1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2017 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) 2014 Integros [integros.com] 29 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved. 30 * Copyright 2017 Nexenta Systems, Inc. 31 */ 32 33 #include <sys/dmu_objset.h> 34 #include <sys/dsl_dataset.h> 35 #include <sys/dsl_dir.h> 36 #include <sys/dsl_prop.h> 37 #include <sys/dsl_synctask.h> 38 #include <sys/dmu_traverse.h> 39 #include <sys/dmu_impl.h> 40 #include <sys/dmu_tx.h> 41 #include <sys/arc.h> 42 #include <sys/zio.h> 43 #include <sys/zap.h> 44 #include <sys/zfeature.h> 45 #include <sys/unique.h> 46 #include <sys/zfs_context.h> 47 #include <sys/zfs_ioctl.h> 48 #include <sys/spa.h> 49 #include <sys/spa_impl.h> 50 #include <sys/vdev.h> 51 #include <sys/zfs_znode.h> 52 #include <sys/zfs_onexit.h> 53 #include <sys/zvol.h> 54 #include <sys/dsl_scan.h> 55 #include <sys/dsl_deadlist.h> 56 #include <sys/dsl_destroy.h> 57 #include <sys/dsl_userhold.h> 58 #include <sys/dsl_bookmark.h> 59 #include <sys/dmu_recv.h> 60 #include <sys/zio_checksum.h> 61 #include <sys/zio_compress.h> 62 #include <zfs_fletcher.h> 63 64 /* 65 * The SPA supports block sizes up to 16MB. However, very large blocks 66 * can have an impact on i/o latency (e.g. tying up a spinning disk for 67 * ~300ms), and also potentially on the memory allocator. Therefore, 68 * we do not allow the recordsize to be set larger than zfs_max_recordsize 69 * (default 1MB). Larger blocks can be created by changing this tunable, 70 * and pools with larger blocks can always be imported and used, regardless 71 * of this setting. 72 */ 73 int zfs_max_recordsize = 1 * 1024 * 1024; 74 75 #define SWITCH64(x, y) \ 76 { \ 77 uint64_t __tmp = (x); \ 78 (x) = (y); \ 79 (y) = __tmp; \ 80 } 81 82 #define DS_REF_MAX (1ULL << 62) 83 84 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds); 85 86 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, 87 uint64_t obj, dmu_tx_t *tx); 88 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, 89 dmu_tx_t *tx); 90 91 extern int spa_asize_inflation; 92 93 static zil_header_t zero_zil; 94 95 /* 96 * Figure out how much of this delta should be propogated to the dsl_dir 97 * layer. If there's a refreservation, that space has already been 98 * partially accounted for in our ancestors. 99 */ 100 static int64_t 101 parent_delta(dsl_dataset_t *ds, int64_t delta) 102 { 103 dsl_dataset_phys_t *ds_phys; 104 uint64_t old_bytes, new_bytes; 105 106 if (ds->ds_reserved == 0) 107 return (delta); 108 109 ds_phys = dsl_dataset_phys(ds); 110 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved); 111 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 112 113 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 114 return (new_bytes - old_bytes); 115 } 116 117 void 118 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 119 { 120 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 121 int compressed = BP_GET_PSIZE(bp); 122 int uncompressed = BP_GET_UCSIZE(bp); 123 int64_t delta; 124 125 dprintf_bp(bp, "ds=%p", ds); 126 127 ASSERT(dmu_tx_is_syncing(tx)); 128 /* It could have been compressed away to nothing */ 129 if (BP_IS_HOLE(bp)) 130 return; 131 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 132 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 133 if (ds == NULL) { 134 dsl_pool_mos_diduse_space(tx->tx_pool, 135 used, compressed, uncompressed); 136 return; 137 } 138 139 ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 140 dmu_buf_will_dirty(ds->ds_dbuf, tx); 141 mutex_enter(&ds->ds_lock); 142 delta = parent_delta(ds, used); 143 dsl_dataset_phys(ds)->ds_referenced_bytes += used; 144 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed; 145 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed; 146 dsl_dataset_phys(ds)->ds_unique_bytes += used; 147 148 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) { 149 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] = 150 B_TRUE; 151 } 152 153 spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp)); 154 if (f != SPA_FEATURE_NONE) 155 ds->ds_feature_activation_needed[f] = B_TRUE; 156 157 mutex_exit(&ds->ds_lock); 158 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 159 compressed, uncompressed, tx); 160 dsl_dir_transfer_space(ds->ds_dir, used - delta, 161 DD_USED_REFRSRV, DD_USED_HEAD, tx); 162 } 163 164 /* 165 * Called when the specified segment has been remapped, and is thus no 166 * longer referenced in the head dataset. The vdev must be indirect. 167 * 168 * If the segment is referenced by a snapshot, put it on the remap deadlist. 169 * Otherwise, add this segment to the obsolete spacemap. 170 */ 171 void 172 dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, uint64_t offset, 173 uint64_t size, uint64_t birth, dmu_tx_t *tx) 174 { 175 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 176 177 ASSERT(dmu_tx_is_syncing(tx)); 178 ASSERT(birth <= tx->tx_txg); 179 ASSERT(!ds->ds_is_snapshot); 180 181 if (birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 182 spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx); 183 } else { 184 blkptr_t fakebp; 185 dva_t *dva = &fakebp.blk_dva[0]; 186 187 ASSERT(ds != NULL); 188 189 mutex_enter(&ds->ds_remap_deadlist_lock); 190 if (!dsl_dataset_remap_deadlist_exists(ds)) { 191 dsl_dataset_create_remap_deadlist(ds, tx); 192 } 193 mutex_exit(&ds->ds_remap_deadlist_lock); 194 195 BP_ZERO(&fakebp); 196 fakebp.blk_birth = birth; 197 DVA_SET_VDEV(dva, vdev); 198 DVA_SET_OFFSET(dva, offset); 199 DVA_SET_ASIZE(dva, size); 200 201 dsl_deadlist_insert(&ds->ds_remap_deadlist, &fakebp, tx); 202 } 203 } 204 205 int 206 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 207 boolean_t async) 208 { 209 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 210 211 int used = bp_get_dsize_sync(spa, bp); 212 int compressed = BP_GET_PSIZE(bp); 213 int uncompressed = BP_GET_UCSIZE(bp); 214 215 if (BP_IS_HOLE(bp)) 216 return (0); 217 218 ASSERT(dmu_tx_is_syncing(tx)); 219 ASSERT(bp->blk_birth <= tx->tx_txg); 220 221 if (ds == NULL) { 222 dsl_free(tx->tx_pool, tx->tx_txg, bp); 223 dsl_pool_mos_diduse_space(tx->tx_pool, 224 -used, -compressed, -uncompressed); 225 return (used); 226 } 227 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 228 229 ASSERT(!ds->ds_is_snapshot); 230 dmu_buf_will_dirty(ds->ds_dbuf, tx); 231 232 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 233 int64_t delta; 234 235 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 236 dsl_free(tx->tx_pool, tx->tx_txg, bp); 237 238 mutex_enter(&ds->ds_lock); 239 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used || 240 !DS_UNIQUE_IS_ACCURATE(ds)); 241 delta = parent_delta(ds, -used); 242 dsl_dataset_phys(ds)->ds_unique_bytes -= used; 243 mutex_exit(&ds->ds_lock); 244 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 245 delta, -compressed, -uncompressed, tx); 246 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 247 DD_USED_REFRSRV, DD_USED_HEAD, tx); 248 } else { 249 dprintf_bp(bp, "putting on dead list: %s", ""); 250 if (async) { 251 /* 252 * We are here as part of zio's write done callback, 253 * which means we're a zio interrupt thread. We can't 254 * call dsl_deadlist_insert() now because it may block 255 * waiting for I/O. Instead, put bp on the deferred 256 * queue and let dsl_pool_sync() finish the job. 257 */ 258 bplist_append(&ds->ds_pending_deadlist, bp); 259 } else { 260 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); 261 } 262 ASSERT3U(ds->ds_prev->ds_object, ==, 263 dsl_dataset_phys(ds)->ds_prev_snap_obj); 264 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0); 265 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 266 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 267 ds->ds_object && bp->blk_birth > 268 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) { 269 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 270 mutex_enter(&ds->ds_prev->ds_lock); 271 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used; 272 mutex_exit(&ds->ds_prev->ds_lock); 273 } 274 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 275 dsl_dir_transfer_space(ds->ds_dir, used, 276 DD_USED_HEAD, DD_USED_SNAP, tx); 277 } 278 } 279 mutex_enter(&ds->ds_lock); 280 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used); 281 dsl_dataset_phys(ds)->ds_referenced_bytes -= used; 282 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed); 283 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed; 284 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed); 285 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed; 286 mutex_exit(&ds->ds_lock); 287 288 return (used); 289 } 290 291 /* 292 * We have to release the fsid syncronously or we risk that a subsequent 293 * mount of the same dataset will fail to unique_insert the fsid. This 294 * failure would manifest itself as the fsid of this dataset changing 295 * between mounts which makes NFS clients quite unhappy. 296 */ 297 static void 298 dsl_dataset_evict_sync(void *dbu) 299 { 300 dsl_dataset_t *ds = dbu; 301 302 ASSERT(ds->ds_owner == NULL); 303 304 unique_remove(ds->ds_fsid_guid); 305 } 306 307 static void 308 dsl_dataset_evict_async(void *dbu) 309 { 310 dsl_dataset_t *ds = dbu; 311 312 ASSERT(ds->ds_owner == NULL); 313 314 ds->ds_dbuf = NULL; 315 316 if (ds->ds_objset != NULL) 317 dmu_objset_evict(ds->ds_objset); 318 319 if (ds->ds_prev) { 320 dsl_dataset_rele(ds->ds_prev, ds); 321 ds->ds_prev = NULL; 322 } 323 324 bplist_destroy(&ds->ds_pending_deadlist); 325 if (dsl_deadlist_is_open(&ds->ds_deadlist)) 326 dsl_deadlist_close(&ds->ds_deadlist); 327 if (dsl_deadlist_is_open(&ds->ds_remap_deadlist)) 328 dsl_deadlist_close(&ds->ds_remap_deadlist); 329 if (ds->ds_dir) 330 dsl_dir_async_rele(ds->ds_dir, ds); 331 332 ASSERT(!list_link_active(&ds->ds_synced_link)); 333 334 list_destroy(&ds->ds_prop_cbs); 335 mutex_destroy(&ds->ds_lock); 336 mutex_destroy(&ds->ds_opening_lock); 337 mutex_destroy(&ds->ds_sendstream_lock); 338 mutex_destroy(&ds->ds_remap_deadlist_lock); 339 zfs_refcount_destroy(&ds->ds_longholds); 340 rrw_destroy(&ds->ds_bp_rwlock); 341 342 kmem_free(ds, sizeof (dsl_dataset_t)); 343 } 344 345 int 346 dsl_dataset_get_snapname(dsl_dataset_t *ds) 347 { 348 dsl_dataset_phys_t *headphys; 349 int err; 350 dmu_buf_t *headdbuf; 351 dsl_pool_t *dp = ds->ds_dir->dd_pool; 352 objset_t *mos = dp->dp_meta_objset; 353 354 if (ds->ds_snapname[0]) 355 return (0); 356 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) 357 return (0); 358 359 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, 360 FTAG, &headdbuf); 361 if (err != 0) 362 return (err); 363 headphys = headdbuf->db_data; 364 err = zap_value_search(dp->dp_meta_objset, 365 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 366 dmu_buf_rele(headdbuf, FTAG); 367 return (err); 368 } 369 370 int 371 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 372 { 373 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 374 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 375 matchtype_t mt = 0; 376 int err; 377 378 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 379 mt = MT_NORMALIZE; 380 381 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 382 value, mt, NULL, 0, NULL); 383 if (err == ENOTSUP && (mt & MT_NORMALIZE)) 384 err = zap_lookup(mos, snapobj, name, 8, 1, value); 385 return (err); 386 } 387 388 int 389 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 390 boolean_t adj_cnt) 