1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or https://opensource.org/licenses/CDDL-1.0. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2020 by Delphix. All rights reserved. 25 * Copyright (c) 2014, Joyent, Inc. All rights reserved. 26 * Copyright (c) 2014 RackTop Systems. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 * Copyright (c) 2016 Actifio, Inc. All rights reserved. 29 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved. 30 * Copyright 2017 Nexenta Systems, Inc. 31 * Copyright (c) 2019, Klara Inc. 32 * Copyright (c) 2019, Allan Jude 33 * Copyright (c) 2020 The FreeBSD Foundation [1] 34 * 35 * [1] Portions of this software were developed by Allan Jude 36 * under sponsorship from the FreeBSD Foundation. 37 */ 38 39 #include <sys/dmu_objset.h> 40 #include <sys/dsl_dataset.h> 41 #include <sys/dsl_dir.h> 42 #include <sys/dsl_prop.h> 43 #include <sys/dsl_synctask.h> 44 #include <sys/dmu_traverse.h> 45 #include <sys/dmu_impl.h> 46 #include <sys/dmu_tx.h> 47 #include <sys/arc.h> 48 #include <sys/zio.h> 49 #include <sys/zap.h> 50 #include <sys/zfeature.h> 51 #include <sys/unique.h> 52 #include <sys/zfs_context.h> 53 #include <sys/zfs_ioctl.h> 54 #include <sys/spa.h> 55 #include <sys/spa_impl.h> 56 #include <sys/vdev.h> 57 #include <sys/zfs_znode.h> 58 #include <sys/zfs_onexit.h> 59 #include <sys/zvol.h> 60 #include <sys/dsl_scan.h> 61 #include <sys/dsl_deadlist.h> 62 #include <sys/dsl_destroy.h> 63 #include <sys/dsl_userhold.h> 64 #include <sys/dsl_bookmark.h> 65 #include <sys/policy.h> 66 #include <sys/dmu_send.h> 67 #include <sys/dmu_recv.h> 68 #include <sys/zio_compress.h> 69 #include <zfs_fletcher.h> 70 #include <sys/zio_checksum.h> 71 #include <sys/brt.h> 72 73 /* 74 * The SPA supports block sizes up to 16MB. However, very large blocks 75 * can have an impact on i/o latency (e.g. tying up a spinning disk for 76 * ~300ms), and also potentially on the memory allocator. Therefore, 77 * we did not allow the recordsize to be set larger than zfs_max_recordsize 78 * (former default: 1MB). Larger blocks could be created by changing this 79 * tunable, and pools with larger blocks could always be imported and used, 80 * regardless of this setting. 81 * 82 * We do, however, still limit it by default to 1M on x86_32, because Linux's 83 * 3/1 memory split doesn't leave much room for 16M chunks. 84 */ 85 #ifdef _ILP32 86 uint_t zfs_max_recordsize = 1 * 1024 * 1024; 87 #else 88 uint_t zfs_max_recordsize = 16 * 1024 * 1024; 89 #endif 90 static int zfs_allow_redacted_dataset_mount = 0; 91 92 int zfs_snapshot_history_enabled = 1; 93 94 #define SWITCH64(x, y) \ 95 { \ 96 uint64_t __tmp = (x); \ 97 (x) = (y); \ 98 (y) = __tmp; \ 99 } 100 101 #define DS_REF_MAX (1ULL << 62) 102 103 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, 104 uint64_t obj, dmu_tx_t *tx); 105 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, 106 dmu_tx_t *tx); 107 108 static void unload_zfeature(dsl_dataset_t *ds, spa_feature_t f); 109 110 extern uint_t spa_asize_inflation; 111 112 static zil_header_t zero_zil; 113 114 /* 115 * Figure out how much of this delta should be propagated to the dsl_dir 116 * layer. If there's a refreservation, that space has already been 117 * partially accounted for in our ancestors. 118 */ 119 static int64_t 120 parent_delta(dsl_dataset_t *ds, int64_t delta) 121 { 122 dsl_dataset_phys_t *ds_phys; 123 uint64_t old_bytes, new_bytes; 124 125 if (ds->ds_reserved == 0) 126 return (delta); 127 128 ds_phys = dsl_dataset_phys(ds); 129 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved); 130 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 131 132 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 133 return (new_bytes - old_bytes); 134 } 135 136 void 137 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 138 { 139 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 140 int used = bp_get_dsize_sync(spa, bp); 141 int compressed = BP_GET_PSIZE(bp); 142 int uncompressed = BP_GET_UCSIZE(bp); 143 int64_t delta; 144 spa_feature_t f; 145 146 dprintf_bp(bp, "ds=%p", ds); 147 148 ASSERT(dmu_tx_is_syncing(tx)); 149 /* It could have been compressed away to nothing */ 150 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp)) 151 return; 152 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 153 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 154 if (ds == NULL) { 155 dsl_pool_mos_diduse_space(tx->tx_pool, 156 used, compressed, uncompressed); 157 return; 158 } 159 160 ASSERT3U(BP_GET_LOGICAL_BIRTH(bp), >, 161 dsl_dataset_phys(ds)->ds_prev_snap_txg); 162 dmu_buf_will_dirty(ds->ds_dbuf, tx); 163 mutex_enter(&ds->ds_lock); 164 delta = parent_delta(ds, used); 165 dsl_dataset_phys(ds)->ds_referenced_bytes += used; 166 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed; 167 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed; 168 dsl_dataset_phys(ds)->ds_unique_bytes += used; 169 170 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) { 171 ds->ds_feature_activation[SPA_FEATURE_LARGE_BLOCKS] = 172 (void *)B_TRUE; 173 } 174 175 176 f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp)); 177 if (f != SPA_FEATURE_NONE) { 178 ASSERT3S(spa_feature_table[f].fi_type, ==, 179 ZFEATURE_TYPE_BOOLEAN); 180 ds->ds_feature_activation[f] = (void *)B_TRUE; 181 } 182 183 f = zio_compress_to_feature(BP_GET_COMPRESS(bp)); 184 if (f != SPA_FEATURE_NONE) { 185 ASSERT3S(spa_feature_table[f].fi_type, ==, 186 ZFEATURE_TYPE_BOOLEAN); 187 ds->ds_feature_activation[f] = (void *)B_TRUE; 188 } 189 190 /* 191 * Track block for livelist, but ignore embedded blocks because 192 * they do not need to be freed. 193 */ 194 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) && 195 BP_GET_LOGICAL_BIRTH(bp) > ds->ds_dir->dd_origin_txg && 196 !(BP_IS_EMBEDDED(bp))) { 197 ASSERT(dsl_dir_is_clone(ds->ds_dir)); 198 ASSERT(spa_feature_is_enabled(spa, 199 SPA_FEATURE_LIVELIST)); 200 bplist_append(&ds->ds_dir->dd_pending_allocs, bp); 201 } 202 203 mutex_exit(&ds->ds_lock); 204 dsl_dir_diduse_transfer_space(ds->ds_dir, delta, 205 compressed, uncompressed, used, 206 DD_USED_REFRSRV, DD_USED_HEAD, tx); 207 } 208 209 /* 210 * Called when the specified segment has been remapped, and is thus no 211 * longer referenced in the head dataset. The vdev must be indirect. 212 * 213 * If the segment is referenced by a snapshot, put it on the remap deadlist. 214 * Otherwise, add this segment to the obsolete spacemap. 215 */ 216 void 217 dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, uint64_t offset, 218 uint64_t size, uint64_t birth, dmu_tx_t *tx) 219 { 220 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 221 222 ASSERT(dmu_tx_is_syncing(tx)); 223 ASSERT(birth <= tx->tx_txg); 224 ASSERT(!ds->ds_is_snapshot); 225 226 if (birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 227 spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx); 228 } else { 229 blkptr_t fakebp; 230 dva_t *dva = &fakebp.blk_dva[0]; 231 232 ASSERT(ds != NULL); 233 234 mutex_enter(&ds->ds_remap_deadlist_lock); 235 if (!dsl_dataset_remap_deadlist_exists(ds)) { 236 dsl_dataset_create_remap_deadlist(ds, tx); 237 } 238 mutex_exit(&ds->ds_remap_deadlist_lock); 239 240 BP_ZERO(&fakebp); 241 BP_SET_LOGICAL_BIRTH(&fakebp, birth); 242 DVA_SET_VDEV(dva, vdev); 243 DVA_SET_OFFSET(dva, offset); 244 DVA_SET_ASIZE(dva, size); 245 dsl_deadlist_insert(&ds->ds_remap_deadlist, &fakebp, B_FALSE, 246 tx); 247 } 248 } 249 250 int 251 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 252 boolean_t async) 253 { 254 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 255 256 int used = bp_get_dsize_sync(spa, bp); 257 int compressed = BP_GET_PSIZE(bp); 258 int uncompressed = BP_GET_UCSIZE(bp); 259 260 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp)) 261 return (0); 262 263 ASSERT(dmu_tx_is_syncing(tx)); 264 ASSERT(BP_GET_LOGICAL_BIRTH(bp) <= tx->tx_txg); 265 266 if (ds == NULL) { 267 dsl_free(tx->tx_pool, tx->tx_txg, bp); 268 dsl_pool_mos_diduse_space(tx->tx_pool, 269 -used, -compressed, -uncompressed); 270 return (used); 271 } 272 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 273 274 ASSERT(!ds->ds_is_snapshot); 275 dmu_buf_will_dirty(ds->ds_dbuf, tx); 276 277 /* 278 * Track block for livelist, but ignore embedded blocks because 279 * they do not need to be freed. 280 */ 281 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) && 282 BP_GET_LOGICAL_BIRTH(bp) > ds->ds_dir->dd_origin_txg && 283 !(BP_IS_EMBEDDED(bp))) { 284 ASSERT(dsl_dir_is_clone(ds->ds_dir)); 285 ASSERT(spa_feature_is_enabled(spa, 286 SPA_FEATURE_LIVELIST)); 287 bplist_append(&ds->ds_dir->dd_pending_frees, bp); 288 } 289 290 if (BP_GET_LOGICAL_BIRTH(bp) > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 291 int64_t delta; 292 293 /* 294 * Put blocks that would create IO on the pool's deadlist for 295 * dsl_process_async_destroys() to find. This is to prevent 296 * zio_free() from creating a ZIO_TYPE_FREE IO for them, which 297 * are very heavy and can lead to out-of-memory conditions if 298 * something tries to free millions of blocks on the same txg. 299 */ 300 boolean_t defer = spa_version(spa) >= SPA_VERSION_DEADLISTS && 301 (BP_IS_GANG(bp) || BP_GET_DEDUP(bp) || 302 brt_maybe_exists(spa, bp)); 303 304 if (defer) { 305 dprintf_bp(bp, "putting on free list: %s", ""); 306 bpobj_enqueue(&ds->ds_dir->dd_pool->dp_free_bpobj, 307 bp, B_FALSE, tx); 308 } else { 309 dprintf_bp(bp, "freeing ds=%llu", 310 (u_longlong_t)ds->ds_object); 311 dsl_free(tx->tx_pool, tx->tx_txg, bp); 312 } 313 314 mutex_enter(&ds->ds_lock); 315 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used || 316 !DS_UNIQUE_IS_ACCURATE(ds)); 317 delta = parent_delta(ds, -used); 318 dsl_dataset_phys(ds)->ds_unique_bytes -= used; 319 mutex_exit(&ds->ds_lock); 320 321 dsl_dir_diduse_transfer_space(ds->ds_dir, 322 delta, -compressed, -uncompressed, -used, 323 DD_USED_REFRSRV, DD_USED_HEAD, tx); 324 325 if (defer) 326 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, 327 DD_USED_HEAD, used, compressed, uncompressed, tx); 328 } else { 329 dprintf_bp(bp, "putting on dead list: %s", ""); 330 if (async) { 331 /* 332 * We are here as part of zio's write done callback, 333 * which means we're a zio interrupt thread. We can't 334 * call dsl_deadlist_insert() now because it may block 335 * waiting for I/O. Instead, put bp on the deferred 336 * queue and let dsl_pool_sync() finish the job. 337 */ 338 bplist_append(&ds->ds_pending_deadlist, bp); 339 } else { 340 dsl_deadlist_insert(&ds->ds_deadlist, bp, B_FALSE, tx); 341 } 342 ASSERT3U(ds->ds_prev->ds_object, ==, 343 dsl_dataset_phys(ds)->ds_prev_snap_obj); 344 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0); 345 /* if (logical birth > prev prev snap txg) prev unique += bs */ 346 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 347 ds->ds_object && BP_GET_LOGICAL_BIRTH(bp) > 348 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) { 349 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 350 mutex_enter(&ds->ds_prev->ds_lock); 351 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used; 352 mutex_exit(&ds->ds_prev->ds_lock); 353 } 354 if (BP_GET_LOGICAL_BIRTH(bp) > ds->ds_dir->dd_origin_txg) { 355 dsl_dir_transfer_space(ds->ds_dir, used, 356 DD_USED_HEAD, DD_USED_SNAP, tx); 357 } 358 } 359 360 dsl_bookmark_block_killed(ds, bp, tx); 361 362 mutex_enter(&ds->ds_lock); 363 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used); 364 dsl_dataset_phys(ds)->ds_referenced_bytes -= used; 365 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed); 366 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed; 367 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed); 368 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed; 369 mutex_exit(&ds->ds_lock); 370 371 return (used); 372 } 373 374 struct feature_type_uint64_array_arg { 375 uint64_t length; 376 uint64_t *array; 377 }; 378 379 static void 380 unload_zfeature(dsl_dataset_t *ds, spa_feature_t f) 381 { 382 switch (spa_feature_table[f].fi_type) { 383 case ZFEATURE_TYPE_BOOLEAN: 384 break; 385 case ZFEATURE_TYPE_UINT64_ARRAY: 386 { 387 struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f]; 388 kmem_free(ftuaa->array, ftuaa->length * sizeof (uint64_t)); 389 kmem_free(ftuaa, sizeof (*ftuaa)); 390 break; 391 } 392 default: 393 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); 394 } 395 } 396 397 static int 398 load_zfeature(objset_t *mos, dsl_dataset_t *ds, spa_feature_t f) 399 { 400 int err = 0; 401 switch (spa_feature_table[f].fi_type) { 402 case ZFEATURE_TYPE_BOOLEAN: 403 err = zap_contains(mos, ds->ds_object, 404 spa_feature_table[f].fi_guid); 405 if (err == 0) { 406 ds->ds_feature[f] = (void *)B_TRUE; 407 } else { 408 ASSERT3U(err, ==, ENOENT); 409 err = 0; 410 } 411 break; 412 case ZFEATURE_TYPE_UINT64_ARRAY: 413 { 414 uint64_t int_size, num_int; 415 uint64_t *data; 416 err = zap_length(mos, ds->ds_object, 417 spa_feature_table[f].fi_guid, &int_size, &num_int); 418 if (err != 0) { 419 ASSERT3U(err, ==, ENOENT); 420 err = 0; 421 break; 422 } 423 ASSERT3U(int_size, ==, sizeof (uint64_t)); 424 data = kmem_alloc(int_size * num_int, KM_SLEEP); 425 VERIFY0(zap_lookup(mos, ds->ds_object, 426 spa_feature_table[f].fi_guid, int_size, num_int, data)); 427 struct feature_type_uint64_array_arg *ftuaa = 428 kmem_alloc(sizeof (*ftuaa), KM_SLEEP); 429 ftuaa->length = num_int; 430 ftuaa->array = data; 431 ds->ds_feature[f] = ftuaa; 432 break; 433 } 434 default: 435 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); 436 } 437 return (err); 438 } 439 440 /* 441 * We have to release the fsid synchronously or we risk that a subsequent 442 * mount of the same dataset will fail to unique_insert the fsid. This 443 * failure would manifest itself as the fsid of this dataset changing 444 * between mounts which makes NFS clients quite unhappy. 445 */ 446 static void 447 dsl_dataset_evict_sync(void *dbu) 448 { 449 dsl_dataset_t *ds = dbu; 450 451 ASSERT(ds->ds_owner == NULL); 452 453 unique_remove(ds->ds_fsid_guid); 454 } 455 456 static void 457 dsl_dataset_evict_async(void *dbu) 458 { 459 dsl_dataset_t *ds = dbu; 460 461 ASSERT(ds->ds_owner == NULL); 462 463 ds->ds_dbuf = NULL; 464 465 if (ds->ds_objset != NULL) 466 dmu_objset_evict(ds->ds_objset); 467 468 if (ds->ds_prev) { 469 dsl_dataset_rele(ds->ds_prev, ds); 470 ds->ds_prev = NULL; 471 } 472 473 dsl_bookmark_fini_ds(ds); 474 475 bplist_destroy(&ds->ds_pending_deadlist); 476 if (dsl_deadlist_is_open(&ds->ds_deadlist)) 477 dsl_deadlist_close(&ds->ds_deadlist); 478 if (dsl_deadlist_is_open(&ds->ds_remap_deadlist)) 479 dsl_deadlist_close(&ds->ds_remap_deadlist); 480 if (ds->ds_dir) 481 dsl_dir_async_rele(ds->ds_dir, ds); 482 483 ASSERT(!list_link_active(&ds->ds_synced_link)); 484 485 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 486 if (dsl_dataset_feature_is_active(ds, f)) 487 unload_zfeature(ds, f); 488 } 489 490 list_destroy(&ds->ds_prop_cbs); 491 mutex_destroy(&ds->ds_lock); 492 mutex_destroy(&ds->ds_opening_lock); 493 mutex_destroy(&ds->ds_sendstream_lock); 494 mutex_destroy(&ds->ds_remap_deadlist_lock); 495 zfs_refcount_destroy(&ds->ds_longholds); 496 rrw_destroy(&ds->ds_bp_rwlock); 497 498 kmem_free(ds, sizeof (dsl_dataset_t)); 499 } 500 501 int 502 dsl_dataset_get_snapname(dsl_dataset_t *ds) 503 { 504 dsl_dataset_phys_t *headphys; 505 int err; 506 dmu_buf_t *headdbuf; 507 dsl_pool_t *dp = ds->ds_dir->dd_pool; 508 objset_t *mos = dp->dp_meta_objset; 509 510 if (ds->ds_snapname[0]) 511 return (0); 512 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) 513 return (0); 514 515 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, 516 FTAG, &headdbuf); 517 if (err != 0) 518 return (err); 519 headphys = headdbuf->db_data; 520 err = zap_value_search(dp->dp_meta_objset, 521 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname, 522 sizeof (ds->ds_snapname)); 523 if (err != 0 && zfs_recover == B_TRUE) { 524 err = 0; 525 (void) snprintf(ds->ds_snapname, sizeof (ds->ds_snapname), 526 "SNAPOBJ=%llu-ERR=%d", 527 (unsigned long long)ds->ds_object, err); 528 } 529 dmu_buf_rele(headdbuf, FTAG); 530 return (err); 531 } 532 533 int 534 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 535 { 536 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 537 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 538 matchtype_t mt = 0; 539 int err; 540 541 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 542 mt = MT_NORMALIZE; 543 544 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 545 value, mt, NULL, 0, NULL); 546 if (err == ENOTSUP && (mt & MT_NORMALIZE)) 547 err = zap_lookup(mos, snapobj, name, 8, 1, value); 548 return (err); 549 } 550 551 int 552 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 553 boolean_t adj_cnt) 554 { 555 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 556 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 557 matchtype_t mt = 0; 558 int err; 559 560 dsl_dir_snap_cmtime_update(ds->ds_dir, tx); 561 562 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 563 mt = MT_NORMALIZE; 564 565 err = zap_remove_norm(mos, snapobj, name, mt, tx); 566 if (err == ENOTSUP && (mt & MT_NORMALIZE)) 567 err = zap_remove(mos, snapobj, name, tx); 568 569 if (err == 0 && adj_cnt) 570 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 571 DD_FIELD_SNAPSHOT_COUNT, tx); 572 573 return (err); 574 } 575 576 boolean_t 577 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, const void *tag) 578 { 579 dmu_buf_t *dbuf = ds->ds_dbuf; 580 boolean_t result = B_FALSE; 581 582 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset, 583 ds->ds_object, DMU_BONUS_BLKID, tag)) { 584 585 if (ds == dmu_buf_get_user(dbuf)) 586 result = B_TRUE; 587 else 588 dmu_buf_rele(dbuf, tag); 589 } 590 591 return (result); 592 } 593 594 int 595 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, const void *tag, 596 dsl_dataset_t **dsp) 597 { 598 objset_t *mos = dp->dp_meta_objset; 599 dmu_buf_t *dbuf; 600 dsl_dataset_t *ds; 601 int err; 602 dmu_object_info_t doi; 603 604 ASSERT(dsl_pool_config_held(dp)); 605 606 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 607 if (err != 0) 608 return (err); 609 610 /* Make sure dsobj has the correct object type. */ 611 dmu_object_info_from_db(dbuf, &doi); 612 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 613 dmu_buf_rele(dbuf, tag); 614 return (SET_ERROR(EINVAL)); 615 } 616 617 ds = dmu_buf_get_user(dbuf); 618 if (ds == NULL) { 619 dsl_dataset_t *winner = NULL; 620 621 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 622 ds->ds_dbuf = dbuf; 623 ds->ds_object = dsobj; 624 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0; 625 list_link_init(&ds->ds_synced_link); 626 627 err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj, 628 NULL, ds, &ds->ds_dir); 629 if (err != 0) { 630 kmem_free(ds, sizeof (dsl_dataset_t)); 631 dmu_buf_rele(dbuf, tag); 632 return (err); 633 } 634 635 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 636 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 637 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 638 mutex_init(&ds->ds_remap_deadlist_lock, 639 NULL, MUTEX_DEFAULT, NULL); 640 rrw_init(&ds->ds_bp_rwlock, B_FALSE); 641 zfs_refcount_create(&ds->ds_longholds); 642 643 bplist_create(&ds->ds_pending_deadlist); 644 645 list_create(&ds->ds_sendstreams, sizeof (dmu_sendstatus_t), 646 offsetof(dmu_sendstatus_t, dss_link)); 647 648 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t), 649 offsetof(dsl_prop_cb_record_t, cbr_ds_node)); 650 651 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 652 spa_feature_t f; 653 654 for (f = 0; f < SPA_FEATURES; f++) { 655 if (!