tx_pool, tx->tx_txg, bp); 287 288 mutex_enter(&ds->ds_lock); 289 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used || 290 !DS_UNIQUE_IS_ACCURATE(ds)); 291 delta = parent_delta(ds, -used); 292 dsl_dataset_phys(ds)->ds_unique_bytes -= used; 293 mutex_exit(&ds->ds_lock); 294 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 295 delta, -compressed, -uncompressed, tx); 296 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 297 DD_USED_REFRSRV, DD_USED_HEAD, tx); 298 } else { 299 dprintf_bp(bp, "putting on dead list: %s", ""); 300 if (async) { 301 /* 302 * We are here as part of zio's write done callback, 303 * which means we're a zio interrupt thread. We can't 304 * call dsl_deadlist_insert() now because it may block 305 * waiting for I/O. Instead, put bp on the deferred 306 * queue and let dsl_pool_sync() finish the job. 307 */ 308 bplist_append(&ds->ds_pending_deadlist, bp); 309 } else { 310 dsl_deadlist_insert(&ds->ds_deadlist, bp, B_FALSE, tx); 311 } 312 ASSERT3U(ds->ds_prev->ds_object, ==, 313 dsl_dataset_phys(ds)->ds_prev_snap_obj); 314 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0); 315 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 316 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 317 ds->ds_object && bp->blk_birth > 318 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) { 319 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 320 mutex_enter(&ds->ds_prev->ds_lock); 321 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used; 322 mutex_exit(&ds->ds_prev->ds_lock); 323 } 324 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 325 dsl_dir_transfer_space(ds->ds_dir, used, 326 DD_USED_HEAD, DD_USED_SNAP, tx); 327 } 328 } 329 330 dsl_bookmark_block_killed(ds, bp, tx); 331 332 mutex_enter(&ds->ds_lock); 333 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used); 334 dsl_dataset_phys(ds)->ds_referenced_bytes -= used; 335 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed); 336 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed; 337 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed); 338 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed; 339 mutex_exit(&ds->ds_lock); 340 341 return (used); 342 } 343 344 struct feature_type_uint64_array_arg { 345 uint64_t length; 346 uint64_t *array; 347 }; 348 349 static void 350 unload_zfeature(dsl_dataset_t *ds, spa_feature_t f) 351 { 352 switch (spa_feature_table[f].fi_type) { 353 case ZFEATURE_TYPE_BOOLEAN: 354 break; 355 case ZFEATURE_TYPE_UINT64_ARRAY: 356 { 357 struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f]; 358 kmem_free(ftuaa->array, ftuaa->length * sizeof (uint64_t)); 359 kmem_free(ftuaa, sizeof (*ftuaa)); 360 break; 361 } 362 default: 363 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); 364 } 365 } 366 367 static int 368 load_zfeature(objset_t *mos, dsl_dataset_t *ds, spa_feature_t f) 369 { 370 int err = 0; 371 switch (spa_feature_table[f].fi_type) { 372 case ZFEATURE_TYPE_BOOLEAN: 373 err = zap_contains(mos, ds->ds_object, 374 spa_feature_table[f].fi_guid); 375 if (err == 0) { 376 ds->ds_feature[f] = (void *)B_TRUE; 377 } else { 378 ASSERT3U(err, ==, ENOENT); 379 err = 0; 380 } 381 break; 382 case ZFEATURE_TYPE_UINT64_ARRAY: 383 { 384 uint64_t int_size, num_int; 385 uint64_t *data; 386 err = zap_length(mos, ds->ds_object, 387 spa_feature_table[f].fi_guid, &int_size, &num_int); 388 if (err != 0) { 389 ASSERT3U(err, ==, ENOENT); 390 err = 0; 391 break; 392 } 393 ASSERT3U(int_size, ==, sizeof (uint64_t)); 394 data = kmem_alloc(int_size * num_int, KM_SLEEP); 395 VERIFY0(zap_lookup(mos, ds->ds_object, 396 spa_feature_table[f].fi_guid, int_size, num_int, data)); 397 struct feature_type_uint64_array_arg *ftuaa = 398 kmem_alloc(sizeof (*ftuaa), KM_SLEEP); 399 ftuaa->length = num_int; 400 ftuaa->array = data; 401 ds->ds_feature[f] = ftuaa; 402 break; 403 } 404 default: 405 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); 406 } 407 return (err); 408 } 409 410 /* 411 * We have to release the fsid synchronously or we risk that a subsequent 412 * mount of the same dataset will fail to unique_insert the fsid. This 413 * failure would manifest itself as the fsid of this dataset changing 414 * between mounts which makes NFS clients quite unhappy. 415 */ 416 static void 417 dsl_dataset_evict_sync(void *dbu) 418 { 419 dsl_dataset_t *ds = dbu; 420 421 ASSERT(ds->ds_owner == NULL); 422 423 unique_remove(ds->ds_fsid_guid); 424 } 425 426 static void 427 dsl_dataset_evict_async(void *dbu) 428 { 429 dsl_dataset_t *ds = dbu; 430 431 ASSERT(ds->ds_owner == NULL); 432 433 ds->ds_dbuf = NULL; 434 435 if (ds->ds_objset != NULL) 436 dmu_objset_evict(ds->ds_objset); 437 438 if (ds->ds_prev) { 439 dsl_dataset_rele(ds->ds_prev, ds); 440 ds->ds_prev = NULL; 441 } 442 443 dsl_bookmark_fini_ds(ds); 444 445 bplist_destroy(&ds->ds_pending_deadlist); 446 if (dsl_deadlist_is_open(&ds->ds_deadlist)) 447 dsl_deadlist_close(&ds->ds_deadlist); 448 if (dsl_deadlist_is_open(&ds->ds_remap_deadlist)) 449 dsl_deadlist_close(&ds->ds_remap_deadlist); 450 if (ds->ds_dir) 451 dsl_dir_async_rele(ds->ds_dir, ds); 452 453 ASSERT(!list_link_active(&ds->ds_synced_link)); 454 455 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 456 if (dsl_dataset_feature_is_active(ds, f)) 457 unload_zfeature(ds, f); 458 } 459 460 list_destroy(&ds->ds_prop_cbs); 461 mutex_destroy(&ds->ds_lock); 462 mutex_destroy(&ds->ds_opening_lock); 463 mutex_destroy(&ds->ds_sendstream_lock); 464 mutex_destroy(&ds->ds_remap_deadlist_lock); 465 zfs_refcount_destroy(&ds->ds_longholds); 466 rrw_destroy(&ds->ds_bp_rwlock); 467 468 kmem_free(ds, sizeof (dsl_dataset_t)); 469 } 470 471 int 472 dsl_dataset_get_snapname(dsl_dataset_t *ds) 473 { 474 dsl_dataset_phys_t *headphys; 475 int err; 476 dmu_buf_t *headdbuf; 477 dsl_pool_t *dp = ds->ds_dir->dd_pool; 478 objset_t *mos = dp->dp_meta_objset; 479 480 if (ds->ds_snapname[0]) 481 return (0); 482 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) 483 return (0); 484 485 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, 486 FTAG, &headdbuf); 487 if (err != 0) 488 return (err); 489 headphys = headdbuf->db_data; 490 err = zap_value_search(dp->dp_meta_objset, 491 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 492 if (err != 0 && zfs_recover == B_TRUE) { 493 err = 0; 494 (void) snprintf(ds->ds_snapname, sizeof (ds->ds_snapname), 495 "SNAPOBJ=%llu-ERR=%d", 496 (unsigned long long)ds->ds_object, err); 497 } 498 dmu_buf_rele(headdbuf, FTAG); 499 return (err); 500 } 501 502 int 503 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 504 { 505 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 506 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 507 matchtype_t mt = 0; 508 int err; 509 510 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 511 mt = MT_NORMALIZE; 512 513 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 514 value, mt, NULL, 0, NULL); 515 if (err == ENOTSUP && (mt & MT_NORMALIZE)) 516 err = zap_lookup(mos, snapobj, name, 8, 1, value); 517 return (err); 518 } 519 520 int 521 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 522 boolean_t adj_cnt) 523 { 524 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 525 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 526 matchtype_t mt = 0; 527 int err; 528 529 dsl_dir_snap_cmtime_update(ds->ds_dir); 530 531 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 532 mt = MT_NORMALIZE; 533 534 err = zap_remove_norm(mos, snapobj, name, mt, tx); 535 if (err == ENOTSUP && (mt & MT_NORMALIZE)) 536 err = zap_remove(mos, snapobj, name, tx); 537 538 if (err == 0 && adj_cnt) 539 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 540 DD_FIELD_SNAPSHOT_COUNT, tx); 541 542 return (err); 543 } 544 545 boolean_t 546 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag) 547 { 548 dmu_buf_t *dbuf = ds->ds_dbuf; 549 boolean_t result = B_FALSE; 550 551 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset, 552 ds->ds_object, DMU_BONUS_BLKID, tag)) { 553 554 if (ds == dmu_buf_get_user(dbuf)) 555 result = B_TRUE; 556 else 557 dmu_buf_rele(dbuf, tag); 558 } 559 560 return (result); 561 } 562 563 int 564 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 565 dsl_dataset_t **dsp) 566 { 567 objset_t *mos = dp->dp_meta_objset; 568 dmu_buf_t *dbuf; 569 dsl_dataset_t *ds; 570 int err; 571 dmu_object_info_t doi; 572 573 ASSERT(dsl_pool_config_held(dp)); 574 575 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 576 if (err != 0) 577 return (err); 578 579 /* Make sure dsobj has the correct object type. */ 580 dmu_object_info_from_db(dbuf, &doi); 581 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 582 dmu_buf_rele(dbuf, tag); 583 return (SET_ERROR(EINVAL)); 584 } 585 586 ds = dmu_buf_get_user(dbuf); 587 if (ds == NULL) { 588 dsl_dataset_t *winner = NULL; 589 590 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 591 ds->ds_dbuf = dbuf; 592 ds->ds_object = dsobj; 593 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0; 594 list_link_init(&ds->ds_synced_link); 595 596 err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj, 597 NULL, ds, &ds->ds_dir); 598 if (err != 0) { 599 kmem_free(ds, sizeof (dsl_dataset_t)); 600 dmu_buf_rele(dbuf, tag); 601 return (err); 602 } 603 604 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 605 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 606 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 607 mutex_init(&ds->ds_remap_deadlist_lock, 608 NULL, MUTEX_DEFAULT, NULL); 609 rrw_init(&ds->ds_bp_rwlock, B_FALSE); 610 zfs_refcount_create(&ds->ds_longholds); 611 612 bplist_create(&ds->ds_pending_deadlist); 613 614 list_create(&ds->ds_sendstreams, sizeof (dmu_sendstatus_t), 615 offsetof(dmu_sendstatus_t, dss_link)); 616 617 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t), 618 offsetof(dsl_prop_cb_record_t, cbr_ds_node)); 619 620 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 621 spa_feature_t f; 622 623 for (f = 0; f < SPA_FEATURES; f++) { 624 if (!(spa_feature_table[f].fi_flags & 625 ZFEATURE_FLAG_PER_DATASET)) 626 continue; 627 err = load_zfeature(mos, ds, f); 628 } 629 } 630 631 if (!ds->ds_is_snapshot) { 632 ds->ds_snapname[0] = '\0'; 633 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 634 err = dsl_dataset_hold_obj(dp, 635 dsl_dataset_phys(ds)->ds_prev_snap_obj, 636 ds, &ds->ds_prev); 637 } 638 err = dsl_bookmark_init_ds(ds); 639 } else { 640 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 641 err = dsl_dataset_get_snapname(ds); 642 if (err == 0 && 643 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { 644 err = zap_count( 645 ds->ds_dir->dd_pool->dp_meta_objset, 646 dsl_dataset_phys(ds)->ds_userrefs_obj, 647 &ds->ds_userrefs); 648 } 649 } 650 651 if (err == 0 && !ds->ds_is_snapshot) { 652 err = dsl_prop_get_int_ds(ds, 653 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 654 &ds->ds_reserved); 655 if (err == 0) { 656 err = dsl_prop_get_int_ds(ds, 657 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 658 &ds->ds_quota); 659 } 660 } else { 661 ds->ds_reserved = ds->ds_quota = 0; 662 } 663 664 if (err == 0 && ds->ds_dir->dd_crypto_obj != 0 && 665 ds->ds_is_snapshot && 666 zap_contains(mos, dsobj, DS_FIELD_IVSET_GUID) != 0) { 667 dp->dp_spa->spa_errata = 668 ZPOOL_ERRATA_ZOL_8308_ENCRYPTION; 669 } 670 671 dsl_deadlist_open(&ds->ds_deadlist, 672 mos, dsl_dataset_phys(ds)->ds_deadlist_obj); 673 uint64_t remap_deadlist_obj = 674 dsl_dataset_get_remap_deadlist_object(ds); 675 if (remap_deadlist_obj != 0) { 676 dsl_deadlist_open(&ds->ds_remap_deadlist, mos, 677 remap_deadlist_obj); 678 } 679 680 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync, 681 dsl_dataset_evict_async, &ds->ds_dbuf); 682 if (err == 0) 683 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu); 684 685 if (err != 0 || winner != NULL) { 686 bplist_destroy(&ds->ds_pending_deadlist); 687 dsl_deadlist_close(&ds->ds_deadlist); 688 if (dsl_deadlist_is_open(&ds->ds_remap_deadlist)) 689 dsl_deadlist_close(&ds->ds_remap_deadlist); 690 dsl_bookmark_fini_ds(ds); 691 if (ds->ds_prev) 692 dsl_dataset_rele(ds->ds_prev, ds); 693 dsl_dir_rele(ds->ds_dir, ds); 694 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 695 if (dsl_dataset_feature_is_active(ds, f)) 696 unload_zfeature(ds, f); 697 } 698 699 list_destroy(&ds->ds_prop_cbs); 700 list_destroy(&ds->ds_sendstreams); 701 mutex_destroy(&ds->ds_lock); 702 mutex_destroy(&ds->ds_opening_lock); 703 mutex_destroy(&ds->ds_sendstream_lock); 704 mutex_destroy(&ds->ds_remap_deadlist_lock); 705 zfs_refcount_destroy(&ds->ds_longholds); 706 rrw_destroy(&ds->ds_bp_rwlock); 707 kmem_free(ds, sizeof (dsl_dataset_t)); 708 if (err != 0) { 709 dmu_buf_rele(dbuf, tag); 710 return (err); 711 } 712 ds = winner; 713 } else { 714 ds->ds_fsid_guid = 715 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); 716 if (ds->ds_fsid_guid != 717 dsl_dataset_phys(ds)->ds_fsid_guid) { 718 zfs_dbgmsg("ds_fsid_guid changed from " 719 "%llx to %llx for pool %s dataset id %llu", 720 (long long) 721 dsl_dataset_phys(ds)->ds_fsid_guid, 722 (long long)ds->ds_fsid_guid, 723 spa_name(dp->dp_spa), 724 (u_longlong_t)dsobj); 725 } 726 } 727 } 728 729 ASSERT3P(ds->ds_dbuf, ==, dbuf); 730 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); 731 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || 732 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 733 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 734 *dsp = ds; 735 736 return (0); 737 } 738 739 int 740 dsl_dataset_create_key_mapping(dsl_dataset_t *ds) 741 { 742 dsl_dir_t *dd = ds->ds_dir; 743 744 if (dd->dd_crypto_obj == 0) 745 return (0); 746 747 return (spa_keystore_create_mapping(dd->dd_pool->dp_spa, 748 ds, ds, &ds->ds_key_mapping)); 749 } 750 751 int 752 dsl_dataset_hold_obj_flags(dsl_pool_t *dp, uint64_t dsobj, 753 ds_hold_flags_t flags, void *tag, dsl_dataset_t **dsp) 754 { 755 int err; 756 757 err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 758 if (err != 0) 759 return (err); 760 761 ASSERT3P(*dsp, !=, NULL); 762 763 if (flags & DS_HOLD_FLAG_DECRYPT) { 764 err = dsl_dataset_create_key_mapping(*dsp); 765 if (err != 0) 766 dsl_dataset_rele(*dsp, tag); 767 } 768 769 return (err); 770 } 771 772 int 773 dsl_dataset_hold_flags(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, 774 void *tag, dsl_dataset_t **dsp) 775 { 776 dsl_dir_t *dd; 777 const char *snapname; 778 uint64_t obj; 779 int err = 0; 780 dsl_dataset_t *ds; 781 782 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 783 if (err != 0) 784 return (err); 785 786 ASSERT(dsl_pool_config_held(dp)); 787 obj = dsl_dir_phys(dd)->dd_head_dataset_obj; 788 if (obj != 0) 789 err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, &ds); 790 else 791 err = SET_ERROR(ENOENT); 792 793 /* we may be looking for a snapshot */ 794 if (err == 0 && snapname != NULL) { 795 dsl_dataset_t *snap_ds; 796 797 if (*snapname++ != '@') { 798 dsl_dataset_rele_flags(ds, flags, tag); 799 dsl_dir_rele(dd, FTAG); 800 return (SET_ERROR(ENOENT)); 801 } 802 803 dprintf("looking for snapshot '%s'\n", snapname); 804 err = dsl_dataset_snap_lookup(ds, snapname, &obj); 805 if (err == 0) { 806 err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, 807 &snap_ds); 808 } 809 dsl_dataset_rele_flags(ds, flags, tag); 810 811 if (err == 0) { 812 mutex_enter(&snap_ds->ds_lock); 813 if (snap_ds->ds_snapname[0] == 0) 814 (void) strlcpy(snap_ds->ds_snapname, snapname, 815 sizeof (snap_ds->ds_snapname)); 816 mutex_exit(&snap_ds->ds_lock); 817 ds = snap_ds; 818 } 819 } 820 if (err == 0) 821 *dsp = ds; 822 dsl_dir_rele(dd, FTAG); 823 return (err); 824 } 825 826 int 827 dsl_dataset_hold(dsl_pool_t *dp, const char *name, void *tag, 828 dsl_dataset_t **dsp) 829 { 830 return (dsl_dataset_hold_flags(dp, name, 0, tag, dsp)); 831 } 832 833 static int 834 dsl_dataset_own_obj_impl(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags, 835 void *tag, boolean_t override, dsl_dataset_t **dsp) 836 { 837 int err = dsl_dataset_hold_obj_flags(dp, dsobj, flags, tag, dsp); 838 if (err != 0) 839 return (err); 840 if (!dsl_dataset_tryown(*dsp, tag, override)) { 841 dsl_dataset_rele_flags(*dsp, flags, tag); 842 *dsp = NULL; 843 return (SET_ERROR(EBUSY)); 844 } 845 return (0); 846 } 847 848 849 int 850 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags, 851 void *tag, dsl_dataset_t **dsp) 852 { 853 return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_FALSE, dsp)); 854 } 855 856 int 857 dsl_dataset_own_obj_force(dsl_pool_t *dp, uint64_t dsobj, 858 ds_hold_flags_t flags, void *tag, dsl_dataset_t **dsp) 859 { 860 return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_TRUE, dsp)); 861 } 862 863 static int 864 dsl_dataset_own_impl(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, 865 void *tag, boolean_t override, dsl_dataset_t **dsp) 866 { 867 int err = dsl_dataset_hold_flags(dp, name, flags, tag, dsp); 868 if (err != 0) 869 return (err); 870 if (!dsl_dataset_tryown(*dsp, tag, override)) { 871 dsl_dataset_rele_flags(*dsp, flags, tag); 872 return (SET_ERROR(EBUSY)); 873 } 874 return (0); 875 } 876 877 int 878 dsl_dataset_own_force(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, 879 void *tag, dsl_dataset_t **dsp) 880 { 881 return (dsl_dataset_own_impl(dp, name, flags, tag, B_TRUE, dsp)); 882 } 883 884 int 885 dsl_dataset_own(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, 886 void *tag, dsl_dataset_t **dsp) 887 { 888 return (dsl_dataset_own_impl(dp, name, flags, tag, B_FALSE, dsp)); 889 } 890 891 /* 892 * See the comment above dsl_pool_hold() for details. In summary, a long 893 * hold is used to prevent destruction of a dataset while the pool hold 894 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 895 * 896 * The dataset and pool must be held when this function is called. After it 897 * is called, the pool hold may be released while the dataset is still held 898 * and accessed. 