391 { 392 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 393 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 394 matchtype_t mt = 0; 395 int err; 396 397 dsl_dir_snap_cmtime_update(ds->ds_dir); 398 399 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 400 mt = MT_NORMALIZE; 401 402 err = zap_remove_norm(mos, snapobj, name, mt, tx); 403 if (err == ENOTSUP && (mt & MT_NORMALIZE)) 404 err = zap_remove(mos, snapobj, name, tx); 405 406 if (err == 0 && adj_cnt) 407 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 408 DD_FIELD_SNAPSHOT_COUNT, tx); 409 410 return (err); 411 } 412 413 boolean_t 414 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag) 415 { 416 dmu_buf_t *dbuf = ds->ds_dbuf; 417 boolean_t result = B_FALSE; 418 419 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset, 420 ds->ds_object, DMU_BONUS_BLKID, tag)) { 421 422 if (ds == dmu_buf_get_user(dbuf)) 423 result = B_TRUE; 424 else 425 dmu_buf_rele(dbuf, tag); 426 } 427 428 return (result); 429 } 430 431 int 432 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 433 dsl_dataset_t **dsp) 434 { 435 objset_t *mos = dp->dp_meta_objset; 436 dmu_buf_t *dbuf; 437 dsl_dataset_t *ds; 438 int err; 439 dmu_object_info_t doi; 440 441 ASSERT(dsl_pool_config_held(dp)); 442 443 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 444 if (err != 0) 445 return (err); 446 447 /* Make sure dsobj has the correct object type. */ 448 dmu_object_info_from_db(dbuf, &doi); 449 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 450 dmu_buf_rele(dbuf, tag); 451 return (SET_ERROR(EINVAL)); 452 } 453 454 ds = dmu_buf_get_user(dbuf); 455 if (ds == NULL) { 456 dsl_dataset_t *winner = NULL; 457 458 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 459 ds->ds_dbuf = dbuf; 460 ds->ds_object = dsobj; 461 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0; 462 463 err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj, 464 NULL, ds, &ds->ds_dir); 465 if (err != 0) { 466 kmem_free(ds, sizeof (dsl_dataset_t)); 467 dmu_buf_rele(dbuf, tag); 468 return (err); 469 } 470 471 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 472 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 473 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 474 mutex_init(&ds->ds_remap_deadlist_lock, 475 NULL, MUTEX_DEFAULT, NULL); 476 rrw_init(&ds->ds_bp_rwlock, B_FALSE); 477 zfs_refcount_create(&ds->ds_longholds); 478 479 bplist_create(&ds->ds_pending_deadlist); 480 481 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 482 offsetof(dmu_sendarg_t, dsa_link)); 483 484 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t), 485 offsetof(dsl_prop_cb_record_t, cbr_ds_node)); 486 487 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 488 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 489 if (!(spa_feature_table[f].fi_flags & 490 ZFEATURE_FLAG_PER_DATASET)) 491 continue; 492 err = zap_contains(mos, dsobj, 493 spa_feature_table[f].fi_guid); 494 if (err == 0) { 495 ds->ds_feature_inuse[f] = B_TRUE; 496 } else { 497 ASSERT3U(err, ==, ENOENT); 498 err = 0; 499 } 500 } 501 } 502 503 if (!ds->ds_is_snapshot) { 504 ds->ds_snapname[0] = '\0'; 505 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 506 err = dsl_dataset_hold_obj(dp, 507 dsl_dataset_phys(ds)->ds_prev_snap_obj, 508 ds, &ds->ds_prev); 509 } 510 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 511 int zaperr = zap_lookup(mos, ds->ds_object, 512 DS_FIELD_BOOKMARK_NAMES, 513 sizeof (ds->ds_bookmarks), 1, 514 &ds->ds_bookmarks); 515 if (zaperr != ENOENT) 516 VERIFY0(zaperr); 517 } 518 } else { 519 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 520 err = dsl_dataset_get_snapname(ds); 521 if (err == 0 && 522 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { 523 err = zap_count( 524 ds->ds_dir->dd_pool->dp_meta_objset, 525 dsl_dataset_phys(ds)->ds_userrefs_obj, 526 &ds->ds_userrefs); 527 } 528 } 529 530 if (err == 0 && !ds->ds_is_snapshot) { 531 err = dsl_prop_get_int_ds(ds, 532 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 533 &ds->ds_reserved); 534 if (err == 0) { 535 err = dsl_prop_get_int_ds(ds, 536 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 537 &ds->ds_quota); 538 } 539 } else { 540 ds->ds_reserved = ds->ds_quota = 0; 541 } 542 543 dsl_deadlist_open(&ds->ds_deadlist, 544 mos, dsl_dataset_phys(ds)->ds_deadlist_obj); 545 uint64_t remap_deadlist_obj = 546 dsl_dataset_get_remap_deadlist_object(ds); 547 if (remap_deadlist_obj != 0) { 548 dsl_deadlist_open(&ds->ds_remap_deadlist, mos, 549 remap_deadlist_obj); 550 } 551 552 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync, 553 dsl_dataset_evict_async, &ds->ds_dbuf); 554 if (err == 0) 555 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu); 556 557 if (err != 0 || winner != NULL) { 558 bplist_destroy(&ds->ds_pending_deadlist); 559 dsl_deadlist_close(&ds->ds_deadlist); 560 if (dsl_deadlist_is_open(&ds->ds_remap_deadlist)) 561 dsl_deadlist_close(&ds->ds_remap_deadlist); 562 if (ds->ds_prev) 563 dsl_dataset_rele(ds->ds_prev, ds); 564 dsl_dir_rele(ds->ds_dir, ds); 565 list_destroy(&ds->ds_prop_cbs); 566 list_destroy(&ds->ds_sendstreams); 567 mutex_destroy(&ds->ds_lock); 568 mutex_destroy(&ds->ds_opening_lock); 569 mutex_destroy(&ds->ds_sendstream_lock); 570 mutex_destroy(&ds->ds_remap_deadlist_lock); 571 zfs_refcount_destroy(&ds->ds_longholds); 572 rrw_destroy(&ds->ds_bp_rwlock); 573 kmem_free(ds, sizeof (dsl_dataset_t)); 574 if (err != 0) { 575 dmu_buf_rele(dbuf, tag); 576 return (err); 577 } 578 ds = winner; 579 } else { 580 ds->ds_fsid_guid = 581 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); 582 if (ds->ds_fsid_guid != 583 dsl_dataset_phys(ds)->ds_fsid_guid) { 584 zfs_dbgmsg("ds_fsid_guid changed from " 585 "%llx to %llx for pool %s dataset id %llu", 586 (long long) 587 dsl_dataset_phys(ds)->ds_fsid_guid, 588 (long long)ds->ds_fsid_guid, 589 spa_name(dp->dp_spa), 590 dsobj); 591 } 592 } 593 } 594 ASSERT3P(ds->ds_dbuf, ==, dbuf); 595 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); 596 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || 597 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 598 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 599 *dsp = ds; 600 return (0); 601 } 602 603 int 604 dsl_dataset_hold(dsl_pool_t *dp, const char *name, 605 void *tag, dsl_dataset_t **dsp) 606 { 607 dsl_dir_t *dd; 608 const char *snapname; 609 uint64_t obj; 610 int err = 0; 611 dsl_dataset_t *ds; 612 613 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 614 if (err != 0) 615 return (err); 616 617 ASSERT(dsl_pool_config_held(dp)); 618 obj = dsl_dir_phys(dd)->dd_head_dataset_obj; 619 if (obj != 0) 620 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 621 else 622 err = SET_ERROR(ENOENT); 623 624 /* we may be looking for a snapshot */ 625 if (err == 0 && snapname != NULL) { 626 dsl_dataset_t *snap_ds; 627 628 if (*snapname++ != '@') { 629 dsl_dataset_rele(ds, tag); 630 dsl_dir_rele(dd, FTAG); 631 return (SET_ERROR(ENOENT)); 632 } 633 634 dprintf("looking for snapshot '%s'\n", snapname); 635 err = dsl_dataset_snap_lookup(ds, snapname, &obj); 636 if (err == 0) 637 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds); 638 dsl_dataset_rele(ds, tag); 639 640 if (err == 0) { 641 mutex_enter(&snap_ds->ds_lock); 642 if (snap_ds->ds_snapname[0] == 0) 643 (void) strlcpy(snap_ds->ds_snapname, snapname, 644 sizeof (snap_ds->ds_snapname)); 645 mutex_exit(&snap_ds->ds_lock); 646 ds = snap_ds; 647 } 648 } 649 if (err == 0) 650 *dsp = ds; 651 dsl_dir_rele(dd, FTAG); 652 return (err); 653 } 654 655 int 656 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 657 void *tag, dsl_dataset_t **dsp) 658 { 659 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 660 if (err != 0) 661 return (err); 662 if (!dsl_dataset_tryown(*dsp, tag)) { 663 dsl_dataset_rele(*dsp, tag); 664 *dsp = NULL; 665 return (SET_ERROR(EBUSY)); 666 } 667 return (0); 668 } 669 670 int 671 dsl_dataset_own(dsl_pool_t *dp, const char *name, 672 void *tag, dsl_dataset_t **dsp) 673 { 674 int err = dsl_dataset_hold(dp, name, tag, dsp); 675 if (err != 0) 676 return (err); 677 if (!dsl_dataset_tryown(*dsp, tag)) { 678 dsl_dataset_rele(*dsp, tag); 679 return (SET_ERROR(EBUSY)); 680 } 681 return (0); 682 } 683 684 /* 685 * See the comment above dsl_pool_hold() for details. In summary, a long 686 * hold is used to prevent destruction of a dataset while the pool hold 687 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 688 * 689 * The dataset and pool must be held when this function is called. After it 690 * is called, the pool hold may be released while the dataset is still held 691 * and accessed. 692 */ 693 void 694 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 695 { 696 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 697 (void) zfs_refcount_add(&ds->ds_longholds, tag); 698 } 699 700 void 701 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 702 { 703 (void) zfs_refcount_remove(&ds->ds_longholds, tag); 704 } 705 706 /* Return B_TRUE if there are any long holds on this dataset. */ 707 boolean_t 708 dsl_dataset_long_held(dsl_dataset_t *ds) 709 { 710 return (!zfs_refcount_is_zero(&ds->ds_longholds)); 711 } 712 713 void 714 dsl_dataset_name(dsl_dataset_t *ds, char *name) 715 { 716 if (ds == NULL) { 717 (void) strcpy(name, "mos"); 718 } else { 719 dsl_dir_name(ds->ds_dir, name); 720 VERIFY0(dsl_dataset_get_snapname(ds)); 721 if (ds->ds_snapname[0]) { 722 VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN), 723 <, ZFS_MAX_DATASET_NAME_LEN); 724 /* 725 * We use a "recursive" mutex so that we 726 * can call dprintf_ds() with ds_lock held. 727 */ 728 if (!MUTEX_HELD(&ds->ds_lock)) { 729 mutex_enter(&ds->ds_lock); 730 VERIFY3U(strlcat(name, ds->ds_snapname, 731 ZFS_MAX_DATASET_NAME_LEN), <, 732 ZFS_MAX_DATASET_NAME_LEN); 733 mutex_exit(&ds->ds_lock); 734 } else { 735 VERIFY3U(strlcat(name, ds->ds_snapname, 736 ZFS_MAX_DATASET_NAME_LEN), <, 737 ZFS_MAX_DATASET_NAME_LEN); 738 } 739 } 740 } 741 } 742 743 int 744 dsl_dataset_namelen(dsl_dataset_t *ds) 745 { 746 VERIFY0(dsl_dataset_get_snapname(ds)); 747 mutex_enter(&ds->ds_lock); 748 int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname); 749 mutex_exit(&ds->ds_lock); 750 return (len); 751 } 752 753 void 754 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 755 { 756 dmu_buf_rele(ds->ds_dbuf, tag); 757 } 758 759 void 760 dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 761 { 762 ASSERT3P(ds->ds_owner, ==, tag); 763 ASSERT(ds->ds_dbuf != NULL); 764 765 mutex_enter(&ds->ds_lock); 766 ds->ds_owner = NULL; 767 mutex_exit(&ds->ds_lock); 768 dsl_dataset_long_rele(ds, tag); 769 dsl_dataset_rele(ds, tag); 770 } 771 772 boolean_t 773 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 774 { 775 boolean_t gotit = FALSE; 776 777 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 778 mutex_enter(&ds->ds_lock); 779 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 780 ds->ds_owner = tag; 781 dsl_dataset_long_hold(ds, tag); 782 gotit = TRUE; 783 } 784 mutex_exit(&ds->ds_lock); 785 return (gotit); 786 } 787 788 boolean_t 789 dsl_dataset_has_owner(dsl_dataset_t *ds) 790 { 791 boolean_t rv; 792 mutex_enter(&ds->ds_lock); 793 rv = (ds->ds_owner != NULL); 794 mutex_exit(&ds->ds_lock); 795 return (rv); 796 } 797 798 static void 799 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx) 800 { 801 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 802 