(spa_feature_table[f].fi_flags & 656 ZFEATURE_FLAG_PER_DATASET)) 657 continue; 658 err = load_zfeature(mos, ds, f); 659 } 660 } 661 662 if (!ds->ds_is_snapshot) { 663 ds->ds_snapname[0] = '\0'; 664 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 665 err = dsl_dataset_hold_obj(dp, 666 dsl_dataset_phys(ds)->ds_prev_snap_obj, 667 ds, &ds->ds_prev); 668 } 669 if (err != 0) 670 goto after_dsl_bookmark_fini; 671 err = dsl_bookmark_init_ds(ds); 672 } else { 673 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 674 err = dsl_dataset_get_snapname(ds); 675 if (err == 0 && 676 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { 677 err = zap_count( 678 ds->ds_dir->dd_pool->dp_meta_objset, 679 dsl_dataset_phys(ds)->ds_userrefs_obj, 680 &ds->ds_userrefs); 681 } 682 } 683 684 if (err == 0 && !ds->ds_is_snapshot) { 685 err = dsl_prop_get_int_ds(ds, 686 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 687 &ds->ds_reserved); 688 if (err == 0) { 689 err = dsl_prop_get_int_ds(ds, 690 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 691 &ds->ds_quota); 692 } 693 } else { 694 ds->ds_reserved = ds->ds_quota = 0; 695 } 696 697 if (err == 0 && ds->ds_dir->dd_crypto_obj != 0 && 698 ds->ds_is_snapshot && 699 zap_contains(mos, dsobj, DS_FIELD_IVSET_GUID) != 0) { 700 dp->dp_spa->spa_errata = 701 ZPOOL_ERRATA_ZOL_8308_ENCRYPTION; 702 } 703 704 dsl_deadlist_open(&ds->ds_deadlist, 705 mos, dsl_dataset_phys(ds)->ds_deadlist_obj); 706 uint64_t remap_deadlist_obj = 707 dsl_dataset_get_remap_deadlist_object(ds); 708 if (remap_deadlist_obj != 0) { 709 dsl_deadlist_open(&ds->ds_remap_deadlist, mos, 710 remap_deadlist_obj); 711 } 712 713 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync, 714 dsl_dataset_evict_async, &ds->ds_dbuf); 715 if (err == 0) 716 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu); 717 718 if (err != 0 || winner != NULL) { 719 dsl_deadlist_close(&ds->ds_deadlist); 720 if (dsl_deadlist_is_open(&ds->ds_remap_deadlist)) 721 dsl_deadlist_close(&ds->ds_remap_deadlist); 722 dsl_bookmark_fini_ds(ds); 723 after_dsl_bookmark_fini: 724 if (ds->ds_prev) 725 dsl_dataset_rele(ds->ds_prev, ds); 726 dsl_dir_rele(ds->ds_dir, ds); 727 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 728 if (dsl_dataset_feature_is_active(ds, f)) 729 unload_zfeature(ds, f); 730 } 731 732 list_destroy(&ds->ds_prop_cbs); 733 list_destroy(&ds->ds_sendstreams); 734 bplist_destroy(&ds->ds_pending_deadlist); 735 mutex_destroy(&ds->ds_lock); 736 mutex_destroy(&ds->ds_opening_lock); 737 mutex_destroy(&ds->ds_sendstream_lock); 738 mutex_destroy(&ds->ds_remap_deadlist_lock); 739 zfs_refcount_destroy(&ds->ds_longholds); 740 rrw_destroy(&ds->ds_bp_rwlock); 741 kmem_free(ds, sizeof (dsl_dataset_t)); 742 if (err != 0) { 743 dmu_buf_rele(dbuf, tag); 744 return (err); 745 } 746 ds = winner; 747 } else { 748 ds->ds_fsid_guid = 749 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); 750 if (ds->ds_fsid_guid != 751 dsl_dataset_phys(ds)->ds_fsid_guid) { 752 zfs_dbgmsg("ds_fsid_guid changed from " 753 "%llx to %llx for pool %s dataset id %llu", 754 (long long) 755 dsl_dataset_phys(ds)->ds_fsid_guid, 756 (long long)ds->ds_fsid_guid, 757 spa_name(dp->dp_spa), 758 (u_longlong_t)dsobj); 759 } 760 } 761 } 762 763 ASSERT3P(ds->ds_dbuf, ==, dbuf); 764 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); 765 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || 766 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 767 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 768 *dsp = ds; 769 770 return (0); 771 } 772 773 int 774 dsl_dataset_create_key_mapping(dsl_dataset_t *ds) 775 { 776 dsl_dir_t *dd = ds->ds_dir; 777 778 if (dd->dd_crypto_obj == 0) 779 return (0); 780 781 return (spa_keystore_create_mapping(dd->dd_pool->dp_spa, 782 ds, ds, &ds->ds_key_mapping)); 783 } 784 785 int 786 dsl_dataset_hold_obj_flags(dsl_pool_t *dp, uint64_t dsobj, 787 ds_hold_flags_t flags, const void *tag, dsl_dataset_t **dsp) 788 { 789 int err; 790 791 err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 792 if (err != 0) 793 return (err); 794 795 ASSERT3P(*dsp, !=, NULL); 796 797 if (flags & DS_HOLD_FLAG_DECRYPT) { 798 err = dsl_dataset_create_key_mapping(*dsp); 799 if (err != 0) 800 dsl_dataset_rele(*dsp, tag); 801 } 802 803 return (err); 804 } 805 806 int 807 dsl_dataset_hold_flags(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, 808 const void *tag, dsl_dataset_t **dsp) 809 { 810 dsl_dir_t *dd; 811 const char *snapname; 812 uint64_t obj; 813 int err = 0; 814 dsl_dataset_t *ds; 815 816 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 817 if (err != 0) 818 return (err); 819 820 ASSERT(dsl_pool_config_held(dp)); 821 obj = dsl_dir_phys(dd)->dd_head_dataset_obj; 822 if (obj != 0) 823 err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, &ds); 824 else 825 err = SET_ERROR(ENOENT); 826 827 /* we may be looking for a snapshot */ 828 if (err == 0 && snapname != NULL) { 829 dsl_dataset_t *snap_ds; 830 831 if (*snapname++ != '@') { 832 dsl_dataset_rele_flags(ds, flags, tag); 833 dsl_dir_rele(dd, FTAG); 834 return (SET_ERROR(ENOENT)); 835 } 836 837 dprintf("looking for snapshot '%s'\n", snapname); 838 err = dsl_dataset_snap_lookup(ds, snapname, &obj); 839 if (err == 0) { 840 err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, 841 &snap_ds); 842 } 843 dsl_dataset_rele_flags(ds, flags, tag); 844 845 if (err == 0) { 846 mutex_enter(&snap_ds->ds_lock); 847 if (snap_ds->ds_snapname[0] == 0) 848 (void) strlcpy(snap_ds->ds_snapname, snapname, 849 sizeof (snap_ds->ds_snapname)); 850 mutex_exit(&snap_ds->ds_lock); 851 ds = snap_ds; 852 } 853 } 854 if (err == 0) 855 *dsp = ds; 856 dsl_dir_rele(dd, FTAG); 857 return (err); 858 } 859 860 int 861 dsl_dataset_hold(dsl_pool_t *dp, const char *name, const void *tag, 862 dsl_dataset_t **dsp) 863 { 864 return (dsl_dataset_hold_flags(dp, name, 0, tag, dsp)); 865 } 866 867 static int 868 dsl_dataset_own_obj_impl(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags, 869 const void *tag, boolean_t override, dsl_dataset_t **dsp) 870 { 871 int err = dsl_dataset_hold_obj_flags(dp, dsobj, flags, tag, dsp); 872 if (err != 0) 873 return (err); 874 if (!dsl_dataset_tryown(*dsp, tag, override)) { 875 dsl_dataset_rele_flags(*dsp, flags, tag); 876 *dsp = NULL; 877 return (SET_ERROR(EBUSY)); 878 } 879 return (0); 880 } 881 882 883 int 884 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags, 885 const void *tag, dsl_dataset_t **dsp) 886 { 887 return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_FALSE, dsp)); 888 } 889 890 int 891 dsl_dataset_own_obj_force(dsl_pool_t *dp, uint64_t dsobj, 892 ds_hold_flags_t flags, const void *tag, dsl_dataset_t **dsp) 893 { 894 return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_TRUE, dsp)); 895 } 896 897 static int 898 dsl_dataset_own_impl(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, 899 const void *tag, boolean_t override, dsl_dataset_t **dsp) 900 { 901 int err = dsl_dataset_hold_flags(dp, name, flags, tag, dsp); 902 if (err != 0) 903 return (err); 904 if (!dsl_dataset_tryown(*dsp, tag, override)) { 905 dsl_dataset_rele_flags(*dsp, flags, tag); 906 return (SET_ERROR(EBUSY)); 907 } 908 return (0); 909 } 910 911 int 912 dsl_dataset_own_force(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, 913 const void *tag, dsl_dataset_t **dsp) 914 { 915 return (dsl_dataset_own_impl(dp, name, flags, tag, B_TRUE, dsp)); 916 } 917 918 int 919 dsl_dataset_own(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags, 920 const void *tag, dsl_dataset_t **dsp) 921 { 922 return (dsl_dataset_own_impl(dp, name, flags, tag, B_FALSE, dsp)); 923 } 924 925 /* 926 * See the comment above dsl_pool_hold() for details. In summary, a long 927 * hold is used to prevent destruction of a dataset while the pool hold 928 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 929 * 930 * The dataset and pool must be held when this function is called. After it 931 * is called, the pool hold may be released while the dataset is still held 932 * and accessed. 933 */ 934 void 935 dsl_dataset_long_hold(dsl_dataset_t *ds, const void *tag) 936 { 937 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 938 (void) zfs_refcount_add(&ds->ds_longholds, tag); 939 } 940 941 void 942 dsl_dataset_long_rele(dsl_dataset_t *ds, const void *tag) 943 { 944 (void) zfs_refcount_remove(&ds->ds_longholds, tag); 945 } 946 947 /* Return B_TRUE if there are any long holds on this dataset. */ 948 boolean_t 949 dsl_dataset_long_held(dsl_dataset_t *ds) 950 { 951 return (!zfs_refcount_is_zero(&ds->ds_longholds)); 952 } 953 954 void 955 dsl_dataset_name(dsl_dataset_t *ds, char *name) 956 { 957 if (ds == NULL) { 958 (void) strlcpy(name, "mos", ZFS_MAX_DATASET_NAME_LEN); 959 } else { 960 dsl_dir_name(ds->ds_dir, name); 961 VERIFY0(dsl_dataset_get_snapname(ds)); 962 if (ds->ds_snapname[0]) { 963 VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN), 964 <, ZFS_MAX_DATASET_NAME_LEN); 965 /* 966 * We use a "recursive" mutex so that we 967 * can call dprintf_ds() with ds_lock held. 968 */ 969 if (!MUTEX_HELD(&ds->ds_lock)) { 970 mutex_enter(&ds->ds_lock); 971 VERIFY3U(strlcat(name, ds->ds_snapname, 972 ZFS_MAX_DATASET_NAME_LEN), <, 973 ZFS_MAX_DATASET_NAME_LEN); 974 mutex_exit(&ds->ds_lock); 975 } else { 976 VERIFY3U(strlcat(name, ds->ds_snapname, 977 ZFS_MAX_DATASET_NAME_LEN), <, 978 ZFS_MAX_DATASET_NAME_LEN); 979 } 980 } 981 } 982 } 983 984 int 985 dsl_dataset_namelen(dsl_dataset_t *ds) 986 { 987 VERIFY0(dsl_dataset_get_snapname(ds)); 988 mutex_enter(&ds->ds_lock); 989 int len = strlen(ds->ds_snapname); 990 mutex_exit(&ds->ds_lock); 991 /* add '@' if ds is a snap */ 992 if (len > 0) 993 len++; 994 len += dsl_dir_namelen(ds->ds_dir); 995 return (len); 996 } 997 998 void 999 dsl_dataset_rele(dsl_dataset_t *ds, const void *tag) 1000 { 1001 dmu_buf_rele(ds->ds_dbuf, tag); 1002 } 1003 1004 void 1005 dsl_dataset_remove_key_mapping(dsl_dataset_t *ds) 1006 { 1007 dsl_dir_t *dd = ds->ds_dir; 1008 1009 if (dd == NULL || dd->dd_crypto_obj == 0) 1010 return; 1011 1012 (void) spa_keystore_remove_mapping(dd->dd_pool->dp_spa, 1013 ds->ds_object, ds); 1014 } 1015 1016 void 1017 dsl_dataset_rele_flags(dsl_dataset_t *ds, ds_hold_flags_t flags, 1018 const void *tag) 1019 { 1020 if (flags & DS_HOLD_FLAG_DECRYPT) 1021 dsl_dataset_remove_key_mapping(ds); 1022 1023 dsl_dataset_rele(ds, tag); 1024 } 1025 1026 void 1027 dsl_dataset_disown(dsl_dataset_t *ds, ds_hold_flags_t flags, const void *tag) 1028 { 1029 ASSERT3P(ds->ds_owner, ==, tag); 1030 ASSERT(ds->ds_dbuf != NULL); 1031 1032 mutex_enter(&ds->ds_lock); 1033 ds->ds_owner = NULL; 1034 mutex_exit(&ds->ds_lock); 1035 dsl_dataset_long_rele(ds, tag); 1036 dsl_dataset_rele_flags(ds, flags, tag); 1037 } 1038 1039 boolean_t 1040 dsl_dataset_tryown(dsl_dataset_t *ds, const void *tag, boolean_t override) 1041 { 1042 boolean_t gotit = FALSE; 1043 1044 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1045 mutex_enter(&ds->ds_lock); 1046 if (ds->ds_owner == NULL && (override || !(DS_IS_INCONSISTENT(ds) || 1047 (dsl_dataset_feature_is_active(ds, 1048 SPA_FEATURE_REDACTED_DATASETS) && 1049 !zfs_allow_redacted_dataset_mount)))) { 1050 ds->ds_owner = tag; 1051 dsl_dataset_long_hold(ds, tag); 1052 gotit = TRUE; 1053 } 1054 mutex_exit(&ds->ds_lock); 1055 return (gotit); 1056 } 1057 1058 boolean_t 1059 dsl_dataset_has_owner(dsl_dataset_t *ds) 1060 { 1061 boolean_t rv; 1062 mutex_enter(&ds->ds_lock); 1063 rv = (ds->ds_owner != NULL); 1064 mutex_exit(&ds->ds_lock); 1065 return (rv); 1066 } 1067 1068 static boolean_t 1069 zfeature_active(spa_feature_t f, void *arg) 1070 { 1071 switch (spa_feature_table[f].fi_type) { 1072 case ZFEATURE_TYPE_BOOLEAN: { 1073 boolean_t val = (boolean_t)(uintptr_t)arg; 1074 ASSERT(val == B_FALSE || val == B_TRUE); 1075 return (val); 1076 } 1077 case ZFEATURE_TYPE_UINT64_ARRAY: 1078 /* 1079 * In this case, arg is a uint64_t array. The feature is active 1080 * if the array is non-null. 1081 */ 1082 return (arg != NULL); 1083 default: 1084 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); 1085 return (B_FALSE); 1086 } 1087 } 1088 1089 boolean_t 1090 dsl_dataset_feature_is_active(dsl_dataset_t *ds, spa_feature_t f) 1091 { 1092 return (zfeature_active(f, ds->ds_feature[f])); 1093 } 1094 1095 /* 1096 * The buffers passed out by this function are references to internal buffers; 1097 * they should not be freed by callers of this function, and they should not be 1098 * used after the dataset has been released. 1099 */ 1100 boolean_t 1101 dsl_dataset_get_uint64_array_feature(dsl_dataset_t *ds, spa_feature_t f, 1102 uint64_t *outlength, uint64_t **outp) 1103 { 1104 VERIFY(spa_feature_table[f].fi_type & ZFEATURE_TYPE_UINT64_ARRAY); 1105 if (!dsl_dataset_feature_is_active(ds, f)) { 1106 return (B_FALSE); 1107 } 1108 struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f]; 1109 *outp = ftuaa->array; 1110 *outlength = ftuaa->length; 1111 return (B_TRUE); 1112 } 1113 1114 void 1115 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, void *arg, 1116 dmu_tx_t *tx) 1117 { 1118 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1119 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 1120 uint64_t zero = 0; 1121 1122 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 1123 1124 spa_feature_incr(spa, f, tx); 1125 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 1126 1127 switch (spa_feature_table[f].fi_type) { 1128 case ZFEATURE_TYPE_BOOLEAN: 1129 ASSERT3S((boolean_t)(uintptr_t)arg, ==, B_TRUE); 1130 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, 1131 sizeof (zero), 1, &zero, tx)); 1132 break; 1133 case ZFEATURE_TYPE_UINT64_ARRAY: 1134 { 1135 struct feature_type_uint64_array_arg *ftuaa = arg; 1136 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, 1137 sizeof (uint64_t), ftuaa->length, ftuaa->array, tx)); 1138 break; 1139 } 1140 default: 1141 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type); 1142 } 1143 } 1144 1145 static void 1146 dsl_dataset_deactivate_feature_impl(dsl_dataset_t *ds, spa_feature_t f, 1147 dmu_tx_t *tx) 1148 { 1149 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 1150 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 1151 uint64_t dsobj = ds->ds_object; 1152 1153 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 1154 1155 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx)); 1156 spa_feature_decr(spa, f, tx); 1157 ds->ds_feature[f] = NULL; 1158 } 1159 1160 void 1161 dsl_dataset_deactivate_feature(dsl_dataset_t *ds, spa_feature_t f, dmu_tx_t *tx) 1162 { 1163 unload_zfeature(ds, f); 1164 dsl_dataset_deactivate_feature_impl(ds, f, tx); 1165 } 1166 1167 uint64_t 1168 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 1169 dsl_crypto_params_t *dcp, uint64_t flags, dmu_tx_t *tx) 1170 { 1171 dsl_pool_t *dp = dd->dd_pool; 1172 dmu_buf_t *dbuf; 1173 dsl_dataset_phys_t *dsphys; 1174 uint64_t dsobj; 1175 objset_t *mos = dp->dp_meta_objset; 1176 1177 if (origin == NULL) 1178 origin = dp->dp_origin_snap; 1179 1180 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 1181 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); 1182 ASSERT(dmu_tx_is_syncing(tx)); 1183 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 1184 1185 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1186 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1187 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1188 dmu_buf_will_dirty(dbuf, tx); 1189 dsphys = dbuf->db_data; 1190 memset(dsphys, 0, sizeof (dsl_dataset_phys_t)); 1191 dsphys->ds_dir_obj = dd->dd_object; 1192 dsphys->ds_flags = flags; 1193 dsphys->ds_fsid_guid = unique_create(); 1194 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1195 sizeof (dsphys->ds_guid)); 1196 dsphys->ds_snapnames_zapobj = 1197 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 1198 DMU_OT_NONE, 0, tx); 1199 dsphys->ds_creation_time = gethrestime_sec(); 1200 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 1201 1202 if (origin == NULL) { 1203 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 1204 } else { 1205 dsl_dataset_t *ohds; /* head of the origin snapshot */ 1206 1207 dsphys->ds_prev_snap_obj = origin->ds_object; 1208 dsphys->ds_prev_snap_txg = 1209 dsl_dataset_phys(origin)->ds_creation_txg; 1210 dsphys->ds_referenced_bytes = 1211 dsl_dataset_phys(origin)->ds_referenced_bytes; 1212 dsphys->ds_compressed_bytes = 1213 dsl_dataset_phys(origin)->ds_compressed_bytes; 1214 dsphys->ds_uncompressed_bytes = 1215 dsl_dataset_phys(origin)->ds_uncompressed_bytes; 1216 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG); 1217 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; 1218 rrw_exit(&origin->ds_bp_rwlock, FTAG); 1219 1220 /* 1221 * Inherit flags that describe the dataset's contents 1222 * (INCONSISTENT) or properties (Case Insensitive). 1223 */ 1224 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & 1225 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 1226 1227 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1228 if (zfeature_active(f, origin->ds_feature[f])) { 1229 dsl_dataset_activate_feature(dsobj, f, 1230 origin->ds_feature[f], tx); 1231 } 1232 } 1233 1234 dmu_buf_will_dirty(origin->ds_dbuf, tx); 1235 dsl_dataset_phys(origin)->ds_num_children++; 1236 1237 VERIFY0(dsl_dataset_hold_obj(dp, 1238 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, 1239 FTAG, &ohds)); 1240 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 1241 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 1242 dsl_dataset_rele(ohds, FTAG); 1243 1244 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 1245 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { 1246 dsl_dataset_phys(origin)->ds_next_clones_obj = 1247 zap_create(mos, 1248 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 1249 } 1250 VERIFY0(zap_add_int(mos, 1251 dsl_dataset_phys(origin)->ds_next_clones_obj, 1252 dsobj, tx)); 1253 } 1254 1255 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1256 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; 1257 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 1258 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 1259 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 1260 dsl_dir_phys(origin->ds_dir)->dd_clones = 1261 zap_create(mos, 1262 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 1263 } 1264 VERIFY0(zap_add_int(mos, 1265 dsl_dir_phys(origin->ds_dir)->dd_clones, 1266 dsobj, tx)); 1267 } 1268 } 1269 1270 /* handle encryption */ 1271 dsl_dataset_create_crypt_sync(dsobj, dd, origin, dcp, tx); 1272 1273 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1274 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1275 1276 dmu_buf_rele(dbuf, FTAG); 1277 1278 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1279 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; 1280 1281 return (dsobj); 1282 } 1283 1284 static void 1285 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 1286 { 1287 objset_t *os; 1288 1289 VERIFY0(dmu_objset_from_ds(ds, &os)); 1290 if (memcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) { 1291 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1292 zio_t *zio; 1293 1294 memset(&os->os_zil_header, 0, sizeof (os->os_zil_header)); 1295 if (os->os_encrypted) 1296 os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_TRUE; 1297 1298 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1299 dsl_dataset_sync(ds, zio, tx); 1300 VERIFY0(zio_wait(zio)); 1301 dsl_dataset_sync_done(ds, tx); 1302 } 1303 } 1304 1305 uint64_t 1306 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 1307 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, 1308 dsl_crypto_params_t *dcp, dmu_tx_t *tx) 1309 { 1310 dsl_pool_t *dp = pdd->dd_pool; 1311 uint64_t dsobj, ddobj; 1312 dsl_dir_t *dd; 1313 1314 ASSERT(dmu_tx_is_syncing(tx)); 1315 ASSERT(lastname[0] != '@'); 1316 /* 1317 * Filesystems will eventually have their origin set to dp_origin_snap, 1318 * but that's taken care of in dsl_dataset_create_sync_dd. When 1319 * creating a filesystem, this function is called with origin equal to 1320 * NULL. 1321 */ 1322 if (origin != NULL) 1323 ASSERT3P(origin, !