899 */ 900 void 901 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 902 { 903 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 904 (void) zfs_refcount_add(&ds->ds_longholds, tag); 905 } 906 907 void 908 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 909 { 910 (void) zfs_refcount_remove(&ds->ds_longholds, tag); 911 } 912 913 /* Return B_TRUE if there are any long holds on this dataset. */ 914 boolean_t 915 dsl_dataset_long_held(dsl_dataset_t *ds) 916 { 917 return (!zfs_refcount_is_zero(&ds->ds_longholds)); 918 } 919 920 void 921 dsl_dataset_name(dsl_dataset_t *ds, char *name) 922 { 923 if (ds == NULL) { 924 (void) strlcpy(name, "mos", ZFS_MAX_DATASET_NAME_LEN); 925 } else { 926 dsl_dir_name(ds->ds_dir, name); 927 VERIFY0(dsl_dataset_get_snapname(ds)); 928 if (ds->ds_snapname[0]) { 929 VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN), 930 <, ZFS_MAX_DATASET_NAME_LEN); 931 /* 932 * We use a "recursive" mutex so that we 933 * can call dprintf_ds() with ds_lock held. 934 */ 935 if (!MUTEX_HELD(&ds->ds_lock)) { 936 mutex_enter(&ds->ds_lock); 937 VERIFY3U(strlcat(name, ds->ds_snapname, 938 ZFS_MAX_DATASET_NAME_LEN), <, 939 ZFS_MAX_DATASET_NAME_LEN); 940 mutex_exit(&ds->ds_lock); 941 } else { 942 VERIFY3U(strlcat(name, ds->ds_snapname, 943 ZFS_MAX_DATASET_NAME_LEN), <, 944 ZFS_MAX_DATASET_NAME_LEN); 945 } 946 } 947 } 948 } 949 950 int 951 dsl_dataset_namelen(dsl_dataset_t *ds) 952 { 953 VERIFY0(dsl_dataset_get_snapname(ds)); 954 mutex_enter(&ds->ds_lock); 955 int len = strlen(ds->ds_snapname); 956 mutex_exit(&ds->ds_lock); 957 /* add '@' if ds is a snap */ 958 if (len > 0) 959 len++; 960 len += dsl_dir_namelen(ds->ds_dir); 961 return (len); 962 } 963 964 void 965 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 966 { 967 dmu_buf_rele(ds->ds_dbuf, tag); 968 } 969 970 void 971 dsl_dataset_remove_key_mapping(dsl_dataset_t *ds) 972 { 973 dsl_dir_t *dd = ds->ds_dir; 974 975 if (dd == NULL || dd->dd_crypto_obj == 0) 976 return; 977 978 (void) spa_keystore_remove_mapping(dd->dd_pool->dp_spa, 979 ds->ds_object, ds); 980 } 981 982 void 983 dsl_dataset_rele_flags(dsl_dataset_t *ds, ds_hold_flags_t flags, void *tag) 984 { 985 if (flags & DS_HOLD_FLAG_DECRYPT) 986 dsl_dataset_remove_key_mapping(ds); 987 988 dsl_dataset_rele(ds, tag); 989 } 990 991 void 992 dsl_dataset_disown(dsl_dataset_t *ds, ds_hold_flags_t flags, void *tag) 993 { 994 ASSERT3P(ds->ds_owner, ==, tag); 995 ASSERT(ds->ds_dbuf != NULL); 996 997 mutex_enter(&ds->ds_lock); 998 ds->ds_owner = NULL; 999 mutex_exit(&ds->ds_lock); 1000 dsl_dataset_long_rele(ds, tag); 1001 dsl_dataset_rele_flags(ds, flags, tag); 1002 } 1003 1004 boolean_t 1005 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag, boolean_t override) 1006 { 1007 boolean_t gotit = FALSE; 1008 1009 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1010 mutex_enter(&ds->ds_lock); 1011 if (ds->ds_owner == NULL && (override || !(DS_IS_INCONSISTENT(ds) || 1012 (dsl_dataset_feature_is_active(ds, 1013 SPA_FEATURE_REDACTED_DATASETS) && 1014 !zfs_allow_redacted_dataset_mount)))) { 1015 ds->ds_owner = tag; 1016 dsl_dataset_long_hold(ds, tag); 1017 gotit = TRUE; 1018 } 1019 mutex_exit(&ds->ds_lock); 1020 return (gotit); 1021 } 1022 1023 boolean_t 1024 dsl_dataset_has_owner(dsl_dataset_t *ds) 1025 { 1026 boolean_t rv; 1027 mutex_enter(&ds->ds_lock); 1028 rv = (ds->ds_owner != NULL); 1029 mutex_exit(&ds->ds_lock); 1030 return (rv); 1031 } 1032 1033 static boolean_t 1034 zfeature_active(spa_feature_t f, void *arg) 1035 { 1036 switch (spa_feature_table[f].fi_type) { 1037 case ZFEATURE_TYPE_BOOLEAN: { 1038 boolean_t val = (boolean_t)(uintptr_t)arg; 1039 ASSERT(val == B_FALSE || val == B_TRUE); 1040 return (val); 1041 } 1042 case ZFEATURE_TYPE_UINT64_ARRAY: 1043 /* 1044 * In this case, arg is a uint64_t array. The feature is active 1045 * if the array is non-null. 1046 */ 1047 return (arg != NULL); 1048 default: 1049 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); 1050 return (B_FALSE); 1051 } 1052 } 1053 1054 boolean_t 1055 dsl_dataset_feature_is_active(dsl_dataset_t *ds, spa_feature_t f) 1056 { 1057 return (zfeature_active(f, ds->ds_feature[f])); 1058 } 1059 1060 /* 1061 * The buffers passed out by this function are references to internal buffers; 1062 * they should not be freed by callers of this function, and they should not be 1063 * used after the dataset has been released. 1064 */ 1065 boolean_t 1066 dsl_dataset_get_uint64_array_feature(dsl_dataset_t *ds, spa_feature_t f, 1067 uint64_t *outlength, uint64_t **outp) 1068 { 1069 VERIFY(spa_feature_table[f].fi_type & ZFEATURE_TYPE_UINT64_ARRAY); 1070 if (!dsl_dataset_feature_is_active(ds, f)) { 1071 return (B_FALSE); 1072 } 1073 struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f]; 1074 *outp = ftuaa->array; 1075 *outlength = ftuaa->length; 1076 return (B_TRUE); 1077 } 1078 1079 void 1080 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, void *arg, 1081 dmu_tx_t *tx) 1082 { 1083 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1084 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 1085 uint64_t zero = 0; 1086 1087 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 1088 1089 spa_feature_incr(spa, f, tx); 1090 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 1091 1092 switch (spa_feature_table[f].fi_type) { 1093 case ZFEATURE_TYPE_BOOLEAN: 1094 ASSERT3S((boolean_t)(uintptr_t)arg, ==, B_TRUE); 1095 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, 1096 sizeof (zero), 1, &zero, tx)); 1097 break; 1098 case ZFEATURE_TYPE_UINT64_ARRAY: 1099 { 1100 struct feature_type_uint64_array_arg *ftuaa = arg; 1101 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, 1102 sizeof (uint64_t), ftuaa->length, ftuaa->array, tx)); 1103 break; 1104 } 1105 default: 1106 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); 1107 } 1108 } 1109 1110 static void 1111 dsl_dataset_deactivate_feature_impl(dsl_dataset_t *ds, spa_feature_t f, 1112 dmu_tx_t *tx) 1113 { 1114 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1115 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 1116 uint64_t dsobj = ds->ds_object; 1117 1118 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 1119 1120 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx)); 1121 spa_feature_decr(spa, f, tx); 1122 ds->ds_feature[f] = NULL; 1123 } 1124 1125 void 1126 dsl_dataset_deactivate_feature(dsl_dataset_t *ds, spa_feature_t f, dmu_tx_t *tx) 1127 { 1128 unload_zfeature(ds, f); 1129 dsl_dataset_deactivate_feature_impl(ds, f, tx); 1130 } 1131 1132 uint64_t 1133 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 1134 dsl_crypto_params_t *dcp, uint64_t flags, dmu_tx_t *tx) 1135 { 1136 dsl_pool_t *dp = dd->dd_pool; 1137 dmu_buf_t *dbuf; 1138 dsl_dataset_phys_t *dsphys; 1139 uint64_t dsobj; 1140 objset_t *mos = dp->dp_meta_objset; 1141 1142 if (origin == NULL) 1143 origin = dp->dp_origin_snap; 1144 1145 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 1146 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); 1147 ASSERT(dmu_tx_is_syncing(tx)); 1148 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 1149 1150 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1151 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1152 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1153 dmu_buf_will_dirty(dbuf, tx); 1154 dsphys = dbuf->db_data; 1155 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1156 dsphys->ds_dir_obj = dd->dd_object; 1157 dsphys->ds_flags = flags; 1158 dsphys->ds_fsid_guid = unique_create(); 1159 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1160 sizeof (dsphys->ds_guid)); 1161 dsphys->ds_snapnames_zapobj = 1162 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 1163 DMU_OT_NONE, 0, tx); 1164 dsphys->ds_creation_time = gethrestime_sec(); 1165 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 1166 1167 if (origin == NULL) { 1168 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 1169 } else { 1170 dsl_dataset_t *ohds; /* head of the origin snapshot */ 1171 1172 dsphys->ds_prev_snap_obj = origin->ds_object; 1173 dsphys->ds_prev_snap_txg = 1174 dsl_dataset_phys(origin)->ds_creation_txg; 1175 dsphys->ds_referenced_bytes = 1176 dsl_dataset_phys(origin)->ds_referenced_bytes; 1177 dsphys->ds_compressed_bytes = 1178 dsl_dataset_phys(origin)->ds_compressed_bytes; 1179 dsphys->ds_uncompressed_bytes = 1180 dsl_dataset_phys(origin)->ds_uncompressed_bytes; 1181 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG); 1182 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; 1183 rrw_exit(&origin->ds_bp_rwlock, FTAG); 1184 1185 /* 1186 * Inherit flags that describe the dataset's contents 1187 * (INCONSISTENT) or properties (Case Insensitive). 1188 */ 1189 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & 1190 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 1191 1192 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1193 if (zfeature_active(f, origin->ds_feature[f])) { 1194 dsl_dataset_activate_feature(dsobj, f, 1195 origin->ds_feature[f], tx); 1196 } 1197 } 1198 1199 dmu_buf_will_dirty(origin->ds_dbuf, tx); 1200 dsl_dataset_phys(origin)->ds_num_children++; 1201 1202 VERIFY0(dsl_dataset_hold_obj(dp, 1203 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, 1204 FTAG, &ohds)); 1205 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 1206 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 1207 dsl_dataset_rele(ohds, FTAG); 1208 1209 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 1210 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { 1211 dsl_dataset_phys(origin)->ds_next_clones_obj = 1212 zap_create(mos, 1213 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 1214 } 1215 VERIFY0(zap_add_int(mos, 1216 dsl_dataset_phys(origin)->ds_next_clones_obj, 1217 dsobj, tx)); 1218 } 1219 1220 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1221 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; 1222 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 1223 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 1224 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 1225 dsl_dir_phys(origin->ds_dir)->dd_clones = 1226 zap_create(mos, 1227 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 1228 } 1229 VERIFY0(zap_add_int(mos, 1230 dsl_dir_phys(origin->ds_dir)->dd_clones, 1231 dsobj, tx)); 1232 } 1233 } 1234 1235 /* handle encryption */ 1236 dsl_dataset_create_crypt_sync(dsobj, dd, origin, dcp, tx); 1237 1238 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1239 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1240 1241 dmu_buf_rele(dbuf, FTAG); 1242 1243 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1244 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; 1245 1246 return (dsobj); 1247 } 1248 1249 static void 1250 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 1251 { 1252 objset_t *os; 1253 1254 VERIFY0(dmu_objset_from_ds(ds, &os)); 1255 if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) { 1256 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1257 zio_t *zio; 1258 1259 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 1260 if (os->os_encrypted) 1261 os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_TRUE; 1262 1263 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1264 dsl_dataset_sync(ds, zio, tx); 1265 VERIFY0(zio_wait(zio)); 1266 1267 /* dsl_dataset_sync_done will drop this reference. */ 1268 dmu_buf_add_ref(ds->ds_dbuf, ds); 1269 dsl_dataset_sync_done(ds, tx); 1270 } 1271 } 1272 1273 uint64_t 1274 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 1275 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, 1276 dsl_crypto_params_t *dcp, dmu_tx_t *tx) 1277 { 1278 dsl_pool_t *dp = pdd->dd_pool; 1279 uint64_t dsobj, ddobj; 1280 dsl_dir_t *dd; 1281 1282 ASSERT(dmu_tx_is_syncing(tx)); 1283 ASSERT(lastname[0] != '@'); 1284 /* 1285 * Filesystems will eventually have their origin set to dp_origin_snap, 1286 * but that's taken care of in dsl_dataset_create_sync_dd. When 1287 * creating a filesystem, this function is called with origin equal to 1288 * NULL. 1289 */ 1290 if (origin != NULL) 1291 ASSERT3P(origin, !=, dp->dp_origin_snap); 1292 1293 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 1294 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 1295 1296 dsobj = dsl_dataset_create_sync_dd(dd, origin, dcp, 1297 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 1298 1299 dsl_deleg_set_create_perms(dd, tx, cr); 1300 1301 /* 1302 * If we are creating a clone and the livelist feature is enabled, 1303 * add the entry DD_FIELD_LIVELIST to ZAP. 1304 */ 1305 if (origin != NULL && 1306 spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LIVELIST)) { 1307 objset_t *mos = dd->dd_pool->dp_meta_objset; 1308 dsl_dir_zapify(dd, tx); 1309 uint64_t obj = dsl_deadlist_alloc(mos, tx); 1310 VERIFY0(zap_add(mos, dd->dd_object, DD_FIELD_LIVELIST, 1311 sizeof (uint64_t), 1, &obj, tx)); 1312 spa_feature_incr(dp->dp_spa, SPA_FEATURE_LIVELIST, tx); 1313 } 1314 1315 /* 1316 * Since we're creating a new node we know it's a leaf, so we can 1317 * initialize the counts if the limit feature is active. 1318 */ 1319 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 1320 uint64_t cnt = 0; 1321 objset_t *os = dd->dd_pool->dp_meta_objset; 1322 1323 dsl_dir_zapify(dd, tx); 1324 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 1325 sizeof (cnt), 1, &cnt, tx)); 1326 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 1327 sizeof (cnt), 1, &cnt, tx)); 1328 } 1329 1330 dsl_dir_rele(dd, FTAG); 1331 1332 /* 1333 * If we are creating a clone, make sure we zero out any stale 1334 * data from the origin snapshots zil header. 1335 */ 1336 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 1337 dsl_dataset_t *ds; 1338 1339 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 1340 dsl_dataset_zero_zil(ds, tx); 1341 dsl_dataset_rele(ds, FTAG); 1342 } 1343 1344 return (dsobj); 1345 } 1346 1347 /* 1348 * The unique space in the head dataset can be calculated by subtracting 1349 * the space used in the most recent snapshot, that is still being used 1350 * in this file system, from the space currently in use. To figure out 1351 * the space in the most recent snapshot still in use, we need to take 1352 * the total space used in the snapshot and subtract out the space that 1353 * has been freed up since the snapshot was taken. 1354 */ 1355 void 1356 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 1357 { 1358 uint64_t mrs_used; 1359 uint64_t dlused, dlcomp, dluncomp; 1360 1361 ASSERT(!ds->ds_is_snapshot); 1362 1363 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) 1364 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; 1365 else 1366 mrs_used = 0; 1367 1368 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 1369 1370 ASSERT3U(dlused, <=, mrs_used); 1371 dsl_dataset_phys(ds)->ds_unique_bytes = 1372 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); 1373 1374 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1375 SPA_VERSION_UNIQUE_ACCURATE) 1376 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1377 } 1378 1379 void 1380 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 1381 dmu_tx_t *tx) 1382 { 1383 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1384 uint64_t count __maybe_unused; 1385 int err; 1386 1387 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); 1388 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1389 obj, tx); 1390 /* 1391 * The err should not be ENOENT, but a bug in a previous version 1392 * of the code could cause upgrade_clones_cb() to not set 1393 * ds_next_snap_obj when it should, leading to a missing entry. 1394 * If we knew that the pool was created after 1395 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 1396 * ENOENT. However, at least we can check that we don't have 1397 * too many entries in the next_clones_obj even after failing to 1398 * remove this one. 1399 */ 1400 if (err != ENOENT) 1401 VERIFY0(err); 1402 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1403 &count)); 1404 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); 1405 } 1406 1407 1408 blkptr_t * 1409 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1410 { 1411 return (&dsl_dataset_phys(ds)->ds_bp); 1412 } 1413 1414 spa_t * 1415 dsl_dataset_get_spa(dsl_dataset_t *ds) 1416 { 1417 return (ds->ds_dir->dd_pool->dp_spa); 1418 } 1419 1420 void 1421 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1422 { 1423 dsl_pool_t *dp; 1424 1425 if (ds == NULL) /* this is the meta-objset */ 1426 return; 1427 1428 ASSERT(ds->ds_objset != NULL); 1429 1430 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) 1431 panic("dirtying snapshot!"); 1432 1433 /* Must not dirty a dataset in the same txg where it got snapshotted. */ 1434 ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 1435 1436 dp = ds->ds_dir->dd_pool; 1437 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1438 objset_t *os = ds->ds_objset; 1439 1440 /* up the hold count until we can be written out */ 1441 dmu_buf_add_ref(ds->ds_dbuf, ds); 1442 1443 /* if this dataset is encrypted, grab a reference to the DCK */ 1444 if (ds->ds_dir->dd_crypto_obj != 0 && 1445 !os->os_raw_receive && 1446 !os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { 1447 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1448 key_mapping_add_ref(ds->ds_key_mapping, ds); 1449 } 1450 } 1451 } 1452 1453 static int 1454 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1455 { 1456 uint64_t asize; 1457 1458 if (!dmu_tx_is_syncing(tx)) 1459 return (0); 1460 1461 /* 1462 * If there's an fs-only reservation, any blocks that might become 1463 * owned by the snapshot dataset must be accommodated by space 1464 * outside of the reservation. 1465 */ 1466 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1467 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); 1468 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1469 return (SET_ERROR(ENOSPC)); 1470 1471 /* 1472 * Propagate any reserved space for this snapshot to other 1473 * snapshot checks in this sync group. 1474 */ 1475 if (asize > 0) 1476 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1477 1478 return (0); 1479 } 1480 1481 int 1482 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1483 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr, proc_t *proc) 1484 { 1485 int error; 1486 uint64_t value; 1487 1488 ds->ds_trysnap_txg = tx->tx_txg; 1489 1490 if (!dmu_tx_is_syncing(tx)) 1491 return (0); 1492 1493 /* 1494 * We don't allow multiple snapshots of the same txg. If there 1495 * is already one, try again. 1496 */ 1497 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) 1498 return (SET_ERROR(EAGAIN)); 1499 1500 /* 1501 * Check for conflicting snapshot name. 1502 */ 1503 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1504 if (error == 0) 1505 return (SET_ERROR(EEXIST)); 1506 if (error != ENOENT) 1507 return (error); 1508 1509 /* 1510 * We don't allow taking snapshots of inconsistent datasets, such as 1511 * those into which we are currently receiving. However, if we are 1512 * creating this snapshot as part of a receive, this check will be 1513 * executed atomically with respect to the completion of the receive 1514 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1515 * case we ignore this, knowing it will be fixed up for us shortly in 1516 * dmu_recv_end_sync(). 1517 */ 1518 if (!recv && DS_IS_INCONSISTENT(ds)) 1519 return (SET_ERROR(EBUSY)); 1520 1521 /* 1522 * Skip the check for temporary snapshots or if we have already checked 1523 * the counts in dsl_dataset_snapshot_check. This means we really only 1524 * check the count here when we're receiving a stream. 