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 803 uint64_t zero = 0; 804 805 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 806 807 spa_feature_incr(spa, f, tx); 808 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 809 810 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, 811 sizeof (zero), 1, &zero, tx)); 812 } 813 814 void 815 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx) 816 { 817 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 818 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 819 820 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 821 822 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx)); 823 spa_feature_decr(spa, f, tx); 824 } 825 826 uint64_t 827 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 828 uint64_t flags, dmu_tx_t *tx) 829 { 830 dsl_pool_t *dp = dd->dd_pool; 831 dmu_buf_t *dbuf; 832 dsl_dataset_phys_t *dsphys; 833 uint64_t dsobj; 834 objset_t *mos = dp->dp_meta_objset; 835 836 if (origin == NULL) 837 origin = dp->dp_origin_snap; 838 839 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 840 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); 841 ASSERT(dmu_tx_is_syncing(tx)); 842 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 843 844 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 845 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 846 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 847 dmu_buf_will_dirty(dbuf, tx); 848 dsphys = dbuf->db_data; 849 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 850 dsphys->ds_dir_obj = dd->dd_object; 851 dsphys->ds_flags = flags; 852 dsphys->ds_fsid_guid = unique_create(); 853 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 854 sizeof (dsphys->ds_guid)); 855 dsphys->ds_snapnames_zapobj = 856 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 857 DMU_OT_NONE, 0, tx); 858 dsphys->ds_creation_time = gethrestime_sec(); 859 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 860 861 if (origin == NULL) { 862 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 863 } else { 864 dsl_dataset_t *ohds; /* head of the origin snapshot */ 865 866 dsphys->ds_prev_snap_obj = origin->ds_object; 867 dsphys->ds_prev_snap_txg = 868 dsl_dataset_phys(origin)->ds_creation_txg; 869 dsphys->ds_referenced_bytes = 870 dsl_dataset_phys(origin)->ds_referenced_bytes; 871 dsphys->ds_compressed_bytes = 872 dsl_dataset_phys(origin)->ds_compressed_bytes; 873 dsphys->ds_uncompressed_bytes = 874 dsl_dataset_phys(origin)->ds_uncompressed_bytes; 875 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG); 876 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; 877 rrw_exit(&origin->ds_bp_rwlock, FTAG); 878 879 /* 880 * Inherit flags that describe the dataset's contents 881 * (INCONSISTENT) or properties (Case Insensitive). 882 */ 883 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & 884 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 885 886 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 887 if (origin->ds_feature_inuse[f]) 888 dsl_dataset_activate_feature(dsobj, f, tx); 889 } 890 891 dmu_buf_will_dirty(origin->ds_dbuf, tx); 892 dsl_dataset_phys(origin)->ds_num_children++; 893 894 VERIFY0(dsl_dataset_hold_obj(dp, 895 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, 896 FTAG, &ohds)); 897 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 898 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 899 dsl_dataset_rele(ohds, FTAG); 900 901 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 902 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { 903 dsl_dataset_phys(origin)->ds_next_clones_obj = 904 zap_create(mos, 905 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 906 } 907 VERIFY0(zap_add_int(mos, 908 dsl_dataset_phys(origin)->ds_next_clones_obj, 909 dsobj, tx)); 910 } 911 912 dmu_buf_will_dirty(dd->dd_dbuf, tx); 913 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; 914 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 915 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 916 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 917 dsl_dir_phys(origin->ds_dir)->dd_clones = 918 zap_create(mos, 919 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 920 } 921 VERIFY0(zap_add_int(mos, 922 dsl_dir_phys(origin->ds_dir)->dd_clones, 923 dsobj, tx)); 924 } 925 } 926 927 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 928 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 929 930 dmu_buf_rele(dbuf, FTAG); 931 932 dmu_buf_will_dirty(dd->dd_dbuf, tx); 933 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; 934 935 return (dsobj); 936 } 937 938 static void 939 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 940 { 941 objset_t *os; 942 943 VERIFY0(dmu_objset_from_ds(ds, &os)); 944 if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) { 945 dsl_pool_t *dp = ds->ds_dir->dd_pool; 946 zio_t *zio; 947 948 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 949 950 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 951 dsl_dataset_sync(ds, zio, tx); 952 VERIFY0(zio_wait(zio)); 953 954 /* dsl_dataset_sync_done will drop this reference. */ 955 dmu_buf_add_ref(ds->ds_dbuf, ds); 956 dsl_dataset_sync_done(ds, tx); 957 } 958 } 959 960 uint64_t 961 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 962 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 963 { 964 dsl_pool_t *dp = pdd->dd_pool; 965 uint64_t dsobj, ddobj; 966 dsl_dir_t *dd; 967 968 ASSERT(dmu_tx_is_syncing(tx)); 969 ASSERT(lastname[0] != '@'); 970 971 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 972 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 973 974 dsobj = dsl_dataset_create_sync_dd(dd, origin, 975 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 976 977 dsl_deleg_set_create_perms(dd, tx, cr); 978 979 /* 980 * Since we're creating a new node we know it's a leaf, so we can 981 * initialize the counts if the limit feature is active. 982 */ 983 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 984 uint64_t cnt = 0; 985 objset_t *os = dd->dd_pool->dp_meta_objset; 986 987 dsl_dir_zapify(dd, tx); 988 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 989 sizeof (cnt), 1, &cnt, tx)); 990 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 991 sizeof (cnt), 1, &cnt, tx)); 992 } 993 994 dsl_dir_rele(dd, FTAG); 995 996 /* 997 * If we are creating a clone, make sure we zero out any stale 998 * data from the origin snapshots zil header. 999 */ 1000 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 1001 dsl_dataset_t *ds; 1002 1003 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 1004 dsl_dataset_zero_zil(ds, tx); 1005 dsl_dataset_rele(ds, FTAG); 1006 } 1007 1008 return (dsobj); 1009 } 1010 1011 /* 1012 * The unique space in the head dataset can be calculated by subtracting 1013 * the space used in the most recent snapshot, that is still being used 1014 * in this file system, from the space currently in use. To figure out 1015 * the space in the most recent snapshot still in use, we need to take 1016 * the total space used in the snapshot and subtract out the space that 1017 * has been freed up since the snapshot was taken. 1018 */ 1019 void 1020 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 1021 { 1022 uint64_t mrs_used; 1023 uint64_t dlused, dlcomp, dluncomp; 1024 1025 ASSERT(!ds->ds_is_snapshot); 1026 1027 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) 1028 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; 1029 else 1030 mrs_used = 0; 1031 1032 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 1033 1034 ASSERT3U(dlused, <=, mrs_used); 1035 dsl_dataset_phys(ds)->ds_unique_bytes = 1036 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); 1037 1038 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1039 SPA_VERSION_UNIQUE_ACCURATE) 1040 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1041 } 1042 1043 void 1044 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 1045 dmu_tx_t *tx) 1046 { 1047 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1048 uint64_t count; 1049 int err; 1050 1051 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); 1052 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1053 obj, tx); 1054 /* 1055 * The err should not be ENOENT, but a bug in a previous version 1056 * of the code could cause upgrade_clones_cb() to not set 1057 * ds_next_snap_obj when it should, leading to a missing entry. 1058 * If we knew that the pool was created after 1059 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 1060 * ENOENT. However, at least we can check that we don't have 1061 * too many entries in the next_clones_obj even after failing to 1062 * remove this one. 1063 */ 1064 if (err != ENOENT) 1065 VERIFY0(err); 1066 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1067 &count)); 1068 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); 1069 } 1070 1071 1072 blkptr_t * 1073 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1074 { 1075 return (&dsl_dataset_phys(ds)->ds_bp); 1076 } 1077 1078 spa_t * 1079 dsl_dataset_get_spa(dsl_dataset_t *ds) 1080 { 1081 return (ds->ds_dir->dd_pool->dp_spa); 1082 } 1083 1084 void 1085 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1086 { 1087 dsl_pool_t *dp; 1088 1089 if (ds == NULL) /* this is the meta-objset */ 1090 return; 1091 1092 ASSERT(ds->ds_objset != NULL); 1093 1094 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) 1095 panic("dirtying snapshot!"); 1096 1097 /* Must not dirty a dataset in the same txg where it got snapshotted. */ 1098 ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 1099 1100 dp = ds->ds_dir->dd_pool; 1101 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1102 /* up the hold count until we can be written out */ 1103 dmu_buf_add_ref(ds->ds_dbuf, ds); 1104 } 1105 } 1106 1107 boolean_t 1108 dsl_dataset_is_dirty(dsl_dataset_t *ds) 1109 { 1110 for (int t = 0; t < TXG_SIZE; t++) { 1111 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1112 ds, t)) 1113 return (B_TRUE); 1114 } 1115 return (B_FALSE); 1116 } 1117 1118 static int 1119 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1120 { 1121 uint64_t asize; 1122 1123 if (!dmu_tx_is_syncing(tx)) 1124 return (0); 1125 1126 /* 1127 * If there's an fs-only reservation, any blocks that might become 1128 * owned by the snapshot dataset must be accommodated by space 1129 * outside of the reservation. 1130 */ 1131 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1132 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); 1133 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1134 return (SET_ERROR(ENOSPC)); 1135 1136 /* 1137 * Propagate any reserved space for this snapshot to other 1138 * snapshot checks in this sync group. 1139 */ 1140 if (asize > 0) 1141 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1142 1143 return (0); 1144 } 1145 1146 int 1147 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1148 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 1149 { 1150 int error; 1151 uint64_t value; 1152 1153 ds->ds_trysnap_txg = tx->tx_txg; 1154 1155 if (!dmu_tx_is_syncing(tx)) 1156 return (0); 1157 1158 /* 1159 * We don't allow multiple snapshots of the same txg. If there 1160 * is already one, try again. 