=, dp->dp_origin_snap); 1324 1325 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 1326 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 1327 1328 dsobj = dsl_dataset_create_sync_dd(dd, origin, dcp, 1329 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 1330 1331 dsl_deleg_set_create_perms(dd, tx, cr); 1332 1333 /* 1334 * If we are creating a clone and the livelist feature is enabled, 1335 * add the entry DD_FIELD_LIVELIST to ZAP. 1336 */ 1337 if (origin != NULL && 1338 spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LIVELIST)) { 1339 objset_t *mos = dd->dd_pool->dp_meta_objset; 1340 dsl_dir_zapify(dd, tx); 1341 uint64_t obj = dsl_deadlist_alloc(mos, tx); 1342 VERIFY0(zap_add(mos, dd->dd_object, DD_FIELD_LIVELIST, 1343 sizeof (uint64_t), 1, &obj, tx)); 1344 spa_feature_incr(dp->dp_spa, SPA_FEATURE_LIVELIST, tx); 1345 } 1346 1347 /* 1348 * Since we're creating a new node we know it's a leaf, so we can 1349 * initialize the counts if the limit feature is active. 1350 */ 1351 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 1352 uint64_t cnt = 0; 1353 objset_t *os = dd->dd_pool->dp_meta_objset; 1354 1355 dsl_dir_zapify(dd, tx); 1356 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 1357 sizeof (cnt), 1, &cnt, tx)); 1358 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 1359 sizeof (cnt), 1, &cnt, tx)); 1360 } 1361 1362 dsl_dir_rele(dd, FTAG); 1363 1364 /* 1365 * If we are creating a clone, make sure we zero out any stale 1366 * data from the origin snapshots zil header. 1367 */ 1368 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 1369 dsl_dataset_t *ds; 1370 1371 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 1372 dsl_dataset_zero_zil(ds, tx); 1373 dsl_dataset_rele(ds, FTAG); 1374 } 1375 1376 return (dsobj); 1377 } 1378 1379 /* 1380 * The unique space in the head dataset can be calculated by subtracting 1381 * the space used in the most recent snapshot, that is still being used 1382 * in this file system, from the space currently in use. To figure out 1383 * the space in the most recent snapshot still in use, we need to take 1384 * the total space used in the snapshot and subtract out the space that 1385 * has been freed up since the snapshot was taken. 1386 */ 1387 void 1388 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 1389 { 1390 uint64_t mrs_used; 1391 uint64_t dlused, dlcomp, dluncomp; 1392 1393 ASSERT(!ds->ds_is_snapshot); 1394 1395 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) 1396 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; 1397 else 1398 mrs_used = 0; 1399 1400 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 1401 1402 ASSERT3U(dlused, <=, mrs_used); 1403 dsl_dataset_phys(ds)->ds_unique_bytes = 1404 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); 1405 1406 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1407 SPA_VERSION_UNIQUE_ACCURATE) 1408 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1409 } 1410 1411 void 1412 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 1413 dmu_tx_t *tx) 1414 { 1415 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1416 uint64_t count __maybe_unused; 1417 int err; 1418 1419 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); 1420 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1421 obj, tx); 1422 /* 1423 * The err should not be ENOENT, but a bug in a previous version 1424 * of the code could cause upgrade_clones_cb() to not set 1425 * ds_next_snap_obj when it should, leading to a missing entry. 1426 * If we knew that the pool was created after 1427 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 1428 * ENOENT. However, at least we can check that we don't have 1429 * too many entries in the next_clones_obj even after failing to 1430 * remove this one. 1431 */ 1432 if (err != ENOENT) 1433 VERIFY0(err); 1434 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1435 &count)); 1436 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); 1437 } 1438 1439 1440 blkptr_t * 1441 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1442 { 1443 return (&dsl_dataset_phys(ds)->ds_bp); 1444 } 1445 1446 spa_t * 1447 dsl_dataset_get_spa(dsl_dataset_t *ds) 1448 { 1449 return (ds->ds_dir->dd_pool->dp_spa); 1450 } 1451 1452 void 1453 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1454 { 1455 dsl_pool_t *dp; 1456 1457 if (ds == NULL) /* this is the meta-objset */ 1458 return; 1459 1460 ASSERT(ds->ds_objset != NULL); 1461 1462 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) 1463 panic("dirtying snapshot!"); 1464 1465 /* Must not dirty a dataset in the same txg where it got snapshotted. */ 1466 ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 1467 1468 dp = ds->ds_dir->dd_pool; 1469 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1470 objset_t *os = ds->ds_objset; 1471 1472 /* up the hold count until we can be written out */ 1473 dmu_buf_add_ref(ds->ds_dbuf, ds); 1474 1475 /* if this dataset is encrypted, grab a reference to the DCK */ 1476 if (ds->ds_dir->dd_crypto_obj != 0 && 1477 !os->os_raw_receive && 1478 !os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { 1479 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1480 key_mapping_add_ref(ds->ds_key_mapping, ds); 1481 } 1482 } 1483 } 1484 1485 static int 1486 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1487 { 1488 uint64_t asize; 1489 1490 if (!dmu_tx_is_syncing(tx)) 1491 return (0); 1492 1493 /* 1494 * If there's an fs-only reservation, any blocks that might become 1495 * owned by the snapshot dataset must be accommodated by space 1496 * outside of the reservation. 1497 */ 1498 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1499 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); 1500 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1501 return (SET_ERROR(ENOSPC)); 1502 1503 /* 1504 * Propagate any reserved space for this snapshot to other 1505 * snapshot checks in this sync group. 1506 */ 1507 if (asize > 0) 1508 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1509 1510 return (0); 1511 } 1512 1513 int 1514 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1515 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr, proc_t *proc) 1516 { 1517 int error; 1518 uint64_t value; 1519 1520 ds->ds_trysnap_txg = tx->tx_txg; 1521 1522 if (!dmu_tx_is_syncing(tx)) 1523 return (0); 1524 1525 /* 1526 * We don't allow multiple snapshots of the same txg. If there 1527 * is already one, try again. 1528 */ 1529 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) 1530 return (SET_ERROR(EAGAIN)); 1531 1532 /* 1533 * Check for conflicting snapshot name. 1534 */ 1535 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1536 if (error == 0) 1537 return (SET_ERROR(EEXIST)); 1538 if (error != ENOENT) 1539 return (error); 1540 1541 /* 1542 * We don't allow taking snapshots of inconsistent datasets, such as 1543 * those into which we are currently receiving. However, if we are 1544 * creating this snapshot as part of a receive, this check will be 1545 * executed atomically with respect to the completion of the receive 1546 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1547 * case we ignore this, knowing it will be fixed up for us shortly in 1548 * dmu_recv_end_sync(). 1549 */ 1550 if (!recv && DS_IS_INCONSISTENT(ds)) 1551 return (SET_ERROR(EBUSY)); 1552 1553 /* 1554 * Skip the check for temporary snapshots or if we have already checked 1555 * the counts in dsl_dataset_snapshot_check. This means we really only 1556 * check the count here when we're receiving a stream. 1557 */ 1558 if (cnt != 0 && cr != NULL) { 1559 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1560 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr, proc); 1561 if (error != 0) 1562 return (error); 1563 } 1564 1565 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1566 if (error != 0) 1567 return (error); 1568 1569 return (0); 1570 } 1571 1572 int 1573 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1574 { 1575 dsl_dataset_snapshot_arg_t *ddsa = arg; 1576 dsl_pool_t *dp = dmu_tx_pool(tx); 1577 nvpair_t *pair; 1578 int rv = 0; 1579 1580 /* 1581 * Pre-compute how many total new snapshots will be created for each 1582 * level in the tree and below. This is needed for validating the 1583 * snapshot limit when either taking a recursive snapshot or when 1584 * taking multiple snapshots. 1585 * 1586 * The problem is that the counts are not actually adjusted when 1587 * we are checking, only when we finally sync. For a single snapshot, 1588 * this is easy, the count will increase by 1 at each node up the tree, 1589 * but its more complicated for the recursive/multiple snapshot case. 1590 * 1591 * The dsl_fs_ss_limit_check function does recursively check the count 1592 * at each level up the tree but since it is validating each snapshot 1593 * independently we need to be sure that we are validating the complete 1594 * count for the entire set of snapshots. We do this by rolling up the 1595 * counts for each component of the name into an nvlist and then 1596 * checking each of those cases with the aggregated count. 1597 * 1598 * This approach properly handles not only the recursive snapshot 1599 * case (where we get all of those on the ddsa_snaps list) but also 1600 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1601 * validate the limit on 'a' using a count of 2). 1602 * 1603 * We validate the snapshot names in the third loop and only report 1604 * name errors once. 1605 */ 1606 if (dmu_tx_is_syncing(tx)) { 1607 char *nm; 1608 nvlist_t *cnt_track = NULL; 1609 cnt_track = fnvlist_alloc(); 1610 1611 nm = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1612 1613 /* Rollup aggregated counts into the cnt_track list */ 1614 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1615 pair != NULL; 1616 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1617 char *pdelim; 1618 uint64_t val; 1619 1620 (void) strlcpy(nm, nvpair_name(pair), MAXPATHLEN); 1621 pdelim = strchr(nm, '@'); 1622 if (pdelim == NULL) 1623 continue; 1624 *pdelim = '\0'; 1625 1626 do { 1627 if (nvlist_lookup_uint64(cnt_track, nm, 1628 &val) == 0) { 1629 /* update existing entry */ 1630 fnvlist_add_uint64(cnt_track, nm, 1631 val + 1); 1632 } else { 1633 /* add to list */ 1634 fnvlist_add_uint64(cnt_track, nm, 1); 1635 } 1636 1637 pdelim = strrchr(nm, '/'); 1638 if (pdelim != NULL) 1639 *pdelim = '\0'; 1640 } while (pdelim != NULL); 1641 } 1642 1643 kmem_free(nm, MAXPATHLEN); 1644 1645 /* Check aggregated counts at each level */ 1646 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1647 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1648 int error = 0; 1649 const char *name; 1650 uint64_t cnt = 0; 1651 dsl_dataset_t *ds; 1652 1653 name = nvpair_name(pair); 1654 cnt = fnvpair_value_uint64(pair); 1655 ASSERT(cnt > 0); 1656 1657 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1658 if (error == 0) { 1659 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1660 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1661 ddsa->ddsa_cr, ddsa->ddsa_proc); 1662 dsl_dataset_rele(ds, FTAG); 1663 } 1664 1665 if (error != 0) { 1666 if (ddsa->ddsa_errors != NULL) 1667 fnvlist_add_int32(ddsa->ddsa_errors, 1668 name, error); 1669 rv = error; 1670 /* only report one error for this check */ 1671 break; 1672 } 1673 } 1674 nvlist_free(cnt_track); 1675 } 1676 1677 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1678 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1679 int error = 0; 1680 dsl_dataset_t *ds; 1681 const char *name, *atp = NULL; 1682 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1683 1684 name = nvpair_name(pair); 1685 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN) 1686 error = SET_ERROR(ENAMETOOLONG); 1687 if (error == 0) { 1688 atp = strchr(name, '@'); 1689 if (atp == NULL) 1690 error = SET_ERROR(EINVAL); 1691 if (error == 0) 1692 (void) strlcpy(dsname, name, atp - name + 1); 1693 } 1694 if (error == 0) 1695 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1696 if (error == 0) { 1697 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1698 error = dsl_dataset_snapshot_check_impl(ds, 1699 atp + 1, tx, B_FALSE, 0, NULL, NULL); 1700 dsl_dataset_rele(ds, FTAG); 1701 } 1702 1703 if (error != 0) { 1704 if (ddsa->ddsa_errors != NULL) { 1705 fnvlist_add_int32(ddsa->ddsa_errors, 1706 name, error); 1707 } 1708 rv = error; 1709 } 1710 } 1711 1712 return (rv); 1713 } 1714 1715 void 1716 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1717 dmu_tx_t *tx) 1718 { 1719 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1720 dmu_buf_t *dbuf; 1721 dsl_dataset_phys_t *dsphys; 1722 uint64_t dsobj, crtxg; 1723 objset_t *mos = dp->dp_meta_objset; 1724 objset_t *os __maybe_unused; 1725 1726 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1727 1728 /* 1729 * If we are on an old pool, the zil must not be active, in which 1730 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1731 */ 1732 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1733 dmu_objset_from_ds(ds, &os) != 0 || 1734 memcmp(&os->os_phys->os_zil_header, &zero_zil, 1735 sizeof (zero_zil)) == 0); 1736 1737 /* Should not snapshot a dirty dataset. */ 1738 ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1739 ds, tx->tx_txg)); 1740 1741 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1742 1743 /* 1744 * The origin's ds_creation_txg has to be < TXG_INITIAL 1745 */ 1746 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1747 crtxg = 1; 1748 else 1749 crtxg = tx->tx_txg; 1750 1751 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1752 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1753 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1754 dmu_buf_will_dirty(dbuf, tx); 1755 dsphys = dbuf->db_data; 1756 memset(dsphys, 0, sizeof (dsl_dataset_phys_t)); 1757 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1758 dsphys->ds_fsid_guid = unique_create(); 1759 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1760 sizeof (dsphys->ds_guid)); 1761 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1762 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 1763 dsphys->ds_next_snap_obj = ds->ds_object; 1764 dsphys->ds_num_children = 1; 1765 dsphys->ds_creation_time = gethrestime_sec(); 1766 dsphys->ds_creation_txg = crtxg; 1767 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; 1768 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; 1769 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; 1770 dsphys->ds_uncompressed_bytes = 1771 dsl_dataset_phys(ds)->ds_uncompressed_bytes; 1772 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; 1773 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1774 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; 1775 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1776 dmu_buf_rele(dbuf, FTAG); 1777 1778 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1779 if (zfeature_active(f, ds->ds_feature[f])) { 1780 dsl_dataset_activate_feature(dsobj, f, 1781 ds->ds_feature[f], tx); 1782 } 1783 } 1784 1785 ASSERT3U(ds->ds_prev != 0, ==, 1786 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); 1787 if (ds->ds_prev) { 1788 uint64_t next_clones_obj = 1789 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; 1790 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1791 ds->ds_object || 1792 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); 1793 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1794 ds->ds_object) { 1795 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1796 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, 1797 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); 1798 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; 1799 } else if (next_clones_obj != 0) { 1800 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1801 dsphys->ds_next_snap_obj, tx); 1802 VERIFY0(zap_add_int(mos, 1803 next_clones_obj, dsobj, tx)); 1804 } 1805 } 1806 1807 /* 1808 * If we have a reference-reservation on this dataset, we will 1809 * need to increase the amount of refreservation being charged 1810 * since our unique space is going to zero. 1811 */ 1812 if (ds->ds_reserved) { 1813 int64_t delta; 1814 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1815 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, 1816 ds->ds_reserved); 1817 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1818 delta, 0, 0, tx); 1819 } 1820 1821 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1822 dsl_dataset_phys(ds)->ds_deadlist_obj = 1823 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, 1824 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 1825 dsl_deadlist_close(&ds->ds_deadlist); 1826 dsl_deadlist_open(&ds->ds_deadlist, mos, 1827 dsl_dataset_phys(ds)->ds_deadlist_obj); 1828 dsl_deadlist_add_key(&ds->ds_deadlist, 1829 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); 1830 dsl_bookmark_snapshotted(ds, tx); 1831 1832 if (dsl_dataset_remap_deadlist_exists(ds)) { 1833 uint64_t remap_deadlist_obj = 1834 dsl_dataset_get_remap_deadlist_object(ds); 1835 /* 1836 * Move the remap_deadlist to the snapshot. The head 1837 * will create a new remap deadlist on demand, from 1838 * dsl_dataset_block_remapped(). 1839 */ 1840 dsl_dataset_unset_remap_deadlist_object(ds, tx); 1841 dsl_deadlist_close(&ds->ds_remap_deadlist); 1842 1843 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 1844 VERIFY0(zap_add(mos, dsobj, DS_FIELD_REMAP_DEADLIST, 1845 sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj, tx)); 1846 } 1847 1848 /* 1849 * Create a ivset guid for this snapshot if the dataset is 1850 * encrypted. This may be overridden by a raw receive. A 1851 * previous implementation of this code did not have this 1852 * field as part of the on-disk format for ZFS encryption 1853 * (see errata #4). As part of the remediation for this 1854 * issue, we ask the user to enable the bookmark_v2 feature 1855 * which is now a dependency of the encryption feature. We 1856 * use this as a heuristic to determine when the user has 1857 * elected to correct any datasets created with the old code. 1858 * As a result, we only do this step if the bookmark_v2 1859 * feature is enabled, which limits the number of states a 1860 * given pool / dataset can be in with regards to terms of 1861 * correcting the issue. 1862 */ 1863 if (ds->ds_dir->dd_crypto_obj != 0 && 1864 spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARK_V2)) { 1865 uint64_t ivset_guid = unique_create(); 1866 1867 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 1868 VERIFY0(zap_add(mos, dsobj, DS_FIELD_IVSET_GUID, 1869 sizeof (ivset_guid), 1, &ivset_guid, tx)); 1870 } 1871 1872 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); 1873 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; 1874 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; 1875 dsl_dataset_phys(ds)->ds_unique_bytes = 0; 1876 1877 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1878 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1879 1880 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1881 snapname, 8, 1, &dsobj, tx)); 1882 1883 if (ds->ds_prev) 1884 dsl_dataset_rele(ds->ds_prev, ds); 1885 VERIFY0(dsl_dataset_hold_obj(dp, 1886 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); 1887 1888 dsl_scan_ds_snapshotted(ds, tx); 1889 1890 dsl_dir_snap_cmtime_update(ds->ds_dir, tx); 1891 1892 if (zfs_snapshot_history_enabled) 1893 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, " "); 1894 } 1895 1896 void 1897 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1898 { 1899 dsl_dataset_snapshot_arg_t *ddsa = arg; 1900 dsl_pool_t *dp = dmu_tx_pool(tx); 1901 nvpair_t *pair; 1902 1903 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1904 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1905 dsl_dataset_t *ds; 1906 const char *name, *atp; 1907 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1908 1909 name = nvpair_name(pair); 1910 atp = strchr(name, '@'); 1911 (void) strlcpy(dsname, name, atp - name + 1); 1912 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1913 1914 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1915 if (ddsa->ddsa_props != NULL) { 1916 dsl_props_set_sync_impl(ds->ds_prev, 1917 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1918 } 1919 dsl_dataset_rele(ds, FTAG); 1920 } 1921 } 1922 1923 /* 1924 * The snapshots must all be in the same pool. 1925 * All-or-nothing: if there are any failures, nothing will be modified. 