1525 */ 1526 if (cnt != 0 && cr != NULL) { 1527 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1528 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr, proc); 1529 if (error != 0) 1530 return (error); 1531 } 1532 1533 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1534 if (error != 0) 1535 return (error); 1536 1537 return (0); 1538 } 1539 1540 int 1541 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1542 { 1543 dsl_dataset_snapshot_arg_t *ddsa = arg; 1544 dsl_pool_t *dp = dmu_tx_pool(tx); 1545 nvpair_t *pair; 1546 int rv = 0; 1547 1548 /* 1549 * Pre-compute how many total new snapshots will be created for each 1550 * level in the tree and below. This is needed for validating the 1551 * snapshot limit when either taking a recursive snapshot or when 1552 * taking multiple snapshots. 1553 * 1554 * The problem is that the counts are not actually adjusted when 1555 * we are checking, only when we finally sync. For a single snapshot, 1556 * this is easy, the count will increase by 1 at each node up the tree, 1557 * but its more complicated for the recursive/multiple snapshot case. 1558 * 1559 * The dsl_fs_ss_limit_check function does recursively check the count 1560 * at each level up the tree but since it is validating each snapshot 1561 * independently we need to be sure that we are validating the complete 1562 * count for the entire set of snapshots. We do this by rolling up the 1563 * counts for each component of the name into an nvlist and then 1564 * checking each of those cases with the aggregated count. 1565 * 1566 * This approach properly handles not only the recursive snapshot 1567 * case (where we get all of those on the ddsa_snaps list) but also 1568 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1569 * validate the limit on 'a' using a count of 2). 1570 * 1571 * We validate the snapshot names in the third loop and only report 1572 * name errors once. 1573 */ 1574 if (dmu_tx_is_syncing(tx)) { 1575 char *nm; 1576 nvlist_t *cnt_track = NULL; 1577 cnt_track = fnvlist_alloc(); 1578 1579 nm = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1580 1581 /* Rollup aggregated counts into the cnt_track list */ 1582 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1583 pair != NULL; 1584 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1585 char *pdelim; 1586 uint64_t val; 1587 1588 (void) strlcpy(nm, nvpair_name(pair), MAXPATHLEN); 1589 pdelim = strchr(nm, '@'); 1590 if (pdelim == NULL) 1591 continue; 1592 *pdelim = '\0'; 1593 1594 do { 1595 if (nvlist_lookup_uint64(cnt_track, nm, 1596 &val) == 0) { 1597 /* update existing entry */ 1598 fnvlist_add_uint64(cnt_track, nm, 1599 val + 1); 1600 } else { 1601 /* add to list */ 1602 fnvlist_add_uint64(cnt_track, nm, 1); 1603 } 1604 1605 pdelim = strrchr(nm, '/'); 1606 if (pdelim != NULL) 1607 *pdelim = '\0'; 1608 } while (pdelim != NULL); 1609 } 1610 1611 kmem_free(nm, MAXPATHLEN); 1612 1613 /* Check aggregated counts at each level */ 1614 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1615 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1616 int error = 0; 1617 char *name; 1618 uint64_t cnt = 0; 1619 dsl_dataset_t *ds; 1620 1621 name = nvpair_name(pair); 1622 cnt = fnvpair_value_uint64(pair); 1623 ASSERT(cnt > 0); 1624 1625 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1626 if (error == 0) { 1627 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1628 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1629 ddsa->ddsa_cr, ddsa->ddsa_proc); 1630 dsl_dataset_rele(ds, FTAG); 1631 } 1632 1633 if (error != 0) { 1634 if (ddsa->ddsa_errors != NULL) 1635 fnvlist_add_int32(ddsa->ddsa_errors, 1636 name, error); 1637 rv = error; 1638 /* only report one error for this check */ 1639 break; 1640 } 1641 } 1642 nvlist_free(cnt_track); 1643 } 1644 1645 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1646 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1647 int error = 0; 1648 dsl_dataset_t *ds; 1649 char *name, *atp = NULL; 1650 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1651 1652 name = nvpair_name(pair); 1653 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN) 1654 error = SET_ERROR(ENAMETOOLONG); 1655 if (error == 0) { 1656 atp = strchr(name, '@'); 1657 if (atp == NULL) 1658 error = SET_ERROR(EINVAL); 1659 if (error == 0) 1660 (void) strlcpy(dsname, name, atp - name + 1); 1661 } 1662 if (error == 0) 1663 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1664 if (error == 0) { 1665 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1666 error = dsl_dataset_snapshot_check_impl(ds, 1667 atp + 1, tx, B_FALSE, 0, NULL, NULL); 1668 dsl_dataset_rele(ds, FTAG); 1669 } 1670 1671 if (error != 0) { 1672 if (ddsa->ddsa_errors != NULL) { 1673 fnvlist_add_int32(ddsa->ddsa_errors, 1674 name, error); 1675 } 1676 rv = error; 1677 } 1678 } 1679 1680 return (rv); 1681 } 1682 1683 void 1684 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1685 dmu_tx_t *tx) 1686 { 1687 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1688 dmu_buf_t *dbuf; 1689 dsl_dataset_phys_t *dsphys; 1690 uint64_t dsobj, crtxg; 1691 objset_t *mos = dp->dp_meta_objset; 1692 static zil_header_t zero_zil __maybe_unused; 1693 objset_t *os __maybe_unused; 1694 1695 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1696 1697 /* 1698 * If we are on an old pool, the zil must not be active, in which 1699 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1700 */ 1701 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1702 dmu_objset_from_ds(ds, &os) != 0 || 1703 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1704 sizeof (zero_zil)) == 0); 1705 1706 /* Should not snapshot a dirty dataset. */ 1707 ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1708 ds, tx->tx_txg)); 1709 1710 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1711 1712 /* 1713 * The origin's ds_creation_txg has to be < TXG_INITIAL 1714 */ 1715 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1716 crtxg = 1; 1717 else 1718 crtxg = tx->tx_txg; 1719 1720 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1721 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1722 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1723 dmu_buf_will_dirty(dbuf, tx); 1724 dsphys = dbuf->db_data; 1725 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1726 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1727 dsphys->ds_fsid_guid = unique_create(); 1728 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1729 sizeof (dsphys->ds_guid)); 1730 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1731 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 1732 dsphys->ds_next_snap_obj = ds->ds_object; 1733 dsphys->ds_num_children = 1; 1734 dsphys->ds_creation_time = gethrestime_sec(); 1735 dsphys->ds_creation_txg = crtxg; 1736 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; 1737 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; 1738 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; 1739 dsphys->ds_uncompressed_bytes = 1740 dsl_dataset_phys(ds)->ds_uncompressed_bytes; 1741 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; 1742 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1743 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; 1744 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1745 dmu_buf_rele(dbuf, FTAG); 1746 1747 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1748 if (zfeature_active(f, ds->ds_feature[f])) { 1749 dsl_dataset_activate_feature(dsobj, f, 1750 ds->ds_feature[f], tx); 1751 } 1752 } 1753 1754 ASSERT3U(ds->ds_prev != 0, ==, 1755 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); 1756 if (ds->ds_prev) { 1757 uint64_t next_clones_obj = 1758 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; 1759 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1760 ds->ds_object || 1761 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); 1762 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1763 ds->ds_object) { 1764 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1765 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, 1766 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); 1767 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; 1768 } else if (next_clones_obj != 0) { 1769 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1770 dsphys->ds_next_snap_obj, tx); 1771 VERIFY0(zap_add_int(mos, 1772 next_clones_obj, dsobj, tx)); 1773 } 1774 } 1775 1776 /* 1777 * If we have a reference-reservation on this dataset, we will 1778 * need to increase the amount of refreservation being charged 1779 * since our unique space is going to zero. 1780 */ 1781 if (ds->ds_reserved) { 1782 int64_t delta; 1783 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1784 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, 1785 ds->ds_reserved); 1786 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1787 delta, 0, 0, tx); 1788 } 1789 1790 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1791 dsl_dataset_phys(ds)->ds_deadlist_obj = 1792 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, 1793 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 1794 dsl_deadlist_close(&ds->ds_deadlist); 1795 dsl_deadlist_open(&ds->ds_deadlist, mos, 1796 dsl_dataset_phys(ds)->ds_deadlist_obj); 1797 dsl_deadlist_add_key(&ds->ds_deadlist, 1798 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); 1799 dsl_bookmark_snapshotted(ds, tx); 1800 1801 if (dsl_dataset_remap_deadlist_exists(ds)) { 1802 uint64_t remap_deadlist_obj = 1803 dsl_dataset_get_remap_deadlist_object(ds); 1804 /* 1805 * Move the remap_deadlist to the snapshot. The head 1806 * will create a new remap deadlist on demand, from 1807 * dsl_dataset_block_remapped(). 1808 */ 1809 dsl_dataset_unset_remap_deadlist_object(ds, tx); 1810 dsl_deadlist_close(&ds->ds_remap_deadlist); 1811 1812 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 1813 VERIFY0(zap_add(mos, dsobj, DS_FIELD_REMAP_DEADLIST, 1814 sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj, tx)); 1815 } 1816 1817 /* 1818 * Create a ivset guid for this snapshot if the dataset is 1819 * encrypted. This may be overridden by a raw receive. A 1820 * previous implementation of this code did not have this 1821 * field as part of the on-disk format for ZFS encryption 1822 * (see errata #4). As part of the remediation for this 1823 * issue, we ask the user to enable the bookmark_v2 feature 1824 * which is now a dependency of the encryption feature. We 1825 * use this as a heuristic to determine when the user has 1826 * elected to correct any datasets created with the old code. 1827 * As a result, we only do this step if the bookmark_v2 1828 * feature is enabled, which limits the number of states a 1829 * given pool / dataset can be in with regards to terms of 1830 * correcting the issue. 1831 */ 1832 if (ds->ds_dir->dd_crypto_obj != 0 && 1833 spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARK_V2)) { 1834 uint64_t ivset_guid = unique_create(); 1835 1836 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 1837 VERIFY0(zap_add(mos, dsobj, DS_FIELD_IVSET_GUID, 1838 sizeof (ivset_guid), 1, &ivset_guid, tx)); 1839 } 1840 1841 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); 1842 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; 1843 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; 1844 dsl_dataset_phys(ds)->ds_unique_bytes = 0; 1845 1846 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1847 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1848 1849 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1850 snapname, 8, 1, &dsobj, tx)); 1851 1852 if (ds->ds_prev) 1853 dsl_dataset_rele(ds->ds_prev, ds); 1854 VERIFY0(dsl_dataset_hold_obj(dp, 1855 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); 1856 1857 dsl_scan_ds_snapshotted(ds, tx); 1858 1859 dsl_dir_snap_cmtime_update(ds->ds_dir); 1860 1861 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, " "); 1862 } 1863 1864 void 1865 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1866 { 1867 dsl_dataset_snapshot_arg_t *ddsa = arg; 1868 dsl_pool_t *dp = dmu_tx_pool(tx); 1869 nvpair_t *pair; 1870 1871 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1872 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1873 dsl_dataset_t *ds; 1874 char *name, *atp; 1875 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1876 1877 name = nvpair_name(pair); 1878 atp = strchr(name, '@'); 1879 (void) strlcpy(dsname, name, atp - name + 1); 1880 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1881 1882 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1883 if (ddsa->ddsa_props != NULL) { 1884 dsl_props_set_sync_impl(ds->ds_prev, 1885 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1886 } 1887 dsl_dataset_rele(ds, FTAG); 1888 } 1889 } 1890 1891 /* 1892 * The snapshots must all be in the same pool. 1893 * All-or-nothing: if there are any failures, nothing will be modified. 1894 */ 1895 int 1896 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1897 { 1898 dsl_dataset_snapshot_arg_t ddsa; 1899 nvpair_t *pair; 1900 boolean_t needsuspend; 1901 int error; 1902 spa_t *spa; 1903 char *firstname; 1904 nvlist_t *suspended = NULL; 1905 1906 pair = nvlist_next_nvpair(snaps, NULL); 1907 if (pair == NULL) 1908 return (0); 1909 firstname = nvpair_name(pair); 1910 1911 error = spa_open(firstname, &spa, FTAG); 1912 if (error != 0) 1913 return (error); 1914 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1915 spa_close(spa, FTAG); 1916 1917 if (needsuspend) { 1918 suspended = fnvlist_alloc(); 1919 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1920 pair = nvlist_next_nvpair(snaps, pair)) { 1921 char fsname[ZFS_MAX_DATASET_NAME_LEN]; 1922 char *snapname = nvpair_name(pair); 1923 char *atp; 1924 void *cookie; 1925 1926 atp = strchr(snapname, '@'); 1927 if (atp == NULL) { 1928 error = SET_ERROR(EINVAL); 1929 break; 1930 } 1931 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1932 1933 error = zil_suspend(fsname, &cookie); 1934 if (error != 0) 1935 break; 1936 fnvlist_add_uint64(suspended, fsname, 1937 (uintptr_t)cookie); 1938 } 1939 } 1940 1941 ddsa.ddsa_snaps = snaps; 1942 ddsa.ddsa_props = props; 1943 ddsa.ddsa_errors = errors; 1944 ddsa.ddsa_cr = CRED(); 1945 ddsa.ddsa_proc = curproc; 1946 1947 if (error == 0) { 1948 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1949 dsl_dataset_snapshot_sync, &ddsa, 1950 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1951 } 1952 1953 if (suspended != NULL) { 1954 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1955 pair = nvlist_next_nvpair(suspended, pair)) { 1956 zil_resume((void *)(uintptr_t) 1957 fnvpair_value_uint64(pair)); 1958 } 1959 fnvlist_free(suspended); 1960 } 1961 1962 if (error == 0) { 1963 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1964 pair = nvlist_next_nvpair(snaps, pair)) { 1965 zvol_create_minor(nvpair_name(pair)); 1966 } 1967 } 1968 1969 return (error); 1970 } 1971 1972 typedef struct dsl_dataset_snapshot_tmp_arg { 1973 const char *ddsta_fsname; 1974 const char *ddsta_snapname; 1975 minor_t ddsta_cleanup_minor; 1976 const char *ddsta_htag; 1977 } dsl_dataset_snapshot_tmp_arg_t; 1978 1979 static int 1980 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1981 { 1982 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1983 dsl_pool_t *dp = dmu_tx_pool(tx); 1984 dsl_dataset_t *ds; 1985 int error; 1986 1987 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1988 if (error != 0) 1989 return (error); 1990 1991 /* NULL cred means no limit check for tmp snapshot */ 1992 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1993 tx, B_FALSE, 0, NULL, NULL); 1994 if (error != 0) { 1995 dsl_dataset_rele(ds, FTAG); 1996 return (error); 1997 } 1998 1999 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 2000 dsl_dataset_rele(ds, FTAG); 2001 return (SET_ERROR(ENOTSUP)); 2002 } 2003 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 2004 B_TRUE, tx); 2005 if (error != 0) { 2006 dsl_dataset_rele(ds, FTAG); 2007 return (error); 2008 } 2009 2010 dsl_dataset_rele(ds, FTAG); 2011 return (0); 2012 } 2013 2014 static void 2015 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 2016 { 2017 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 2018 dsl_pool_t *dp = dmu_tx_pool(tx); 2019 dsl_dataset_t *ds = NULL; 2020 2021 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 2022 2023 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 2024 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 2025 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 2026 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 2027 2028 dsl_dataset_rele(ds, FTAG); 2029 } 2030 2031 int 2032 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 2033 minor_t cleanup_minor, const char *htag) 2034 { 2035 dsl_dataset_snapshot_tmp_arg_t ddsta; 2036 int error; 2037 spa_t *spa; 2038 boolean_t needsuspend; 2039 void *cookie; 2040 2041 ddsta.ddsta_fsname = fsname; 2042 ddsta.ddsta_snapname = snapname; 2043 ddsta.ddsta_cleanup_minor = cleanup_minor; 2044 ddsta.ddsta_htag = htag; 2045 2046 error = spa_open(fsname, &spa, FTAG); 2047 if (error != 0) 2048 return (error); 2049 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 2050 spa_close(spa, FTAG); 2051 2052 if (needsuspend) { 2053 error = zil_suspend(fsname, &cookie); 2054 if (error != 0) 2055 return (error); 2056 } 2057 2058 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 2059 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 2060 2061 if (needsuspend) 2062 zil_resume(cookie); 2063 return (error); 2064 } 2065 2066 void 2067 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 2068 { 2069 ASSERT(dmu_tx_is_syncing(tx)); 2070 ASSERT(ds->ds_objset != NULL); 2071 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); 2072 2073 /* 2074 * in case we had to change ds_fsid_guid when we opened it, 2075 * sync it out now. 2076 */ 2077 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2078 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; 2079 2080 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) { 2081 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 2082 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1, 2083 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx)); 2084 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 2085 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1, 2086 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx)); 2087 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 2088 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1, 2089 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx)); 2090 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0; 2091 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0; 2092 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0; 2093 } 2094 2095 dmu_objset_sync(ds->ds_objset, zio, tx); 2096 2097 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 2098 if (zfeature_active(f, ds->ds_feature_activation[f])) { 2099 if (zfeature_active(f, ds->ds_feature[f])) 2100 continue; 2101 dsl_dataset_activate_feature(ds->ds_object, f, 2102 ds->ds_feature_activation[f], tx); 2103 ds->ds_feature[f] = ds->ds_feature_activation[f]; 2104 } 2105 } 2106 } 2107 2108 /* 2109 * Check if the percentage of blocks shared between the clone and the 2110 * snapshot (as opposed to those that are clone only) is below a certain 2111 * threshold 2112 */ 2113 static boolean_t 2114 dsl_livelist_should_disable(dsl_dataset_t *ds) 2115 { 2116 uint64_t used, referenced; 2117 int percent_shared; 2118 2119 used = dsl_dir_get_usedds(ds->ds_dir); 2120 referenced = dsl_get_referenced(ds); 2121 ASSERT3U(referenced, >=, 0); 2122 ASSERT3U(used, >=, 0); 2123 if (referenced == 0) 2124 return (B_FALSE); 2125 percent_shared = (100 * (referenced - used)) / referenced; 2126 if (percent_shared <= zfs_livelist_min_percent_shared) 2127 return (B_TRUE); 2128 return (B_FALSE); 2129 } 2130 2131 /* 2132 * Check if it is possible to combine two livelist entries into one. 2133 * This is the case if the combined number of 'live' blkptrs (ALLOCs that 2134 * don't have a matching FREE) is under the maximum sublist size. 2135 * We check this by subtracting twice the total number of frees from the total 2136 * number of blkptrs. FREEs are counted twice because each FREE blkptr 2137 * will cancel out an ALLOC blkptr when the livelist is processed. 