1161 */ 1162 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) 1163 return (SET_ERROR(EAGAIN)); 1164 1165 /* 1166 * Check for conflicting snapshot name. 1167 */ 1168 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1169 if (error == 0) 1170 return (SET_ERROR(EEXIST)); 1171 if (error != ENOENT) 1172 return (error); 1173 1174 /* 1175 * We don't allow taking snapshots of inconsistent datasets, such as 1176 * those into which we are currently receiving. However, if we are 1177 * creating this snapshot as part of a receive, this check will be 1178 * executed atomically with respect to the completion of the receive 1179 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1180 * case we ignore this, knowing it will be fixed up for us shortly in 1181 * dmu_recv_end_sync(). 1182 */ 1183 if (!recv && DS_IS_INCONSISTENT(ds)) 1184 return (SET_ERROR(EBUSY)); 1185 1186 /* 1187 * Skip the check for temporary snapshots or if we have already checked 1188 * the counts in dsl_dataset_snapshot_check. This means we really only 1189 * check the count here when we're receiving a stream. 1190 */ 1191 if (cnt != 0 && cr != NULL) { 1192 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1193 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 1194 if (error != 0) 1195 return (error); 1196 } 1197 1198 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1199 if (error != 0) 1200 return (error); 1201 1202 return (0); 1203 } 1204 1205 int 1206 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1207 { 1208 dsl_dataset_snapshot_arg_t *ddsa = arg; 1209 dsl_pool_t *dp = dmu_tx_pool(tx); 1210 nvpair_t *pair; 1211 int rv = 0; 1212 1213 /* 1214 * Pre-compute how many total new snapshots will be created for each 1215 * level in the tree and below. This is needed for validating the 1216 * snapshot limit when either taking a recursive snapshot or when 1217 * taking multiple snapshots. 1218 * 1219 * The problem is that the counts are not actually adjusted when 1220 * we are checking, only when we finally sync. For a single snapshot, 1221 * this is easy, the count will increase by 1 at each node up the tree, 1222 * but its more complicated for the recursive/multiple snapshot case. 1223 * 1224 * The dsl_fs_ss_limit_check function does recursively check the count 1225 * at each level up the tree but since it is validating each snapshot 1226 * independently we need to be sure that we are validating the complete 1227 * count for the entire set of snapshots. We do this by rolling up the 1228 * counts for each component of the name into an nvlist and then 1229 * checking each of those cases with the aggregated count. 1230 * 1231 * This approach properly handles not only the recursive snapshot 1232 * case (where we get all of those on the ddsa_snaps list) but also 1233 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1234 * validate the limit on 'a' using a count of 2). 1235 * 1236 * We validate the snapshot names in the third loop and only report 1237 * name errors once. 1238 */ 1239 if (dmu_tx_is_syncing(tx)) { 1240 nvlist_t *cnt_track = NULL; 1241 cnt_track = fnvlist_alloc(); 1242 1243 /* Rollup aggregated counts into the cnt_track list */ 1244 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1245 pair != NULL; 1246 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1247 char *pdelim; 1248 uint64_t val; 1249 char nm[MAXPATHLEN]; 1250 1251 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1252 pdelim = strchr(nm, '@'); 1253 if (pdelim == NULL) 1254 continue; 1255 *pdelim = '\0'; 1256 1257 do { 1258 if (nvlist_lookup_uint64(cnt_track, nm, 1259 &val) == 0) { 1260 /* update existing entry */ 1261 fnvlist_add_uint64(cnt_track, nm, 1262 val + 1); 1263 } else { 1264 /* add to list */ 1265 fnvlist_add_uint64(cnt_track, nm, 1); 1266 } 1267 1268 pdelim = strrchr(nm, '/'); 1269 if (pdelim != NULL) 1270 *pdelim = '\0'; 1271 } while (pdelim != NULL); 1272 } 1273 1274 /* Check aggregated counts at each level */ 1275 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1276 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1277 int error = 0; 1278 char *name; 1279 uint64_t cnt = 0; 1280 dsl_dataset_t *ds; 1281 1282 name = nvpair_name(pair); 1283 cnt = fnvpair_value_uint64(pair); 1284 ASSERT(cnt > 0); 1285 1286 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1287 if (error == 0) { 1288 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1289 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1290 ddsa->ddsa_cr); 1291 dsl_dataset_rele(ds, FTAG); 1292 } 1293 1294 if (error != 0) { 1295 if (ddsa->ddsa_errors != NULL) 1296 fnvlist_add_int32(ddsa->ddsa_errors, 1297 name, error); 1298 rv = error; 1299 /* only report one error for this check */ 1300 break; 1301 } 1302 } 1303 nvlist_free(cnt_track); 1304 } 1305 1306 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1307 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1308 int error = 0; 1309 dsl_dataset_t *ds; 1310 char *name, *atp; 1311 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1312 1313 name = nvpair_name(pair); 1314 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN) 1315 error = SET_ERROR(ENAMETOOLONG); 1316 if (error == 0) { 1317 atp = strchr(name, '@'); 1318 if (atp == NULL) 1319 error = SET_ERROR(EINVAL); 1320 if (error == 0) 1321 (void) strlcpy(dsname, name, atp - name + 1); 1322 } 1323 if (error == 0) 1324 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1325 if (error == 0) { 1326 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1327 error = dsl_dataset_snapshot_check_impl(ds, 1328 atp + 1, tx, B_FALSE, 0, NULL); 1329 dsl_dataset_rele(ds, FTAG); 1330 } 1331 1332 if (error != 0) { 1333 if (ddsa->ddsa_errors != NULL) { 1334 fnvlist_add_int32(ddsa->ddsa_errors, 1335 name, error); 1336 } 1337 rv = error; 1338 } 1339 } 1340 1341 return (rv); 1342 } 1343 1344 void 1345 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1346 dmu_tx_t *tx) 1347 { 1348 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1349 dmu_buf_t *dbuf; 1350 dsl_dataset_phys_t *dsphys; 1351 uint64_t dsobj, crtxg; 1352 objset_t *mos = dp->dp_meta_objset; 1353 objset_t *os; 1354 1355 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1356 1357 /* 1358 * If we are on an old pool, the zil must not be active, in which 1359 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1360 */ 1361 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1362 dmu_objset_from_ds(ds, &os) != 0 || 1363 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1364 sizeof (zero_zil)) == 0); 1365 1366 /* Should not snapshot a dirty dataset. */ 1367 ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1368 ds, tx->tx_txg)); 1369 1370 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1371 1372 /* 1373 * The origin's ds_creation_txg has to be < TXG_INITIAL 1374 */ 1375 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1376 crtxg = 1; 1377 else 1378 crtxg = tx->tx_txg; 1379 1380 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1381 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1382 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1383 dmu_buf_will_dirty(dbuf, tx); 1384 dsphys = dbuf->db_data; 1385 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1386 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1387 dsphys->ds_fsid_guid = unique_create(); 1388 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1389 sizeof (dsphys->ds_guid)); 1390 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1391 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 1392 dsphys->ds_next_snap_obj = ds->ds_object; 1393 dsphys->ds_num_children = 1; 1394 dsphys->ds_creation_time = gethrestime_sec(); 1395 dsphys->ds_creation_txg = crtxg; 1396 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; 1397 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; 1398 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; 1399 dsphys->ds_uncompressed_bytes = 1400 dsl_dataset_phys(ds)->ds_uncompressed_bytes; 1401 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; 1402 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1403 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; 1404 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1405 dmu_buf_rele(dbuf, FTAG); 1406 1407 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1408 if (ds->ds_feature_inuse[f]) 1409 dsl_dataset_activate_feature(dsobj, f, tx); 1410 } 1411 1412 ASSERT3U(ds->ds_prev != 0, ==, 1413 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); 1414 if (ds->ds_prev) { 1415 uint64_t next_clones_obj = 1416 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; 1417 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1418 ds->ds_object || 1419 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); 1420 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1421 ds->ds_object) { 1422 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1423 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, 1424 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); 1425 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; 1426 } else if (next_clones_obj != 0) { 1427 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1428 dsphys->ds_next_snap_obj, tx); 1429 VERIFY0(zap_add_int(mos, 1430 next_clones_obj, dsobj, tx)); 1431 } 1432 } 1433 1434 /* 1435 * If we have a reference-reservation on this dataset, we will 1436 * need to increase the amount of refreservation being charged 1437 * since our unique space is going to zero. 1438 */ 1439 if (ds->ds_reserved) { 1440 int64_t delta; 1441 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1442 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, 1443 ds->ds_reserved); 1444 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1445 delta, 0, 0, tx); 1446 } 1447 1448 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1449 dsl_dataset_phys(ds)->ds_deadlist_obj = 1450 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, 1451 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 1452 dsl_deadlist_close(&ds->ds_deadlist); 1453 dsl_deadlist_open(&ds->ds_deadlist, mos, 1454 dsl_dataset_phys(ds)->ds_deadlist_obj); 1455 dsl_deadlist_add_key(&ds->ds_deadlist, 1456 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); 1457 1458 if (dsl_dataset_remap_deadlist_exists(ds)) { 1459 uint64_t remap_deadlist_obj = 1460 dsl_dataset_get_remap_deadlist_object(ds); 1461 /* 1462 * Move the remap_deadlist to the snapshot. The head 1463 * will create a new remap deadlist on demand, from 1464 * dsl_dataset_block_remapped(). 1465 */ 1466 dsl_dataset_unset_remap_deadlist_object(ds, tx); 1467 dsl_deadlist_close(&ds->ds_remap_deadlist); 1468 1469 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 1470 VERIFY0(zap_add(mos, dsobj, DS_FIELD_REMAP_DEADLIST, 1471 sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj, tx)); 1472 } 1473 1474 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); 1475 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; 1476 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; 1477 dsl_dataset_phys(ds)->ds_unique_bytes = 0; 1478 1479 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1480 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1481 1482 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1483 snapname, 8, 1, &dsobj, tx)); 1484 1485 if (ds->ds_prev) 1486 dsl_dataset_rele(ds->ds_prev, ds); 1487 VERIFY0(dsl_dataset_hold_obj(dp, 1488 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); 1489 1490 dsl_scan_ds_snapshotted(ds, tx); 1491 1492 dsl_dir_snap_cmtime_update(ds->ds_dir); 1493 1494 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1495 } 1496 1497 void 1498 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1499 { 1500 dsl_dataset_snapshot_arg_t *ddsa = arg; 1501 dsl_pool_t *dp = dmu_tx_pool(tx); 1502 nvpair_t *pair; 1503 1504 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1505 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1506 dsl_dataset_t *ds; 1507 char *name, *atp; 1508 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1509 1510 name = nvpair_name(pair); 1511 atp = strchr(name, '@'); 1512 (void) strlcpy(dsname, name, atp - name + 1); 1513 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1514 1515 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1516 if (ddsa->ddsa_props != NULL) { 1517 dsl_props_set_sync_impl(ds->ds_prev, 1518 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1519 } 1520 dsl_dataset_rele(ds, FTAG); 1521 } 1522 } 1523 1524 /* 1525 * The snapshots must all be in the same pool. 1526 * All-or-nothing: if there are any failures, nothing will be modified. 