1926 */ 1927 int 1928 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1929 { 1930 dsl_dataset_snapshot_arg_t ddsa; 1931 nvpair_t *pair; 1932 boolean_t needsuspend; 1933 int error; 1934 spa_t *spa; 1935 const char *firstname; 1936 nvlist_t *suspended = NULL; 1937 1938 pair = nvlist_next_nvpair(snaps, NULL); 1939 if (pair == NULL) 1940 return (0); 1941 firstname = nvpair_name(pair); 1942 1943 error = spa_open(firstname, &spa, FTAG); 1944 if (error != 0) 1945 return (error); 1946 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1947 spa_close(spa, FTAG); 1948 1949 if (needsuspend) { 1950 suspended = fnvlist_alloc(); 1951 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1952 pair = nvlist_next_nvpair(snaps, pair)) { 1953 char fsname[ZFS_MAX_DATASET_NAME_LEN]; 1954 const char *snapname = nvpair_name(pair); 1955 const char *atp; 1956 void *cookie; 1957 1958 atp = strchr(snapname, '@'); 1959 if (atp == NULL) { 1960 error = SET_ERROR(EINVAL); 1961 break; 1962 } 1963 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1964 1965 error = zil_suspend(fsname, &cookie); 1966 if (error != 0) 1967 break; 1968 fnvlist_add_uint64(suspended, fsname, 1969 (uintptr_t)cookie); 1970 } 1971 } 1972 1973 ddsa.ddsa_snaps = snaps; 1974 ddsa.ddsa_props = props; 1975 ddsa.ddsa_errors = errors; 1976 ddsa.ddsa_cr = CRED(); 1977 ddsa.ddsa_proc = curproc; 1978 1979 if (error == 0) { 1980 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1981 dsl_dataset_snapshot_sync, &ddsa, 1982 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1983 } 1984 1985 if (suspended != NULL) { 1986 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1987 pair = nvlist_next_nvpair(suspended, pair)) { 1988 zil_resume((void *)(uintptr_t) 1989 fnvpair_value_uint64(pair)); 1990 } 1991 fnvlist_free(suspended); 1992 } 1993 1994 if (error == 0) { 1995 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1996 pair = nvlist_next_nvpair(snaps, pair)) { 1997 zvol_create_minor(nvpair_name(pair)); 1998 } 1999 } 2000 2001 return (error); 2002 } 2003 2004 typedef struct dsl_dataset_snapshot_tmp_arg { 2005 const char *ddsta_fsname; 2006 const char *ddsta_snapname; 2007 minor_t ddsta_cleanup_minor; 2008 const char *ddsta_htag; 2009 } dsl_dataset_snapshot_tmp_arg_t; 2010 2011 static int 2012 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 2013 { 2014 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 2015 dsl_pool_t *dp = dmu_tx_pool(tx); 2016 dsl_dataset_t *ds; 2017 int error; 2018 2019 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 2020 if (error != 0) 2021 return (error); 2022 2023 /* NULL cred means no limit check for tmp snapshot */ 2024 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 2025 tx, B_FALSE, 0, NULL, NULL); 2026 if (error != 0) { 2027 dsl_dataset_rele(ds, FTAG); 2028 return (error); 2029 } 2030 2031 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 2032 dsl_dataset_rele(ds, FTAG); 2033 return (SET_ERROR(ENOTSUP)); 2034 } 2035 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 2036 B_TRUE, tx); 2037 if (error != 0) { 2038 dsl_dataset_rele(ds, FTAG); 2039 return (error); 2040 } 2041 2042 dsl_dataset_rele(ds, FTAG); 2043 return (0); 2044 } 2045 2046 static void 2047 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 2048 { 2049 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 2050 dsl_pool_t *dp = dmu_tx_pool(tx); 2051 dsl_dataset_t *ds = NULL; 2052 2053 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 2054 2055 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 2056 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 2057 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 2058 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 2059 2060 dsl_dataset_rele(ds, FTAG); 2061 } 2062 2063 int 2064 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 2065 minor_t cleanup_minor, const char *htag) 2066 { 2067 dsl_dataset_snapshot_tmp_arg_t ddsta; 2068 int error; 2069 spa_t *spa; 2070 boolean_t needsuspend; 2071 void *cookie; 2072 2073 ddsta.ddsta_fsname = fsname; 2074 ddsta.ddsta_snapname = snapname; 2075 ddsta.ddsta_cleanup_minor = cleanup_minor; 2076 ddsta.ddsta_htag = htag; 2077 2078 error = spa_open(fsname, &spa, FTAG); 2079 if (error != 0) 2080 return (error); 2081 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 2082 spa_close(spa, FTAG); 2083 2084 if (needsuspend) { 2085 error = zil_suspend(fsname, &cookie); 2086 if (error != 0) 2087 return (error); 2088 } 2089 2090 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 2091 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 2092 2093 if (needsuspend) 2094 zil_resume(cookie); 2095 return (error); 2096 } 2097 2098 /* Nonblocking dataset sync. Assumes dataset:objset is always 1:1 */ 2099 void 2100 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *rio, dmu_tx_t *tx) 2101 { 2102 ASSERT(dmu_tx_is_syncing(tx)); 2103 ASSERT(ds->ds_objset != NULL); 2104 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); 2105 2106 /* 2107 * in case we had to change ds_fsid_guid when we opened it, 2108 * sync it out now. 2109 */ 2110 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2111 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; 2112 2113 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) { 2114 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 2115 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1, 2116 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx)); 2117 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 2118 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1, 2119 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx)); 2120 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 2121 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1, 2122 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx)); 2123 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0; 2124 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0; 2125 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0; 2126 } 2127 2128 dmu_objset_sync(ds->ds_objset, rio, tx); 2129 } 2130 2131 /* 2132 * Check if the percentage of blocks shared between the clone and the 2133 * snapshot (as opposed to those that are clone only) is below a certain 2134 * threshold 2135 */ 2136 static boolean_t 2137 dsl_livelist_should_disable(dsl_dataset_t *ds) 2138 { 2139 uint64_t used, referenced; 2140 int percent_shared; 2141 2142 used = dsl_dir_get_usedds(ds->ds_dir); 2143 referenced = dsl_get_referenced(ds); 2144 if (referenced == 0) 2145 return (B_FALSE); 2146 percent_shared = (100 * (referenced - used)) / referenced; 2147 if (percent_shared <= zfs_livelist_min_percent_shared) 2148 return (B_TRUE); 2149 return (B_FALSE); 2150 } 2151 2152 /* 2153 * Check if it is possible to combine two livelist entries into one. 2154 * This is the case if the combined number of 'live' blkptrs (ALLOCs that 2155 * don't have a matching FREE) is under the maximum sublist size. 2156 * We check this by subtracting twice the total number of frees from the total 2157 * number of blkptrs. FREEs are counted twice because each FREE blkptr 2158 * will cancel out an ALLOC blkptr when the livelist is processed. 2159 */ 2160 static boolean_t 2161 dsl_livelist_should_condense(dsl_deadlist_entry_t *first, 2162 dsl_deadlist_entry_t *next) 2163 { 2164 uint64_t total_free = first->dle_bpobj.bpo_phys->bpo_num_freed + 2165 next->dle_bpobj.bpo_phys->bpo_num_freed; 2166 uint64_t total_entries = first->dle_bpobj.bpo_phys->bpo_num_blkptrs + 2167 next->dle_bpobj.bpo_phys->bpo_num_blkptrs; 2168 if ((total_entries - (2 * total_free)) < zfs_livelist_max_entries) 2169 return (B_TRUE); 2170 return (B_FALSE); 2171 } 2172 2173 typedef struct try_condense_arg { 2174 spa_t *spa; 2175 dsl_dataset_t *ds; 2176 } try_condense_arg_t; 2177 2178 /* 2179 * Iterate over the livelist entries, searching for a pair to condense. 2180 * A nonzero return value means stop, 0 means keep looking. 2181 */ 2182 static int 2183 dsl_livelist_try_condense(void *arg, dsl_deadlist_entry_t *first) 2184 { 2185 try_condense_arg_t *tca = arg; 2186 spa_t *spa = tca->spa; 2187 dsl_dataset_t *ds = tca->ds; 2188 dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist; 2189 dsl_deadlist_entry_t *next; 2190 2191 /* The condense thread has not yet been created at import */ 2192 if (spa->spa_livelist_condense_zthr == NULL) 2193 return (1); 2194 2195 /* A condense is already in progress */ 2196 if (spa->spa_to_condense.ds != NULL) 2197 return (1); 2198 2199 next = AVL_NEXT(&ll->dl_tree, &first->dle_node); 2200 /* The livelist has only one entry - don't condense it */ 2201 if (next == NULL) 2202 return (1); 2203 2204 /* Next is the newest entry - don't condense it */ 2205 if (AVL_NEXT(&ll->dl_tree, &next->dle_node) == NULL) 2206 return (1); 2207 2208 /* This pair is not ready to condense but keep looking */ 2209 if (!dsl_livelist_should_condense(first, next)) 2210 return (0); 2211 2212 /* 2213 * Add a ref to prevent the dataset from being evicted while 2214 * the condense zthr or synctask are running. Ref will be 2215 * released at the end of the condense synctask 2216 */ 2217 dmu_buf_add_ref(ds->ds_dbuf, spa); 2218 2219 spa->spa_to_condense.ds = ds; 2220 spa->spa_to_condense.first = first; 2221 spa->spa_to_condense.next = next; 2222 spa->spa_to_condense.syncing = B_FALSE; 2223 spa->spa_to_condense.cancelled = B_FALSE; 2224 2225 zthr_wakeup(spa->spa_livelist_condense_zthr); 2226 return (1); 2227 } 2228 2229 static void 2230 dsl_flush_pending_livelist(dsl_dataset_t *ds, dmu_tx_t *tx) 2231 { 2232 dsl_dir_t *dd = ds->ds_dir; 2233 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 2234 dsl_deadlist_entry_t *last = dsl_deadlist_last(&dd->dd_livelist); 2235 2236 /* Check if we need to add a new sub-livelist */ 2237 if (last == NULL) { 2238 /* The livelist is empty */ 2239 dsl_deadlist_add_key(&dd->dd_livelist, 2240 tx->tx_txg - 1, tx); 2241 } else if (spa_sync_pass(spa) == 1) { 2242 /* 2243 * Check if the newest entry is full. If it is, make a new one. 2244 * We only do this once per sync because we could overfill a 2245 * sublist in one sync pass and don't want to add another entry 2246 * for a txg that is already represented. This ensures that 2247 * blkptrs born in the same txg are stored in the same sublist. 2248 */ 2249 bpobj_t bpobj = last->dle_bpobj; 2250 uint64_t all = bpobj.bpo_phys->bpo_num_blkptrs; 2251 uint64_t free = bpobj.bpo_phys->bpo_num_freed; 2252 uint64_t alloc = all - free; 2253 if (alloc > zfs_livelist_max_entries) { 2254 dsl_deadlist_add_key(&dd->dd_livelist, 2255 tx->tx_txg - 1, tx); 2256 } 2257 } 2258 2259 /* Insert each entry into the on-disk livelist */ 2260 bplist_iterate(&dd->dd_pending_allocs, 2261 dsl_deadlist_insert_alloc_cb, &dd->dd_livelist, tx); 2262 bplist_iterate(&dd->dd_pending_frees, 2263 dsl_deadlist_insert_free_cb, &dd->dd_livelist, tx); 2264 2265 /* Attempt to condense every pair of adjacent entries */ 2266 try_condense_arg_t arg = { 2267 .spa = spa, 2268 .ds = ds 2269 }; 2270 dsl_deadlist_iterate(&dd->dd_livelist, dsl_livelist_try_condense, 2271 &arg); 2272 } 2273 2274 void 2275 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx) 2276 { 2277 objset_t *os = ds->ds_objset; 2278 2279 bplist_iterate(&ds->ds_pending_deadlist, 2280 dsl_deadlist_insert_alloc_cb, &ds->ds_deadlist, tx); 2281 2282 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist)) { 2283 dsl_flush_pending_livelist(ds, tx); 2284 if (dsl_livelist_should_disable(ds)) { 2285 dsl_dir_remove_livelist(ds->ds_dir, tx, B_TRUE); 2286 } 2287 } 2288 2289 dsl_bookmark_sync_done(ds, tx); 2290 2291 multilist_destroy(&os->os_synced_dnodes); 2292 2293 if (os->os_encrypted) 2294 os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_FALSE; 2295 else 2296 ASSERT0(os->os_next_write_raw[tx->tx_txg & TXG_MASK]); 2297 2298 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 2299 if (zfeature_active(f, 2300 ds->ds_feature_activation[f])) { 2301 if (zfeature_active(f, ds->ds_feature[f])) 2302 continue; 2303 dsl_dataset_activate_feature(ds->ds_object, f, 2304 ds->ds_feature_activation[f], tx); 2305 ds->ds_feature[f] = ds->ds_feature_activation[f]; 2306 } 2307 } 2308 2309 ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx))); 2310 } 2311 2312 int 2313 get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val) 2314 { 2315 uint64_t count = 0; 2316 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 2317 zap_cursor_t zc; 2318 zap_attribute_t *za; 2319 2320 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 2321 2322 /* 2323 * There may be missing entries in ds_next_clones_obj 2324 * due to a bug in a previous version of the code. 2325 * Only trust it if it has the right number of entries. 2326 */ 2327 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { 2328 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 2329 &count)); 2330 } 2331 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) { 2332 return (SET_ERROR(ENOENT)); 2333 } 2334 2335 za = zap_attribute_alloc(); 2336 for (zap_cursor_init(&zc, mos, 2337 dsl_dataset_phys(ds)->ds_next_clones_obj); 2338 zap_cursor_retrieve(&zc, za) == 0; 2339 zap_cursor_advance(&zc)) { 2340 dsl_dataset_t *clone; 2341 char buf[ZFS_MAX_DATASET_NAME_LEN]; 2342 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 2343 za->za_first_integer, FTAG, &clone)); 2344 dsl_dir_name(clone->ds_dir, buf); 2345 fnvlist_add_boolean(val, buf); 2346 dsl_dataset_rele(clone, FTAG); 2347 } 2348 zap_cursor_fini(&zc); 2349 zap_attribute_free(za); 2350 return (0); 2351 } 2352 2353 void 2354 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 2355 { 2356 nvlist_t *propval = fnvlist_alloc(); 2357 nvlist_t *val = fnvlist_alloc(); 2358 2359 if (get_clones_stat_impl(ds, val) == 0) { 2360 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 2361 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), 2362 propval); 2363 } 2364 2365 nvlist_free(val); 2366 nvlist_free(propval); 2367 } 2368 2369 static char * 2370 get_receive_resume_token_impl(dsl_dataset_t *ds) 2371 { 2372 if (!dsl_dataset_has_resume_receive_state(ds)) 2373 return (NULL); 2374 2375 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2376 char *str; 2377 void *packed; 2378 uint8_t *compressed; 2379 uint64_t val; 2380 nvlist_t *token_nv = fnvlist_alloc(); 2381 size_t packed_size, compressed_size; 2382 2383 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2384 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) { 2385 fnvlist_add_uint64(token_nv, "fromguid", val); 2386 } 2387 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2388 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) { 2389 fnvlist_add_uint64(token_nv, "object", val); 2390 } 2391 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2392 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) { 2393 fnvlist_add_uint64(token_nv, "offset", val); 2394 } 2395 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2396 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) { 2397 fnvlist_add_uint64(token_nv, "bytes", val); 2398 } 2399 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2400 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) { 2401 fnvlist_add_uint64(token_nv, "toguid", val); 2402 } 2403 char buf[MAXNAMELEN]; 2404 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 2405 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) { 2406 fnvlist_add_string(token_nv, "toname", buf); 2407 } 2408 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2409 DS_FIELD_RESUME_LARGEBLOCK) == 0) { 2410 fnvlist_add_boolean(token_nv, "largeblockok"); 2411 } 2412 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2413 DS_FIELD_RESUME_EMBEDOK) == 0) { 2414 fnvlist_add_boolean(token_nv, "embedok"); 2415 } 2416 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2417 DS_FIELD_RESUME_COMPRESSOK) == 0) { 2418 fnvlist_add_boolean(token_nv, "compressok"); 2419 } 2420 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2421 DS_FIELD_RESUME_RAWOK) == 0) { 2422 fnvlist_add_boolean(token_nv, "rawok"); 2423 } 2424 if (dsl_dataset_feature_is_active(ds, 2425 SPA_FEATURE_REDACTED_DATASETS)) { 2426 uint64_t num_redact_snaps = 0; 2427 uint64_t *redact_snaps = NULL; 2428 VERIFY3B(dsl_dataset_get_uint64_array_feature(ds, 2429 SPA_FEATURE_REDACTED_DATASETS, &num_redact_snaps, 2430 &redact_snaps), ==, B_TRUE); 2431 fnvlist_add_uint64_array(token_nv, "redact_snaps", 2432 redact_snaps, num_redact_snaps); 2433 } 2434 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 2435 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS) == 0) { 2436 uint64_t num_redact_snaps = 0, int_size = 0; 2437 uint64_t *redact_snaps = NULL; 2438 VERIFY0(zap_length(dp->dp_meta_objset, ds->ds_object, 2439 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, &int_size, 2440 &num_redact_snaps)); 2441 ASSERT3U(int_size, ==, sizeof (uint64_t)); 2442 2443 redact_snaps = kmem_alloc(int_size * num_redact_snaps, 2444 KM_SLEEP); 2445 VERIFY0(zap_lookup(dp->dp_meta_objset, ds->ds_object, 2446 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, int_size, 2447 num_redact_snaps, redact_snaps)); 2448 fnvlist_add_uint64_array(token_nv, "book_redact_snaps", 2449 redact_snaps, num_redact_snaps); 2450 kmem_free(redact_snaps, int_size * num_redact_snaps); 2451 } 2452 packed = fnvlist_pack(token_nv, &packed_size); 2453 fnvlist_free(token_nv); 2454 compressed = kmem_alloc(packed_size, KM_SLEEP); 2455 2456 /* Call compress function directly to avoid hole detection. */ 2457 abd_t pabd, cabd; 2458 abd_get_from_buf_struct(&pabd, packed, packed_size); 2459 abd_get_from_buf_struct(&cabd, compressed, packed_size); 2460 compressed_size = zfs_gzip_compress(&pabd, &cabd, 2461 packed_size, packed_size, 6); 2462 abd_free(&cabd); 2463 abd_free(&pabd); 2464 2465 zio_cksum_t cksum; 2466 fletcher_4_native_varsize(compressed, compressed_size, &cksum); 2467 2468 size_t alloc_size = compressed_size * 2 + 1; 2469 str = kmem_alloc(alloc_size, KM_SLEEP); 2470 for (int i = 0; i < compressed_size; i++) { 2471 size_t offset = i * 2; 2472 (void) snprintf(str + offset, alloc_size - offset, 2473 "%02x", compressed[i]); 2474 } 2475 str[compressed_size * 2] = '\0'; 2476 char *propval = kmem_asprintf("%u-%llx-%llx-%s", 2477 ZFS_SEND_RESUME_TOKEN_VERSION, 2478 (longlong_t)cksum.zc_word[0], 2479 (longlong_t)packed_size, str); 2480 kmem_free(packed, packed_size); 2481 kmem_free(str, alloc_size); 2482 kmem_free(compressed, packed_size); 2483 return (propval); 2484 } 2485 2486 /* 2487 * Returns a string that represents the receive resume state token. It should 2488 * be freed with strfree(). NULL is returned if no resume state is present. 