2138 */ 2139 static boolean_t 2140 dsl_livelist_should_condense(dsl_deadlist_entry_t *first, 2141 dsl_deadlist_entry_t *next) 2142 { 2143 uint64_t total_free = first->dle_bpobj.bpo_phys->bpo_num_freed + 2144 next->dle_bpobj.bpo_phys->bpo_num_freed; 2145 uint64_t total_entries = first->dle_bpobj.bpo_phys->bpo_num_blkptrs + 2146 next->dle_bpobj.bpo_phys->bpo_num_blkptrs; 2147 if ((total_entries - (2 * total_free)) < zfs_livelist_max_entries) 2148 return (B_TRUE); 2149 return (B_FALSE); 2150 } 2151 2152 typedef struct try_condense_arg { 2153 spa_t *spa; 2154 dsl_dataset_t *ds; 2155 } try_condense_arg_t; 2156 2157 /* 2158 * Iterate over the livelist entries, searching for a pair to condense. 2159 * A nonzero return value means stop, 0 means keep looking. 2160 */ 2161 static int 2162 dsl_livelist_try_condense(void *arg, dsl_deadlist_entry_t *first) 2163 { 2164 try_condense_arg_t *tca = arg; 2165 spa_t *spa = tca->spa; 2166 dsl_dataset_t *ds = tca->ds; 2167 dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist; 2168 dsl_deadlist_entry_t *next; 2169 2170 /* The condense thread has not yet been created at import */ 2171 if (spa->spa_livelist_condense_zthr == NULL) 2172 return (1); 2173 2174 /* A condense is already in progress */ 2175 if (spa->spa_to_condense.ds != NULL) 2176 return (1); 2177 2178 next = AVL_NEXT(&ll->dl_tree, &first->dle_node); 2179 /* The livelist has only one entry - don't condense it */ 2180 if (next == NULL) 2181 return (1); 2182 2183 /* Next is the newest entry - don't condense it */ 2184 if (AVL_NEXT(&ll->dl_tree, &next->dle_node) == NULL) 2185 return (1); 2186 2187 /* This pair is not ready to condense but keep looking */ 2188 if (!dsl_livelist_should_condense(first, next)) 2189 return (0); 2190 2191 /* 2192 * Add a ref to prevent the dataset from being evicted while 2193 * the condense zthr or synctask are running. Ref will be 2194 * released at the end of the condense synctask 2195 */ 2196 dmu_buf_add_ref(ds->ds_dbuf, spa); 2197 2198 spa->spa_to_condense.ds = ds; 2199 spa->spa_to_condense.first = first; 2200 spa->spa_to_condense.next = next; 2201 spa->spa_to_condense.syncing = B_FALSE; 2202 spa->spa_to_condense.cancelled = B_FALSE; 2203 2204 zthr_wakeup(spa->spa_livelist_condense_zthr); 2205 return (1); 2206 } 2207 2208 static void 2209 dsl_flush_pending_livelist(dsl_dataset_t *ds, dmu_tx_t *tx) 2210 { 2211 dsl_dir_t *dd = ds->ds_dir; 2212 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 2213 dsl_deadlist_entry_t *last = dsl_deadlist_last(&dd->dd_livelist); 2214 2215 /* Check if we need to add a new sub-livelist */ 2216 if (last == NULL) { 2217 /* The livelist is empty */ 2218 dsl_deadlist_add_key(&dd->dd_livelist, 2219 tx->tx_txg - 1, tx); 2220 } else if (spa_sync_pass(spa) == 1) { 2221 /* 2222 * Check if the newest entry is full. If it is, make a new one. 2223 * We only do this once per sync because we could overfill a 2224 * sublist in one sync pass and don't want to add another entry 2225 * for a txg that is already represented. This ensures that 2226 * blkptrs born in the same txg are stored in the same sublist. 2227 */ 2228 bpobj_t bpobj = last->dle_bpobj; 2229 uint64_t all = bpobj.bpo_phys->bpo_num_blkptrs; 2230 uint64_t free = bpobj.bpo_phys->bpo_num_freed; 2231 uint64_t alloc = all - free; 2232 if (alloc > zfs_livelist_max_entries) { 2233 dsl_deadlist_add_key(&dd->dd_livelist, 2234 tx->tx_txg - 1, tx); 2235 } 2236 } 2237 2238 /* Insert each entry into the on-disk livelist */ 2239 bplist_iterate(&dd->dd_pending_allocs, 2240 dsl_deadlist_insert_alloc_cb, &dd->dd_livelist, tx); 2241 bplist_iterate(&dd->dd_pending_frees, 2242 dsl_deadlist_insert_free_cb, &dd->dd_livelist, tx); 2243 2244 /* Attempt to condense every pair of adjacent entries */ 2245 try_condense_arg_t arg = { 2246 .spa = spa, 2247 .ds = ds 2248 }; 2249 dsl_deadlist_iterate(&dd->dd_livelist, dsl_livelist_try_condense, 2250 &arg); 2251 } 2252 2253 void 2254 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx) 2255 { 2256 objset_t *os = ds->ds_objset; 2257 2258 bplist_iterate(&ds->ds_pending_deadlist, 2259 dsl_deadlist_insert_alloc_cb, &ds->ds_deadlist, tx); 2260 2261 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist)) { 2262 dsl_flush_pending_livelist(ds, tx); 2263 if (dsl_livelist_should_disable(ds)) { 2264 dsl_dir_remove_livelist(ds->ds_dir, tx, B_TRUE); 2265 } 2266 } 2267 2268 dsl_bookmark_sync_done(ds, tx); 2269 2270 multilist_destroy(&os->os_synced_dnodes); 2271 2272 if (os->os_encrypted) 2273 os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_FALSE; 2274 else 2275 ASSERT0(os->os_next_write_raw[tx->tx_txg & TXG_MASK]); 2276 2277 ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx))); 2278 2279 dmu_buf_rele(ds->ds_dbuf, ds); 2280 } 2281 2282 int 2283 get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val) 2284 { 2285 uint64_t count = 0; 2286 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 2287 zap_cursor_t zc; 2288 zap_attribute_t za; 2289 2290 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 2291 2292 /* 2293 * There may be missing entries in ds_next_clones_obj 2294 * due to a bug in a previous version of the code. 2295 * Only trust it if it has the right number of entries. 2296 */ 2297 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { 2298 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 2299 &count)); 2300 } 2301 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) { 2302 return (SET_ERROR(ENOENT)); 2303 } 2304 for (zap_cursor_init(&zc, mos, 2305 dsl_dataset_phys(ds)->ds_next_clones_obj); 2306 zap_cursor_retrieve(&zc, &za) == 0; 2307 zap_cursor_advance(&zc)) { 2308 dsl_dataset_t *clone; 2309 char buf[ZFS_MAX_DATASET_NAME_LEN]; 2310 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 2311 za.za_first_integer, FTAG, &clone)); 2312 dsl_dir_name(clone->ds_dir, buf); 2313 fnvlist_add_boolean(val, buf); 2314 dsl_dataset_rele(clone, FTAG); 2315 } 2316 zap_cursor_fini(&zc); 2317 return (0); 2318 } 2319 2320 void 2321 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 2322 { 2323 nvlist_t *propval = fnvlist_alloc(); 2324 nvlist_t *val = fnvlist_alloc(); 2325 2326 if (get_clones_stat_impl(ds, val) == 0) { 2327 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 2328 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), 2329 propval); 2330 } 2331 2332 nvlist_free(val); 2333 nvlist_free(propval); 2334 } 2335 2336 /* 2337 * Returns a string that represents the receive resume stats token. It should 2338 * be freed with strfree(). 2339 */ 2340 char * 2341 get_receive_resume_stats_impl(dsl_dataset_t *ds) 2342 { 2343 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2344 2345 if (dsl_dataset_has_resume_receive_state(ds)) { 2346 char *str; 2347 void *packed; 2348 uint8_t *compressed; 2349 uint64_t val; 2350 nvlist_t *token_nv = fnvlist_alloc(); 2351 size_t packed_size, compressed_size; 2352 2353 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2354 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) { 2355 fnvlist_add_uint64(token_nv, "fromguid", val); 2356 } 2357 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2358 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) { 2359 fnvlist_add_uint64(token_nv, "object", val); 2360 } 2361 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2362 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) { 2363 fnvlist_add_uint64(token_nv, "offset", val); 2364 } 2365 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2366 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) { 2367 fnvlist_add_uint64(token_nv, "bytes", val); 2368 } 2369 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2370 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) { 2371 fnvlist_add_uint64(token_nv, "toguid", val); 2372 } 2373 char buf[MAXNAMELEN]; 2374 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2375 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) { 2376 fnvlist_add_string(token_nv, "toname", buf); 2377 } 2378 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2379 DS_FIELD_RESUME_LARGEBLOCK) == 0) { 2380 fnvlist_add_boolean(token_nv, "largeblockok"); 2381 } 2382 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2383 DS_FIELD_RESUME_EMBEDOK) == 0) { 2384 fnvlist_add_boolean(token_nv, "embedok"); 2385 } 2386 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2387 DS_FIELD_RESUME_COMPRESSOK) == 0) { 2388 fnvlist_add_boolean(token_nv, "compressok"); 2389 } 2390 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2391 DS_FIELD_RESUME_RAWOK) == 0) { 2392 fnvlist_add_boolean(token_nv, "rawok"); 2393 } 2394 if (dsl_dataset_feature_is_active(ds, 2395 SPA_FEATURE_REDACTED_DATASETS)) { 2396 uint64_t num_redact_snaps; 2397 uint64_t *redact_snaps; 2398 VERIFY(dsl_dataset_get_uint64_array_feature(ds, 2399 SPA_FEATURE_REDACTED_DATASETS, &num_redact_snaps, 2400 &redact_snaps)); 2401 fnvlist_add_uint64_array(token_nv, "redact_snaps", 2402 redact_snaps, num_redact_snaps); 2403 } 2404 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2405 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS) == 0) { 2406 uint64_t num_redact_snaps, int_size; 2407 uint64_t *redact_snaps; 2408 VERIFY0(zap_length(dp->dp_meta_objset, ds->ds_object, 2409 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, &int_size, 2410 &num_redact_snaps)); 2411 ASSERT3U(int_size, ==, sizeof (uint64_t)); 2412 2413 redact_snaps = kmem_alloc(int_size * num_redact_snaps, 2414 KM_SLEEP); 2415 VERIFY0(zap_lookup(dp->dp_meta_objset, ds->ds_object, 2416 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, int_size, 2417 num_redact_snaps, redact_snaps)); 2418 fnvlist_add_uint64_array(token_nv, "book_redact_snaps", 2419 redact_snaps, num_redact_snaps); 2420 kmem_free(redact_snaps, int_size * num_redact_snaps); 2421 } 2422 packed = fnvlist_pack(token_nv, &packed_size); 2423 fnvlist_free(token_nv); 2424 compressed = kmem_alloc(packed_size, KM_SLEEP); 2425 2426 compressed_size = gzip_compress(packed, compressed, 2427 packed_size, packed_size, 6); 2428 2429 zio_cksum_t cksum; 2430 fletcher_4_native_varsize(compressed, compressed_size, &cksum); 2431 2432 size_t alloc_size = compressed_size * 2 + 1; 2433 str = kmem_alloc(alloc_size, KM_SLEEP); 2434 for (int i = 0; i < compressed_size; i++) { 2435 size_t offset = i * 2; 2436 (void) snprintf(str + offset, alloc_size - offset, 2437 "%02x", compressed[i]); 2438 } 2439 str[compressed_size * 2] = '\0'; 2440 char *propval = kmem_asprintf("%u-%llx-%llx-%s", 2441 ZFS_SEND_RESUME_TOKEN_VERSION, 2442 (longlong_t)cksum.zc_word[0], 2443 (longlong_t)packed_size, str); 2444 kmem_free(packed, packed_size); 2445 kmem_free(str, alloc_size); 2446 kmem_free(compressed, packed_size); 2447 return (propval); 2448 } 2449 return (kmem_strdup("")); 2450 } 2451 2452 /* 2453 * Returns a string that represents the receive resume stats token of the 2454 * dataset's child. It should be freed with strfree(). 2455 */ 2456 char * 2457 get_child_receive_stats(dsl_dataset_t *ds) 2458 { 2459 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 2460 dsl_dataset_t *recv_ds; 2461 dsl_dataset_name(ds, recvname); 2462 if (strlcat(recvname, "/", sizeof (recvname)) < 2463 sizeof (recvname) && 2464 strlcat(recvname, recv_clone_name, sizeof (recvname)) < 2465 sizeof (recvname) && 2466 dsl_dataset_hold(ds->ds_dir->dd_pool, recvname, FTAG, 2467 &recv_ds) == 0) { 2468 char *propval = get_receive_resume_stats_impl(recv_ds); 2469 dsl_dataset_rele(recv_ds, FTAG); 2470 return (propval); 2471 } 2472 return (kmem_strdup("")); 2473 } 2474 2475 static void 2476 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv) 2477 { 2478 char *propval = get_receive_resume_stats_impl(ds); 2479 if (strcmp(propval, "") != 0) { 2480 dsl_prop_nvlist_add_string(nv, 2481 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval); 2482 } else { 2483 char *childval = get_child_receive_stats(ds); 2484 if (strcmp(childval, "") != 0) { 2485 dsl_prop_nvlist_add_string(nv, 2486 ZFS_PROP_RECEIVE_RESUME_TOKEN, childval); 2487 } 2488 kmem_strfree(childval); 2489 } 2490 kmem_strfree(propval); 2491 } 2492 2493 uint64_t 2494 dsl_get_refratio(dsl_dataset_t *ds) 2495 { 2496 uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : 2497 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / 2498 dsl_dataset_phys(ds)->ds_compressed_bytes); 2499 return (ratio); 2500 } 2501 2502 uint64_t 2503 dsl_get_logicalreferenced(dsl_dataset_t *ds) 2504 { 2505 return (dsl_dataset_phys(ds)->ds_uncompressed_bytes); 2506 } 2507 2508 uint64_t 2509 dsl_get_compressratio(dsl_dataset_t *ds) 2510 { 2511 if (ds->ds_is_snapshot) { 2512 return (dsl_get_refratio(ds)); 2513 } else { 2514 dsl_dir_t *dd = ds->ds_dir; 2515 mutex_enter(&dd->dd_lock); 2516 uint64_t val = dsl_dir_get_compressratio(dd); 2517 mutex_exit(&dd->dd_lock); 2518 return (val); 2519 } 2520 } 2521 2522 uint64_t 2523 dsl_get_used(dsl_dataset_t *ds) 2524 { 2525 if (ds->ds_is_snapshot) { 2526 return (dsl_dataset_phys(ds)->ds_unique_bytes); 2527 } else { 2528 dsl_dir_t *dd = ds->ds_dir; 2529 mutex_enter(&dd->dd_lock); 2530 uint64_t val = dsl_dir_get_used(dd); 2531 mutex_exit(&dd->dd_lock); 2532 return (val); 2533 } 2534 } 2535 2536 uint64_t 2537 dsl_get_creation(dsl_dataset_t *ds) 2538 { 2539 return (dsl_dataset_phys(ds)->ds_creation_time); 2540 } 2541 2542 uint64_t 2543 dsl_get_creationtxg(dsl_dataset_t *ds) 2544 { 2545 return (dsl_dataset_phys(ds)->ds_creation_txg); 2546 } 2547 2548 uint64_t 2549 dsl_get_refquota(dsl_dataset_t *ds) 2550 { 2551 return (ds->ds_quota); 2552 } 2553 2554 uint64_t 2555 dsl_get_refreservation(dsl_dataset_t *ds) 2556 { 2557 return (ds->ds_reserved); 2558 } 2559 2560 uint64_t 2561 dsl_get_guid(dsl_dataset_t *ds) 2562 { 2563 return (dsl_dataset_phys(ds)->ds_guid); 2564 } 2565 2566 uint64_t 2567 dsl_get_unique(dsl_dataset_t *ds) 2568 { 2569 return (dsl_dataset_phys(ds)->ds_unique_bytes); 2570 } 2571 2572 uint64_t 2573 dsl_get_objsetid(dsl_dataset_t *ds) 2574 { 2575 return (ds->ds_object); 2576 } 2577 2578 uint64_t 2579 dsl_get_userrefs(dsl_dataset_t *ds) 2580 { 2581 return (ds->ds_userrefs); 2582 } 2583 2584 uint64_t 2585 dsl_get_defer_destroy(dsl_dataset_t *ds) 2586 { 2587 return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 2588 } 2589 2590 uint64_t 2591 dsl_get_referenced(dsl_dataset_t *ds) 2592 { 2593 return (dsl_dataset_phys(ds)->ds_referenced_bytes); 2594 } 2595 2596 uint64_t 2597 dsl_get_numclones(dsl_dataset_t *ds) 2598 { 2599 ASSERT(ds->ds_is_snapshot); 2600 return (dsl_dataset_phys(ds)->ds_num_children - 1); 2601 } 2602 2603 uint64_t 2604 dsl_get_inconsistent(dsl_dataset_t *ds) 2605 { 2606 return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ? 2607 1 : 0); 2608 } 2609 2610 uint64_t 2611 dsl_get_redacted(dsl_dataset_t *ds) 2612 { 2613 return (dsl_dataset_feature_is_active(ds, 2614 SPA_FEATURE_REDACTED_DATASETS)); 2615 } 2616 2617 uint64_t 2618 dsl_get_available(dsl_dataset_t *ds) 2619 { 2620 uint64_t refdbytes = dsl_get_referenced(ds); 2621 uint64_t availbytes = dsl_dir_space_available(ds->ds_dir, 2622 NULL, 0, TRUE); 2623 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 2624 availbytes += 2625 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 2626 } 2627 if (ds->ds_quota != 0) { 2628 /* 2629 * Adjust available bytes according to refquota 2630 */ 2631 if (refdbytes < ds->ds_quota) { 2632 availbytes = MIN(availbytes, 2633 ds->ds_quota - refdbytes); 2634 } else { 2635 availbytes = 0; 2636 } 2637 } 2638 return (availbytes); 2639 } 2640 2641 int 2642 dsl_get_written(dsl_dataset_t *ds, uint64_t *written) 2643 { 2644 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2645 dsl_dataset_t *prev; 2646 int err = dsl_dataset_hold_obj(dp, 2647 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 2648 if (err == 0) { 2649 uint64_t comp, uncomp; 2650 err = dsl_dataset_space_written(prev, ds, written, 2651 &comp, &uncomp); 2652 dsl_dataset_rele(prev, FTAG); 2653 } 2654 return (err); 2655 } 2656 2657 /* 2658 * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN. 2659 */ 2660 int 2661 dsl_get_prev_snap(dsl_dataset_t *ds, char *snap) 2662 { 2663 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2664 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 2665 dsl_dataset_name(ds->ds_prev, snap); 2666 return (0); 2667 } else { 2668 return (SET_ERROR(ENOENT)); 2669 } 2670 } 2671 2672 void 2673 dsl_get_redact_snaps(dsl_dataset_t *ds, nvlist_t *propval) 2674 { 2675 uint64_t nsnaps; 2676 uint64_t *snaps; 2677 if (dsl_dataset_get_uint64_array_feature(ds, 2678 SPA_FEATURE_REDACTED_DATASETS, &nsnaps, &snaps)) { 2679 fnvlist_add_uint64_array(propval, ZPROP_VALUE, snaps, 2680 nsnaps); 2681 } 2682 } 2683 2684 /* 2685 * Returns the mountpoint property and source for the given dataset in the value 2686 * and source buffers. The value buffer must be at least as large as MAXPATHLEN 2687 * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN. 2688 * Returns 0 on success and an error on failure. 