1527 */ 1528 int 1529 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1530 { 1531 dsl_dataset_snapshot_arg_t ddsa; 1532 nvpair_t *pair; 1533 boolean_t needsuspend; 1534 int error; 1535 spa_t *spa; 1536 char *firstname; 1537 nvlist_t *suspended = NULL; 1538 1539 pair = nvlist_next_nvpair(snaps, NULL); 1540 if (pair == NULL) 1541 return (0); 1542 firstname = nvpair_name(pair); 1543 1544 error = spa_open(firstname, &spa, FTAG); 1545 if (error != 0) 1546 return (error); 1547 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1548 spa_close(spa, FTAG); 1549 1550 if (needsuspend) { 1551 suspended = fnvlist_alloc(); 1552 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1553 pair = nvlist_next_nvpair(snaps, pair)) { 1554 char fsname[ZFS_MAX_DATASET_NAME_LEN]; 1555 char *snapname = nvpair_name(pair); 1556 char *atp; 1557 void *cookie; 1558 1559 atp = strchr(snapname, '@'); 1560 if (atp == NULL) { 1561 error = SET_ERROR(EINVAL); 1562 break; 1563 } 1564 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1565 1566 error = zil_suspend(fsname, &cookie); 1567 if (error != 0) 1568 break; 1569 fnvlist_add_uint64(suspended, fsname, 1570 (uintptr_t)cookie); 1571 } 1572 } 1573 1574 ddsa.ddsa_snaps = snaps; 1575 ddsa.ddsa_props = props; 1576 ddsa.ddsa_errors = errors; 1577 ddsa.ddsa_cr = CRED(); 1578 1579 if (error == 0) { 1580 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1581 dsl_dataset_snapshot_sync, &ddsa, 1582 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1583 } 1584 1585 if (suspended != NULL) { 1586 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1587 pair = nvlist_next_nvpair(suspended, pair)) { 1588 zil_resume((void *)(uintptr_t) 1589 fnvpair_value_uint64(pair)); 1590 } 1591 fnvlist_free(suspended); 1592 } 1593 1594 return (error); 1595 } 1596 1597 typedef struct dsl_dataset_snapshot_tmp_arg { 1598 const char *ddsta_fsname; 1599 const char *ddsta_snapname; 1600 minor_t ddsta_cleanup_minor; 1601 const char *ddsta_htag; 1602 } dsl_dataset_snapshot_tmp_arg_t; 1603 1604 static int 1605 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1606 { 1607 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1608 dsl_pool_t *dp = dmu_tx_pool(tx); 1609 dsl_dataset_t *ds; 1610 int error; 1611 1612 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1613 if (error != 0) 1614 return (error); 1615 1616 /* NULL cred means no limit check for tmp snapshot */ 1617 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1618 tx, B_FALSE, 0, NULL); 1619 if (error != 0) { 1620 dsl_dataset_rele(ds, FTAG); 1621 return (error); 1622 } 1623 1624 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1625 dsl_dataset_rele(ds, FTAG); 1626 return (SET_ERROR(ENOTSUP)); 1627 } 1628 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1629 B_TRUE, tx); 1630 if (error != 0) { 1631 dsl_dataset_rele(ds, FTAG); 1632 return (error); 1633 } 1634 1635 dsl_dataset_rele(ds, FTAG); 1636 return (0); 1637 } 1638 1639 static void 1640 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1641 { 1642 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1643 dsl_pool_t *dp = dmu_tx_pool(tx); 1644 dsl_dataset_t *ds; 1645 1646 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1647 1648 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1649 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1650 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1651 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1652 1653 dsl_dataset_rele(ds, FTAG); 1654 } 1655 1656 int 1657 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1658 minor_t cleanup_minor, const char *htag) 1659 { 1660 dsl_dataset_snapshot_tmp_arg_t ddsta; 1661 int error; 1662 spa_t *spa; 1663 boolean_t needsuspend; 1664 void *cookie; 1665 1666 ddsta.ddsta_fsname = fsname; 1667 ddsta.ddsta_snapname = snapname; 1668 ddsta.ddsta_cleanup_minor = cleanup_minor; 1669 ddsta.ddsta_htag = htag; 1670 1671 error = spa_open(fsname, &spa, FTAG); 1672 if (error != 0) 1673 return (error); 1674 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1675 spa_close(spa, FTAG); 1676 1677 if (needsuspend) { 1678 error = zil_suspend(fsname, &cookie); 1679 if (error != 0) 1680 return (error); 1681 } 1682 1683 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1684 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 1685 1686 if (needsuspend) 1687 zil_resume(cookie); 1688 return (error); 1689 } 1690 1691 void 1692 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1693 { 1694 ASSERT(dmu_tx_is_syncing(tx)); 1695 ASSERT(ds->ds_objset != NULL); 1696 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); 1697 1698 /* 1699 * in case we had to change ds_fsid_guid when we opened it, 1700 * sync it out now. 1701 */ 1702 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1703 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; 1704 1705 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) { 1706 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1707 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1, 1708 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx)); 1709 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1710 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1, 1711 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx)); 1712 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1713 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1, 1714 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx)); 1715 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0; 1716 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0; 1717 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0; 1718 } 1719 1720 dmu_objset_sync(ds->ds_objset, zio, tx); 1721 1722 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1723 if (ds->ds_feature_activation_needed[f]) { 1724 if (ds->ds_feature_inuse[f]) 1725 continue; 1726 dsl_dataset_activate_feature(ds->ds_object, f, tx); 1727 ds->ds_feature_inuse[f] = B_TRUE; 1728 } 1729 } 1730 } 1731 1732 static int 1733 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 1734 { 1735 dsl_deadlist_t *dl = arg; 1736 dsl_deadlist_insert(dl, bp, tx); 1737 return (0); 1738 } 1739 1740 void 1741 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx) 1742 { 1743 objset_t *os = ds->ds_objset; 1744 1745 bplist_iterate(&ds->ds_pending_deadlist, 1746 deadlist_enqueue_cb, &ds->ds_deadlist, tx); 1747 1748 if (os->os_synced_dnodes != NULL) { 1749 multilist_destroy(os->os_synced_dnodes); 1750 os->os_synced_dnodes = NULL; 1751 } 1752 1753 ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx))); 1754 1755 dmu_buf_rele(ds->ds_dbuf, ds); 1756 } 1757 1758 int 1759 get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val) 1760 { 1761 uint64_t count = 0; 1762 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1763 zap_cursor_t zc; 1764 zap_attribute_t za; 1765 1766 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1767 1768 /* 1769 * There may be missing entries in ds_next_clones_obj 1770 * due to a bug in a previous version of the code. 1771 * Only trust it if it has the right number of entries. 1772 */ 1773 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { 1774 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1775 &count)); 1776 } 1777 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) { 1778 return (ENOENT); 1779 } 1780 for (zap_cursor_init(&zc, mos, 1781 dsl_dataset_phys(ds)->ds_next_clones_obj); 1782 zap_cursor_retrieve(&zc, &za) == 0; 1783 zap_cursor_advance(&zc)) { 1784 dsl_dataset_t *clone; 1785 char buf[ZFS_MAX_DATASET_NAME_LEN]; 1786 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1787 za.za_first_integer, FTAG, &clone)); 1788 dsl_dir_name(clone->ds_dir, buf); 1789 fnvlist_add_boolean(val, buf); 1790 dsl_dataset_rele(clone, FTAG); 1791 } 1792 zap_cursor_fini(&zc); 1793 return (0); 1794 } 1795 1796 void 1797 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1798 { 1799 nvlist_t *propval = fnvlist_alloc(); 1800 nvlist_t *val; 1801 1802 /* 1803 * We use nvlist_alloc() instead of fnvlist_alloc() because the 1804 * latter would allocate the list with NV_UNIQUE_NAME flag. 1805 * As a result, every time a clone name is appended to the list 1806 * it would be (linearly) searched for for a duplicate name. 1807 * We already know that all clone names must be unique and we 1808 * want avoid the quadratic complexity of double-checking that 1809 * because we can have a large number of clones. 1810 */ 1811 VERIFY0(nvlist_alloc(&val, 0, KM_SLEEP)); 1812 1813 if (get_clones_stat_impl(ds, val) == 0) { 1814 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1815 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), 1816 propval); 1817 } 1818 1819 nvlist_free(val); 1820 nvlist_free(propval); 1821 } 1822 1823 /* 1824 * Returns a string that represents the receive resume stats token. It should 1825 * be freed with strfree(). 1826 */ 1827 char * 1828 get_receive_resume_stats_impl(dsl_dataset_t *ds) 1829 { 1830 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1831 1832 if (dsl_dataset_has_resume_receive_state(ds)) { 1833 char *str; 1834 void *packed; 1835 uint8_t *compressed; 1836 uint64_t val; 1837 nvlist_t *token_nv = fnvlist_alloc(); 1838 size_t packed_size, compressed_size; 1839 1840 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1841 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) { 1842 fnvlist_add_uint64(token_nv, "fromguid", val); 1843 } 1844 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1845 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) { 1846 fnvlist_add_uint64(token_nv, "object", val); 1847 } 1848 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1849 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) { 1850 fnvlist_add_uint64(token_nv, "offset", val); 1851 } 1852 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1853 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) { 1854 fnvlist_add_uint64(token_nv, "bytes", val); 1855 } 1856 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1857 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) { 1858 fnvlist_add_uint64(token_nv, "toguid", val); 1859 } 1860 char buf[256]; 1861 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1862 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) { 1863 fnvlist_add_string(token_nv, "toname", buf); 1864 } 1865 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1866 DS_FIELD_RESUME_LARGEBLOCK) == 0) { 1867 fnvlist_add_boolean(token_nv, "largeblockok"); 1868 } 1869 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1870 DS_FIELD_RESUME_EMBEDOK) == 0) { 1871 fnvlist_add_boolean(token_nv, "embedok"); 1872 } 1873 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1874 DS_FIELD_RESUME_COMPRESSOK) == 0) { 1875 fnvlist_add_boolean(token_nv, "compressok"); 1876 } 1877 packed = fnvlist_pack(token_nv, &packed_size); 1878 fnvlist_free(token_nv); 1879 compressed = kmem_alloc(packed_size, KM_SLEEP); 1880 1881 compressed_size = gzip_compress(packed, compressed, 1882 packed_size, packed_size, 6); 1883 1884 zio_cksum_t cksum; 1885 fletcher_4_native(compressed, compressed_size, NULL, &cksum); 1886 1887 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP); 1888 for (int i = 0; i < compressed_size; i++) { 1889 (void) sprintf(str + i * 2, "%02x", compressed[i]); 1890 } 1891 str[compressed_size * 2] = '\0'; 1892 char *propval = kmem_asprintf("%u-%llx-%llx-%s", 1893 ZFS_SEND_RESUME_TOKEN_VERSION, 1894 (longlong_t)cksum.zc_word[0], 1895 (longlong_t)packed_size, str); 1896 kmem_free(packed, packed_size); 1897 kmem_free(str, compressed_size * 2 + 1); 1898 kmem_free(compressed, packed_size); 1899 return (propval); 1900 } 1901 return (strdup("")); 1902 } 1903 1904 /* 1905 * Returns a string that represents the receive resume stats token of the 1906 * dataset's child. It should be freed with strfree(). 