2489 */ 2490 char * 2491 get_receive_resume_token(dsl_dataset_t *ds) 2492 { 2493 /* 2494 * A failed "newfs" (e.g. full) resumable receive leaves 2495 * the stats set on this dataset. Check here for the prop. 2496 */ 2497 char *token = get_receive_resume_token_impl(ds); 2498 if (token != NULL) 2499 return (token); 2500 /* 2501 * A failed incremental resumable receive leaves the 2502 * stats set on our child named "%recv". Check the child 2503 * for the prop. 2504 */ 2505 /* 6 extra bytes for /%recv */ 2506 char name[ZFS_MAX_DATASET_NAME_LEN + 6]; 2507 dsl_dataset_t *recv_ds; 2508 dsl_dataset_name(ds, name); 2509 if (strlcat(name, "/", sizeof (name)) < sizeof (name) && 2510 strlcat(name, recv_clone_name, sizeof (name)) < sizeof (name) && 2511 dsl_dataset_hold(ds->ds_dir->dd_pool, name, FTAG, &recv_ds) == 0) { 2512 token = get_receive_resume_token_impl(recv_ds); 2513 dsl_dataset_rele(recv_ds, FTAG); 2514 } 2515 return (token); 2516 } 2517 2518 uint64_t 2519 dsl_get_refratio(dsl_dataset_t *ds) 2520 { 2521 uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : 2522 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / 2523 dsl_dataset_phys(ds)->ds_compressed_bytes); 2524 return (ratio); 2525 } 2526 2527 uint64_t 2528 dsl_get_logicalreferenced(dsl_dataset_t *ds) 2529 { 2530 return (dsl_dataset_phys(ds)->ds_uncompressed_bytes); 2531 } 2532 2533 uint64_t 2534 dsl_get_compressratio(dsl_dataset_t *ds) 2535 { 2536 if (ds->ds_is_snapshot) { 2537 return (dsl_get_refratio(ds)); 2538 } else { 2539 dsl_dir_t *dd = ds->ds_dir; 2540 mutex_enter(&dd->dd_lock); 2541 uint64_t val = dsl_dir_get_compressratio(dd); 2542 mutex_exit(&dd->dd_lock); 2543 return (val); 2544 } 2545 } 2546 2547 uint64_t 2548 dsl_get_used(dsl_dataset_t *ds) 2549 { 2550 if (ds->ds_is_snapshot) { 2551 return (dsl_dataset_phys(ds)->ds_unique_bytes); 2552 } else { 2553 dsl_dir_t *dd = ds->ds_dir; 2554 mutex_enter(&dd->dd_lock); 2555 uint64_t val = dsl_dir_get_used(dd); 2556 mutex_exit(&dd->dd_lock); 2557 return (val); 2558 } 2559 } 2560 2561 uint64_t 2562 dsl_get_creation(dsl_dataset_t *ds) 2563 { 2564 return (dsl_dataset_phys(ds)->ds_creation_time); 2565 } 2566 2567 uint64_t 2568 dsl_get_creationtxg(dsl_dataset_t *ds) 2569 { 2570 return (dsl_dataset_phys(ds)->ds_creation_txg); 2571 } 2572 2573 uint64_t 2574 dsl_get_refquota(dsl_dataset_t *ds) 2575 { 2576 return (ds->ds_quota); 2577 } 2578 2579 uint64_t 2580 dsl_get_refreservation(dsl_dataset_t *ds) 2581 { 2582 return (ds->ds_reserved); 2583 } 2584 2585 uint64_t 2586 dsl_get_guid(dsl_dataset_t *ds) 2587 { 2588 return (dsl_dataset_phys(ds)->ds_guid); 2589 } 2590 2591 uint64_t 2592 dsl_get_unique(dsl_dataset_t *ds) 2593 { 2594 return (dsl_dataset_phys(ds)->ds_unique_bytes); 2595 } 2596 2597 uint64_t 2598 dsl_get_objsetid(dsl_dataset_t *ds) 2599 { 2600 return (ds->ds_object); 2601 } 2602 2603 uint64_t 2604 dsl_get_userrefs(dsl_dataset_t *ds) 2605 { 2606 return (ds->ds_userrefs); 2607 } 2608 2609 uint64_t 2610 dsl_get_defer_destroy(dsl_dataset_t *ds) 2611 { 2612 return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 2613 } 2614 2615 uint64_t 2616 dsl_get_referenced(dsl_dataset_t *ds) 2617 { 2618 return (dsl_dataset_phys(ds)->ds_referenced_bytes); 2619 } 2620 2621 uint64_t 2622 dsl_get_numclones(dsl_dataset_t *ds) 2623 { 2624 ASSERT(ds->ds_is_snapshot); 2625 return (dsl_dataset_phys(ds)->ds_num_children - 1); 2626 } 2627 2628 uint64_t 2629 dsl_get_inconsistent(dsl_dataset_t *ds) 2630 { 2631 return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ? 2632 1 : 0); 2633 } 2634 2635 uint64_t 2636 dsl_get_redacted(dsl_dataset_t *ds) 2637 { 2638 return (dsl_dataset_feature_is_active(ds, 2639 SPA_FEATURE_REDACTED_DATASETS)); 2640 } 2641 2642 uint64_t 2643 dsl_get_available(dsl_dataset_t *ds) 2644 { 2645 uint64_t refdbytes = dsl_get_referenced(ds); 2646 uint64_t availbytes = dsl_dir_space_available(ds->ds_dir, 2647 NULL, 0, TRUE); 2648 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 2649 availbytes += 2650 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 2651 } 2652 if (ds->ds_quota != 0) { 2653 /* 2654 * Adjust available bytes according to refquota 2655 */ 2656 if (refdbytes < ds->ds_quota) { 2657 availbytes = MIN(availbytes, 2658 ds->ds_quota - refdbytes); 2659 } else { 2660 availbytes = 0; 2661 } 2662 } 2663 return (availbytes); 2664 } 2665 2666 int 2667 dsl_get_written(dsl_dataset_t *ds, uint64_t *written) 2668 { 2669 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2670 dsl_dataset_t *prev; 2671 int err = dsl_dataset_hold_obj(dp, 2672 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 2673 if (err == 0) { 2674 uint64_t comp, uncomp; 2675 err = dsl_dataset_space_written(prev, ds, written, 2676 &comp, &uncomp); 2677 dsl_dataset_rele(prev, FTAG); 2678 } 2679 return (err); 2680 } 2681 2682 /* 2683 * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN. 2684 */ 2685 int 2686 dsl_get_prev_snap(dsl_dataset_t *ds, char *snap) 2687 { 2688 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2689 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 2690 dsl_dataset_name(ds->ds_prev, snap); 2691 return (0); 2692 } else { 2693 return (SET_ERROR(ENOENT)); 2694 } 2695 } 2696 2697 void 2698 dsl_get_redact_snaps(dsl_dataset_t *ds, nvlist_t *propval) 2699 { 2700 uint64_t nsnaps; 2701 uint64_t *snaps; 2702 if (dsl_dataset_get_uint64_array_feature(ds, 2703 SPA_FEATURE_REDACTED_DATASETS, &nsnaps, &snaps)) { 2704 fnvlist_add_uint64_array(propval, ZPROP_VALUE, snaps, 2705 nsnaps); 2706 } 2707 } 2708 2709 /* 2710 * Returns the mountpoint property and source for the given dataset in the value 2711 * and source buffers. The value buffer must be at least as large as MAXPATHLEN 2712 * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN. 2713 * Returns 0 on success and an error on failure. 2714 */ 2715 int 2716 dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value, 2717 char *source) 2718 { 2719 int error; 2720 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2721 2722 /* Retrieve the mountpoint value stored in the zap object */ 2723 error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1, 2724 ZAP_MAXVALUELEN, value, source); 2725 if (error != 0) { 2726 return (error); 2727 } 2728 2729 /* 2730 * Process the dsname and source to find the full mountpoint string. 2731 * Can be skipped for 'legacy' or 'none'. 2732 */ 2733 if (value[0] == '/') { 2734 char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP); 2735 char *root = buf; 2736 const char *relpath; 2737 2738 /* 2739 * If we inherit the mountpoint, even from a dataset 2740 * with a received value, the source will be the path of 2741 * the dataset we inherit from. If source is 2742 * ZPROP_SOURCE_VAL_RECVD, the received value is not 2743 * inherited. 2744 */ 2745 if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) { 2746 relpath = ""; 2747 } else { 2748 ASSERT0(strncmp(dsname, source, strlen(source))); 2749 relpath = dsname + strlen(source); 2750 if (relpath[0] == '/') 2751 relpath++; 2752 } 2753 2754 spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN); 2755 2756 /* 2757 * Special case an alternate root of '/'. This will 2758 * avoid having multiple leading slashes in the 2759 * mountpoint path. 2760 */ 2761 if (strcmp(root, "/") == 0) 2762 root++; 2763 2764 /* 2765 * If the mountpoint is '/' then skip over this 2766 * if we are obtaining either an alternate root or 2767 * an inherited mountpoint. 2768 */ 2769 char *mnt = value; 2770 if (value[1] == '\0' && (root[0] != '\0' || 2771 relpath[0] != '\0')) 2772 mnt = value + 1; 2773 2774 mnt = kmem_strdup(mnt); 2775 2776 if (relpath[0] == '\0') { 2777 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s", 2778 root, mnt); 2779 } else { 2780 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s", 2781 root, mnt, relpath[0] == '@' ? "" : "/", 2782 relpath); 2783 } 2784 kmem_free(buf, ZAP_MAXVALUELEN); 2785 kmem_strfree(mnt); 2786 } 2787 2788 return (0); 2789 } 2790 2791 void 2792 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 2793 { 2794 dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool; 2795 2796 ASSERT(dsl_pool_config_held(dp)); 2797 2798 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, 2799 dsl_get_refratio(ds)); 2800 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 2801 dsl_get_logicalreferenced(ds)); 2802 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 2803 dsl_get_compressratio(ds)); 2804 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 2805 dsl_get_used(ds)); 2806 2807 if (ds->ds_is_snapshot) { 2808 get_clones_stat(ds, nv); 2809 } else { 2810 char buf[ZFS_MAX_DATASET_NAME_LEN]; 2811 if (dsl_get_prev_snap(ds, buf) == 0) 2812 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, 2813 buf); 2814 dsl_dir_stats(ds->ds_dir, nv); 2815 } 2816 2817 nvlist_t *propval = fnvlist_alloc(); 2818 dsl_get_redact_snaps(ds, propval); 2819 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_REDACT_SNAPS), 2820 propval); 2821 nvlist_free(propval); 2822 2823 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, 2824 dsl_get_available(ds)); 2825 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, 2826 dsl_get_referenced(ds)); 2827 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 2828 dsl_get_creation(ds)); 2829 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 2830 dsl_get_creationtxg(ds)); 2831 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 2832 dsl_get_refquota(ds)); 2833 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 2834 dsl_get_refreservation(ds)); 2835 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 2836 dsl_get_guid(ds)); 2837 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 2838 dsl_get_unique(ds)); 2839 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 2840 dsl_get_objsetid(ds)); 2841 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 2842 dsl_get_userrefs(ds)); 2843 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 2844 dsl_get_defer_destroy(ds)); 2845 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOTS_CHANGED, 2846 dsl_dir_snap_cmtime(ds->ds_dir).tv_sec); 2847 dsl_dataset_crypt_stats(ds, nv); 2848 2849 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 2850 uint64_t written; 2851 if (dsl_get_written(ds, &written) == 0) { 2852 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 2853 written); 2854 } 2855 } 2856 2857 if (!dsl_dataset_is_snapshot(ds)) { 2858 char *token = get_receive_resume_token(ds); 2859 if (token != NULL) { 2860 dsl_prop_nvlist_add_string(nv, 2861 ZFS_PROP_RECEIVE_RESUME_TOKEN, token); 2862 kmem_strfree(token); 2863 } 2864 } 2865 } 2866 2867 void 2868 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 2869 { 2870 dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool; 2871 ASSERT(dsl_pool_config_held(dp)); 2872 2873 stat->dds_creation_txg = dsl_get_creationtxg(ds); 2874 stat->dds_inconsistent = dsl_get_inconsistent(ds); 2875 stat->dds_guid = dsl_get_guid(ds); 2876 stat->dds_redacted = dsl_get_redacted(ds); 2877 stat->dds_origin[0] = '\0'; 2878 if (ds->ds_is_snapshot) { 2879 stat->dds_is_snapshot = B_TRUE; 2880 stat->dds_num_clones = dsl_get_numclones(ds); 2881 } else { 2882 stat->dds_is_snapshot = B_FALSE; 2883 stat->dds_num_clones = 0; 2884 2885 if (dsl_dir_is_clone(ds->ds_dir)) { 2886 dsl_dir_get_origin(ds->ds_dir, stat->dds_origin); 2887 } 2888 } 2889 } 2890 2891 uint64_t 2892 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 2893 { 2894 return (ds->ds_fsid_guid); 2895 } 2896 2897 void 2898 dsl_dataset_space(dsl_dataset_t *ds, 2899 uint64_t *refdbytesp, uint64_t *availbytesp, 2900 uint64_t *usedobjsp, uint64_t *availobjsp) 2901 { 2902 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; 2903 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 2904 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) 2905 *availbytesp += 2906 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 2907 if (ds->ds_quota != 0) { 2908 /* 2909 * Adjust available bytes according to refquota 2910 */ 2911 if (*refdbytesp < ds->ds_quota) 2912 *availbytesp = MIN(*availbytesp, 2913 ds->ds_quota - *refdbytesp); 2914 else 2915 *availbytesp = 0; 2916 } 2917 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2918 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); 2919 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2920 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 2921 } 2922 2923 boolean_t 2924 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 2925 { 2926 dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool; 2927 uint64_t birth; 2928 2929 ASSERT(dsl_pool_config_held(dp)); 2930 if (snap == NULL) 2931 return (B_FALSE); 2932 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2933 birth = BP_GET_LOGICAL_BIRTH(dsl_dataset_get_blkptr(ds)); 2934 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2935 if (birth > dsl_dataset_phys(snap)->ds_creation_txg) { 2936 objset_t *os, *os_snap; 2937 /* 2938 * It may be that only the ZIL differs, because it was 2939 * reset in the head. Don't count that as being 2940 * modified. 2941 */ 2942 if (dmu_objset_from_ds(ds, &os) != 0) 2943 return (B_TRUE); 2944 if (dmu_objset_from_ds(snap, &os_snap) != 0) 2945 return (B_TRUE); 2946 return (memcmp(&os->os_phys->os_meta_dnode, 2947 &os_snap->os_phys->os_meta_dnode, 2948 sizeof (os->os_phys->os_meta_dnode)) != 0); 2949 } 2950 return (B_FALSE); 2951 } 2952 2953 static int 2954 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 2955 dsl_dataset_t *hds, void *arg) 2956 { 2957 (void) dp; 2958 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2959 int error; 2960 uint64_t val; 2961 2962 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2963 if (error != 0) { 2964 /* ignore nonexistent snapshots */ 2965 return (error == ENOENT ? 0 : error); 2966 } 2967 2968 /* new name should not exist */ 2969 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 2970 if (error == 0) 2971 error = SET_ERROR(EEXIST); 2972 else if (error == ENOENT) 2973 error = 0; 2974 2975 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2976 if (dsl_dir_namelen(hds->ds_dir) + 1 + 2977 strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN) 2978 error = SET_ERROR(ENAMETOOLONG); 2979 2980 return (error); 2981 } 2982 2983 int 2984 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 2985 { 2986 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2987 dsl_pool_t *dp = dmu_tx_pool(tx); 2988 dsl_dataset_t *hds; 2989 int error; 2990 2991 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 2992 if (error != 0) 2993 return (error); 2994 2995 if (ddrsa->ddrsa_recursive) { 2996 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 2997 dsl_dataset_rename_snapshot_check_impl, ddrsa, 2998 DS_FIND_CHILDREN); 2999 } else { 3000 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 3001 } 3002 dsl_dataset_rele(hds, FTAG); 3003 return (error); 3004 } 3005 3006 static int 3007 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 3008 dsl_dataset_t *hds, void *arg) 3009 { 3010 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 3011 dsl_dataset_t *ds; 3012 uint64_t val; 3013 dmu_tx_t *tx = ddrsa->ddrsa_tx; 3014 char *oldname, *newname; 3015 int error; 3016 3017 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 3018 ASSERT(error == 0 || error == ENOENT); 3019 if (error == ENOENT) { 3020 /* ignore nonexistent snapshots */ 3021 return (0); 3022 } 3023 3024 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 3025 3026 /* log before we change the name */ 3027 spa_history_log_internal_ds(ds, "rename", tx, 3028 "-> @%s", ddrsa->ddrsa_newsnapname); 3029 3030 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 3031 B_FALSE)); 3032 mutex_enter(&ds->ds_lock); 3033 (void) strlcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname, 3034 sizeof (ds->ds_snapname)); 3035 mutex_exit(&ds->ds_lock); 3036 VERIFY0(zap_add(dp->dp_meta_objset, 3037 dsl_dataset_phys(hds)->ds_snapnames_zapobj, 3038 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 3039 3040 oldname = kmem_asprintf("%s@%s", ddrsa->ddrsa_fsname, 3041 ddrsa->ddrsa_oldsnapname); 3042 newname = kmem_asprintf("%s@%s", ddrsa->ddrsa_fsname, 3043 ddrsa->ddrsa_newsnapname); 3044 zvol_rename_minors(dp->dp_spa, oldname, newname, B_TRUE); 3045 kmem_strfree(oldname); 3046 kmem_strfree(newname); 3047 3048 dsl_dataset_rele(ds, FTAG); 3049 return (0); 3050 } 3051 3052 void 3053 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 3054 { 3055 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 3056 dsl_pool_t *dp = dmu_tx_pool(tx); 3057 dsl_dataset_t *hds = NULL; 3058 3059 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 3060 ddrsa->ddrsa_tx = tx; 3061 if (ddrsa->ddrsa_recursive) { 3062 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 3063 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 3064 DS_FIND_CHILDREN)); 3065 } else { 3066 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 3067 } 3068 dsl_dataset_rele(hds, FTAG); 3069 } 3070 3071 int 3072 dsl_dataset_rename_snapshot(const char *fsname, 3073 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 3074 { 3075 dsl_dataset_rename_snapshot_arg_t ddrsa; 3076 3077 ddrsa.ddrsa_fsname = fsname; 3078 ddrsa.ddrsa_oldsnapname = oldsnapname; 3079 ddrsa.ddrsa_newsnapname = newsnapname; 3080 ddrsa.ddrsa_recursive = recursive; 3081 3082 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 3083 dsl_dataset_rename_snapshot_sync, &ddrsa, 3084 1, ZFS_SPACE_CHECK_RESERVED)); 3085 } 3086 3087 /* 3088 * If we're doing an ownership handoff, we need to make sure that there is 3089 * only one long hold on the dataset. We're not allowed to change anything here 3090 * so we don't permanently release the long hold or regular hold here. We want 3091 * to do this only when syncing to avoid the dataset unexpectedly going away 3092 * when we release the long hold. 3093 */ 3094 static int 3095 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 3096 { 3097 boolean_t held = B_FALSE; 3098 3099 if (!dmu_tx_is_syncing(tx)) 3100 return (0); 3101 3102 dsl_dir_t *dd = ds->ds_dir; 3103 mutex_enter(&dd->dd_activity_lock); 3104 uint64_t holds = zfs_refcount_count(&ds->ds_longholds) - 3105 (owner != NULL ? 1 : 0); 3106 /* 3107 * The value of dd_activity_waiters can chance as soon as we drop the 3108 * lock, but we're fine with that; new waiters coming in or old 3109 * waiters leaving doesn't cause problems, since we're going to cancel 3110 * waiters later anyway. The goal of this check is to verify that no 3111 * non-waiters have long-holds, and all new long-holds will be 3112 * prevented because we're holding the pool config as writer. 3113 */ 3114 if (holds != dd->dd_activity_waiters) 3115 held = B_TRUE; 3116 mutex_exit(&dd->dd_activity_lock); 3117 3118 if (held) 3119 return (SET_ERROR(EBUSY)); 3120 3121 return (0); 3122 } 3123 3124 int 3125 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 3126 { 3127 dsl_dataset_rollback_arg_t *ddra = arg; 3128 dsl_pool_t *dp = dmu_tx_pool(tx); 3129 dsl_dataset_t *ds; 3130 int64_t unused_refres_delta; 3131 int error; 3132 3133 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 3134 if (error != 0) 3135 return (error); 3136 3137 /* must not be a snapshot */ 3138 if (ds->ds_is_snapshot) { 3139 dsl_dataset_rele(ds, FTAG); 3140 return (SET_ERROR(EINVAL)); 3141 } 3142 3143 /* must have a most recent snapshot */ 3144 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { 3145 dsl_dataset_rele(ds, FTAG); 3146 return (SET_ERROR(ESRCH)); 3147 } 3148 3149 /* 3150 * No rollback to a snapshot created in the current txg, because 3151 * the rollback may dirty the dataset and create blocks that are 3152 * not reachable from the rootbp while having a birth txg that 3153 * falls into the snapshot's range. 3154 */ 3155 if (dmu_tx_is_syncing(tx) && 3156 dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) { 3157 dsl_dataset_rele(ds, FTAG); 3158 return (SET_ERROR(EAGAIN)); 3159 } 3160 3161 /* 3162 * If the expected target snapshot is specified, then check that 3163 * the latest snapshot is it. 3164 */ 3165 if (ddra->ddra_tosnap != NULL) { 3166 dsl_dataset_t *snapds; 3167 3168 /* Check if the target snapshot exists at all. */ 3169 error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds); 3170 if (error != 0) { 3171 /* 3172 * ESRCH is used to signal that the target snapshot does 3173 * not exist, while ENOENT is used to report that 3174 * the rolled back dataset does not exist. 3175 * ESRCH is also used to cover other cases where the 3176 * target snapshot is not related to the dataset being 3177 * rolled back such as being in a different pool. 3178 */ 3179 if (error == ENOENT || error == EXDEV) 3180 error = SET_ERROR(ESRCH); 3181 dsl_dataset_rele(ds, FTAG); 3182 return (error); 3183 } 3184 ASSERT(snapds->ds_is_snapshot); 3185 3186 /* Check if the snapshot is the latest snapshot indeed. */ 3187 if (snapds != ds->ds_prev) { 3188 /* 3189 * Distinguish between the case where the only problem 3190 * is intervening snapshots (EEXIST) vs the snapshot 3191 * not being a valid target for rollback (ESRCH). 3192 */ 3193 if (snapds->ds_dir == ds->ds_dir || 3194 (dsl_dir_is_clone(ds->ds_dir) && 3195 dsl_dir_phys(ds->ds_dir)->dd_origin_obj == 3196 snapds->ds_object)) { 3197 error = SET_ERROR(EEXIST); 3198 } else { 3199 error = SET_ERROR(ESRCH); 3200 } 3201 dsl_dataset_rele(snapds, FTAG); 3202 dsl_dataset_rele(ds, FTAG); 3203 return (error); 3204 } 3205 dsl_dataset_rele(snapds, FTAG); 3206 } 3207 3208 /* must not have any bookmarks after the most recent snapshot */ 3209 if (dsl_bookmark_latest_txg(ds) > 3210 dsl_dataset_phys(ds)->ds_prev_snap_txg) { 3211 dsl_dataset_rele(ds, FTAG); 3212 return (SET_ERROR(EEXIST)); 3213 } 3214 3215 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 3216 if (error != 0) { 3217 dsl_dataset_rele(ds, FTAG); 3218 return (error); 3219 } 3220 3221 /* 3222 * Check if the snap we are rolling back to uses more than 3223 * the refquota. 