2689 */ 2690 int 2691 dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value, 2692 char *source) 2693 { 2694 int error; 2695 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2696 2697 /* Retrieve the mountpoint value stored in the zap object */ 2698 error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1, 2699 ZAP_MAXVALUELEN, value, source); 2700 if (error != 0) { 2701 return (error); 2702 } 2703 2704 /* 2705 * Process the dsname and source to find the full mountpoint string. 2706 * Can be skipped for 'legacy' or 'none'. 2707 */ 2708 if (value[0] == '/') { 2709 char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP); 2710 char *root = buf; 2711 const char *relpath; 2712 2713 /* 2714 * If we inherit the mountpoint, even from a dataset 2715 * with a received value, the source will be the path of 2716 * the dataset we inherit from. If source is 2717 * ZPROP_SOURCE_VAL_RECVD, the received value is not 2718 * inherited. 2719 */ 2720 if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) { 2721 relpath = ""; 2722 } else { 2723 ASSERT0(strncmp(dsname, source, strlen(source))); 2724 relpath = dsname + strlen(source); 2725 if (relpath[0] == '/') 2726 relpath++; 2727 } 2728 2729 spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN); 2730 2731 /* 2732 * Special case an alternate root of '/'. This will 2733 * avoid having multiple leading slashes in the 2734 * mountpoint path. 2735 */ 2736 if (strcmp(root, "/") == 0) 2737 root++; 2738 2739 /* 2740 * If the mountpoint is '/' then skip over this 2741 * if we are obtaining either an alternate root or 2742 * an inherited mountpoint. 2743 */ 2744 char *mnt = value; 2745 if (value[1] == '\0' && (root[0] != '\0' || 2746 relpath[0] != '\0')) 2747 mnt = value + 1; 2748 2749 if (relpath[0] == '\0') { 2750 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s", 2751 root, mnt); 2752 } else { 2753 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s", 2754 root, mnt, relpath[0] == '@' ? "" : "/", 2755 relpath); 2756 } 2757 kmem_free(buf, ZAP_MAXVALUELEN); 2758 } 2759 2760 return (0); 2761 } 2762 2763 void 2764 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 2765 { 2766 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2767 2768 ASSERT(dsl_pool_config_held(dp)); 2769 2770 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, 2771 dsl_get_refratio(ds)); 2772 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 2773 dsl_get_logicalreferenced(ds)); 2774 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 2775 dsl_get_compressratio(ds)); 2776 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 2777 dsl_get_used(ds)); 2778 2779 if (ds->ds_is_snapshot) { 2780 get_clones_stat(ds, nv); 2781 } else { 2782 char buf[ZFS_MAX_DATASET_NAME_LEN]; 2783 if (dsl_get_prev_snap(ds, buf) == 0) 2784 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, 2785 buf); 2786 dsl_dir_stats(ds->ds_dir, nv); 2787 } 2788 2789 nvlist_t *propval = fnvlist_alloc(); 2790 dsl_get_redact_snaps(ds, propval); 2791 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_REDACT_SNAPS), 2792 propval); 2793 nvlist_free(propval); 2794 2795 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, 2796 dsl_get_available(ds)); 2797 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, 2798 dsl_get_referenced(ds)); 2799 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 2800 dsl_get_creation(ds)); 2801 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 2802 dsl_get_creationtxg(ds)); 2803 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 2804 dsl_get_refquota(ds)); 2805 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 2806 dsl_get_refreservation(ds)); 2807 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 2808 dsl_get_guid(ds)); 2809 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 2810 dsl_get_unique(ds)); 2811 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 2812 dsl_get_objsetid(ds)); 2813 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 2814 dsl_get_userrefs(ds)); 2815 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 2816 dsl_get_defer_destroy(ds)); 2817 dsl_dataset_crypt_stats(ds, nv); 2818 2819 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 2820 uint64_t written; 2821 if (dsl_get_written(ds, &written) == 0) { 2822 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 2823 written); 2824 } 2825 } 2826 2827 if (!dsl_dataset_is_snapshot(ds)) { 2828 /* 2829 * A failed "newfs" (e.g. full) resumable receive leaves 2830 * the stats set on this dataset. Check here for the prop. 2831 */ 2832 get_receive_resume_stats(ds, nv); 2833 2834 /* 2835 * A failed incremental resumable receive leaves the 2836 * stats set on our child named "%recv". Check the child 2837 * for the prop. 2838 */ 2839 /* 6 extra bytes for /%recv */ 2840 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 2841 dsl_dataset_t *recv_ds; 2842 dsl_dataset_name(ds, recvname); 2843 if (strlcat(recvname, "/", sizeof (recvname)) < 2844 sizeof (recvname) && 2845 strlcat(recvname, recv_clone_name, sizeof (recvname)) < 2846 sizeof (recvname) && 2847 dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) { 2848 get_receive_resume_stats(recv_ds, nv); 2849 dsl_dataset_rele(recv_ds, FTAG); 2850 } 2851 } 2852 } 2853 2854 void 2855 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 2856 { 2857 dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool; 2858 ASSERT(dsl_pool_config_held(dp)); 2859 2860 stat->dds_creation_txg = dsl_get_creationtxg(ds); 2861 stat->dds_inconsistent = dsl_get_inconsistent(ds); 2862 stat->dds_guid = dsl_get_guid(ds); 2863 stat->dds_redacted = dsl_get_redacted(ds); 2864 stat->dds_origin[0] = '\0'; 2865 if (ds->ds_is_snapshot) { 2866 stat->dds_is_snapshot = B_TRUE; 2867 stat->dds_num_clones = dsl_get_numclones(ds); 2868 } else { 2869 stat->dds_is_snapshot = B_FALSE; 2870 stat->dds_num_clones = 0; 2871 2872 if (dsl_dir_is_clone(ds->ds_dir)) { 2873 dsl_dir_get_origin(ds->ds_dir, stat->dds_origin); 2874 } 2875 } 2876 } 2877 2878 uint64_t 2879 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 2880 { 2881 return (ds->ds_fsid_guid); 2882 } 2883 2884 void 2885 dsl_dataset_space(dsl_dataset_t *ds, 2886 uint64_t *refdbytesp, uint64_t *availbytesp, 2887 uint64_t *usedobjsp, uint64_t *availobjsp) 2888 { 2889 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; 2890 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 2891 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) 2892 *availbytesp += 2893 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 2894 if (ds->ds_quota != 0) { 2895 /* 2896 * Adjust available bytes according to refquota 2897 */ 2898 if (*refdbytesp < ds->ds_quota) 2899 *availbytesp = MIN(*availbytesp, 2900 ds->ds_quota - *refdbytesp); 2901 else 2902 *availbytesp = 0; 2903 } 2904 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2905 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); 2906 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2907 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 2908 } 2909 2910 boolean_t 2911 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 2912 { 2913 dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool; 2914 uint64_t birth; 2915 2916 ASSERT(dsl_pool_config_held(dp)); 2917 if (snap == NULL) 2918 return (B_FALSE); 2919 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2920 birth = dsl_dataset_get_blkptr(ds)->blk_birth; 2921 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2922 if (birth > dsl_dataset_phys(snap)->ds_creation_txg) { 2923 objset_t *os, *os_snap; 2924 /* 2925 * It may be that only the ZIL differs, because it was 2926 * reset in the head. Don't count that as being 2927 * modified. 2928 */ 2929 if (dmu_objset_from_ds(ds, &os) != 0) 2930 return (B_TRUE); 2931 if (dmu_objset_from_ds(snap, &os_snap) != 0) 2932 return (B_TRUE); 2933 return (bcmp(&os->os_phys->os_meta_dnode, 2934 &os_snap->os_phys->os_meta_dnode, 2935 sizeof (os->os_phys->os_meta_dnode)) != 0); 2936 } 2937 return (B_FALSE); 2938 } 2939 2940 typedef struct dsl_dataset_rename_snapshot_arg { 2941 const char *ddrsa_fsname; 2942 const char *ddrsa_oldsnapname; 2943 const char *ddrsa_newsnapname; 2944 boolean_t ddrsa_recursive; 2945 dmu_tx_t *ddrsa_tx; 2946 } dsl_dataset_rename_snapshot_arg_t; 2947 2948 /* ARGSUSED */ 2949 static int 2950 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 2951 dsl_dataset_t *hds, void *arg) 2952 { 2953 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2954 int error; 2955 uint64_t val; 2956 2957 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2958 if (error != 0) { 2959 /* ignore nonexistent snapshots */ 2960 return (error == ENOENT ? 0 : error); 2961 } 2962 2963 /* new name should not exist */ 2964 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 2965 if (error == 0) 2966 error = SET_ERROR(EEXIST); 2967 else if (error == ENOENT) 2968 error = 0; 2969 2970 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2971 if (dsl_dir_namelen(hds->ds_dir) + 1 + 2972 strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN) 2973 error = SET_ERROR(ENAMETOOLONG); 2974 2975 return (error); 2976 } 2977 2978 static int 2979 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 2980 { 2981 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2982 dsl_pool_t *dp = dmu_tx_pool(tx); 2983 dsl_dataset_t *hds; 2984 int error; 2985 2986 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 2987 if (error != 0) 2988 return (error); 2989 2990 if (ddrsa->ddrsa_recursive) { 2991 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 2992 dsl_dataset_rename_snapshot_check_impl, ddrsa, 2993 DS_FIND_CHILDREN); 2994 } else { 2995 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 2996 } 2997 dsl_dataset_rele(hds, FTAG); 2998 return (error); 2999 } 3000 3001 static int 3002 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 3003 dsl_dataset_t *hds, void *arg) 3004 { 3005 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 3006 dsl_dataset_t *ds; 3007 uint64_t val; 3008 dmu_tx_t *tx = ddrsa->ddrsa_tx; 3009 int error; 3010 3011 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 3012 ASSERT(error == 0 || error == ENOENT); 3013 if (error == ENOENT) { 3014 /* ignore nonexistent snapshots */ 3015 return (0); 3016 } 3017 3018 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 3019 3020 /* log before we change the name */ 3021 spa_history_log_internal_ds(ds, "rename", tx, 3022 "-> @%s", ddrsa->ddrsa_newsnapname); 3023 3024 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 3025 B_FALSE)); 3026 mutex_enter(&ds->ds_lock); 3027 (void) strlcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname, 3028 sizeof (ds->ds_snapname)); 3029 mutex_exit(&ds->ds_lock); 3030 VERIFY0(zap_add(dp->dp_meta_objset, 3031 dsl_dataset_phys(hds)->ds_snapnames_zapobj, 3032 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 3033 zvol_rename_minors(dp->dp_spa, ddrsa->ddrsa_oldsnapname, 3034 ddrsa->ddrsa_newsnapname, B_TRUE); 3035 3036 dsl_dataset_rele(ds, FTAG); 3037 return (0); 3038 } 3039 3040 static void 3041 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 3042 { 3043 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 3044 dsl_pool_t *dp = dmu_tx_pool(tx); 3045 dsl_dataset_t *hds = NULL; 3046 3047 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 3048 ddrsa->ddrsa_tx = tx; 3049 if (ddrsa->ddrsa_recursive) { 3050 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 3051 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 3052 DS_FIND_CHILDREN)); 3053 } else { 3054 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 3055 } 3056 dsl_dataset_rele(hds, FTAG); 3057 } 3058 3059 int 3060 dsl_dataset_rename_snapshot(const char *fsname, 3061 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 3062 { 3063 dsl_dataset_rename_snapshot_arg_t ddrsa; 3064 3065 ddrsa.ddrsa_fsname = fsname; 3066 ddrsa.ddrsa_oldsnapname = oldsnapname; 3067 ddrsa.ddrsa_newsnapname = newsnapname; 3068 ddrsa.ddrsa_recursive = recursive; 3069 3070 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 3071 dsl_dataset_rename_snapshot_sync, &ddrsa, 3072 1, ZFS_SPACE_CHECK_RESERVED)); 3073 } 3074 3075 /* 3076 * If we're doing an ownership handoff, we need to make sure that there is 3077 * only one long hold on the dataset. We're not allowed to change anything here 3078 * so we don't permanently release the long hold or regular hold here. We want 3079 * to do this only when syncing to avoid the dataset unexpectedly going away 3080 * when we release the long hold. 3081 */ 3082 static int 3083 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 3084 { 3085 boolean_t held = B_FALSE; 3086 3087 if (!dmu_tx_is_syncing(tx)) 3088 return (0); 3089 3090 dsl_dir_t *dd = ds->ds_dir; 3091 mutex_enter(&dd->dd_activity_lock); 3092 uint64_t holds = zfs_refcount_count(&ds->ds_longholds) - 3093 (owner != NULL ? 1 : 0); 3094 /* 3095 * The value of dd_activity_waiters can chance as soon as we drop the 3096 * lock, but we're fine with that; new waiters coming in or old 3097 * waiters leaving doesn't cause problems, since we're going to cancel 3098 * waiters later anyway. The goal of this check is to verify that no 3099 * non-waiters have long-holds, and all new long-holds will be 3100 * prevented because we're holding the pool config as writer. 3101 */ 3102 if (holds != dd->dd_activity_waiters) 3103 held = B_TRUE; 3104 mutex_exit(&dd->dd_activity_lock); 3105 3106 if (held) 3107 return (SET_ERROR(EBUSY)); 3108 3109 return (0); 3110 } 3111 3112 int 3113 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 3114 { 3115 dsl_dataset_rollback_arg_t *ddra = arg; 3116 dsl_pool_t *dp = dmu_tx_pool(tx); 3117 dsl_dataset_t *ds; 3118 int64_t unused_refres_delta; 3119 int error; 3120 3121 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 3122 if (error != 0) 3123 return (error); 3124 3125 /* must not be a snapshot */ 3126 if (ds->ds_is_snapshot) { 3127 dsl_dataset_rele(ds, FTAG); 3128 return (SET_ERROR(EINVAL)); 3129 } 3130 3131 /* must have a most recent snapshot */ 3132 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { 3133 dsl_dataset_rele(ds, FTAG); 3134 return (SET_ERROR(ESRCH)); 3135 } 3136 3137 /* 3138 * No rollback to a snapshot created in the current txg, because 3139 * the rollback may dirty the dataset and create blocks that are 3140 * not reachable from the rootbp while having a birth txg that 3141 * falls into the snapshot's range. 3142 */ 3143 if (dmu_tx_is_syncing(tx) && 3144 dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) { 3145 dsl_dataset_rele(ds, FTAG); 3146 return (SET_ERROR(EAGAIN)); 3147 } 3148 3149 /* 3150 * If the expected target snapshot is specified, then check that 3151 * the latest snapshot is it. 3152 */ 3153 if (ddra->ddra_tosnap != NULL) { 3154 dsl_dataset_t *snapds; 3155 3156 /* Check if the target snapshot exists at all. */ 3157 error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds); 3158 if (error != 0) { 3159 /* 3160 * ESRCH is used to signal that the target snapshot does 3161 * not exist, while ENOENT is used to report that 3162 * the rolled back dataset does not exist. 3163 * ESRCH is also used to cover other cases where the 3164 * target snapshot is not related to the dataset being 3165 * rolled back such as being in a different pool. 3166 */ 3167 if (error == ENOENT || error == EXDEV) 3168 error = SET_ERROR(ESRCH); 3169 dsl_dataset_rele(ds, FTAG); 3170 return (error); 3171 } 3172 ASSERT(snapds->ds_is_snapshot); 3173 3174 /* Check if the snapshot is the latest snapshot indeed. */ 3175 if (snapds != ds->ds_prev) { 3176 /* 3177 * Distinguish between the case where the only problem 3178 * is intervening snapshots (EEXIST) vs the snapshot 3179 * not being a valid target for rollback (ESRCH). 3180 */ 3181 if (snapds->ds_dir == ds->ds_dir || 3182 (dsl_dir_is_clone(ds->ds_dir) && 3183 dsl_dir_phys(ds->ds_dir)->dd_origin_obj == 3184 snapds->ds_object)) { 3185 error = SET_ERROR(EEXIST); 3186 } else { 3187 error = SET_ERROR(ESRCH); 3188 } 3189 dsl_dataset_rele(snapds, FTAG); 3190 dsl_dataset_rele(ds, FTAG); 3191 return (error); 3192 } 3193 dsl_dataset_rele(snapds, FTAG); 3194 } 3195 3196 /* must not have any bookmarks after the most recent snapshot */ 3197 if (dsl_bookmark_latest_txg(ds) > 3198 dsl_dataset_phys(ds)->ds_prev_snap_txg) { 3199 dsl_dataset_rele(ds, FTAG); 3200 return (SET_ERROR(EEXIST)); 3201 } 3202 3203 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 3204 if (error != 0) { 3205 dsl_dataset_rele(ds, FTAG); 3206 return (error); 3207 } 3208 3209 /* 3210 * Check if the snap we are rolling back to uses more than 3211 * the refquota. 3212 */ 3213 if (ds->ds_quota != 0 && 3214 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { 3215 dsl_dataset_rele(ds, FTAG); 3216 return (SET_ERROR(EDQUOT)); 3217 } 3218 3219 /* 3220 * When we do the clone swap, we will temporarily use more space 3221 * due to the refreservation (the head will no longer have any 3222 * unique space, so the entire amount of the refreservation will need 3223 * to be free). We will immediately destroy the clone, freeing 3224 * this space, but the freeing happens over many txg's. 3225 */ 3226 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 3227 dsl_dataset_phys(ds)->ds_unique_bytes); 3228 3229 if (unused_refres_delta > 0 && 3230 unused_refres_delta > 3231 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 3232 dsl_dataset_rele(ds, FTAG); 3233 return (SET_ERROR(ENOSPC)); 3234 } 3235 3236 dsl_dataset_rele(ds, FTAG); 3237 return (0); 3238 } 3239 3240 void 3241 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 3242 { 3243 dsl_dataset_rollback_arg_t *ddra = arg; 3244 dsl_pool_t *dp = dmu_tx_pool(tx); 3245 dsl_dataset_t *ds, *clone; 3246 uint64_t cloneobj; 3247 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 3248 3249 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 3250 3251 dsl_dataset_name(ds->ds_prev, namebuf); 3252 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 3253 3254 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 3255 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, NULL, tx); 3256 3257 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 3258 3259 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 3260 dsl_dataset_zero_zil(ds, tx); 3261 3262 dsl_destroy_head_sync_impl(clone, tx); 3263 3264 dsl_dataset_rele(clone, FTAG); 3265 dsl_dataset_rele(ds, FTAG); 3266 } 3267 3268 /* 3269 * Rolls back the given filesystem or volume to the most recent snapshot. 3270 * The name of the most recent snapshot will be returned under key "target" 3271 * in the result nvlist. 3272 * 3273 * If owner != NULL: 3274 * - The existing dataset MUST be owned by the specified owner at entry 3275 * - Upon return, dataset will still be held by the same owner, whether we 3276 * succeed or not. 3277 * 3278 * This mode is required any time the existing filesystem is mounted. See 3279 * notes above zfs_suspend_fs() for further details. 