1907 */ 1908 char * 1909 get_child_receive_stats(dsl_dataset_t *ds) 1910 { 1911 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 1912 dsl_dataset_t *recv_ds; 1913 dsl_dataset_name(ds, recvname); 1914 if (strlcat(recvname, "/", sizeof (recvname)) < 1915 sizeof (recvname) && 1916 strlcat(recvname, recv_clone_name, sizeof (recvname)) < 1917 sizeof (recvname) && 1918 dsl_dataset_hold(ds->ds_dir->dd_pool, recvname, FTAG, 1919 &recv_ds) == 0) { 1920 char *propval = get_receive_resume_stats_impl(recv_ds); 1921 dsl_dataset_rele(recv_ds, FTAG); 1922 return (propval); 1923 } 1924 return (strdup("")); 1925 } 1926 1927 static void 1928 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv) 1929 { 1930 char *propval = get_receive_resume_stats_impl(ds); 1931 if (strcmp(propval, "") != 0) { 1932 dsl_prop_nvlist_add_string(nv, 1933 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval); 1934 } else { 1935 char *childval = get_child_receive_stats(ds); 1936 if (strcmp(childval, "") != 0) { 1937 dsl_prop_nvlist_add_string(nv, 1938 ZFS_PROP_RECEIVE_RESUME_TOKEN, childval); 1939 } 1940 strfree(childval); 1941 } 1942 strfree(propval); 1943 } 1944 1945 uint64_t 1946 dsl_get_refratio(dsl_dataset_t *ds) 1947 { 1948 uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : 1949 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / 1950 dsl_dataset_phys(ds)->ds_compressed_bytes); 1951 return (ratio); 1952 } 1953 1954 uint64_t 1955 dsl_get_logicalreferenced(dsl_dataset_t *ds) 1956 { 1957 return (dsl_dataset_phys(ds)->ds_uncompressed_bytes); 1958 } 1959 1960 uint64_t 1961 dsl_get_compressratio(dsl_dataset_t *ds) 1962 { 1963 if (ds->ds_is_snapshot) { 1964 return (dsl_get_refratio(ds)); 1965 } else { 1966 dsl_dir_t *dd = ds->ds_dir; 1967 mutex_enter(&dd->dd_lock); 1968 uint64_t val = dsl_dir_get_compressratio(dd); 1969 mutex_exit(&dd->dd_lock); 1970 return (val); 1971 } 1972 } 1973 1974 uint64_t 1975 dsl_get_used(dsl_dataset_t *ds) 1976 { 1977 if (ds->ds_is_snapshot) { 1978 return (dsl_dataset_phys(ds)->ds_unique_bytes); 1979 } else { 1980 dsl_dir_t *dd = ds->ds_dir; 1981 mutex_enter(&dd->dd_lock); 1982 uint64_t val = dsl_dir_get_used(dd); 1983 mutex_exit(&dd->dd_lock); 1984 return (val); 1985 } 1986 } 1987 1988 uint64_t 1989 dsl_get_creation(dsl_dataset_t *ds) 1990 { 1991 return (dsl_dataset_phys(ds)->ds_creation_time); 1992 } 1993 1994 uint64_t 1995 dsl_get_creationtxg(dsl_dataset_t *ds) 1996 { 1997 return (dsl_dataset_phys(ds)->ds_creation_txg); 1998 } 1999 2000 uint64_t 2001 dsl_get_refquota(dsl_dataset_t *ds) 2002 { 2003 return (ds->ds_quota); 2004 } 2005 2006 uint64_t 2007 dsl_get_refreservation(dsl_dataset_t *ds) 2008 { 2009 return (ds->ds_reserved); 2010 } 2011 2012 uint64_t 2013 dsl_get_guid(dsl_dataset_t *ds) 2014 { 2015 return (dsl_dataset_phys(ds)->ds_guid); 2016 } 2017 2018 uint64_t 2019 dsl_get_unique(dsl_dataset_t *ds) 2020 { 2021 return (dsl_dataset_phys(ds)->ds_unique_bytes); 2022 } 2023 2024 uint64_t 2025 dsl_get_objsetid(dsl_dataset_t *ds) 2026 { 2027 return (ds->ds_object); 2028 } 2029 2030 uint64_t 2031 dsl_get_userrefs(dsl_dataset_t *ds) 2032 { 2033 return (ds->ds_userrefs); 2034 } 2035 2036 uint64_t 2037 dsl_get_defer_destroy(dsl_dataset_t *ds) 2038 { 2039 return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 2040 } 2041 2042 uint64_t 2043 dsl_get_referenced(dsl_dataset_t *ds) 2044 { 2045 return (dsl_dataset_phys(ds)->ds_referenced_bytes); 2046 } 2047 2048 uint64_t 2049 dsl_get_numclones(dsl_dataset_t *ds) 2050 { 2051 ASSERT(ds->ds_is_snapshot); 2052 return (dsl_dataset_phys(ds)->ds_num_children - 1); 2053 } 2054 2055 uint64_t 2056 dsl_get_inconsistent(dsl_dataset_t *ds) 2057 { 2058 return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ? 2059 1 : 0); 2060 } 2061 2062 uint64_t 2063 dsl_get_available(dsl_dataset_t *ds) 2064 { 2065 uint64_t refdbytes = dsl_get_referenced(ds); 2066 uint64_t availbytes = dsl_dir_space_available(ds->ds_dir, 2067 NULL, 0, TRUE); 2068 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 2069 availbytes += 2070 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 2071 } 2072 if (ds->ds_quota != 0) { 2073 /* 2074 * Adjust available bytes according to refquota 2075 */ 2076 if (refdbytes < ds->ds_quota) { 2077 availbytes = MIN(availbytes, 2078 ds->ds_quota - refdbytes); 2079 } else { 2080 availbytes = 0; 2081 } 2082 } 2083 return (availbytes); 2084 } 2085 2086 int 2087 dsl_get_written(dsl_dataset_t *ds, uint64_t *written) 2088 { 2089 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2090 dsl_dataset_t *prev; 2091 int err = dsl_dataset_hold_obj(dp, 2092 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 2093 if (err == 0) { 2094 uint64_t comp, uncomp; 2095 err = dsl_dataset_space_written(prev, ds, written, 2096 &comp, &uncomp); 2097 dsl_dataset_rele(prev, FTAG); 2098 } 2099 return (err); 2100 } 2101 2102 /* 2103 * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN. 2104 */ 2105 int 2106 dsl_get_prev_snap(dsl_dataset_t *ds, char *snap) 2107 { 2108 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2109 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 2110 dsl_dataset_name(ds->ds_prev, snap); 2111 return (0); 2112 } else { 2113 return (ENOENT); 2114 } 2115 } 2116 2117 /* 2118 * Returns the mountpoint property and source for the given dataset in the value 2119 * and source buffers. The value buffer must be at least as large as MAXPATHLEN 2120 * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN. 2121 * Returns 0 on success and an error on failure. 2122 */ 2123 int 2124 dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value, 2125 char *source) 2126 { 2127 int error; 2128 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2129 2130 /* Retrieve the mountpoint value stored in the zap opbject */ 2131 error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1, 2132 ZAP_MAXVALUELEN, value, source); 2133 if (error != 0) { 2134 return (error); 2135 } 2136 2137 /* 2138 * Process the dsname and source to find the full mountpoint string. 2139 * Can be skipped for 'legacy' or 'none'. 2140 */ 2141 if (value[0] == '/') { 2142 char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP); 2143 char *root = buf; 2144 const char *relpath; 2145 2146 /* 2147 * If we inherit the mountpoint, even from a dataset 2148 * with a received value, the source will be the path of 2149 * the dataset we inherit from. If source is 2150 * ZPROP_SOURCE_VAL_RECVD, the received value is not 2151 * inherited. 2152 */ 2153 if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) { 2154 relpath = ""; 2155 } else { 2156 ASSERT0(strncmp(dsname, source, strlen(source))); 2157 relpath = dsname + strlen(source); 2158 if (relpath[0] == '/') 2159 relpath++; 2160 } 2161 2162 spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN); 2163 2164 /* 2165 * Special case an alternate root of '/'. This will 2166 * avoid having multiple leading slashes in the 2167 * mountpoint path. 2168 */ 2169 if (strcmp(root, "/") == 0) 2170 root++; 2171 2172 /* 2173 * If the mountpoint is '/' then skip over this 2174 * if we are obtaining either an alternate root or 2175 * an inherited mountpoint. 2176 */ 2177 char *mnt = value; 2178 if (value[1] == '\0' && (root[0] != '\0' || 2179 relpath[0] != '\0')) 2180 mnt = value + 1; 2181 2182 if (relpath[0] == '\0') { 2183 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s", 2184 root, mnt); 2185 } else { 2186 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s", 2187 root, mnt, relpath[0] == '@' ? "" : "/", 2188 relpath); 2189 } 2190 kmem_free(buf, ZAP_MAXVALUELEN); 2191 } 2192 2193 return (0); 2194 } 2195 2196 void 2197 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 2198 { 2199 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2200 2201 ASSERT(dsl_pool_config_held(dp)); 2202 2203 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, 2204 dsl_get_refratio(ds)); 2205 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 2206 dsl_get_logicalreferenced(ds)); 2207 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 2208 dsl_get_compressratio(ds)); 2209 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 2210 dsl_get_used(ds)); 2211 2212 if (ds->ds_is_snapshot) { 2213 get_clones_stat(ds, nv); 2214 } else { 2215 char buf[ZFS_MAX_DATASET_NAME_LEN]; 2216 if (dsl_get_prev_snap(ds, buf) == 0) 2217 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, 2218 buf); 2219 dsl_dir_stats(ds->ds_dir, nv); 2220 } 2221 2222 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, 2223 dsl_get_available(ds)); 2224 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, 2225 dsl_get_referenced(ds)); 2226 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 2227 dsl_get_creation(ds)); 2228 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 2229 dsl_get_creationtxg(ds)); 2230 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 2231 dsl_get_refquota(ds)); 2232 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 2233 dsl_get_refreservation(ds)); 2234 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 2235 dsl_get_guid(ds)); 2236 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 2237 dsl_get_unique(ds)); 2238 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 2239 dsl_get_objsetid(ds)); 2240 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 2241 dsl_get_userrefs(ds)); 2242 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 2243 dsl_get_defer_destroy(ds)); 2244 2245 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 2246 uint64_t written; 2247 if (dsl_get_written(ds, &written) == 0) { 2248 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 2249 written); 2250 } 2251 } 2252 2253 if (!dsl_dataset_is_snapshot(ds)) { 2254 /* 2255 * A failed "newfs" (e.g. full) resumable receive leaves 2256 * the stats set on this dataset. Check here for the prop. 2257 */ 2258 get_receive_resume_stats(ds, nv); 2259 2260 /* 2261 * A failed incremental resumable receive leaves the 2262 * stats set on our child named "%recv". Check the child 2263 * for the prop. 2264 */ 2265 /* 6 extra bytes for /%recv */ 2266 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 2267 dsl_dataset_t *recv_ds; 2268 dsl_dataset_name(ds, recvname); 2269 if (strlcat(recvname, "/", sizeof (recvname)) < 2270 sizeof (recvname) && 2271 strlcat(recvname, recv_clone_name, sizeof (recvname)) < 2272 sizeof (recvname) && 2273 dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) { 2274 get_receive_resume_stats(recv_ds, nv); 2275 dsl_dataset_rele(recv_ds, FTAG); 2276 } 2277 } 2278 } 2279 2280 void 2281 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 2282 { 2283 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2284 ASSERT(dsl_pool_config_held(dp)); 2285 2286 stat->dds_creation_txg = dsl_get_creationtxg(ds); 2287 stat->dds_inconsistent = dsl_get_inconsistent(ds); 2288 stat->dds_guid = dsl_get_guid(ds); 2289 stat->dds_origin[0] = '\0'; 2290 if (ds->ds_is_snapshot) { 2291 stat->dds_is_snapshot = B_TRUE; 2292 stat->dds_num_clones = dsl_get_numclones(ds); 2293 } else { 2294 stat->dds_is_snapshot = B_FALSE; 2295 stat->dds_num_clones = 0; 2296 2297 if (dsl_dir_is_clone(ds->ds_dir)) { 2298 dsl_dir_get_origin(ds->ds_dir, stat->dds_origin); 2299 } 2300 } 2301 } 2302 2303 uint64_t 2304 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 2305 { 2306 return (ds->ds_fsid_guid); 2307 } 2308 2309 void 2310 dsl_dataset_space(dsl_dataset_t *ds, 2311 uint64_t *refdbytesp, uint64_t *availbytesp, 2312 uint64_t *usedobjsp, uint64_t *availobjsp) 2313 { 2314 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; 2315 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 2316 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) 2317 *availbytesp += 2318 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 2319 if (ds->ds_quota != 0) { 2320 /* 2321 * Adjust available bytes according to refquota 2322 */ 2323 if (*refdbytesp < ds->ds_quota) 2324 *availbytesp = MIN(*availbytesp, 2325 ds->ds_quota - *refdbytesp); 2326 else 2327 *availbytesp = 0; 2328 } 2329 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2330 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); 2331 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2332 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 2333 } 2334 2335 boolean_t 2336 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 2337 { 2338 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2339 uint64_t birth; 2340 2341 ASSERT(dsl_pool_config_held(dp)); 2342 if (snap == NULL) 2343 return (B_FALSE); 2344 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2345 birth = dsl_dataset_get_blkptr(ds)->blk_birth; 2346 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2347 if (birth > dsl_dataset_phys(snap)->ds_creation_txg) { 2348 objset_t *os, *os_snap; 2349 /* 2350 * It may be that only the ZIL differs, because it was 2351 * reset in the head. Don't count that as being 2352 * modified. 2353 */ 2354 if (dmu_objset_from_ds(ds, &os) != 0) 2355 return (B_TRUE); 2356 if (dmu_objset_from_ds(snap, &os_snap) != 0) 2357 return (B_TRUE); 2358 return (bcmp(&os->os_phys->os_meta_dnode, 2359 &os_snap->os_phys->os_meta_dnode, 2360 sizeof (os->os_phys->os_meta_dnode)) != 0); 2361 } 2362 return (B_FALSE); 2363 } 2364 2365 typedef struct dsl_dataset_rename_snapshot_arg { 2366 const char *ddrsa_fsname; 2367 const char *ddrsa_oldsnapname; 2368 const char *ddrsa_newsnapname; 2369 boolean_t ddrsa_recursive; 2370 dmu_tx_t *ddrsa_tx; 2371 } dsl_dataset_rename_snapshot_arg_t; 2372 2373 /* ARGSUSED */ 2374 static int 2375 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 2376 dsl_dataset_t *hds, void *arg) 2377 { 2378 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2379 int error; 2380 uint64_t val; 2381 2382 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2383 if (error != 0) { 2384 /* ignore nonexistent snapshots */ 2385 return (error == ENOENT ? 0 : error); 2386 } 2387 2388 /* new name should not exist */ 2389 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 2390 if (error == 0) 2391 error = SET_ERROR(EEXIST); 2392 else if (error == ENOENT) 2393 error = 0; 2394 2395 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2396 if (dsl_dir_namelen(hds->ds_dir) + 1 + 2397 strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN) 2398 error = SET_ERROR(ENAMETOOLONG); 2399 2400 return (error); 2401 } 2402 2403 static int 2404 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 2405 { 2406 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2407 dsl_pool_t *dp = dmu_tx_pool(tx); 2408 dsl_dataset_t *hds; 2409 int error; 2410 2411 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 2412 if (error != 0) 2413 return (error); 2414 2415 if (ddrsa->ddrsa_recursive) { 2416 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 2417 dsl_dataset_rename_snapshot_check_impl, ddrsa, 2418 DS_FIND_CHILDREN); 2419 } else { 2420 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 2421 } 2422 dsl_dataset_rele(hds, FTAG); 2423 return (error); 2424 } 2425 2426 static int 2427 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 2428 dsl_dataset_t *hds, void *arg) 2429 { 2430 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2431 dsl_dataset_t *ds; 2432 uint64_t val; 2433 dmu_tx_t *tx = ddrsa->ddrsa_tx; 2434 int error; 2435 2436 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2437 ASSERT(error == 0 || error == ENOENT); 2438 if (error == ENOENT) { 2439 /* ignore nonexistent snapshots */ 2440 return (0); 2441 } 2442 2443 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 2444 2445 /* log before we change the name */ 2446 spa_history_log_internal_ds(ds, "rename", tx, 2447 "-> @%s", ddrsa->ddrsa_newsnapname); 2448 2449 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 2450 B_FALSE)); 2451 mutex_enter(&ds->ds_lock); 2452 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 2453 mutex_exit(&ds->ds_lock); 2454 VERIFY0(zap_add(dp->dp_meta_objset, 2455 dsl_dataset_phys(hds)->ds_snapnames_zapobj, 2456 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 2457 2458 dsl_dataset_rele(ds, FTAG); 2459 return (0); 2460 } 2461 2462 static void 2463 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 2464 { 2465 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2466 dsl_pool_t *dp = dmu_tx_pool(tx); 2467 dsl_dataset_t *hds; 2468 2469 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 2470 ddrsa->ddrsa_tx = tx; 2471 if (ddrsa->ddrsa_recursive) { 2472 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 2473 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 2474 DS_FIND_CHILDREN)); 2475 } else { 2476 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 2477 } 2478 dsl_dataset_rele(hds, FTAG); 2479 } 2480 2481 int 2482 dsl_dataset_rename_snapshot(const char *fsname, 2483 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 2484 { 2485 dsl_dataset_rename_snapshot_arg_t ddrsa; 2486 2487 ddrsa.ddrsa_fsname = fsname; 2488 ddrsa.ddrsa_oldsnapname = oldsnapname; 2489 ddrsa.ddrsa_newsnapname = newsnapname; 2490 ddrsa.ddrsa_recursive = recursive; 2491 2492 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 2493 dsl_dataset_rename_snapshot_sync, &ddrsa, 2494 1, ZFS_SPACE_CHECK_RESERVED)); 2495 } 2496 2497 /* 2498 * If we're doing an ownership handoff, we need to make sure that there is 2499 * only one long hold on the dataset. We're not allowed to change anything here 2500 * so we don't permanently release the long hold or regular hold here. We want 2501 * to do this only when syncing to avoid the dataset unexpectedly going away 2502 * when we release the long hold. 2503 */ 2504 static int 2505 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 2506 { 2507 boolean_t held; 2508 2509 if (!dmu_tx_is_syncing(tx)) 2510 return (0); 2511 2512 if (owner != NULL) { 2513 VERIFY3P(ds->ds_owner, ==, owner); 2514 dsl_dataset_long_rele(ds, owner); 2515 } 2516 2517 held = dsl_dataset_long_held(ds); 2518 2519 if (owner != NULL) 2520 dsl_dataset_long_hold(ds, owner); 2521 2522 if (held) 2523 return (SET_ERROR(EBUSY)); 2524 2525 return (0); 2526 } 2527 2528 int 2529 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 2530 { 2531 dsl_dataset_rollback_arg_t *ddra = arg; 2532 dsl_pool_t *dp = dmu_tx_pool(tx); 2533 dsl_dataset_t *ds; 2534 int64_t unused_refres_delta; 2535 int error; 2536 2537 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 2538 if (error != 0) 2539 return (error); 2540 2541 /* must not be a snapshot */ 2542 if (ds->ds_is_snapshot) { 2543 dsl_dataset_rele(ds, FTAG); 2544 return (SET_ERROR(EINVAL)); 2545 } 2546 2547 /* must have a most recent snapshot */ 2548 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { 2549 dsl_dataset_rele(ds, FTAG); 2550 return (SET_ERROR(ESRCH)); 2551 } 2552 2553 /* 2554 * No rollback to a snapshot created in the current txg, because 2555 * the rollback may dirty the dataset and create blocks that are 2556 * not reachable from the rootbp while having a birth txg that 2557 * falls into the snapshot's range. 2558 */ 2559 if (dmu_tx_is_syncing(tx) && 2560 dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) { 2561 dsl_dataset_rele(ds, FTAG); 2562 return (SET_ERROR(EAGAIN)); 2563 } 2564 2565 /* 2566 * If the expected target snapshot is specified, then check that 2567 * the latest snapshot is it. 2568 */ 2569 if (ddra->ddra_tosnap != NULL) { 2570 dsl_dataset_t *snapds; 2571 2572 /* Check if the target snapshot exists at all. */ 2573 error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds); 2574 if (error != 0) { 2575 /* 2576 * ESRCH is used to signal that the target snapshot does 2577 * not exist, while ENOENT is used to report that 2578 * the rolled back dataset does not exist. 2579 * ESRCH is also used to cover other cases where the 2580 * target snapshot is not related to the dataset being 2581 * rolled back such as being in a different pool. 2582 */ 2583 if (error == ENOENT || error == EXDEV) 2584 error = SET_ERROR(ESRCH); 2585 dsl_dataset_rele(ds, FTAG); 2586 return (error); 2587 } 2588 ASSERT(snapds->ds_is_snapshot); 2589 2590 /* Check if the snapshot is the latest snapshot indeed. */ 2591 if (snapds != ds->ds_prev) { 2592 /* 2593 * Distinguish between the case where the only problem 2594 * is intervening snapshots (EEXIST) vs the snapshot 2595 * not being a valid target for rollback (ESRCH). 2596 */ 2597 if (snapds->ds_dir == ds->ds_dir || 2598 (dsl_dir_is_clone(ds->ds_dir) && 2599 dsl_dir_phys(ds->ds_dir)->dd_origin_obj == 2600 snapds->ds_object)) { 2601 error = SET_ERROR(EEXIST); 2602 } else { 2603 error = SET_ERROR(ESRCH); 2604 } 2605 dsl_dataset_rele(snapds, FTAG); 2606 dsl_dataset_rele(ds, FTAG); 2607 return (error); 2608 } 2609 dsl_dataset_rele(snapds, FTAG); 2610 } 2611 2612 /* must not have any bookmarks after the most recent snapshot */ 2613 nvlist_t *proprequest = fnvlist_alloc(); 2614 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2615 nvlist_t *bookmarks = fnvlist_alloc(); 2616 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 2617 fnvlist_free(proprequest); 2618 if (error != 0) { 2619 dsl_dataset_rele(ds, FTAG); 2620 return (error); 2621 } 2622 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 2623 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 2624 nvlist_t *valuenv = 2625 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 2626 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2627 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 2628 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 2629 fnvlist_free(bookmarks); 2630 dsl_dataset_rele(ds, FTAG); 2631 return (SET_ERROR(EEXIST)); 2632 } 2633 } 2634 fnvlist_free(bookmarks); 2635 2636 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 2637 if (error != 0) { 2638 dsl_dataset_rele(ds, FTAG); 2639 return (error); 2640 } 2641 2642 /* 2643 * Check if the snap we are rolling back to uses more than 2644 * the refquota. 