3224 */ 3225 if (ds->ds_quota != 0 && 3226 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { 3227 dsl_dataset_rele(ds, FTAG); 3228 return (SET_ERROR(EDQUOT)); 3229 } 3230 3231 /* 3232 * When we do the clone swap, we will temporarily use more space 3233 * due to the refreservation (the head will no longer have any 3234 * unique space, so the entire amount of the refreservation will need 3235 * to be free). We will immediately destroy the clone, freeing 3236 * this space, but the freeing happens over many txg's. 3237 */ 3238 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 3239 dsl_dataset_phys(ds)->ds_unique_bytes); 3240 3241 if (unused_refres_delta > 0 && 3242 unused_refres_delta > 3243 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 3244 dsl_dataset_rele(ds, FTAG); 3245 return (SET_ERROR(ENOSPC)); 3246 } 3247 3248 dsl_dataset_rele(ds, FTAG); 3249 return (0); 3250 } 3251 3252 void 3253 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 3254 { 3255 dsl_dataset_rollback_arg_t *ddra = arg; 3256 dsl_pool_t *dp = dmu_tx_pool(tx); 3257 dsl_dataset_t *ds, *clone; 3258 uint64_t cloneobj; 3259 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 3260 3261 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 3262 3263 dsl_dataset_name(ds->ds_prev, namebuf); 3264 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 3265 3266 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 3267 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, NULL, tx); 3268 3269 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 3270 3271 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 3272 dsl_dataset_zero_zil(ds, tx); 3273 3274 dsl_destroy_head_sync_impl(clone, tx); 3275 3276 dsl_dataset_rele(clone, FTAG); 3277 dsl_dataset_rele(ds, FTAG); 3278 } 3279 3280 /* 3281 * Rolls back the given filesystem or volume to the most recent snapshot. 3282 * The name of the most recent snapshot will be returned under key "target" 3283 * in the result nvlist. 3284 * 3285 * If owner != NULL: 3286 * - The existing dataset MUST be owned by the specified owner at entry 3287 * - Upon return, dataset will still be held by the same owner, whether we 3288 * succeed or not. 3289 * 3290 * This mode is required any time the existing filesystem is mounted. See 3291 * notes above zfs_suspend_fs() for further details. 3292 */ 3293 int 3294 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner, 3295 nvlist_t *result) 3296 { 3297 dsl_dataset_rollback_arg_t ddra; 3298 3299 ddra.ddra_fsname = fsname; 3300 ddra.ddra_tosnap = tosnap; 3301 ddra.ddra_owner = owner; 3302 ddra.ddra_result = result; 3303 3304 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 3305 dsl_dataset_rollback_sync, &ddra, 3306 1, ZFS_SPACE_CHECK_RESERVED)); 3307 } 3308 3309 struct promotenode { 3310 list_node_t link; 3311 dsl_dataset_t *ds; 3312 }; 3313 3314 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 3315 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 3316 const void *tag); 3317 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, const void *tag); 3318 3319 int 3320 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 3321 { 3322 dsl_dataset_promote_arg_t *ddpa = arg; 3323 dsl_pool_t *dp = dmu_tx_pool(tx); 3324 dsl_dataset_t *hds; 3325 struct promotenode *snap; 3326 int err; 3327 uint64_t unused; 3328 uint64_t ss_mv_cnt; 3329 size_t max_snap_len; 3330 boolean_t conflicting_snaps; 3331 3332 err = promote_hold(ddpa, dp, FTAG); 3333 if (err != 0) 3334 return (err); 3335 3336 hds = ddpa->ddpa_clone; 3337 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; 3338 3339 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { 3340 promote_rele(ddpa, FTAG); 3341 return (SET_ERROR(EXDEV)); 3342 } 3343 3344 snap = list_head(&ddpa->shared_snaps); 3345 if (snap == NULL) { 3346 err = SET_ERROR(ENOENT); 3347 goto out; 3348 } 3349 dsl_dataset_t *const origin_ds = snap->ds; 3350 3351 /* 3352 * Encrypted clones share a DSL Crypto Key with their origin's dsl dir. 3353 * When doing a promote we must make sure the encryption root for 3354 * both the target and the target's origin does not change to avoid 3355 * needing to rewrap encryption keys 3356 */ 3357 err = dsl_dataset_promote_crypt_check(hds->ds_dir, origin_ds->ds_dir); 3358 if (err != 0) 3359 goto out; 3360 3361 /* 3362 * Compute and check the amount of space to transfer. Since this is 3363 * so expensive, don't do the preliminary check. 3364 */ 3365 if (!dmu_tx_is_syncing(tx)) { 3366 promote_rele(ddpa, FTAG); 3367 return (0); 3368 } 3369 3370 /* compute origin's new unique space */ 3371 snap = list_tail(&ddpa->clone_snaps); 3372 ASSERT(snap != NULL); 3373 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 3374 origin_ds->ds_object); 3375 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 3376 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, 3377 &ddpa->unique, &unused, &unused); 3378 3379 /* 3380 * Walk the snapshots that we are moving 3381 * 3382 * Compute space to transfer. Consider the incremental changes 3383 * to used by each snapshot: 3384 * (my used) = (prev's used) + (blocks born) - (blocks killed) 3385 * So each snapshot gave birth to: 3386 * (blocks born) = (my used) - (prev's used) + (blocks killed) 3387 * So a sequence would look like: 3388 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 3389 * Which simplifies to: 3390 * uN + kN + kN-1 + ... + k1 + k0 3391 * Note however, if we stop before we reach the ORIGIN we get: 3392 * uN + kN + kN-1 + ... + kM - uM-1 3393 */ 3394 conflicting_snaps = B_FALSE; 3395 ss_mv_cnt = 0; 3396 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; 3397 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; 3398 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; 3399 for (snap = list_head(&ddpa->shared_snaps); snap; 3400 snap = list_next(&ddpa->shared_snaps, snap)) { 3401 uint64_t val, dlused, dlcomp, dluncomp; 3402 dsl_dataset_t *ds = snap->ds; 3403 3404 ss_mv_cnt++; 3405 3406 /* 3407 * If there are long holds, we won't be able to evict 3408 * the objset. 3409 */ 3410 if (dsl_dataset_long_held(ds)) { 3411 err = SET_ERROR(EBUSY); 3412 goto out; 3413 } 3414 3415 /* Check that the snapshot name does not conflict */ 3416 VERIFY0(dsl_dataset_get_snapname(ds)); 3417 if (strlen(ds->ds_snapname) >= max_snap_len) { 3418 err = SET_ERROR(ENAMETOOLONG); 3419 goto out; 3420 } 3421 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 3422 if (err == 0) { 3423 fnvlist_add_boolean(ddpa->err_ds, 3424 snap->ds->ds_snapname); 3425 conflicting_snaps = B_TRUE; 3426 } else if (err != ENOENT) { 3427 goto out; 3428 } 3429 3430 /* The very first snapshot does not have a deadlist */ 3431 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) 3432 continue; 3433 3434 dsl_deadlist_space(&ds->ds_deadlist, 3435 &dlused, &dlcomp, &dluncomp); 3436 ddpa->used += dlused; 3437 ddpa->comp += dlcomp; 3438 ddpa->uncomp += dluncomp; 3439 } 3440 3441 /* 3442 * Check that bookmarks that are being transferred don't have 3443 * name conflicts. 3444 */ 3445 for (dsl_bookmark_node_t *dbn = avl_first(&origin_ds->ds_bookmarks); 3446 dbn != NULL && dbn->dbn_phys.zbm_creation_txg <= 3447 dsl_dataset_phys(origin_ds)->ds_creation_txg; 3448 dbn = AVL_NEXT(&origin_ds->ds_bookmarks, dbn)) { 3449 if (strlen(dbn->dbn_name) >= max_snap_len) { 3450 err = SET_ERROR(ENAMETOOLONG); 3451 goto out; 3452 } 3453 zfs_bookmark_phys_t bm; 3454 err = dsl_bookmark_lookup_impl(ddpa->ddpa_clone, 3455 dbn->dbn_name, &bm); 3456 3457 if (err == 0) { 3458 fnvlist_add_boolean(ddpa->err_ds, dbn->dbn_name); 3459 conflicting_snaps = B_TRUE; 3460 } else if (err == ESRCH) { 3461 err = 0; 3462 } 3463 if (err != 0) { 3464 goto out; 3465 } 3466 } 3467 3468 /* 3469 * In order to return the full list of conflicting snapshots, we check 3470 * whether there was a conflict after traversing all of them. 3471 */ 3472 if (conflicting_snaps) { 3473 err = SET_ERROR(EEXIST); 3474 goto out; 3475 } 3476 3477 /* 3478 * If we are a clone of a clone then we never reached ORIGIN, 3479 * so we need to subtract out the clone origin's used space. 3480 */ 3481 if (ddpa->origin_origin) { 3482 ddpa->used -= 3483 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; 3484 ddpa->comp -= 3485 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; 3486 ddpa->uncomp -= 3487 dsl_dataset_phys(ddpa->origin_origin)-> 3488 ds_uncompressed_bytes; 3489 } 3490 3491 /* Check that there is enough space and limit headroom here */ 3492 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 3493 0, ss_mv_cnt, ddpa->used, ddpa->cr, ddpa->proc); 3494 if (err != 0) 3495 goto out; 3496 3497 /* 3498 * Compute the amounts of space that will be used by snapshots 3499 * after the promotion (for both origin and clone). For each, 3500 * it is the amount of space that will be on all of their 3501 * deadlists (that was not born before their new origin). 3502 */ 3503 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 3504 uint64_t space; 3505 3506 /* 3507 * Note, typically this will not be a clone of a clone, 3508 * so dd_origin_txg will be < TXG_INITIAL, so 3509 * these snaplist_space() -> dsl_deadlist_space_range() 3510 * calls will be fast because they do not have to 3511 * iterate over all bps. 3512 */ 3513 snap = list_head(&ddpa->origin_snaps); 3514 if (snap == NULL) { 3515 err = SET_ERROR(ENOENT); 3516 goto out; 3517 } 3518 err = snaplist_space(&ddpa->shared_snaps, 3519 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 3520 if (err != 0) 3521 goto out; 3522 3523 err = snaplist_space(&ddpa->clone_snaps, 3524 snap->ds->ds_dir->dd_origin_txg, &space); 3525 if (err != 0) 3526 goto out; 3527 ddpa->cloneusedsnap += space; 3528 } 3529 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & 3530 DD_FLAG_USED_BREAKDOWN) { 3531 err = snaplist_space(&ddpa->origin_snaps, 3532 dsl_dataset_phys(origin_ds)->ds_creation_txg, 3533 &ddpa->originusedsnap); 3534 if (err != 0) 3535 goto out; 3536 } 3537 3538 out: 3539 promote_rele(ddpa, FTAG); 3540 return (err); 3541 } 3542 3543 void 3544 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 3545 { 3546 dsl_dataset_promote_arg_t *ddpa = arg; 3547 dsl_pool_t *dp = dmu_tx_pool(tx); 3548 dsl_dataset_t *hds; 3549 struct promotenode *snap; 3550 dsl_dataset_t *origin_ds; 3551 dsl_dataset_t *origin_head; 3552 dsl_dir_t *dd; 3553 dsl_dir_t *odd = NULL; 3554 uint64_t oldnext_obj; 3555 int64_t delta; 3556 3557 ASSERT(nvlist_empty(ddpa->err_ds)); 3558 3559 VERIFY0(promote_hold(ddpa, dp, FTAG)); 3560 hds = ddpa->ddpa_clone; 3561 3562 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); 3563 3564 snap = list_head(&ddpa->shared_snaps); 3565 origin_ds = snap->ds; 3566 dd = hds->ds_dir; 3567 3568 snap = list_head(&ddpa->origin_snaps); 3569 origin_head = snap->ds; 3570 3571 /* 3572 * We need to explicitly open odd, since origin_ds's dd will be 3573 * changing. 3574 */ 3575 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 3576 NULL, FTAG, &odd)); 3577 3578 dsl_dataset_promote_crypt_sync(hds->ds_dir, odd, tx); 3579 3580 /* change origin's next snap */ 3581 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 3582 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; 3583 snap = list_tail(&ddpa->clone_snaps); 3584 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 3585 origin_ds->ds_object); 3586 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; 3587 3588 /* change the origin's next clone */ 3589 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { 3590 dsl_dataset_remove_from_next_clones(origin_ds, 3591 snap->ds->ds_object, tx); 3592 VERIFY0(zap_add_int(dp->dp_meta_objset, 3593 dsl_dataset_phys(origin_ds)->ds_next_clones_obj, 3594 oldnext_obj, tx)); 3595 } 3596 3597 /* change origin */ 3598 dmu_buf_will_dirty(dd->dd_dbuf, tx); 3599 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); 3600 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; 3601 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 3602 dmu_buf_will_dirty(odd->dd_dbuf, tx); 3603 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; 3604 origin_head->ds_dir->dd_origin_txg = 3605 dsl_dataset_phys(origin_ds)->ds_creation_txg; 3606 3607 /* change dd_clone entries */ 3608 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 3609 VERIFY0(zap_remove_int(dp->dp_meta_objset, 3610 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); 3611 VERIFY0(zap_add_int(dp->dp_meta_objset, 3612 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 3613 hds->ds_object, tx)); 3614 3615 VERIFY0(zap_remove_int(dp->dp_meta_objset, 3616 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 3617 origin_head->ds_object, tx)); 3618 if (dsl_dir_phys(dd)->dd_clones == 0) { 3619 dsl_dir_phys(dd)->dd_clones = 3620 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, 3621 DMU_OT_NONE, 0, tx); 3622 } 3623 VERIFY0(zap_add_int(dp->dp_meta_objset, 3624 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); 3625 } 3626 3627 /* 3628 * Move bookmarks to this dir. 3629 */ 3630 dsl_bookmark_node_t *dbn_next; 3631 for (dsl_bookmark_node_t *dbn = avl_first(&origin_head->ds_bookmarks); 3632 dbn != NULL && dbn->dbn_phys.zbm_creation_txg <= 3633 dsl_dataset_phys(origin_ds)->ds_creation_txg; 3634 dbn = dbn_next) { 3635 dbn_next = AVL_NEXT(&origin_head->ds_bookmarks, dbn); 3636 3637 avl_remove(&origin_head->ds_bookmarks, dbn); 3638 VERIFY0(zap_remove(dp->dp_meta_objset, 3639 origin_head->ds_bookmarks_obj, dbn->dbn_name, tx)); 3640 3641 dsl_bookmark_node_add(hds, dbn, tx); 3642 } 3643 3644 dsl_bookmark_next_changed(hds, origin_ds, tx); 3645 3646 /* move snapshots to this dir */ 3647 for (snap = list_head(&ddpa->shared_snaps); snap; 3648 snap = list_next(&ddpa->shared_snaps, snap)) { 3649 dsl_dataset_t *ds = snap->ds; 3650 3651 /* 3652 * Property callbacks are registered to a particular 3653 * dsl_dir. Since ours is changing, evict the objset 3654 * so that they will be unregistered from the old dsl_dir. 3655 */ 3656 if (ds->ds_objset) { 3657 dmu_objset_evict(ds->ds_objset); 3658 ds->ds_objset = NULL; 3659 } 3660 3661 /* move snap name entry */ 3662 VERIFY0(dsl_dataset_get_snapname(ds)); 3663 VERIFY0(dsl_dataset_snap_remove(origin_head, 3664 ds->ds_snapname, tx, B_TRUE)); 3665 VERIFY0(zap_add(dp->dp_meta_objset, 3666 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 3667 8, 1, &ds->ds_object, tx)); 3668 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 3669 DD_FIELD_SNAPSHOT_COUNT, tx); 3670 3671 /* change containing dsl_dir */ 3672 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3673 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); 3674 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; 3675 ASSERT3P(ds->ds_dir, ==, odd); 3676 dsl_dir_rele(ds->ds_dir, ds); 3677 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 3678 NULL, ds, &ds->ds_dir)); 3679 3680 /* move any clone references */ 3681 if (dsl_dataset_phys(ds)->ds_next_clones_obj && 3682 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 3683 zap_cursor_t zc; 3684 zap_attribute_t *za = zap_attribute_alloc(); 3685 3686 for (zap_cursor_init(&zc, dp->dp_meta_objset, 3687 dsl_dataset_phys(ds)->ds_next_clones_obj); 3688 zap_cursor_retrieve(&zc, za) == 0; 3689 zap_cursor_advance(&zc)) { 3690 dsl_dataset_t *cnds; 3691 uint64_t o; 3692 3693 if (za->za_first_integer == oldnext_obj) { 3694 /* 3695 * We've already moved the 3696 * origin's reference. 3697 */ 3698 continue; 3699 } 3700 3701 VERIFY0(dsl_dataset_hold_obj(dp, 3702 za->za_first_integer, FTAG, &cnds)); 3703 o = dsl_dir_phys(cnds->ds_dir)-> 3704 dd_head_dataset_obj; 3705 3706 VERIFY0(zap_remove_int(dp->dp_meta_objset, 3707 dsl_dir_phys(odd)->dd_clones, o, tx)); 3708 VERIFY0(zap_add_int(dp->dp_meta_objset, 3709 dsl_dir_phys(dd)->dd_clones, o, tx)); 3710 dsl_dataset_rele(cnds, FTAG); 3711 } 3712 zap_cursor_fini(&zc); 3713 zap_attribute_free(za); 3714 } 3715 3716 ASSERT(!dsl_prop_hascb(ds)); 3717 } 3718 3719 /* 3720 * Change space accounting. 3721 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 3722 * both be valid, or both be 0 (resulting in delta == 0). This 3723 * is true for each of {clone,origin} independently. 3724 */ 3725 3726 delta = ddpa->cloneusedsnap - 3727 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; 3728 ASSERT3S(delta, >=, 0); 3729 ASSERT3U(ddpa->used, >=, delta); 3730 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 3731 dsl_dir_diduse_space(dd, DD_USED_HEAD, 3732 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 3733 3734 delta = ddpa->originusedsnap - 3735 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; 3736 ASSERT3S(delta, <=, 0); 3737 ASSERT3U(ddpa->used, >=, -delta); 3738 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 3739 dsl_dir_diduse_space(odd, DD_USED_HEAD, 3740 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 3741 3742 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; 3743 3744 /* 3745 * Since livelists are specific to a clone's origin txg, they 3746 * are no longer accurate. Destroy the livelist from the clone being 3747 * promoted. If the origin dataset is a clone, destroy its livelist 3748 * as well. 3749 */ 3750 dsl_dir_remove_livelist(dd, tx, B_TRUE); 3751 dsl_dir_remove_livelist(odd, tx, B_TRUE); 3752 3753 /* log history record */ 3754 spa_history_log_internal_ds(hds, "promote", tx, " "); 3755 3756 dsl_dir_rele(odd, FTAG); 3757 3758 /* 3759 * Transfer common error blocks from old head to new head, before 3760 * calling promote_rele() on ddpa since we need to dereference 3761 * origin_head and hds. 3762 */ 3763 if (spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_HEAD_ERRLOG)) { 3764 uint64_t old_head = origin_head->ds_object; 3765 uint64_t new_head = hds->ds_object; 3766 spa_swap_errlog(dp->dp_spa, new_head, old_head, tx); 3767 } 3768 3769 promote_rele(ddpa, FTAG); 3770 } 3771 3772 /* 3773 * Make a list of dsl_dataset_t's for the snapshots between first_obj 3774 * (exclusive) and last_obj (inclusive). The list will be in reverse 3775 * order (last_obj will be the list_head()). If first_obj == 0, do all 3776 * snapshots back to this dataset's origin. 