3280 */ 3281 int 3282 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner, 3283 nvlist_t *result) 3284 { 3285 dsl_dataset_rollback_arg_t ddra; 3286 3287 ddra.ddra_fsname = fsname; 3288 ddra.ddra_tosnap = tosnap; 3289 ddra.ddra_owner = owner; 3290 ddra.ddra_result = result; 3291 3292 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 3293 dsl_dataset_rollback_sync, &ddra, 3294 1, ZFS_SPACE_CHECK_RESERVED)); 3295 } 3296 3297 struct promotenode { 3298 list_node_t link; 3299 dsl_dataset_t *ds; 3300 }; 3301 3302 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 3303 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 3304 void *tag); 3305 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 3306 3307 int 3308 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 3309 { 3310 dsl_dataset_promote_arg_t *ddpa = arg; 3311 dsl_pool_t *dp = dmu_tx_pool(tx); 3312 dsl_dataset_t *hds; 3313 struct promotenode *snap; 3314 dsl_dataset_t *origin_ds, *origin_head; 3315 int err; 3316 uint64_t unused; 3317 uint64_t ss_mv_cnt; 3318 size_t max_snap_len; 3319 boolean_t conflicting_snaps; 3320 3321 err = promote_hold(ddpa, dp, FTAG); 3322 if (err != 0) 3323 return (err); 3324 3325 hds = ddpa->ddpa_clone; 3326 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; 3327 3328 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { 3329 promote_rele(ddpa, FTAG); 3330 return (SET_ERROR(EXDEV)); 3331 } 3332 3333 snap = list_head(&ddpa->shared_snaps); 3334 origin_head = snap->ds; 3335 if (snap == NULL) { 3336 err = SET_ERROR(ENOENT); 3337 goto out; 3338 } 3339 origin_ds = snap->ds; 3340 3341 /* 3342 * Encrypted clones share a DSL Crypto Key with their origin's dsl dir. 3343 * When doing a promote we must make sure the encryption root for 3344 * both the target and the target's origin does not change to avoid 3345 * needing to rewrap encryption keys 3346 */ 3347 err = dsl_dataset_promote_crypt_check(hds->ds_dir, origin_ds->ds_dir); 3348 if (err != 0) 3349 goto out; 3350 3351 /* 3352 * Compute and check the amount of space to transfer. Since this is 3353 * so expensive, don't do the preliminary check. 3354 */ 3355 if (!dmu_tx_is_syncing(tx)) { 3356 promote_rele(ddpa, FTAG); 3357 return (0); 3358 } 3359 3360 /* compute origin's new unique space */ 3361 snap = list_tail(&ddpa->clone_snaps); 3362 ASSERT(snap != NULL); 3363 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 3364 origin_ds->ds_object); 3365 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 3366 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, 3367 &ddpa->unique, &unused, &unused); 3368 3369 /* 3370 * Walk the snapshots that we are moving 3371 * 3372 * Compute space to transfer. Consider the incremental changes 3373 * to used by each snapshot: 3374 * (my used) = (prev's used) + (blocks born) - (blocks killed) 3375 * So each snapshot gave birth to: 3376 * (blocks born) = (my used) - (prev's used) + (blocks killed) 3377 * So a sequence would look like: 3378 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 3379 * Which simplifies to: 3380 * uN + kN + kN-1 + ... + k1 + k0 3381 * Note however, if we stop before we reach the ORIGIN we get: 3382 * uN + kN + kN-1 + ... + kM - uM-1 3383 */ 3384 conflicting_snaps = B_FALSE; 3385 ss_mv_cnt = 0; 3386 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; 3387 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; 3388 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; 3389 for (snap = list_head(&ddpa->shared_snaps); snap; 3390 snap = list_next(&ddpa->shared_snaps, snap)) { 3391 uint64_t val, dlused, dlcomp, dluncomp; 3392 dsl_dataset_t *ds = snap->ds; 3393 3394 ss_mv_cnt++; 3395 3396 /* 3397 * If there are long holds, we won't be able to evict 3398 * the objset. 3399 */ 3400 if (dsl_dataset_long_held(ds)) { 3401 err = SET_ERROR(EBUSY); 3402 goto out; 3403 } 3404 3405 /* Check that the snapshot name does not conflict */ 3406 VERIFY0(dsl_dataset_get_snapname(ds)); 3407 if (strlen(ds->ds_snapname) >= max_snap_len) { 3408 err = SET_ERROR(ENAMETOOLONG); 3409 goto out; 3410 } 3411 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 3412 if (err == 0) { 3413 fnvlist_add_boolean(ddpa->err_ds, 3414 snap->ds->ds_snapname); 3415 conflicting_snaps = B_TRUE; 3416 } else if (err != ENOENT) { 3417 goto out; 3418 } 3419 3420 /* The very first snapshot does not have a deadlist */ 3421 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) 3422 continue; 3423 3424 dsl_deadlist_space(&ds->ds_deadlist, 3425 &dlused, &dlcomp, &dluncomp); 3426 ddpa->used += dlused; 3427 ddpa->comp += dlcomp; 3428 ddpa->uncomp += dluncomp; 3429 } 3430 3431 /* 3432 * Check that bookmarks that are being transferred don't have 3433 * name conflicts. 3434 */ 3435 for (dsl_bookmark_node_t *dbn = avl_first(&origin_head->ds_bookmarks); 3436 dbn != NULL && dbn->dbn_phys.zbm_creation_txg <= 3437 dsl_dataset_phys(origin_ds)->ds_creation_txg; 3438 dbn = AVL_NEXT(&origin_head->ds_bookmarks, dbn)) { 3439 if (strlen(dbn->dbn_name) >= max_snap_len) { 3440 err = SET_ERROR(ENAMETOOLONG); 3441 goto out; 3442 } 3443 zfs_bookmark_phys_t bm; 3444 err = dsl_bookmark_lookup_impl(ddpa->ddpa_clone, 3445 dbn->dbn_name, &bm); 3446 3447 if (err == 0) { 3448 fnvlist_add_boolean(ddpa->err_ds, dbn->dbn_name); 3449 conflicting_snaps = B_TRUE; 3450 } else if (err == ESRCH) { 3451 err = 0; 3452 } else if (err != 0) { 3453 goto out; 3454 } 3455 } 3456 3457 /* 3458 * In order to return the full list of conflicting snapshots, we check 3459 * whether there was a conflict after traversing all of them. 3460 */ 3461 if (conflicting_snaps) { 3462 err = SET_ERROR(EEXIST); 3463 goto out; 3464 } 3465 3466 /* 3467 * If we are a clone of a clone then we never reached ORIGIN, 3468 * so we need to subtract out the clone origin's used space. 3469 */ 3470 if (ddpa->origin_origin) { 3471 ddpa->used -= 3472 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; 3473 ddpa->comp -= 3474 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; 3475 ddpa->uncomp -= 3476 dsl_dataset_phys(ddpa->origin_origin)-> 3477 ds_uncompressed_bytes; 3478 } 3479 3480 /* Check that there is enough space and limit headroom here */ 3481 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 3482 0, ss_mv_cnt, ddpa->used, ddpa->cr, ddpa->proc); 3483 if (err != 0) 3484 goto out; 3485 3486 /* 3487 * Compute the amounts of space that will be used by snapshots 3488 * after the promotion (for both origin and clone). For each, 3489 * it is the amount of space that will be on all of their 3490 * deadlists (that was not born before their new origin). 3491 */ 3492 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 3493 uint64_t space; 3494 3495 /* 3496 * Note, typically this will not be a clone of a clone, 3497 * so dd_origin_txg will be < TXG_INITIAL, so 3498 * these snaplist_space() -> dsl_deadlist_space_range() 3499 * calls will be fast because they do not have to 3500 * iterate over all bps. 3501 */ 3502 snap = list_head(&ddpa->origin_snaps); 3503 if (snap == NULL) { 3504 err = SET_ERROR(ENOENT); 3505 goto out; 3506 } 3507 err = snaplist_space(&ddpa->shared_snaps, 3508 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 3509 if (err != 0) 3510 goto out; 3511 3512 err = snaplist_space(&ddpa->clone_snaps, 3513 snap->ds->ds_dir->dd_origin_txg, &space); 3514 if (err != 0) 3515 goto out; 3516 ddpa->cloneusedsnap += space; 3517 } 3518 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & 3519 DD_FLAG_USED_BREAKDOWN) { 3520 err = snaplist_space(&ddpa->origin_snaps, 3521 dsl_dataset_phys(origin_ds)->ds_creation_txg, 3522 &ddpa->originusedsnap); 3523 if (err != 0) 3524 goto out; 3525 } 3526 3527 out: 3528 promote_rele(ddpa, FTAG); 3529 return (err); 3530 } 3531 3532 void 3533 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 3534 { 3535 dsl_dataset_promote_arg_t *ddpa = arg; 3536 dsl_pool_t *dp = dmu_tx_pool(tx); 3537 dsl_dataset_t *hds; 3538 struct promotenode *snap; 3539 dsl_dataset_t *origin_ds; 3540 dsl_dataset_t *origin_head; 3541 dsl_dir_t *dd; 3542 dsl_dir_t *odd = NULL; 3543 uint64_t oldnext_obj; 3544 int64_t delta; 3545 3546 ASSERT(nvlist_empty(ddpa->err_ds)); 3547 3548 VERIFY0(promote_hold(ddpa, dp, FTAG)); 3549 hds = ddpa->ddpa_clone; 3550 3551 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); 3552 3553 snap = list_head(&ddpa->shared_snaps); 3554 origin_ds = snap->ds; 3555 dd = hds->ds_dir; 3556 3557 snap = list_head(&ddpa->origin_snaps); 3558 origin_head = snap->ds; 3559 3560 /* 3561 * We need to explicitly open odd, since origin_ds's dd will be 3562 * changing. 3563 */ 3564 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 3565 NULL, FTAG, &odd)); 3566 3567 dsl_dataset_promote_crypt_sync(hds->ds_dir, odd, tx); 3568 3569 /* change origin's next snap */ 3570 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 3571 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; 3572 snap = list_tail(&ddpa->clone_snaps); 3573 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 3574 origin_ds->ds_object); 3575 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; 3576 3577 /* change the origin's next clone */ 3578 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { 3579 dsl_dataset_remove_from_next_clones(origin_ds, 3580 snap->ds->ds_object, tx); 3581 VERIFY0(zap_add_int(dp->dp_meta_objset, 3582 dsl_dataset_phys(origin_ds)->ds_next_clones_obj, 3583 oldnext_obj, tx)); 3584 } 3585 3586 /* change origin */ 3587 dmu_buf_will_dirty(dd->dd_dbuf, tx); 3588 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); 3589 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; 3590 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 3591 dmu_buf_will_dirty(odd->dd_dbuf, tx); 3592 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; 3593 origin_head->ds_dir->dd_origin_txg = 3594 dsl_dataset_phys(origin_ds)->ds_creation_txg; 3595 3596 /* change dd_clone entries */ 3597 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 3598 VERIFY0(zap_remove_int(dp->dp_meta_objset, 3599 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); 3600 VERIFY0(zap_add_int(dp->dp_meta_objset, 3601 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 3602 hds->ds_object, tx)); 3603 3604 VERIFY0(zap_remove_int(dp->dp_meta_objset, 3605 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 3606 origin_head->ds_object, tx)); 3607 if (dsl_dir_phys(dd)->dd_clones == 0) { 3608 dsl_dir_phys(dd)->dd_clones = 3609 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, 3610 DMU_OT_NONE, 0, tx); 3611 } 3612 VERIFY0(zap_add_int(dp->dp_meta_objset, 3613 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); 3614 } 3615 3616 /* 3617 * Move bookmarks to this dir. 3618 */ 3619 dsl_bookmark_node_t *dbn_next; 3620 for (dsl_bookmark_node_t *dbn = avl_first(&origin_head->ds_bookmarks); 3621 dbn != NULL && dbn->dbn_phys.zbm_creation_txg <= 3622 dsl_dataset_phys(origin_ds)->ds_creation_txg; 3623 dbn = dbn_next) { 3624 dbn_next = AVL_NEXT(&origin_head->ds_bookmarks, dbn); 3625 3626 avl_remove(&origin_head->ds_bookmarks, dbn); 3627 VERIFY0(zap_remove(dp->dp_meta_objset, 3628 origin_head->ds_bookmarks_obj, dbn->dbn_name, tx)); 3629 3630 dsl_bookmark_node_add(hds, dbn, tx); 3631 } 3632 3633 dsl_bookmark_next_changed(hds, origin_ds, tx); 3634 3635 /* move snapshots to this dir */ 3636 for (snap = list_head(&ddpa->shared_snaps); snap; 3637 snap = list_next(&ddpa->shared_snaps, snap)) { 3638 dsl_dataset_t *ds = snap->ds; 3639 3640 /* 3641 * Property callbacks are registered to a particular 3642 * dsl_dir. Since ours is changing, evict the objset 3643 * so that they will be unregistered from the old dsl_dir. 3644 */ 3645 if (ds->ds_objset) { 3646 dmu_objset_evict(ds->ds_objset); 3647 ds->ds_objset = NULL; 3648 } 3649 3650 /* move snap name entry */ 3651 VERIFY0(dsl_dataset_get_snapname(ds)); 3652 VERIFY0(dsl_dataset_snap_remove(origin_head, 3653 ds->ds_snapname, tx, B_TRUE)); 3654 VERIFY0(zap_add(dp->dp_meta_objset, 3655 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 3656 8, 1, &ds->ds_object, tx)); 3657 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 3658 DD_FIELD_SNAPSHOT_COUNT, tx); 3659 3660 /* change containing dsl_dir */ 3661 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3662 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); 3663 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; 3664 ASSERT3P(ds->ds_dir, ==, odd); 3665 dsl_dir_rele(ds->ds_dir, ds); 3666 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 3667 NULL, ds, &ds->ds_dir)); 3668 3669 /* move any clone references */ 3670 if (dsl_dataset_phys(ds)->ds_next_clones_obj && 3671 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 3672 zap_cursor_t zc; 3673 zap_attribute_t za; 3674 3675 for (zap_cursor_init(&zc, dp->dp_meta_objset, 3676 dsl_dataset_phys(ds)->ds_next_clones_obj); 3677 zap_cursor_retrieve(&zc, &za) == 0; 3678 zap_cursor_advance(&zc)) { 3679 dsl_dataset_t *cnds; 3680 uint64_t o; 3681 3682 if (za.za_first_integer == oldnext_obj) { 3683 /* 3684 * We've already moved the 3685 * origin's reference. 3686 */ 3687 continue; 3688 } 3689 3690 VERIFY0(dsl_dataset_hold_obj(dp, 3691 za.za_first_integer, FTAG, &cnds)); 3692 o = dsl_dir_phys(cnds->ds_dir)-> 3693 dd_head_dataset_obj; 3694 3695 VERIFY0(zap_remove_int(dp->dp_meta_objset, 3696 dsl_dir_phys(odd)->dd_clones, o, tx)); 3697 VERIFY0(zap_add_int(dp->dp_meta_objset, 3698 dsl_dir_phys(dd)->dd_clones, o, tx)); 3699 dsl_dataset_rele(cnds, FTAG); 3700 } 3701 zap_cursor_fini(&zc); 3702 } 3703 3704 ASSERT(!dsl_prop_hascb(ds)); 3705 } 3706 3707 /* 3708 * Change space accounting. 3709 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 3710 * both be valid, or both be 0 (resulting in delta == 0). This 3711 * is true for each of {clone,origin} independently. 3712 */ 3713 3714 delta = ddpa->cloneusedsnap - 3715 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; 3716 ASSERT3S(delta, >=, 0); 3717 ASSERT3U(ddpa->used, >=, delta); 3718 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 3719 dsl_dir_diduse_space(dd, DD_USED_HEAD, 3720 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 3721 3722 delta = ddpa->originusedsnap - 3723 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; 3724 ASSERT3S(delta, <=, 0); 3725 ASSERT3U(ddpa->used, >=, -delta); 3726 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 3727 dsl_dir_diduse_space(odd, DD_USED_HEAD, 3728 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 3729 3730 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; 3731 3732 /* 3733 * Since livelists are specific to a clone's origin txg, they 3734 * are no longer accurate. Destroy the livelist from the clone being 3735 * promoted. If the origin dataset is a clone, destroy its livelist 3736 * as well. 3737 */ 3738 dsl_dir_remove_livelist(dd, tx, B_TRUE); 3739 dsl_dir_remove_livelist(odd, tx, B_TRUE); 3740 3741 /* log history record */ 3742 spa_history_log_internal_ds(hds, "promote", tx, " "); 3743 3744 dsl_dir_rele(odd, FTAG); 3745 promote_rele(ddpa, FTAG); 3746 } 3747 3748 /* 3749 * Make a list of dsl_dataset_t's for the snapshots between first_obj 3750 * (exclusive) and last_obj (inclusive). The list will be in reverse 3751 * order (last_obj will be the list_head()). If first_obj == 0, do all 3752 * snapshots back to this dataset's origin. 3753 */ 3754 static int 3755 snaplist_make(dsl_pool_t *dp, 3756 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 3757 { 3758 uint64_t obj = last_obj; 3759 3760 list_create(l, sizeof (struct promotenode), 3761 offsetof(struct promotenode, link)); 3762 3763 while (obj != first_obj) { 3764 dsl_dataset_t *ds; 3765 struct promotenode *snap; 3766 int err; 3767 3768 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 3769 ASSERT(err != ENOENT); 3770 if (err != 0) 3771 return (err); 3772 3773 if (first_obj == 0) 3774 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; 3775 3776 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 3777 snap->ds = ds; 3778 list_insert_tail(l, snap); 3779 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3780 } 3781 3782 return (0); 3783 } 3784 3785 static int 3786 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 3787 { 3788 struct promotenode *snap; 3789 3790 *spacep = 0; 3791 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 3792 uint64_t used, comp, uncomp; 3793 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 3794 mintxg, UINT64_MAX, &used, &comp, &uncomp); 3795 *spacep += used; 3796 } 3797 return (0); 3798 } 3799 3800 static void 3801 snaplist_destroy(list_t *l, void *tag) 3802 { 3803 struct promotenode *snap; 3804 3805 if (l == NULL || !list_link_active(&l->list_head)) 3806 return; 3807 3808 while ((snap = list_tail(l)) != NULL) { 3809 list_remove(l, snap); 3810 dsl_dataset_rele(snap->ds, tag); 3811 kmem_free(snap, sizeof (*snap)); 3812 } 3813 list_destroy(l); 3814 } 3815 3816 static int 3817 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 3818 { 3819 int error; 3820 dsl_dir_t *dd; 3821 struct promotenode *snap; 3822 3823 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 3824 &ddpa->ddpa_clone); 3825 if (error != 0) 3826 return (error); 3827 dd = ddpa->ddpa_clone->ds_dir; 3828 3829 if (ddpa->ddpa_clone->ds_is_snapshot || 3830 !dsl_dir_is_clone(dd)) { 3831 dsl_dataset_rele(ddpa->ddpa_clone, tag); 3832 return (SET_ERROR(EINVAL)); 3833 } 3834 3835 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, 3836 &ddpa->shared_snaps, tag); 3837 if (error != 0) 3838 goto out; 3839 3840 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 3841 &ddpa->clone_snaps, tag); 3842 if (error != 0) 3843 goto out; 3844 3845 snap = list_head(&ddpa->shared_snaps); 3846 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); 3847 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, 3848 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, 3849 &ddpa->origin_snaps, tag); 3850 if (error != 0) 3851 goto out; 3852 3853 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { 3854 error = dsl_dataset_hold_obj(dp, 3855 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, 3856 tag, &ddpa->origin_origin); 3857 if (error != 0) 3858 goto out; 3859 } 3860 out: 3861 if (error != 0) 3862 promote_rele(ddpa, tag); 3863 return (error); 3864 } 3865 3866 static void 3867 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 3868 { 3869 snaplist_destroy(&ddpa->shared_snaps, tag); 3870 snaplist_destroy(&ddpa->clone_snaps, tag); 3871 snaplist_destroy(&ddpa->origin_snaps, tag); 3872 if (ddpa->origin_origin != NULL) 3873 dsl_dataset_rele(ddpa->origin_origin, tag); 3874 dsl_dataset_rele(ddpa->ddpa_clone, tag); 3875 } 3876 3877 /* 3878 * Promote a clone. 3879 * 3880 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 3881 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.) 3882 */ 3883 int 3884 dsl_dataset_promote(const char *name, char *conflsnap) 3885 { 3886 dsl_dataset_promote_arg_t ddpa = { 0 }; 3887 uint64_t numsnaps; 3888 int error; 3889 nvpair_t *snap_pair; 3890 objset_t *os; 3891 3892 /* 3893 * We will modify space proportional to the number of 3894 * snapshots. Compute numsnaps. 3895 */ 3896 error = dmu_objset_hold(name, FTAG, &os); 3897 if (error != 0) 3898 return (error); 3899 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 3900 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, 3901 &numsnaps); 3902 dmu_objset_rele(os, FTAG); 3903 if (error != 0) 3904 return (error); 3905 3906 ddpa.ddpa_clonename = name; 3907 ddpa.err_ds = fnvlist_alloc(); 3908 ddpa.cr = CRED(); 3909 ddpa.proc = curproc; 3910 3911 error = dsl_sync_task(name, dsl_dataset_promote_check, 3912 dsl_dataset_promote_sync, &ddpa, 3913 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED); 3914 3915 /* 3916 * Return the first conflicting snapshot found. 3917 */ 3918 snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL); 3919 if (snap_pair != NULL && conflsnap != NULL) 3920 (void) strlcpy(conflsnap, nvpair_name(snap_pair), 3921 ZFS_MAX_DATASET_NAME_LEN); 3922 3923 fnvlist_free(ddpa.err_ds); 3924 return (error); 3925 } 3926 3927 int 3928 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 3929 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 3930 { 3931 /* 3932 * "slack" factor for received datasets with refquota set on them. 3933 * See the bottom of this function for details on its use. 3934 */ 3935 uint64_t refquota_slack = (uint64_t)DMU_MAX_ACCESS * 3936 spa_asize_inflation; 3937 int64_t unused_refres_delta; 3938 3939 /* they should both be heads */ 3940 if (clone->ds_is_snapshot || 3941 origin_head->ds_is_snapshot) 3942 return (SET_ERROR(EINVAL)); 3943 3944 /* if we are not forcing, the branch point should be just before them */ 3945 if (!force && clone->ds_prev != origin_head->ds_prev) 3946 return (SET_ERROR(EINVAL)); 3947 3948 /* clone should be the clone (unless they are unrelated) */ 3949 if (clone->ds_prev != NULL && 3950 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 3951 origin_head->ds_dir != clone->ds_prev->ds_dir) 3952 return (SET_ERROR(EINVAL)); 3953 3954 /* the clone should be a child of the origin */ 3955 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 3956 return (SET_ERROR(EINVAL)); 3957 3958 /* origin_head shouldn't be modified unless 'force' */ 3959 if (!force && 3960 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 3961 return (SET_ERROR(ETXTBSY)); 3962 3963 /* origin_head should have no long holds (e.g. is not mounted) */ 3964 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 3965 return (SET_ERROR(EBUSY)); 3966 3967 /* check amount of any unconsumed refreservation */ 3968 unused_refres_delta = 3969 (int64_t)MIN(origin_head->ds_reserved, 3970 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 3971 (int64_t)MIN(origin_head->ds_reserved, 3972 dsl_dataset_phys(clone)->ds_unique_bytes); 3973 3974 if (unused_refres_delta > 0 && 3975 unused_refres_delta > 3976 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 3977 return (SET_ERROR(ENOSPC)); 3978 3979 /* 3980 * The clone can't be too much over the head's refquota. 3981 * 3982 * To ensure that the entire refquota can be used, we allow one 3983 * transaction to exceed the refquota. Therefore, this check 3984 * needs to also allow for the space referenced to be more than the 3985 * refquota. The maximum amount of space that one transaction can use 3986 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this 3987 * overage ensures that we are able to receive a filesystem that 3988 * exceeds the refquota on the source system. 3989 * 3990 * So that overage is the refquota_slack we use below. 3991 */ 3992 if (origin_head->ds_quota != 0 && 3993 dsl_dataset_phys(clone)->ds_referenced_bytes > 3994 origin_head->ds_quota + refquota_slack) 3995 return (SET_ERROR(EDQUOT)); 3996 3997 return (0); 3998 } 3999 4000 static void 4001 dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone, 4002 dsl_dataset_t *origin, dmu_tx_t *tx) 4003 { 4004 uint64_t clone_remap_dl_obj, origin_remap_dl_obj; 4005 dsl_pool_t *dp = dmu_tx_pool(tx); 4006 4007 ASSERT(dsl_pool_sync_context(dp)); 4008 4009 clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone); 4010 origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin); 4011 4012 if (clone_remap_dl_obj != 0) { 4013 dsl_deadlist_close(&clone->ds_remap_deadlist); 4014 dsl_dataset_unset_remap_deadlist_object(clone, tx); 4015 } 4016 if (origin_remap_dl_obj != 0) { 4017 dsl_deadlist_close(&origin->ds_remap_deadlist); 4018 dsl_dataset_unset_remap_deadlist_object(origin, tx); 4019 } 4020 4021 if (clone_remap_dl_obj != 0) { 4022 dsl_dataset_set_remap_deadlist_object(origin, 4023 clone_remap_dl_obj, tx); 4024 dsl_deadlist_open(&origin->ds_remap_deadlist, 4025 dp->dp_meta_objset, clone_remap_dl_obj); 4026 } 4027 if (origin_remap_dl_obj != 0) { 4028 dsl_dataset_set_remap_deadlist_object(clone, 4029 origin_remap_dl_obj, tx); 4030 dsl_deadlist_open(&clone->ds_remap_deadlist, 4031 dp->dp_meta_objset, origin_remap_dl_obj); 4032 } 4033 } 4034 4035 void 4036 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 4037 dsl_dataset_t *origin_head, dmu_tx_t *tx) 4038 { 4039 dsl_pool_t *dp = dmu_tx_pool(tx); 4040 int64_t unused_refres_delta; 4041 4042 ASSERT(clone->ds_reserved == 0); 4043 /* 4044 * NOTE: On DEBUG kernels there could be a race between this and 4045 * the check function if spa_asize_inflation is adjusted... 4046 */ 4047 ASSERT(origin_head->ds_quota == 0 || 4048 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota + 4049 DMU_MAX_ACCESS * spa_asize_inflation); 4050 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 4051 4052 dsl_dir_cancel_waiters(origin_head->ds_dir); 4053 4054 /* 4055 * Swap per-dataset feature flags. 4056 */ 4057 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 4058 if (!(spa_feature_table[f].fi_flags & 4059 ZFEATURE_FLAG_PER_DATASET)) { 4060 ASSERT(!dsl_dataset_feature_is_active(clone, f)); 4061 ASSERT(!dsl_dataset_feature_is_active(origin_head, f)); 4062 continue; 4063 } 4064 4065 boolean_t clone_inuse = dsl_dataset_feature_is_active(clone, f); 4066 void *clone_feature = clone->ds_feature[f]; 4067 boolean_t origin_head_inuse = 4068 dsl_dataset_feature_is_active(origin_head, f); 4069 void *origin_head_feature = origin_head->ds_feature[f]; 4070 4071 if (clone_inuse) 4072 dsl_dataset_deactivate_feature_impl(clone, f, tx); 4073 if (origin_head_inuse) 4074 dsl_dataset_deactivate_feature_impl(origin_head, f, tx); 4075 4076 if (clone_inuse) { 4077 dsl_dataset_activate_feature(origin_head->ds_object, f, 4078 clone_feature, tx); 4079 origin_head->ds_feature[f] = clone_feature; 4080 } 4081 if (origin_head_inuse) { 4082 dsl_dataset_activate_feature(clone->ds_object, f, 4083 origin_head_feature, tx); 4084 clone->ds_feature[f] = origin_head_feature; 4085 } 4086 } 4087 4088 dmu_buf_will_dirty(clone->ds_dbuf, tx); 4089 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 4090 4091 if (clone->ds_objset != NULL) { 4092 dmu_objset_evict(clone->ds_objset); 4093 clone->ds_objset = NULL; 4094 } 4095 4096 if (origin_head->ds_objset != NULL) { 4097 dmu_objset_evict(origin_head->ds_objset); 4098 origin_head->ds_objset = NULL; 4099 } 4100 4101 unused_refres_delta = 4102 (int64_t)MIN(origin_head->ds_reserved, 4103 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 4104 (int64_t)MIN(origin_head->ds_reserved, 4105 dsl_dataset_phys(clone)->ds_unique_bytes); 4106 4107 /* 4108 * Reset origin's unique bytes. 4109 */ 4110 { 4111 dsl_dataset_t *origin = clone->ds_prev; 4112 uint64_t comp, uncomp; 4113 4114 dmu_buf_will_dirty(origin->ds_dbuf, tx); 4115 dsl_deadlist_space_range(&clone->ds_deadlist, 4116 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, 4117 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); 4118 } 4119 4120 /* swap blkptrs */ 4121 { 4122 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG); 4123 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG); 4124 blkptr_t tmp; 4125 tmp = dsl_dataset_phys(origin_head)->ds_bp; 4126 dsl_dataset_phys(origin_head)->ds_bp = 4127 dsl_dataset_phys(clone)->ds_bp; 4128 dsl_dataset_phys(clone)->ds_bp = tmp; 4129 rrw_exit(&origin_head->ds_bp_rwlock, FTAG); 4130 rrw_exit(&clone->ds_bp_rwlock, FTAG); 4131 } 4132 4133 /* set dd_*_bytes */ 4134 { 4135 int64_t dused, dcomp, duncomp; 4136 uint64_t cdl_used, cdl_comp, cdl_uncomp; 4137 uint64_t odl_used, odl_comp, odl_uncomp; 4138 4139 ASSERT3U(dsl_dir_phys(clone->ds_dir)-> 4140 dd_used_breakdown[DD_USED_SNAP], ==, 0); 4141 4142 dsl_deadlist_space(&clone->ds_deadlist, 4143 &cdl_used, &cdl_comp, &cdl_uncomp); 4144 dsl_deadlist_space(&origin_head->ds_deadlist, 4145 &odl_used, &odl_comp, &odl_uncomp); 4146 4147 dused = dsl_dataset_phys(clone)->ds_referenced_bytes + 4148 cdl_used - 4149 (dsl_dataset_phys(origin_head)->ds_referenced_bytes + 4150 odl_used); 4151 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + 4152 cdl_comp - 4153 (dsl_dataset_phys(origin_head)->ds_compressed_bytes + 4154 odl_comp); 4155 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + 4156 cdl_uncomp - 4157 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + 4158 odl_uncomp); 4159 4160 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 4161 dused, dcomp, duncomp, tx); 4162 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 4163 -dused, -dcomp, -duncomp, tx); 4164 4165 /* 4166 * The difference in the space used by snapshots is the 4167 * difference in snapshot space due to the head's 4168 * deadlist (since that's the only thing that's 4169 * changing that affects the snapused). 4170 */ 4171 dsl_deadlist_space_range(&clone->ds_deadlist, 4172 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 4173 &cdl_used, &cdl_comp, &cdl_uncomp); 4174 dsl_deadlist_space_range(&origin_head->ds_deadlist, 4175 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 4176 &odl_used, &odl_comp, &odl_uncomp); 4177 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 4178 DD_USED_HEAD, DD_USED_SNAP, tx); 4179 } 4180 4181 /* swap ds_*_bytes */ 4182 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, 4183 dsl_dataset_phys(clone)->ds_referenced_bytes); 4184 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, 4185 dsl_dataset_phys(clone)->ds_compressed_bytes); 4186 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, 4187 dsl_dataset_phys(clone)->ds_uncompressed_bytes); 4188 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, 4189 dsl_dataset_phys(clone)->ds_unique_bytes); 4190 4191 /* apply any parent delta for change in unconsumed refreservation */ 4192 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 4193 unused_refres_delta, 0, 0, tx); 4194 4195 /* 4196 * Swap deadlists. 4197 */ 4198 dsl_deadlist_close(&clone->ds_deadlist); 4199 dsl_deadlist_close(&origin_head->ds_deadlist); 4200 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, 4201 dsl_dataset_phys(clone)->ds_deadlist_obj); 4202 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 4203 dsl_dataset_phys(clone)->ds_deadlist_obj); 4204 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 4205 dsl_dataset_phys(origin_head)->ds_deadlist_obj); 4206 dsl_dataset_swap_remap_deadlists(clone, origin_head, tx); 4207 4208 /* 4209 * If there is a bookmark at the origin, its "next dataset" is 4210 * changing, so we need to reset its FBN. 4211 */ 4212 dsl_bookmark_next_changed(origin_head, origin_head->ds_prev, tx); 4213 4214 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 4215 4216 /* 4217 * Destroy any livelists associated with the clone or the origin, 4218 * since after the swap the corresponding livelists are no longer 4219 * valid. 4220 */ 4221 dsl_dir_remove_livelist(clone->ds_dir, tx, B_TRUE); 4222 dsl_dir_remove_livelist(origin_head->ds_dir, tx, B_TRUE); 4223 4224 spa_history_log_internal_ds(clone, "clone swap", tx, 4225 "parent=%s", origin_head->ds_dir->dd_myname); 4226 } 4227 4228 /* 4229 * Given a pool name and a dataset object number in that pool, 4230 * return the name of that dataset. 4231 */ 4232 int 4233 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 4234 { 4235 dsl_pool_t *dp; 4236 dsl_dataset_t *ds; 4237 int error; 4238 4239 error = dsl_pool_hold(pname, FTAG, &dp); 4240 if (error != 0) 4241 return (error); 4242 4243 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 4244 if (error == 0) { 4245 dsl_dataset_name(ds, buf); 4246 dsl_dataset_rele(ds, FTAG); 4247 } 4248 dsl_pool_rele(dp, FTAG); 4249 4250 return (error); 4251 } 4252 4253 int 4254 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 4255 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 4256 { 4257 int error = 0; 4258 4259 ASSERT3S(asize, >, 0); 4260 4261 /* 4262 * *ref_rsrv is the portion of asize that will come from any 4263 * unconsumed refreservation space. 4264 */ 4265 *ref_rsrv = 0; 4266 4267 mutex_enter(&ds->ds_lock); 4268 /* 4269 * Make a space adjustment for reserved bytes. 4270 */ 4271 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 4272 ASSERT3U(*used, >=, 4273 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 4274 *used -= 4275 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 4276 *ref_rsrv = 4277 asize - MIN(asize, parent_delta(ds, asize + inflight)); 4278 } 4279 4280 if (!check_quota || ds->ds_quota == 0) { 4281 mutex_exit(&ds->ds_lock); 4282 return (0); 4283 } 4284 /* 4285 * If they are requesting more space, and our current estimate 4286 * is over quota, they get to try again unless the actual 4287 * on-disk is over quota and there are no pending changes (which 4288 * may free up space for us). 4289 */ 4290 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= 4291 ds->ds_quota) { 4292 if (inflight > 0 || 4293 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) 4294 error = SET_ERROR(ERESTART); 4295 else 4296 error = SET_ERROR(EDQUOT); 4297 } 4298 mutex_exit(&ds->ds_lock); 4299 4300 return (error); 4301 } 4302 4303 typedef struct dsl_dataset_set_qr_arg { 4304 const char *ddsqra_name; 4305 zprop_source_t ddsqra_source; 4306 uint64_t ddsqra_value; 4307 } dsl_dataset_set_qr_arg_t; 4308 4309 4310 /* ARGSUSED */ 4311 static int 4312 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 4313 { 4314 dsl_dataset_set_qr_arg_t *ddsqra = arg; 4315 dsl_pool_t *dp = dmu_tx_pool(tx); 4316 dsl_dataset_t *ds; 4317 int error; 4318 uint64_t newval; 4319 4320 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 4321 return (SET_ERROR(ENOTSUP)); 4322 4323 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 4324 if (error != 0) 4325 return (error); 4326 4327 if (ds->ds_is_snapshot) { 4328 dsl_dataset_rele(ds, FTAG); 4329 return (SET_ERROR(EINVAL)); 4330 } 4331 4332 error = dsl_prop_predict(ds->ds_dir, 4333 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 4334 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 4335 if (error != 0) { 4336 dsl_dataset_rele(ds, FTAG); 4337 return (error); 4338 } 4339 4340 if (newval == 0) { 4341 dsl_dataset_rele(ds, FTAG); 4342 return (0); 4343 } 4344 4345 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || 4346 newval < ds->ds_reserved) { 4347 dsl_dataset_rele(ds, FTAG); 4348 return (SET_ERROR(ENOSPC)); 4349 } 4350 4351 dsl_dataset_rele(ds, FTAG); 4352 return (0); 4353 } 4354 4355 static void 4356 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 4357 { 4358 dsl_dataset_set_qr_arg_t *ddsqra = arg; 4359 dsl_pool_t *dp = dmu_tx_pool(tx); 4360 dsl_dataset_t *ds = NULL; 4361 uint64_t newval; 4362 4363 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 4364 4365 dsl_prop_set_sync_impl(ds, 4366 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 4367 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 4368 &ddsqra->ddsqra_value, tx); 4369 4370 VERIFY0(dsl_prop_get_int_ds(ds, 4371 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 4372 4373 if (ds->ds_quota != newval) { 4374 dmu_buf_will_dirty(ds->ds_dbuf, tx); 4375 ds->ds_quota = newval; 4376 } 4377 dsl_dataset_rele(ds, FTAG); 4378 } 4379 4380 int 4381 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 4382 uint64_t refquota) 4383 { 4384 dsl_dataset_set_qr_arg_t ddsqra; 4385 4386 ddsqra.ddsqra_name = dsname; 4387 ddsqra.ddsqra_source = source; 4388 ddsqra.ddsqra_value = refquota; 4389 4390 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 4391 dsl_dataset_set_refquota_sync, &ddsqra, 0, 4392 ZFS_SPACE_CHECK_EXTRA_RESERVED)); 4393 } 4394 4395 static int 4396 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 4397 { 4398 dsl_dataset_set_qr_arg_t *ddsqra = arg; 4399 dsl_pool_t *dp = dmu_tx_pool(tx); 4400 dsl_dataset_t *ds; 4401 int error; 4402 uint64_t newval, unique; 4403 4404 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 4405 return (SET_ERROR(ENOTSUP)); 4406 4407 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 4408 if (error != 0) 4409 return (error); 4410 4411 if (ds->ds_is_snapshot) { 4412 dsl_dataset_rele(ds, FTAG); 4413 return (SET_ERROR(EINVAL)); 4414 } 4415 4416 error = dsl_prop_predict(ds->ds_dir, 4417 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 4418 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 4419 if (error != 0) { 4420 dsl_dataset_rele(ds, FTAG); 4421 return (error); 4422 } 4423 4424 /* 4425 * If we are doing the preliminary check in open context, the 4426 * space estimates may be inaccurate. 