3777 */ 3778 static int 3779 snaplist_make(dsl_pool_t *dp, 3780 uint64_t first_obj, uint64_t last_obj, list_t *l, const void *tag) 3781 { 3782 uint64_t obj = last_obj; 3783 3784 list_create(l, sizeof (struct promotenode), 3785 offsetof(struct promotenode, link)); 3786 3787 while (obj != first_obj) { 3788 dsl_dataset_t *ds; 3789 struct promotenode *snap; 3790 int err; 3791 3792 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 3793 ASSERT(err != ENOENT); 3794 if (err != 0) 3795 return (err); 3796 3797 if (first_obj == 0) 3798 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; 3799 3800 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 3801 snap->ds = ds; 3802 list_insert_tail(l, snap); 3803 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3804 } 3805 3806 return (0); 3807 } 3808 3809 static int 3810 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 3811 { 3812 struct promotenode *snap; 3813 3814 *spacep = 0; 3815 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 3816 uint64_t used, comp, uncomp; 3817 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 3818 mintxg, UINT64_MAX, &used, &comp, &uncomp); 3819 *spacep += used; 3820 } 3821 return (0); 3822 } 3823 3824 static void 3825 snaplist_destroy(list_t *l, const void *tag) 3826 { 3827 struct promotenode *snap; 3828 3829 if (l == NULL || !list_link_active(&l->list_head)) 3830 return; 3831 3832 while ((snap = list_remove_tail(l)) != NULL) { 3833 dsl_dataset_rele(snap->ds, tag); 3834 kmem_free(snap, sizeof (*snap)); 3835 } 3836 list_destroy(l); 3837 } 3838 3839 static int 3840 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, const void *tag) 3841 { 3842 int error; 3843 dsl_dir_t *dd; 3844 struct promotenode *snap; 3845 3846 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 3847 &ddpa->ddpa_clone); 3848 if (error != 0) 3849 return (error); 3850 dd = ddpa->ddpa_clone->ds_dir; 3851 3852 if (ddpa->ddpa_clone->ds_is_snapshot || 3853 !dsl_dir_is_clone(dd)) { 3854 dsl_dataset_rele(ddpa->ddpa_clone, tag); 3855 return (SET_ERROR(EINVAL)); 3856 } 3857 3858 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, 3859 &ddpa->shared_snaps, tag); 3860 if (error != 0) 3861 goto out; 3862 3863 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 3864 &ddpa->clone_snaps, tag); 3865 if (error != 0) 3866 goto out; 3867 3868 snap = list_head(&ddpa->shared_snaps); 3869 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); 3870 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, 3871 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, 3872 &ddpa->origin_snaps, tag); 3873 if (error != 0) 3874 goto out; 3875 3876 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { 3877 error = dsl_dataset_hold_obj(dp, 3878 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, 3879 tag, &ddpa->origin_origin); 3880 if (error != 0) 3881 goto out; 3882 } 3883 out: 3884 if (error != 0) 3885 promote_rele(ddpa, tag); 3886 return (error); 3887 } 3888 3889 static void 3890 promote_rele(dsl_dataset_promote_arg_t *ddpa, const void *tag) 3891 { 3892 snaplist_destroy(&ddpa->shared_snaps, tag); 3893 snaplist_destroy(&ddpa->clone_snaps, tag); 3894 snaplist_destroy(&ddpa->origin_snaps, tag); 3895 if (ddpa->origin_origin != NULL) 3896 dsl_dataset_rele(ddpa->origin_origin, tag); 3897 dsl_dataset_rele(ddpa->ddpa_clone, tag); 3898 } 3899 3900 /* 3901 * Promote a clone. 3902 * 3903 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 3904 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.) 3905 */ 3906 int 3907 dsl_dataset_promote(const char *name, char *conflsnap) 3908 { 3909 dsl_dataset_promote_arg_t ddpa = { 0 }; 3910 uint64_t numsnaps; 3911 int error; 3912 nvpair_t *snap_pair; 3913 objset_t *os; 3914 3915 /* 3916 * We will modify space proportional to the number of 3917 * snapshots. Compute numsnaps. 3918 */ 3919 error = dmu_objset_hold(name, FTAG, &os); 3920 if (error != 0) 3921 return (error); 3922 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 3923 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, 3924 &numsnaps); 3925 dmu_objset_rele(os, FTAG); 3926 if (error != 0) 3927 return (error); 3928 3929 ddpa.ddpa_clonename = name; 3930 ddpa.err_ds = fnvlist_alloc(); 3931 ddpa.cr = CRED(); 3932 ddpa.proc = curproc; 3933 3934 error = dsl_sync_task(name, dsl_dataset_promote_check, 3935 dsl_dataset_promote_sync, &ddpa, 3936 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED); 3937 3938 /* 3939 * Return the first conflicting snapshot found. 3940 */ 3941 snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL); 3942 if (snap_pair != NULL && conflsnap != NULL) 3943 (void) strlcpy(conflsnap, nvpair_name(snap_pair), 3944 ZFS_MAX_DATASET_NAME_LEN); 3945 3946 fnvlist_free(ddpa.err_ds); 3947 return (error); 3948 } 3949 3950 int 3951 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 3952 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 3953 { 3954 /* 3955 * "slack" factor for received datasets with refquota set on them. 3956 * See the bottom of this function for details on its use. 3957 */ 3958 uint64_t refquota_slack = (uint64_t)DMU_MAX_ACCESS * 3959 spa_asize_inflation; 3960 int64_t unused_refres_delta; 3961 3962 /* they should both be heads */ 3963 if (clone->ds_is_snapshot || 3964 origin_head->ds_is_snapshot) 3965 return (SET_ERROR(EINVAL)); 3966 3967 /* if we are not forcing, the branch point should be just before them */ 3968 if (!force && clone->ds_prev != origin_head->ds_prev) 3969 return (SET_ERROR(EINVAL)); 3970 3971 /* clone should be the clone (unless they are unrelated) */ 3972 if (clone->ds_prev != NULL && 3973 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 3974 origin_head->ds_dir != clone->ds_prev->ds_dir) 3975 return (SET_ERROR(EINVAL)); 3976 3977 /* the clone should be a child of the origin */ 3978 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 3979 return (SET_ERROR(EINVAL)); 3980 3981 /* origin_head shouldn't be modified unless 'force' */ 3982 if (!force && 3983 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 3984 return (SET_ERROR(ETXTBSY)); 3985 3986 /* origin_head should have no long holds (e.g. is not mounted) */ 3987 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 3988 return (SET_ERROR(EBUSY)); 3989 3990 /* check amount of any unconsumed refreservation */ 3991 unused_refres_delta = 3992 (int64_t)MIN(origin_head->ds_reserved, 3993 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 3994 (int64_t)MIN(origin_head->ds_reserved, 3995 dsl_dataset_phys(clone)->ds_unique_bytes); 3996 3997 if (unused_refres_delta > 0 && 3998 unused_refres_delta > 3999 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 4000 return (SET_ERROR(ENOSPC)); 4001 4002 /* 4003 * The clone can't be too much over the head's refquota. 4004 * 4005 * To ensure that the entire refquota can be used, we allow one 4006 * transaction to exceed the refquota. Therefore, this check 4007 * needs to also allow for the space referenced to be more than the 4008 * refquota. The maximum amount of space that one transaction can use 4009 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this 4010 * overage ensures that we are able to receive a filesystem that 4011 * exceeds the refquota on the source system. 4012 * 4013 * So that overage is the refquota_slack we use below. 4014 */ 4015 if (origin_head->ds_quota != 0 && 4016 dsl_dataset_phys(clone)->ds_referenced_bytes > 4017 origin_head->ds_quota + refquota_slack) 4018 return (SET_ERROR(EDQUOT)); 4019 4020 return (0); 4021 } 4022 4023 static void 4024 dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone, 4025 dsl_dataset_t *origin, dmu_tx_t *tx) 4026 { 4027 uint64_t clone_remap_dl_obj, origin_remap_dl_obj; 4028 dsl_pool_t *dp = dmu_tx_pool(tx); 4029 4030 ASSERT(dsl_pool_sync_context(dp)); 4031 4032 clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone); 4033 origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin); 4034 4035 if (clone_remap_dl_obj != 0) { 4036 dsl_deadlist_close(&clone->ds_remap_deadlist); 4037 dsl_dataset_unset_remap_deadlist_object(clone, tx); 4038 } 4039 if (origin_remap_dl_obj != 0) { 4040 dsl_deadlist_close(&origin->ds_remap_deadlist); 4041 dsl_dataset_unset_remap_deadlist_object(origin, tx); 4042 } 4043 4044 if (clone_remap_dl_obj != 0) { 4045 dsl_dataset_set_remap_deadlist_object(origin, 4046 clone_remap_dl_obj, tx); 4047 dsl_deadlist_open(&origin->ds_remap_deadlist, 4048 dp->dp_meta_objset, clone_remap_dl_obj); 4049 } 4050 if (origin_remap_dl_obj != 0) { 4051 dsl_dataset_set_remap_deadlist_object(clone, 4052 origin_remap_dl_obj, tx); 4053 dsl_deadlist_open(&clone->ds_remap_deadlist, 4054 dp->dp_meta_objset, origin_remap_dl_obj); 4055 } 4056 } 4057 4058 void 4059 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 4060 dsl_dataset_t *origin_head, dmu_tx_t *tx) 4061 { 4062 dsl_pool_t *dp = dmu_tx_pool(tx); 4063 int64_t unused_refres_delta; 4064 4065 ASSERT(clone->ds_reserved == 0); 4066 /* 4067 * NOTE: On DEBUG kernels there could be a race between this and 4068 * the check function if spa_asize_inflation is adjusted... 4069 */ 4070 ASSERT(origin_head->ds_quota == 0 || 4071 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota + 4072 DMU_MAX_ACCESS * spa_asize_inflation); 4073 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 4074 4075 dsl_dir_cancel_waiters(origin_head->ds_dir); 4076 4077 /* 4078 * Swap per-dataset feature flags. 4079 */ 4080 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 4081 if (!(spa_feature_table[f].fi_flags & 4082 ZFEATURE_FLAG_PER_DATASET)) { 4083 ASSERT(!dsl_dataset_feature_is_active(clone, f)); 4084 ASSERT(!dsl_dataset_feature_is_active(origin_head, f)); 4085 continue; 4086 } 4087 4088 boolean_t clone_inuse = dsl_dataset_feature_is_active(clone, f); 4089 void *clone_feature = clone->ds_feature[f]; 4090 boolean_t origin_head_inuse = 4091 dsl_dataset_feature_is_active(origin_head, f); 4092 void *origin_head_feature = origin_head->ds_feature[f]; 4093 4094 if (clone_inuse) 4095 dsl_dataset_deactivate_feature_impl(clone, f, tx); 4096 if (origin_head_inuse) 4097 dsl_dataset_deactivate_feature_impl(origin_head, f, tx); 4098 4099 if (clone_inuse) { 4100 dsl_dataset_activate_feature(origin_head->ds_object, f, 4101 clone_feature, tx); 4102 origin_head->ds_feature[f] = clone_feature; 4103 } 4104 if (origin_head_inuse) { 4105 dsl_dataset_activate_feature(clone->ds_object, f, 4106 origin_head_feature, tx); 4107 clone->ds_feature[f] = origin_head_feature; 4108 } 4109 } 4110 4111 dmu_buf_will_dirty(clone->ds_dbuf, tx); 4112 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 4113 4114 if (clone->ds_objset != NULL) { 4115 dmu_objset_evict(clone->ds_objset); 4116 clone->ds_objset = NULL; 4117 } 4118 4119 if (origin_head->ds_objset != NULL) { 4120 dmu_objset_evict(origin_head->ds_objset); 4121 origin_head->ds_objset = NULL; 4122 } 4123 4124 unused_refres_delta = 4125 (int64_t)MIN(origin_head->ds_reserved, 4126 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 4127 (int64_t)MIN(origin_head->ds_reserved, 4128 dsl_dataset_phys(clone)->ds_unique_bytes); 4129 4130 /* 4131 * Reset origin's unique bytes. 4132 */ 4133 { 4134 dsl_dataset_t *origin = clone->ds_prev; 4135 uint64_t comp, uncomp; 4136 4137 dmu_buf_will_dirty(origin->ds_dbuf, tx); 4138 dsl_deadlist_space_range(&clone->ds_deadlist, 4139 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, 4140 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); 4141 } 4142 4143 /* swap blkptrs */ 4144 { 4145 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG); 4146 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG); 4147 blkptr_t tmp; 4148 tmp = dsl_dataset_phys(origin_head)->ds_bp; 4149 dsl_dataset_phys(origin_head)->ds_bp = 4150 dsl_dataset_phys(clone)->ds_bp; 4151 dsl_dataset_phys(clone)->ds_bp = tmp; 4152 rrw_exit(&origin_head->ds_bp_rwlock, FTAG); 4153 rrw_exit(&clone->ds_bp_rwlock, FTAG); 4154 } 4155 4156 /* set dd_*_bytes */ 4157 { 4158 int64_t dused, dcomp, duncomp; 4159 uint64_t cdl_used, cdl_comp, cdl_uncomp; 4160 uint64_t odl_used, odl_comp, odl_uncomp; 4161 4162 ASSERT3U(dsl_dir_phys(clone->ds_dir)-> 4163 dd_used_breakdown[DD_USED_SNAP], ==, 0); 4164 4165 dsl_deadlist_space(&clone->ds_deadlist, 4166 &cdl_used, &cdl_comp, &cdl_uncomp); 4167 dsl_deadlist_space(&origin_head->ds_deadlist, 4168 &odl_used, &odl_comp, &odl_uncomp); 4169 4170 dused = dsl_dataset_phys(clone)->ds_referenced_bytes + 4171 cdl_used - 4172 (dsl_dataset_phys(origin_head)->ds_referenced_bytes + 4173 odl_used); 4174 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + 4175 cdl_comp - 4176 (dsl_dataset_phys(origin_head)->ds_compressed_bytes + 4177 odl_comp); 4178 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + 4179 cdl_uncomp - 4180 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + 4181 odl_uncomp); 4182 4183 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 4184 dused, dcomp, duncomp, tx); 4185 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 4186 -dused, -dcomp, -duncomp, tx); 4187 4188 /* 4189 * The difference in the space used by snapshots is the 4190 * difference in snapshot space due to the head's 4191 * deadlist (since that's the only thing that's 4192 * changing that affects the snapused). 4193 */ 4194 dsl_deadlist_space_range(&clone->ds_deadlist, 4195 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 4196 &cdl_used, &cdl_comp, &cdl_uncomp); 4197 dsl_deadlist_space_range(&origin_head->ds_deadlist, 4198 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 4199 &odl_used, &odl_comp, &odl_uncomp); 4200 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 4201 DD_USED_HEAD, DD_USED_SNAP, tx); 4202 } 4203 4204 /* swap ds_*_bytes */ 4205 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, 4206 dsl_dataset_phys(clone)->ds_referenced_bytes); 4207 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, 4208 dsl_dataset_phys(clone)->ds_compressed_bytes); 4209 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, 4210 dsl_dataset_phys(clone)->ds_uncompressed_bytes); 4211 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, 4212 dsl_dataset_phys(clone)->ds_unique_bytes); 4213 4214 /* apply any parent delta for change in unconsumed refreservation */ 4215 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 4216 unused_refres_delta, 0, 0, tx); 4217 4218 /* 4219 * Swap deadlists. 4220 */ 4221 dsl_deadlist_close(&clone->ds_deadlist); 4222 dsl_deadlist_close(&origin_head->ds_deadlist); 4223 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, 4224 dsl_dataset_phys(clone)->ds_deadlist_obj); 4225 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 4226 dsl_dataset_phys(clone)->ds_deadlist_obj); 4227 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 4228 dsl_dataset_phys(origin_head)->ds_deadlist_obj); 4229 dsl_dataset_swap_remap_deadlists(clone, origin_head, tx); 4230 4231 /* 4232 * If there is a bookmark at the origin, its "next dataset" is 4233 * changing, so we need to reset its FBN. 4234 */ 4235 dsl_bookmark_next_changed(origin_head, origin_head->ds_prev, tx); 4236 4237 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 4238 4239 /* 4240 * Destroy any livelists associated with the clone or the origin, 4241 * since after the swap the corresponding livelists are no longer 4242 * valid. 4243 */ 4244 dsl_dir_remove_livelist(clone->ds_dir, tx, B_TRUE); 4245 dsl_dir_remove_livelist(origin_head->ds_dir, tx, B_TRUE); 4246 4247 spa_history_log_internal_ds(clone, "clone swap", tx, 4248 "parent=%s", origin_head->ds_dir->dd_myname); 4249 } 4250 4251 /* 4252 * Given a pool name and a dataset object number in that pool, 4253 * return the name of that dataset. 4254 */ 4255 int 4256 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 4257 { 4258 dsl_pool_t *dp; 4259 dsl_dataset_t *ds; 4260 int error; 4261 4262 error = dsl_pool_hold(pname, FTAG, &dp); 4263 if (error != 0) 4264 return (error); 4265 4266 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 4267 if (error == 0) { 4268 dsl_dataset_name(ds, buf); 4269 dsl_dataset_rele(ds, FTAG); 4270 } 4271 dsl_pool_rele(dp, FTAG); 4272 4273 return (error); 4274 } 4275 4276 int 4277 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 4278 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 4279 { 4280 int error = 0; 4281 4282 ASSERT3S(asize, >, 0); 4283 4284 /* 4285 * *ref_rsrv is the portion of asize that will come from any 4286 * unconsumed refreservation space. 4287 */ 4288 *ref_rsrv = 0; 4289 4290 mutex_enter(&ds->ds_lock); 4291 /* 4292 * Make a space adjustment for reserved bytes. 4293 */ 4294 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 4295 ASSERT3U(*used, >=, 4296 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 4297 *used -= 4298 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 4299 *ref_rsrv = 4300 asize - MIN(asize, parent_delta(ds, asize + inflight)); 4301 } 4302 4303 if (!check_quota || ds->ds_quota == 0) { 4304 mutex_exit(&ds->ds_lock); 4305 return (0); 4306 } 4307 /* 4308 * If they are requesting more space, and our current estimate 4309 * is over quota, they get to try again unless the actual 4310 * on-disk is over quota and there are no pending changes (which 4311 * may free up space for us). 4312 */ 4313 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= 4314 ds->ds_quota) { 4315 if (inflight > 0 || 4316 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) 4317 error = SET_ERROR(ERESTART); 4318 else 4319 error = SET_ERROR(EDQUOT); 4320 } 4321 mutex_exit(&ds->ds_lock); 4322 4323 return (error); 4324 } 4325 4326 typedef struct dsl_dataset_set_qr_arg { 4327 const char *ddsqra_name; 4328 zprop_source_t ddsqra_source; 4329 uint64_t ddsqra_value; 4330 } dsl_dataset_set_qr_arg_t; 4331 4332 4333 static int 4334 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 4335 { 4336 dsl_dataset_set_qr_arg_t *ddsqra = arg; 4337 dsl_pool_t *dp = dmu_tx_pool(tx); 4338 dsl_dataset_t *ds; 4339 int error; 4340 uint64_t newval; 4341 4342 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 4343 return (SET_ERROR(ENOTSUP)); 4344 4345 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 4346 if (error != 0) 4347 return (error); 4348 4349 if (ds->ds_is_snapshot) { 4350 dsl_dataset_rele(ds, FTAG); 4351 return (SET_ERROR(EINVAL)); 4352 } 4353 4354 error = dsl_prop_predict(ds->ds_dir, 4355 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 4356 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 4357 if (error != 0) { 4358 dsl_dataset_rele(ds, FTAG); 4359 return (error); 4360 } 4361 4362 if (newval == 0) { 4363 dsl_dataset_rele(ds, FTAG); 4364 return (0); 4365 } 4366 4367 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || 4368 newval < ds->ds_reserved) { 4369 dsl_dataset_rele(ds, FTAG); 4370 return (SET_ERROR(ENOSPC)); 4371 } 4372 4373 dsl_dataset_rele(ds, FTAG); 4374 return (0); 4375 } 4376 4377 static void 4378 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 4379 { 4380 dsl_dataset_set_qr_arg_t *ddsqra = arg; 4381 dsl_pool_t *dp = dmu_tx_pool(tx); 4382 dsl_dataset_t *ds = NULL; 4383 uint64_t newval; 4384 4385 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 4386 4387 dsl_prop_set_sync_impl(ds, 4388 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 4389 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 4390 &ddsqra->ddsqra_value, tx); 4391 4392 VERIFY0(dsl_prop_get_int_ds(ds, 4393 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 4394 4395 if (ds->ds_quota != newval) { 4396 dmu_buf_will_dirty(ds->ds_dbuf, tx); 4397 ds->ds_quota = newval; 4398 } 4399 dsl_dataset_rele(ds, FTAG); 4400 } 4401 4402 int 4403 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 4404 uint64_t refquota) 4405 { 4406 dsl_dataset_set_qr_arg_t ddsqra; 4407 4408 ddsqra.ddsqra_name = dsname; 4409 ddsqra.ddsqra_source = source; 4410 ddsqra.ddsqra_value = refquota; 4411 4412 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 4413 dsl_dataset_set_refquota_sync, &ddsqra, 0, 4414 ZFS_SPACE_CHECK_EXTRA_RESERVED)); 4415 } 4416 4417 static int 4418 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 4419 { 4420 dsl_dataset_set_qr_arg_t *ddsqra = arg; 4421 dsl_pool_t *dp = dmu_tx_pool(tx); 4422 dsl_dataset_t *ds; 4423 int error; 4424 uint64_t newval, unique; 4425 4426 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 4427 return (SET_ERROR(ENOTSUP)); 4428 4429 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 4430 if (error != 0) 4431 return (error); 4432 4433 if (ds->ds_is_snapshot) { 4434 dsl_dataset_rele(ds, FTAG); 4435 return (SET_ERROR(EINVAL)); 4436 } 4437 4438 error = dsl_prop_predict(ds->ds_dir, 4439 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 4440 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 4441 if (error != 0) { 4442 dsl_dataset_rele(ds, FTAG); 4443 return (error); 4444 } 4445 4446 /* 4447 * If we are doing the preliminary check in open context, the 4448 * space estimates may be inaccurate. 