4427 */ 4428 if (!dmu_tx_is_syncing(tx)) { 4429 dsl_dataset_rele(ds, FTAG); 4430 return (0); 4431 } 4432 4433 mutex_enter(&ds->ds_lock); 4434 if (!DS_UNIQUE_IS_ACCURATE(ds)) 4435 dsl_dataset_recalc_head_uniq(ds); 4436 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 4437 mutex_exit(&ds->ds_lock); 4438 4439 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 4440 uint64_t delta = MAX(unique, newval) - 4441 MAX(unique, ds->ds_reserved); 4442 4443 if (delta > 4444 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 4445 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 4446 dsl_dataset_rele(ds, FTAG); 4447 return (SET_ERROR(ENOSPC)); 4448 } 4449 } 4450 4451 dsl_dataset_rele(ds, FTAG); 4452 return (0); 4453 } 4454 4455 void 4456 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 4457 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 4458 { 4459 uint64_t newval; 4460 uint64_t unique; 4461 int64_t delta; 4462 4463 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 4464 source, sizeof (value), 1, &value, tx); 4465 4466 VERIFY0(dsl_prop_get_int_ds(ds, 4467 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 4468 4469 dmu_buf_will_dirty(ds->ds_dbuf, tx); 4470 mutex_enter(&ds->ds_dir->dd_lock); 4471 mutex_enter(&ds->ds_lock); 4472 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 4473 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 4474 delta = MAX(0, (int64_t)(newval - unique)) - 4475 MAX(0, (int64_t)(ds->ds_reserved - unique)); 4476 ds->ds_reserved = newval; 4477 mutex_exit(&ds->ds_lock); 4478 4479 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 4480 mutex_exit(&ds->ds_dir->dd_lock); 4481 } 4482 4483 static void 4484 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 4485 { 4486 dsl_dataset_set_qr_arg_t *ddsqra = arg; 4487 dsl_pool_t *dp = dmu_tx_pool(tx); 4488 dsl_dataset_t *ds = NULL; 4489 4490 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 4491 dsl_dataset_set_refreservation_sync_impl(ds, 4492 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 4493 dsl_dataset_rele(ds, FTAG); 4494 } 4495 4496 int 4497 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 4498 uint64_t refreservation) 4499 { 4500 dsl_dataset_set_qr_arg_t ddsqra; 4501 4502 ddsqra.ddsqra_name = dsname; 4503 ddsqra.ddsqra_source = source; 4504 ddsqra.ddsqra_value = refreservation; 4505 4506 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 4507 dsl_dataset_set_refreservation_sync, &ddsqra, 0, 4508 ZFS_SPACE_CHECK_EXTRA_RESERVED)); 4509 } 4510 4511 typedef struct dsl_dataset_set_compression_arg { 4512 const char *ddsca_name; 4513 zprop_source_t ddsca_source; 4514 uint64_t ddsca_value; 4515 } dsl_dataset_set_compression_arg_t; 4516 4517 /* ARGSUSED */ 4518 static int 4519 dsl_dataset_set_compression_check(void *arg, dmu_tx_t *tx) 4520 { 4521 dsl_dataset_set_compression_arg_t *ddsca = arg; 4522 dsl_pool_t *dp = dmu_tx_pool(tx); 4523 4524 uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value); 4525 spa_feature_t f = zio_compress_to_feature(compval); 4526 4527 if (f == SPA_FEATURE_NONE) 4528 return (SET_ERROR(EINVAL)); 4529 4530 if (!spa_feature_is_enabled(dp->dp_spa, f)) 4531 return (SET_ERROR(ENOTSUP)); 4532 4533 return (0); 4534 } 4535 4536 static void 4537 dsl_dataset_set_compression_sync(void *arg, dmu_tx_t *tx) 4538 { 4539 dsl_dataset_set_compression_arg_t *ddsca = arg; 4540 dsl_pool_t *dp = dmu_tx_pool(tx); 4541 dsl_dataset_t *ds = NULL; 4542 4543 uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value); 4544 spa_feature_t f = zio_compress_to_feature(compval); 4545 ASSERT3S(spa_feature_table[f].fi_type, ==, ZFEATURE_TYPE_BOOLEAN); 4546 4547 VERIFY0(dsl_dataset_hold(dp, ddsca->ddsca_name, FTAG, &ds)); 4548 if (zfeature_active(f, ds->ds_feature[f]) != B_TRUE) { 4549 ds->ds_feature_activation[f] = (void *)B_TRUE; 4550 dsl_dataset_activate_feature(ds->ds_object, f, 4551 ds->ds_feature_activation[f], tx); 4552 ds->ds_feature[f] = ds->ds_feature_activation[f]; 4553 } 4554 dsl_dataset_rele(ds, FTAG); 4555 } 4556 4557 int 4558 dsl_dataset_set_compression(const char *dsname, zprop_source_t source, 4559 uint64_t compression) 4560 { 4561 dsl_dataset_set_compression_arg_t ddsca; 4562 4563 /* 4564 * The sync task is only required for zstd in order to activate 4565 * the feature flag when the property is first set. 4566 */ 4567 if (ZIO_COMPRESS_ALGO(compression) != ZIO_COMPRESS_ZSTD) 4568 return (0); 4569 4570 ddsca.ddsca_name = dsname; 4571 ddsca.ddsca_source = source; 4572 ddsca.ddsca_value = compression; 4573 4574 return (dsl_sync_task(dsname, dsl_dataset_set_compression_check, 4575 dsl_dataset_set_compression_sync, &ddsca, 0, 4576 ZFS_SPACE_CHECK_EXTRA_RESERVED)); 4577 } 4578 4579 /* 4580 * Return (in *usedp) the amount of space referenced by "new" that was not 4581 * referenced at the time the bookmark corresponds to. "New" may be a 4582 * snapshot or a head. The bookmark must be before new, in 4583 * new's filesystem (or its origin) -- caller verifies this. 4584 * 4585 * The written space is calculated by considering two components: First, we 4586 * ignore any freed space, and calculate the written as new's used space 4587 * minus old's used space. Next, we add in the amount of space that was freed 4588 * between the two time points, thus reducing new's used space relative to 4589 * old's. Specifically, this is the space that was born before 4590 * zbm_creation_txg, and freed before new (ie. on new's deadlist or a 4591 * previous deadlist). 4592 * 4593 * space freed [---------------------] 4594 * snapshots ---O-------O--------O-------O------ 4595 * bookmark new 4596 * 4597 * Note, the bookmark's zbm_*_bytes_refd must be valid, but if the HAS_FBN 4598 * flag is not set, we will calculate the freed_before_next based on the 4599 * next snapshot's deadlist, rather than using zbm_*_freed_before_next_snap. 4600 */ 4601 static int 4602 dsl_dataset_space_written_impl(zfs_bookmark_phys_t *bmp, 4603 dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4604 { 4605 int err = 0; 4606 dsl_pool_t *dp = new->ds_dir->dd_pool; 4607 4608 ASSERT(dsl_pool_config_held(dp)); 4609 if (dsl_dataset_is_snapshot(new)) { 4610 ASSERT3U(bmp->zbm_creation_txg, <, 4611 dsl_dataset_phys(new)->ds_creation_txg); 4612 } 4613 4614 *usedp = 0; 4615 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; 4616 *usedp -= bmp->zbm_referenced_bytes_refd; 4617 4618 *compp = 0; 4619 *compp += dsl_dataset_phys(new)->ds_compressed_bytes; 4620 *compp -= bmp->zbm_compressed_bytes_refd; 4621 4622 *uncompp = 0; 4623 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; 4624 *uncompp -= bmp->zbm_uncompressed_bytes_refd; 4625 4626 dsl_dataset_t *snap = new; 4627 4628 while (dsl_dataset_phys(snap)->ds_prev_snap_txg > 4629 bmp->zbm_creation_txg) { 4630 uint64_t used, comp, uncomp; 4631 4632 dsl_deadlist_space_range(&snap->ds_deadlist, 4633 0, bmp->zbm_creation_txg, 4634 &used, &comp, &uncomp); 4635 *usedp += used; 4636 *compp += comp; 4637 *uncompp += uncomp; 4638 4639 uint64_t snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 4640 if (snap != new) 4641 dsl_dataset_rele(snap, FTAG); 4642 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 4643 if (err != 0) 4644 break; 4645 } 4646 4647 /* 4648 * We might not have the FBN if we are calculating written from 4649 * a snapshot (because we didn't know the correct "next" snapshot 4650 * until now). 4651 */ 4652 if (bmp->zbm_flags & ZBM_FLAG_HAS_FBN) { 4653 *usedp += bmp->zbm_referenced_freed_before_next_snap; 4654 *compp += bmp->zbm_compressed_freed_before_next_snap; 4655 *uncompp += bmp->zbm_uncompressed_freed_before_next_snap; 4656 } else { 4657 ASSERT3U(dsl_dataset_phys(snap)->ds_prev_snap_txg, ==, 4658 bmp->zbm_creation_txg); 4659 uint64_t used, comp, uncomp; 4660 dsl_deadlist_space(&snap->ds_deadlist, &used, &comp, &uncomp); 4661 *usedp += used; 4662 *compp += comp; 4663 *uncompp += uncomp; 4664 } 4665 if (snap != new) 4666 dsl_dataset_rele(snap, FTAG); 4667 return (err); 4668 } 4669 4670 /* 4671 * Return (in *usedp) the amount of space written in new that was not 4672 * present at the time the bookmark corresponds to. New may be a 4673 * snapshot or the head. Old must be a bookmark before new, in 4674 * new's filesystem (or its origin) -- caller verifies this. 4675 */ 4676 int 4677 dsl_dataset_space_written_bookmark(zfs_bookmark_phys_t *bmp, 4678 dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4679 { 4680 if (!(bmp->zbm_flags & ZBM_FLAG_HAS_FBN)) 4681 return (SET_ERROR(ENOTSUP)); 4682 return (dsl_dataset_space_written_impl(bmp, new, 4683 usedp, compp, uncompp)); 4684 } 4685 4686 /* 4687 * Return (in *usedp) the amount of space written in new that is not 4688 * present in oldsnap. New may be a snapshot or the head. Old must be 4689 * a snapshot before new, in new's filesystem (or its origin). If not then 4690 * fail and return EINVAL. 4691 */ 4692 int 4693 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 4694 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4695 { 4696 if (!dsl_dataset_is_before(new, oldsnap, 0)) 4697 return (SET_ERROR(EINVAL)); 4698 4699 zfs_bookmark_phys_t zbm = { 0 }; 4700 dsl_dataset_phys_t *dsp = dsl_dataset_phys(oldsnap); 4701 zbm.zbm_guid = dsp->ds_guid; 4702 zbm.zbm_creation_txg = dsp->ds_creation_txg; 4703 zbm.zbm_creation_time = dsp->ds_creation_time; 4704 zbm.zbm_referenced_bytes_refd = dsp->ds_referenced_bytes; 4705 zbm.zbm_compressed_bytes_refd = dsp->ds_compressed_bytes; 4706 zbm.zbm_uncompressed_bytes_refd = dsp->ds_uncompressed_bytes; 4707 4708 /* 4709 * If oldsnap is the origin (or origin's origin, ...) of new, 4710 * we can't easily calculate the effective FBN. Therefore, 4711 * we do not set ZBM_FLAG_HAS_FBN, so that the _impl will calculate 4712 * it relative to the correct "next": the next snapshot towards "new", 4713 * rather than the next snapshot in oldsnap's dsl_dir. 4714 */ 4715 return (dsl_dataset_space_written_impl(&zbm, new, 4716 usedp, compp, uncompp)); 4717 } 4718 4719 /* 4720 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 4721 * lastsnap, and all snapshots in between are deleted. 4722 * 4723 * blocks that would be freed [---------------------------] 4724 * snapshots ---O-------O--------O-------O--------O 4725 * firstsnap lastsnap 4726 * 4727 * This is the set of blocks that were born after the snap before firstsnap, 4728 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 4729 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 4730 * We calculate this by iterating over the relevant deadlists (from the snap 4731 * after lastsnap, backward to the snap after firstsnap), summing up the 4732 * space on the deadlist that was born after the snap before firstsnap. 4733 */ 4734 int 4735 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 4736 dsl_dataset_t *lastsnap, 4737 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4738 { 4739 int err = 0; 4740 uint64_t snapobj; 4741 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 4742 4743 ASSERT(firstsnap->ds_is_snapshot); 4744 ASSERT(lastsnap->ds_is_snapshot); 4745 4746 /* 4747 * Check that the snapshots are in the same dsl_dir, and firstsnap 4748 * is before lastsnap. 4749 */ 4750 if (firstsnap->ds_dir != lastsnap->ds_dir || 4751 dsl_dataset_phys(firstsnap)->ds_creation_txg > 4752 dsl_dataset_phys(lastsnap)->ds_creation_txg) 4753 return (SET_ERROR(EINVAL)); 4754 4755 *usedp = *compp = *uncompp = 0; 4756 4757 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; 4758 while (snapobj != firstsnap->ds_object) { 4759 dsl_dataset_t *ds; 4760 uint64_t used, comp, uncomp; 4761 4762 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 4763 if (err != 0) 4764 break; 4765 4766 dsl_deadlist_space_range(&ds->ds_deadlist, 4767 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, 4768 &used, &comp, &uncomp); 4769 *usedp += used; 4770 *compp += comp; 4771 *uncompp += uncomp; 4772 4773 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 4774 ASSERT3U(snapobj, !=, 0); 4775 dsl_dataset_rele(ds, FTAG); 4776 } 4777 return (err); 4778 } 4779 4780 /* 4781 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 4782 * For example, they could both be snapshots of the same filesystem, and 4783 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 4784 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 4785 * filesystem. Or 'earlier' could be the origin's origin. 4786 * 4787 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 4788 */ 4789 boolean_t 4790 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 4791 uint64_t earlier_txg) 4792 { 4793 dsl_pool_t *dp = later->ds_dir->dd_pool; 4794 int error; 4795 boolean_t ret; 4796 4797 ASSERT(dsl_pool_config_held(dp)); 4798 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); 4799 4800 if (earlier_txg == 0) 4801 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; 4802 4803 if (later->ds_is_snapshot && 4804 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) 4805 return (B_FALSE); 4806 4807 if (later->ds_dir == earlier->ds_dir) 4808 return (B_TRUE); 4809 4810 /* 4811 * We check dd_origin_obj explicitly here rather than using 4812 * dsl_dir_is_clone() so that we will return TRUE if "earlier" 4813 * is $ORIGIN@$ORIGIN. dsl_dataset_space_written() depends on 4814 * this behavior. 4815 */ 4816 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == 0) 4817 return (B_FALSE); 4818 4819 dsl_dataset_t *origin; 4820 error = dsl_dataset_hold_obj(dp, 4821 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); 4822 if (error != 0) 4823 return (B_FALSE); 4824 if (dsl_dataset_phys(origin)->ds_creation_txg == earlier_txg && 4825 origin->ds_dir == earlier->ds_dir) { 4826 dsl_dataset_rele(origin, FTAG); 4827 return (B_TRUE); 4828 } 4829 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 4830 dsl_dataset_rele(origin, FTAG); 4831 return (ret); 4832 } 4833 4834 void 4835 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 4836 { 4837 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 4838 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 4839 } 4840 4841 boolean_t 4842 dsl_dataset_is_zapified(dsl_dataset_t *ds) 4843 { 4844 dmu_object_info_t doi; 4845 4846 dmu_object_info_from_db(ds->ds_dbuf, &doi); 4847 return (doi.doi_type == DMU_OTN_ZAP_METADATA); 4848 } 4849 4850 boolean_t 4851 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds) 4852 { 4853 return (dsl_dataset_is_zapified(ds) && 4854 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset, 4855 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0); 4856 } 4857 4858 uint64_t 4859 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds) 4860 { 4861 uint64_t remap_deadlist_obj; 4862 int err; 4863 4864 if (!dsl_dataset_is_zapified(ds)) 4865 return (0); 4866 4867 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, 4868 DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1, 4869 &remap_deadlist_obj); 4870 4871 if (err != 0) { 4872 VERIFY3S(err, ==, ENOENT); 4873 return (0); 4874 } 4875 4876 ASSERT(remap_deadlist_obj != 0); 4877 return (remap_deadlist_obj); 4878 } 4879 4880 boolean_t 4881 dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds) 4882 { 4883 EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist), 4884 dsl_dataset_get_remap_deadlist_object(ds) != 0); 4885 return (dsl_deadlist_is_open(&ds->ds_remap_deadlist)); 4886 } 4887 4888 static void 4889 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj, 4890 dmu_tx_t *tx) 4891 { 4892 ASSERT(obj != 0); 4893 dsl_dataset_zapify(ds, tx); 4894 VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, 4895 DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx)); 4896 } 4897 4898 static void 4899 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx) 4900 { 4901 VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset, 4902 ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx)); 4903 } 4904 4905 void 4906 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx) 4907 { 4908 uint64_t remap_deadlist_object; 4909 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 4910 4911 ASSERT(dmu_tx_is_syncing(tx)); 4912 ASSERT(dsl_dataset_remap_deadlist_exists(ds)); 4913 4914 remap_deadlist_object = ds->ds_remap_deadlist.dl_object; 4915 dsl_deadlist_close(&ds->ds_remap_deadlist); 4916 dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx); 4917 dsl_dataset_unset_remap_deadlist_object(ds, tx); 4918 spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); 4919 } 4920 4921 void 4922 dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx) 4923 { 4924 uint64_t remap_deadlist_obj; 4925 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 4926 4927 ASSERT(dmu_tx_is_syncing(tx)); 4928 ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock)); 4929 /* 4930 * Currently we only create remap deadlists when there are indirect 4931 * vdevs with referenced mappings. 4932 */ 4933 ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL)); 4934 4935 remap_deadlist_obj = dsl_deadlist_clone( 4936 &ds->ds_deadlist, UINT64_MAX, 4937 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 4938 dsl_dataset_set_remap_deadlist_object(ds, 4939 remap_deadlist_obj, tx); 4940 dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa), 4941 remap_deadlist_obj); 4942 spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); 4943 } 4944 4945 void 4946 dsl_dataset_activate_redaction(dsl_dataset_t *ds, uint64_t *redact_snaps, 4947 uint64_t num_redact_snaps, dmu_tx_t *tx) 4948 { 4949 uint64_t dsobj = ds->ds_object; 4950 struct feature_type_uint64_array_arg *ftuaa = 4951 kmem_zalloc(sizeof (*ftuaa), KM_SLEEP); 4952 ftuaa->length = (int64_t)num_redact_snaps; 4953 if (num_redact_snaps > 0) { 4954 ftuaa->array = kmem_alloc(num_redact_snaps * sizeof (uint64_t), 4955 KM_SLEEP); 4956 bcopy(redact_snaps, ftuaa->array, num_redact_snaps * 4957 sizeof (uint64_t)); 4958 } 4959 dsl_dataset_activate_feature(dsobj, SPA_FEATURE_REDACTED_DATASETS, 4960 ftuaa, tx); 4961 ds->ds_feature[SPA_FEATURE_REDACTED_DATASETS] = ftuaa; 4962 } 4963 4964 /* BEGIN CSTYLED */ 4965 #if defined(_LP64) 4966 #define RECORDSIZE_PERM ZMOD_RW 4967 #else 4968 /* Limited to 1M on 32-bit platforms due to lack of virtual address space */ 4969 #define RECORDSIZE_PERM ZMOD_RD 4970 #endif 4971 ZFS_MODULE_PARAM(zfs, zfs_, max_recordsize, INT, RECORDSIZE_PERM, 4972 "Max allowed record size"); 4973 4974 ZFS_MODULE_PARAM(zfs, zfs_, allow_redacted_dataset_mount, INT, ZMOD_RW, 4975 "Allow mounting of redacted datasets"); 4976 /* END CSTYLED */ 4977 4978 EXPORT_SYMBOL(dsl_dataset_hold); 4979 EXPORT_SYMBOL(dsl_dataset_hold_flags); 4980 EXPORT_SYMBOL(dsl_dataset_hold_obj); 4981 EXPORT_SYMBOL(dsl_dataset_hold_obj_flags); 4982 EXPORT_SYMBOL(dsl_dataset_own); 4983 EXPORT_SYMBOL(dsl_dataset_own_obj); 4984 EXPORT_SYMBOL(dsl_dataset_name); 4985 EXPORT_SYMBOL(dsl_dataset_rele); 4986 EXPORT_SYMBOL(dsl_dataset_rele_flags); 4987 EXPORT_SYMBOL(dsl_dataset_disown); 4988 EXPORT_SYMBOL(dsl_dataset_tryown); 4989 EXPORT_SYMBOL(dsl_dataset_create_sync); 4990 EXPORT_SYMBOL(dsl_dataset_create_sync_dd); 4991 EXPORT_SYMBOL(dsl_dataset_snapshot_check); 4992 EXPORT_SYMBOL(dsl_dataset_snapshot_sync); 4993 EXPORT_SYMBOL(dsl_dataset_promote); 4994 EXPORT_SYMBOL(dsl_dataset_user_hold); 4995 EXPORT_SYMBOL(dsl_dataset_user_release); 4996 EXPORT_SYMBOL(dsl_dataset_get_holds); 4997 EXPORT_SYMBOL(dsl_dataset_get_blkptr); 4998 EXPORT_SYMBOL(dsl_dataset_get_spa); 4999 EXPORT_SYMBOL(dsl_dataset_modified_since_snap); 5000 EXPORT_SYMBOL(dsl_dataset_space_written); 5001 EXPORT_SYMBOL(dsl_dataset_space_wouldfree); 5002 EXPORT_SYMBOL(dsl_dataset_sync); 5003 EXPORT_SYMBOL(dsl_dataset_block_born); 5004 EXPORT_SYMBOL(dsl_dataset_block_kill); 5005 EXPORT_SYMBOL(dsl_dataset_dirty); 5006 EXPORT_SYMBOL(dsl_dataset_stats); 5007 EXPORT_SYMBOL(dsl_dataset_fast_stat); 5008 EXPORT_SYMBOL(dsl_dataset_space); 5009 EXPORT_SYMBOL(dsl_dataset_fsid_guid); 5010 EXPORT_SYMBOL(dsl_dsobj_to_dsname); 5011 EXPORT_SYMBOL(dsl_dataset_check_quota); 5012 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl); 5013 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl); 5014