4449 */ 4450 if (!dmu_tx_is_syncing(tx)) { 4451 dsl_dataset_rele(ds, FTAG); 4452 return (0); 4453 } 4454 4455 mutex_enter(&ds->ds_lock); 4456 if (!DS_UNIQUE_IS_ACCURATE(ds)) 4457 dsl_dataset_recalc_head_uniq(ds); 4458 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 4459 mutex_exit(&ds->ds_lock); 4460 4461 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 4462 uint64_t delta = MAX(unique, newval) - 4463 MAX(unique, ds->ds_reserved); 4464 4465 if (delta > 4466 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 4467 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 4468 dsl_dataset_rele(ds, FTAG); 4469 return (SET_ERROR(ENOSPC)); 4470 } 4471 } 4472 4473 dsl_dataset_rele(ds, FTAG); 4474 return (0); 4475 } 4476 4477 void 4478 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 4479 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 4480 { 4481 uint64_t newval; 4482 uint64_t unique; 4483 int64_t delta; 4484 4485 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 4486 source, sizeof (value), 1, &value, tx); 4487 4488 VERIFY0(dsl_prop_get_int_ds(ds, 4489 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 4490 4491 dmu_buf_will_dirty(ds->ds_dbuf, tx); 4492 mutex_enter(&ds->ds_dir->dd_lock); 4493 mutex_enter(&ds->ds_lock); 4494 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 4495 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 4496 delta = MAX(0, (int64_t)(newval - unique)) - 4497 MAX(0, (int64_t)(ds->ds_reserved - unique)); 4498 ds->ds_reserved = newval; 4499 mutex_exit(&ds->ds_lock); 4500 4501 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 4502 mutex_exit(&ds->ds_dir->dd_lock); 4503 } 4504 4505 static void 4506 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 4507 { 4508 dsl_dataset_set_qr_arg_t *ddsqra = arg; 4509 dsl_pool_t *dp = dmu_tx_pool(tx); 4510 dsl_dataset_t *ds = NULL; 4511 4512 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 4513 dsl_dataset_set_refreservation_sync_impl(ds, 4514 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 4515 dsl_dataset_rele(ds, FTAG); 4516 } 4517 4518 int 4519 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 4520 uint64_t refreservation) 4521 { 4522 dsl_dataset_set_qr_arg_t ddsqra; 4523 4524 ddsqra.ddsqra_name = dsname; 4525 ddsqra.ddsqra_source = source; 4526 ddsqra.ddsqra_value = refreservation; 4527 4528 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 4529 dsl_dataset_set_refreservation_sync, &ddsqra, 0, 4530 ZFS_SPACE_CHECK_EXTRA_RESERVED)); 4531 } 4532 4533 typedef struct dsl_dataset_set_compression_arg { 4534 const char *ddsca_name; 4535 zprop_source_t ddsca_source; 4536 uint64_t ddsca_value; 4537 } dsl_dataset_set_compression_arg_t; 4538 4539 static int 4540 dsl_dataset_set_compression_check(void *arg, dmu_tx_t *tx) 4541 { 4542 dsl_dataset_set_compression_arg_t *ddsca = arg; 4543 dsl_pool_t *dp = dmu_tx_pool(tx); 4544 4545 uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value); 4546 spa_feature_t f = zio_compress_to_feature(compval); 4547 4548 if (f == SPA_FEATURE_NONE) 4549 return (SET_ERROR(EINVAL)); 4550 4551 if (!spa_feature_is_enabled(dp->dp_spa, f)) 4552 return (SET_ERROR(ENOTSUP)); 4553 4554 return (0); 4555 } 4556 4557 static void 4558 dsl_dataset_set_compression_sync(void *arg, dmu_tx_t *tx) 4559 { 4560 dsl_dataset_set_compression_arg_t *ddsca = arg; 4561 dsl_pool_t *dp = dmu_tx_pool(tx); 4562 dsl_dataset_t *ds = NULL; 4563 4564 uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value); 4565 spa_feature_t f = zio_compress_to_feature(compval); 4566 ASSERT3S(f, !=, SPA_FEATURE_NONE); 4567 ASSERT3S(spa_feature_table[f].fi_type, ==, ZFEATURE_TYPE_BOOLEAN); 4568 4569 VERIFY0(dsl_dataset_hold(dp, ddsca->ddsca_name, FTAG, &ds)); 4570 if (zfeature_active(f, ds->ds_feature[f]) != B_TRUE) { 4571 ds->ds_feature_activation[f] = (void *)B_TRUE; 4572 dsl_dataset_activate_feature(ds->ds_object, f, 4573 ds->ds_feature_activation[f], tx); 4574 ds->ds_feature[f] = ds->ds_feature_activation[f]; 4575 } 4576 dsl_dataset_rele(ds, FTAG); 4577 } 4578 4579 int 4580 dsl_dataset_set_compression(const char *dsname, zprop_source_t source, 4581 uint64_t compression) 4582 { 4583 dsl_dataset_set_compression_arg_t ddsca; 4584 4585 /* 4586 * The sync task is only required for zstd in order to activate 4587 * the feature flag when the property is first set. 4588 */ 4589 if (ZIO_COMPRESS_ALGO(compression) != ZIO_COMPRESS_ZSTD) 4590 return (0); 4591 4592 ddsca.ddsca_name = dsname; 4593 ddsca.ddsca_source = source; 4594 ddsca.ddsca_value = compression; 4595 4596 return (dsl_sync_task(dsname, dsl_dataset_set_compression_check, 4597 dsl_dataset_set_compression_sync, &ddsca, 0, 4598 ZFS_SPACE_CHECK_EXTRA_RESERVED)); 4599 } 4600 4601 /* 4602 * Return (in *usedp) the amount of space referenced by "new" that was not 4603 * referenced at the time the bookmark corresponds to. "New" may be a 4604 * snapshot or a head. The bookmark must be before new, in 4605 * new's filesystem (or its origin) -- caller verifies this. 4606 * 4607 * The written space is calculated by considering two components: First, we 4608 * ignore any freed space, and calculate the written as new's used space 4609 * minus old's used space. Next, we add in the amount of space that was freed 4610 * between the two time points, thus reducing new's used space relative to 4611 * old's. Specifically, this is the space that was born before 4612 * zbm_creation_txg, and freed before new (ie. on new's deadlist or a 4613 * previous deadlist). 4614 * 4615 * space freed [---------------------] 4616 * snapshots ---O-------O--------O-------O------ 4617 * bookmark new 4618 * 4619 * Note, the bookmark's zbm_*_bytes_refd must be valid, but if the HAS_FBN 4620 * flag is not set, we will calculate the freed_before_next based on the 4621 * next snapshot's deadlist, rather than using zbm_*_freed_before_next_snap. 4622 */ 4623 static int 4624 dsl_dataset_space_written_impl(zfs_bookmark_phys_t *bmp, 4625 dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4626 { 4627 int err = 0; 4628 dsl_pool_t *dp = new->ds_dir->dd_pool; 4629 4630 ASSERT(dsl_pool_config_held(dp)); 4631 if (dsl_dataset_is_snapshot(new)) { 4632 ASSERT3U(bmp->zbm_creation_txg, <, 4633 dsl_dataset_phys(new)->ds_creation_txg); 4634 } 4635 4636 *usedp = 0; 4637 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; 4638 *usedp -= bmp->zbm_referenced_bytes_refd; 4639 4640 *compp = 0; 4641 *compp += dsl_dataset_phys(new)->ds_compressed_bytes; 4642 *compp -= bmp->zbm_compressed_bytes_refd; 4643 4644 *uncompp = 0; 4645 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; 4646 *uncompp -= bmp->zbm_uncompressed_bytes_refd; 4647 4648 dsl_dataset_t *snap = new; 4649 4650 while (dsl_dataset_phys(snap)->ds_prev_snap_txg > 4651 bmp->zbm_creation_txg) { 4652 uint64_t used, comp, uncomp; 4653 4654 dsl_deadlist_space_range(&snap->ds_deadlist, 4655 0, bmp->zbm_creation_txg, 4656 &used, &comp, &uncomp); 4657 *usedp += used; 4658 *compp += comp; 4659 *uncompp += uncomp; 4660 4661 uint64_t snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 4662 if (snap != new) 4663 dsl_dataset_rele(snap, FTAG); 4664 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 4665 if (err != 0) 4666 break; 4667 } 4668 4669 /* 4670 * We might not have the FBN if we are calculating written from 4671 * a snapshot (because we didn't know the correct "next" snapshot 4672 * until now). 4673 */ 4674 if (bmp->zbm_flags & ZBM_FLAG_HAS_FBN) { 4675 *usedp += bmp->zbm_referenced_freed_before_next_snap; 4676 *compp += bmp->zbm_compressed_freed_before_next_snap; 4677 *uncompp += bmp->zbm_uncompressed_freed_before_next_snap; 4678 } else { 4679 ASSERT3U(dsl_dataset_phys(snap)->ds_prev_snap_txg, ==, 4680 bmp->zbm_creation_txg); 4681 uint64_t used, comp, uncomp; 4682 dsl_deadlist_space(&snap->ds_deadlist, &used, &comp, &uncomp); 4683 *usedp += used; 4684 *compp += comp; 4685 *uncompp += uncomp; 4686 } 4687 if (snap != new) 4688 dsl_dataset_rele(snap, FTAG); 4689 return (err); 4690 } 4691 4692 /* 4693 * Return (in *usedp) the amount of space written in new that was not 4694 * present at the time the bookmark corresponds to. New may be a 4695 * snapshot or the head. Old must be a bookmark before new, in 4696 * new's filesystem (or its origin) -- caller verifies this. 4697 */ 4698 int 4699 dsl_dataset_space_written_bookmark(zfs_bookmark_phys_t *bmp, 4700 dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4701 { 4702 if (!(bmp->zbm_flags & ZBM_FLAG_HAS_FBN)) 4703 return (SET_ERROR(ENOTSUP)); 4704 return (dsl_dataset_space_written_impl(bmp, new, 4705 usedp, compp, uncompp)); 4706 } 4707 4708 /* 4709 * Return (in *usedp) the amount of space written in new that is not 4710 * present in oldsnap. New may be a snapshot or the head. Old must be 4711 * a snapshot before new, in new's filesystem (or its origin). If not then 4712 * fail and return EINVAL. 4713 */ 4714 int 4715 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 4716 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4717 { 4718 if (!dsl_dataset_is_before(new, oldsnap, 0)) 4719 return (SET_ERROR(EINVAL)); 4720 4721 zfs_bookmark_phys_t zbm = { 0 }; 4722 dsl_dataset_phys_t *dsp = dsl_dataset_phys(oldsnap); 4723 zbm.zbm_guid = dsp->ds_guid; 4724 zbm.zbm_creation_txg = dsp->ds_creation_txg; 4725 zbm.zbm_creation_time = dsp->ds_creation_time; 4726 zbm.zbm_referenced_bytes_refd = dsp->ds_referenced_bytes; 4727 zbm.zbm_compressed_bytes_refd = dsp->ds_compressed_bytes; 4728 zbm.zbm_uncompressed_bytes_refd = dsp->ds_uncompressed_bytes; 4729 4730 /* 4731 * If oldsnap is the origin (or origin's origin, ...) of new, 4732 * we can't easily calculate the effective FBN. Therefore, 4733 * we do not set ZBM_FLAG_HAS_FBN, so that the _impl will calculate 4734 * it relative to the correct "next": the next snapshot towards "new", 4735 * rather than the next snapshot in oldsnap's dsl_dir. 4736 */ 4737 return (dsl_dataset_space_written_impl(&zbm, new, 4738 usedp, compp, uncompp)); 4739 } 4740 4741 /* 4742 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 4743 * lastsnap, and all snapshots in between are deleted. 4744 * 4745 * blocks that would be freed [---------------------------] 4746 * snapshots ---O-------O--------O-------O--------O 4747 * firstsnap lastsnap 4748 * 4749 * This is the set of blocks that were born after the snap before firstsnap, 4750 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 4751 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 4752 * We calculate this by iterating over the relevant deadlists (from the snap 4753 * after lastsnap, backward to the snap after firstsnap), summing up the 4754 * space on the deadlist that was born after the snap before firstsnap. 4755 */ 4756 int 4757 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 4758 dsl_dataset_t *lastsnap, 4759 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4760 { 4761 int err = 0; 4762 uint64_t snapobj; 4763 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 4764 4765 ASSERT(firstsnap->ds_is_snapshot); 4766 ASSERT(lastsnap->ds_is_snapshot); 4767 4768 /* 4769 * Check that the snapshots are in the same dsl_dir, and firstsnap 4770 * is before lastsnap. 4771 */ 4772 if (firstsnap->ds_dir != lastsnap->ds_dir || 4773 dsl_dataset_phys(firstsnap)->ds_creation_txg > 4774 dsl_dataset_phys(lastsnap)->ds_creation_txg) 4775 return (SET_ERROR(EINVAL)); 4776 4777 *usedp = *compp = *uncompp = 0; 4778 4779 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; 4780 while (snapobj != firstsnap->ds_object) { 4781 dsl_dataset_t *ds; 4782 uint64_t used, comp, uncomp; 4783 4784 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 4785 if (err != 0) 4786 break; 4787 4788 dsl_deadlist_space_range(&ds->ds_deadlist, 4789 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, 4790 &used, &comp, &uncomp); 4791 *usedp += used; 4792 *compp += comp; 4793 *uncompp += uncomp; 4794 4795 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 4796 ASSERT3U(snapobj, !=, 0); 4797 dsl_dataset_rele(ds, FTAG); 4798 } 4799 return (err); 4800 } 4801 4802 /* 4803 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 4804 * For example, they could both be snapshots of the same filesystem, and 4805 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 4806 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 4807 * filesystem. Or 'earlier' could be the origin's origin. 4808 * 4809 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 4810 */ 4811 boolean_t 4812 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 4813 uint64_t earlier_txg) 4814 { 4815 dsl_pool_t *dp = later->ds_dir->dd_pool; 4816 int error; 4817 boolean_t ret; 4818 4819 ASSERT(dsl_pool_config_held(dp)); 4820 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); 4821 4822 if (earlier_txg == 0) 4823 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; 4824 4825 if (later->ds_is_snapshot && 4826 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) 4827 return (B_FALSE); 4828 4829 if (later->ds_dir == earlier->ds_dir) 4830 return (B_TRUE); 4831 4832 /* 4833 * We check dd_origin_obj explicitly here rather than using 4834 * dsl_dir_is_clone() so that we will return TRUE if "earlier" 4835 * is $ORIGIN@$ORIGIN. dsl_dataset_space_written() depends on 4836 * this behavior. 4837 */ 4838 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == 0) 4839 return (B_FALSE); 4840 4841 dsl_dataset_t *origin; 4842 error = dsl_dataset_hold_obj(dp, 4843 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); 4844 if (error != 0) 4845 return (B_FALSE); 4846 if (dsl_dataset_phys(origin)->ds_creation_txg == earlier_txg && 4847 origin->ds_dir == earlier->ds_dir) { 4848 dsl_dataset_rele(origin, FTAG); 4849 return (B_TRUE); 4850 } 4851 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 4852 dsl_dataset_rele(origin, FTAG); 4853 return (ret); 4854 } 4855 4856 void 4857 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 4858 { 4859 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 4860 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 4861 } 4862 4863 boolean_t 4864 dsl_dataset_is_zapified(dsl_dataset_t *ds) 4865 { 4866 dmu_object_info_t doi; 4867 4868 dmu_object_info_from_db(ds->ds_dbuf, &doi); 4869 return (doi.doi_type == DMU_OTN_ZAP_METADATA); 4870 } 4871 4872 boolean_t 4873 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds) 4874 { 4875 return (dsl_dataset_is_zapified(ds) && 4876 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset, 4877 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0); 4878 } 4879 4880 uint64_t 4881 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds) 4882 { 4883 uint64_t remap_deadlist_obj; 4884 int err; 4885 4886 if (!dsl_dataset_is_zapified(ds)) 4887 return (0); 4888 4889 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, 4890 DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1, 4891 &remap_deadlist_obj); 4892 4893 if (err != 0) { 4894 VERIFY3S(err, ==, ENOENT); 4895 return (0); 4896 } 4897 4898 ASSERT(remap_deadlist_obj != 0); 4899 return (remap_deadlist_obj); 4900 } 4901 4902 boolean_t 4903 dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds) 4904 { 4905 EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist), 4906 dsl_dataset_get_remap_deadlist_object(ds) != 0); 4907 return (dsl_deadlist_is_open(&ds->ds_remap_deadlist)); 4908 } 4909 4910 static void 4911 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj, 4912 dmu_tx_t *tx) 4913 { 4914 ASSERT(obj != 0); 4915 dsl_dataset_zapify(ds, tx); 4916 VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, 4917 DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx)); 4918 } 4919 4920 static void 4921 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx) 4922 { 4923 VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset, 4924 ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx)); 4925 } 4926 4927 void 4928 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx) 4929 { 4930 uint64_t remap_deadlist_object; 4931 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 4932 4933 ASSERT(dmu_tx_is_syncing(tx)); 4934 ASSERT(dsl_dataset_remap_deadlist_exists(ds)); 4935 4936 remap_deadlist_object = ds->ds_remap_deadlist.dl_object; 4937 dsl_deadlist_close(&ds->ds_remap_deadlist); 4938 dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx); 4939 dsl_dataset_unset_remap_deadlist_object(ds, tx); 4940 spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); 4941 } 4942 4943 void 4944 dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx) 4945 { 4946 uint64_t remap_deadlist_obj; 4947 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 4948 4949 ASSERT(dmu_tx_is_syncing(tx)); 4950 ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock)); 4951 /* 4952 * Currently we only create remap deadlists when there are indirect 4953 * vdevs with referenced mappings. 4954 */ 4955 ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL)); 4956 4957 remap_deadlist_obj = dsl_deadlist_clone( 4958 &ds->ds_deadlist, UINT64_MAX, 4959 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 4960 dsl_dataset_set_remap_deadlist_object(ds, 4961 remap_deadlist_obj, tx); 4962 dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa), 4963 remap_deadlist_obj); 4964 spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); 4965 } 4966 4967 void 4968 dsl_dataset_activate_redaction(dsl_dataset_t *ds, uint64_t *redact_snaps, 4969 uint64_t num_redact_snaps, dmu_tx_t *tx) 4970 { 4971 uint64_t dsobj = ds->ds_object; 4972 struct feature_type_uint64_array_arg *ftuaa = 4973 kmem_zalloc(sizeof (*ftuaa), KM_SLEEP); 4974 ftuaa->length = (int64_t)num_redact_snaps; 4975 if (num_redact_snaps > 0) { 4976 ftuaa->array = kmem_alloc(num_redact_snaps * sizeof (uint64_t), 4977 KM_SLEEP); 4978 memcpy(ftuaa->array, redact_snaps, num_redact_snaps * 4979 sizeof (uint64_t)); 4980 } 4981 dsl_dataset_activate_feature(dsobj, SPA_FEATURE_REDACTED_DATASETS, 4982 ftuaa, tx); 4983 ds->ds_feature[SPA_FEATURE_REDACTED_DATASETS] = ftuaa; 4984 } 4985 4986 /* 4987 * Find and return (in *oldest_dsobj) the oldest snapshot of the dsobj 4988 * dataset whose birth time is >= min_txg. 4989 */ 4990 int 4991 dsl_dataset_oldest_snapshot(spa_t *spa, uint64_t head_ds, uint64_t min_txg, 4992 uint64_t *oldest_dsobj) 4993 { 4994 dsl_dataset_t *ds; 4995 dsl_pool_t *dp = spa->spa_dsl_pool; 4996 4997 int error = dsl_dataset_hold_obj(dp, head_ds, FTAG, &ds); 4998 if (error != 0) 4999 return (error); 5000 5001 uint64_t prev_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 5002 uint64_t prev_obj_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 5003 5004 while (prev_obj != 0 && min_txg < prev_obj_txg) { 5005 dsl_dataset_rele(ds, FTAG); 5006 if ((error = dsl_dataset_hold_obj(dp, prev_obj, 5007 FTAG, &ds)) != 0) 5008 return (error); 5009 prev_obj_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 5010 prev_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 5011 } 5012 *oldest_dsobj = ds->ds_object; 5013 dsl_dataset_rele(ds, FTAG); 5014 return (0); 5015 } 5016 5017 ZFS_MODULE_PARAM(zfs, zfs_, max_recordsize, UINT, ZMOD_RW, 5018 "Max allowed record size"); 5019 5020 ZFS_MODULE_PARAM(zfs, zfs_, allow_redacted_dataset_mount, INT, ZMOD_RW, 5021 "Allow mounting of redacted datasets"); 5022 5023 ZFS_MODULE_PARAM(zfs, zfs_, snapshot_history_enabled, INT, ZMOD_RW, 5024 "Include snapshot events in pool history/events"); 5025 5026 EXPORT_SYMBOL(dsl_dataset_hold); 5027 EXPORT_SYMBOL(dsl_dataset_hold_flags); 5028 EXPORT_SYMBOL(dsl_dataset_hold_obj); 5029 EXPORT_SYMBOL(dsl_dataset_hold_obj_flags); 5030 EXPORT_SYMBOL(dsl_dataset_own); 5031 EXPORT_SYMBOL(dsl_dataset_own_obj); 5032 EXPORT_SYMBOL(dsl_dataset_name); 5033 EXPORT_SYMBOL(dsl_dataset_rele); 5034 EXPORT_SYMBOL(dsl_dataset_rele_flags); 5035 EXPORT_SYMBOL(dsl_dataset_disown); 5036 EXPORT_SYMBOL(dsl_dataset_tryown); 5037 EXPORT_SYMBOL(dsl_dataset_create_sync); 5038 EXPORT_SYMBOL(dsl_dataset_create_sync_dd); 5039 EXPORT_SYMBOL(dsl_dataset_snapshot_check); 5040 EXPORT_SYMBOL(dsl_dataset_snapshot_sync); 5041 EXPORT_SYMBOL(dsl_dataset_promote); 5042 EXPORT_SYMBOL(dsl_dataset_user_hold); 5043 EXPORT_SYMBOL(dsl_dataset_user_release); 5044 EXPORT_SYMBOL(dsl_dataset_get_holds); 5045 EXPORT_SYMBOL(dsl_dataset_get_blkptr); 5046 EXPORT_SYMBOL(dsl_dataset_get_spa); 5047 EXPORT_SYMBOL(dsl_dataset_modified_since_snap); 5048 EXPORT_SYMBOL(dsl_dataset_space_written); 5049 EXPORT_SYMBOL(dsl_dataset_space_wouldfree); 5050 EXPORT_SYMBOL(dsl_dataset_sync); 5051 EXPORT_SYMBOL(dsl_dataset_block_born); 5052 EXPORT_SYMBOL(dsl_dataset_block_kill); 5053 EXPORT_SYMBOL(dsl_dataset_dirty); 5054 EXPORT_SYMBOL(dsl_dataset_stats); 5055 EXPORT_SYMBOL(dsl_dataset_fast_stat); 5056 EXPORT_SYMBOL(dsl_dataset_space); 5057 EXPORT_SYMBOL(dsl_dataset_fsid_guid); 5058 EXPORT_SYMBOL(dsl_dsobj_to_dsname); 5059 EXPORT_SYMBOL(dsl_dataset_check_quota); 5060 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl); 5061 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl); 5062