2645 */ 2646 if (ds->ds_quota != 0 && 2647 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { 2648 dsl_dataset_rele(ds, FTAG); 2649 return (SET_ERROR(EDQUOT)); 2650 } 2651 2652 /* 2653 * When we do the clone swap, we will temporarily use more space 2654 * due to the refreservation (the head will no longer have any 2655 * unique space, so the entire amount of the refreservation will need 2656 * to be free). We will immediately destroy the clone, freeing 2657 * this space, but the freeing happens over many txg's. 2658 */ 2659 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 2660 dsl_dataset_phys(ds)->ds_unique_bytes); 2661 2662 if (unused_refres_delta > 0 && 2663 unused_refres_delta > 2664 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 2665 dsl_dataset_rele(ds, FTAG); 2666 return (SET_ERROR(ENOSPC)); 2667 } 2668 2669 dsl_dataset_rele(ds, FTAG); 2670 return (0); 2671 } 2672 2673 void 2674 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 2675 { 2676 dsl_dataset_rollback_arg_t *ddra = arg; 2677 dsl_pool_t *dp = dmu_tx_pool(tx); 2678 dsl_dataset_t *ds, *clone; 2679 uint64_t cloneobj; 2680 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 2681 2682 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 2683 2684 dsl_dataset_name(ds->ds_prev, namebuf); 2685 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 2686 2687 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 2688 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 2689 2690 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 2691 2692 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 2693 dsl_dataset_zero_zil(ds, tx); 2694 2695 dsl_destroy_head_sync_impl(clone, tx); 2696 2697 dsl_dataset_rele(clone, FTAG); 2698 dsl_dataset_rele(ds, FTAG); 2699 } 2700 2701 /* 2702 * Rolls back the given filesystem or volume to the most recent snapshot. 2703 * The name of the most recent snapshot will be returned under key "target" 2704 * in the result nvlist. 2705 * 2706 * If owner != NULL: 2707 * - The existing dataset MUST be owned by the specified owner at entry 2708 * - Upon return, dataset will still be held by the same owner, whether we 2709 * succeed or not. 2710 * 2711 * This mode is required any time the existing filesystem is mounted. See 2712 * notes above zfs_suspend_fs() for further details. 2713 */ 2714 int 2715 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner, 2716 nvlist_t *result) 2717 { 2718 dsl_dataset_rollback_arg_t ddra; 2719 2720 ddra.ddra_fsname = fsname; 2721 ddra.ddra_tosnap = tosnap; 2722 ddra.ddra_owner = owner; 2723 ddra.ddra_result = result; 2724 2725 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 2726 dsl_dataset_rollback_sync, &ddra, 2727 1, ZFS_SPACE_CHECK_RESERVED)); 2728 } 2729 2730 struct promotenode { 2731 list_node_t link; 2732 dsl_dataset_t *ds; 2733 }; 2734 2735 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2736 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2737 void *tag); 2738 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2739 2740 int 2741 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2742 { 2743 dsl_dataset_promote_arg_t *ddpa = arg; 2744 dsl_pool_t *dp = dmu_tx_pool(tx); 2745 dsl_dataset_t *hds; 2746 struct promotenode *snap; 2747 dsl_dataset_t *origin_ds; 2748 int err; 2749 uint64_t unused; 2750 uint64_t ss_mv_cnt; 2751 size_t max_snap_len; 2752 boolean_t conflicting_snaps; 2753 2754 err = promote_hold(ddpa, dp, FTAG); 2755 if (err != 0) 2756 return (err); 2757 2758 hds = ddpa->ddpa_clone; 2759 snap = list_head(&ddpa->shared_snaps); 2760 origin_ds = snap->ds; 2761 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; 2762 2763 snap = list_head(&ddpa->origin_snaps); 2764 2765 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { 2766 promote_rele(ddpa, FTAG); 2767 return (SET_ERROR(EXDEV)); 2768 } 2769 2770 /* 2771 * Compute and check the amount of space to transfer. Since this is 2772 * so expensive, don't do the preliminary check. 2773 */ 2774 if (!dmu_tx_is_syncing(tx)) { 2775 promote_rele(ddpa, FTAG); 2776 return (0); 2777 } 2778 2779 /* compute origin's new unique space */ 2780 snap = list_tail(&ddpa->clone_snaps); 2781 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2782 origin_ds->ds_object); 2783 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2784 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, 2785 &ddpa->unique, &unused, &unused); 2786 2787 /* 2788 * Walk the snapshots that we are moving 2789 * 2790 * Compute space to transfer. Consider the incremental changes 2791 * to used by each snapshot: 2792 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2793 * So each snapshot gave birth to: 2794 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2795 * So a sequence would look like: 2796 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2797 * Which simplifies to: 2798 * uN + kN + kN-1 + ... + k1 + k0 2799 * Note however, if we stop before we reach the ORIGIN we get: 2800 * uN + kN + kN-1 + ... + kM - uM-1 2801 */ 2802 conflicting_snaps = B_FALSE; 2803 ss_mv_cnt = 0; 2804 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; 2805 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; 2806 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; 2807 for (snap = list_head(&ddpa->shared_snaps); snap; 2808 snap = list_next(&ddpa->shared_snaps, snap)) { 2809 uint64_t val, dlused, dlcomp, dluncomp; 2810 dsl_dataset_t *ds = snap->ds; 2811 2812 ss_mv_cnt++; 2813 2814 /* 2815 * If there are long holds, we won't be able to evict 2816 * the objset. 2817 */ 2818 if (dsl_dataset_long_held(ds)) { 2819 err = SET_ERROR(EBUSY); 2820 goto out; 2821 } 2822 2823 /* Check that the snapshot name does not conflict */ 2824 VERIFY0(dsl_dataset_get_snapname(ds)); 2825 if (strlen(ds->ds_snapname) >= max_snap_len) { 2826 err = SET_ERROR(ENAMETOOLONG); 2827 goto out; 2828 } 2829 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2830 if (err == 0) { 2831 fnvlist_add_boolean(ddpa->err_ds, 2832 snap->ds->ds_snapname); 2833 conflicting_snaps = B_TRUE; 2834 } else if (err != ENOENT) { 2835 goto out; 2836 } 2837 2838 /* The very first snapshot does not have a deadlist */ 2839 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) 2840 continue; 2841 2842 dsl_deadlist_space(&ds->ds_deadlist, 2843 &dlused, &dlcomp, &dluncomp); 2844 ddpa->used += dlused; 2845 ddpa->comp += dlcomp; 2846 ddpa->uncomp += dluncomp; 2847 } 2848 2849 /* 2850 * In order to return the full list of conflicting snapshots, we check 2851 * whether there was a conflict after traversing all of them. 2852 */ 2853 if (conflicting_snaps) { 2854 err = SET_ERROR(EEXIST); 2855 goto out; 2856 } 2857 2858 /* 2859 * If we are a clone of a clone then we never reached ORIGIN, 2860 * so we need to subtract out the clone origin's used space. 2861 */ 2862 if (ddpa->origin_origin) { 2863 ddpa->used -= 2864 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; 2865 ddpa->comp -= 2866 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; 2867 ddpa->uncomp -= 2868 dsl_dataset_phys(ddpa->origin_origin)-> 2869 ds_uncompressed_bytes; 2870 } 2871 2872 /* Check that there is enough space and limit headroom here */ 2873 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2874 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2875 if (err != 0) 2876 goto out; 2877 2878 /* 2879 * Compute the amounts of space that will be used by snapshots 2880 * after the promotion (for both origin and clone). For each, 2881 * it is the amount of space that will be on all of their 2882 * deadlists (that was not born before their new origin). 2883 */ 2884 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2885 uint64_t space; 2886 2887 /* 2888 * Note, typically this will not be a clone of a clone, 2889 * so dd_origin_txg will be < TXG_INITIAL, so 2890 * these snaplist_space() -> dsl_deadlist_space_range() 2891 * calls will be fast because they do not have to 2892 * iterate over all bps. 2893 */ 2894 snap = list_head(&ddpa->origin_snaps); 2895 err = snaplist_space(&ddpa->shared_snaps, 2896 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2897 if (err != 0) 2898 goto out; 2899 2900 err = snaplist_space(&ddpa->clone_snaps, 2901 snap->ds->ds_dir->dd_origin_txg, &space); 2902 if (err != 0) 2903 goto out; 2904 ddpa->cloneusedsnap += space; 2905 } 2906 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & 2907 DD_FLAG_USED_BREAKDOWN) { 2908 err = snaplist_space(&ddpa->origin_snaps, 2909 dsl_dataset_phys(origin_ds)->ds_creation_txg, 2910 &ddpa->originusedsnap); 2911 if (err != 0) 2912 goto out; 2913 } 2914 2915 out: 2916 promote_rele(ddpa, FTAG); 2917 return (err); 2918 } 2919 2920 void 2921 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2922 { 2923 dsl_dataset_promote_arg_t *ddpa = arg; 2924 dsl_pool_t *dp = dmu_tx_pool(tx); 2925 dsl_dataset_t *hds; 2926 struct promotenode *snap; 2927 dsl_dataset_t *origin_ds; 2928 dsl_dataset_t *origin_head; 2929 dsl_dir_t *dd; 2930 dsl_dir_t *odd = NULL; 2931 uint64_t oldnext_obj; 2932 int64_t delta; 2933 2934 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2935 hds = ddpa->ddpa_clone; 2936 2937 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); 2938 2939 snap = list_head(&ddpa->shared_snaps); 2940 origin_ds = snap->ds; 2941 dd = hds->ds_dir; 2942 2943 snap = list_head(&ddpa->origin_snaps); 2944 origin_head = snap->ds; 2945 2946 /* 2947 * We need to explicitly open odd, since origin_ds's dd will be 2948 * changing. 2949 */ 2950 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2951 NULL, FTAG, &odd)); 2952 2953 /* change origin's next snap */ 2954 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2955 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; 2956 snap = list_tail(&ddpa->clone_snaps); 2957 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2958 origin_ds->ds_object); 2959 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; 2960 2961 /* change the origin's next clone */ 2962 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { 2963 dsl_dataset_remove_from_next_clones(origin_ds, 2964 snap->ds->ds_object, tx); 2965 VERIFY0(zap_add_int(dp->dp_meta_objset, 2966 dsl_dataset_phys(origin_ds)->ds_next_clones_obj, 2967 oldnext_obj, tx)); 2968 } 2969 2970 /* change origin */ 2971 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2972 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); 2973 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; 2974 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2975 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2976 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; 2977 origin_head->ds_dir->dd_origin_txg = 2978 dsl_dataset_phys(origin_ds)->ds_creation_txg; 2979 2980 /* change dd_clone entries */ 2981 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2982 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2983 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); 2984 VERIFY0(zap_add_int(dp->dp_meta_objset, 2985 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2986 hds->ds_object, tx)); 2987 2988 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2989 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2990 origin_head->ds_object, tx)); 2991 if (dsl_dir_phys(dd)->dd_clones == 0) { 2992 dsl_dir_phys(dd)->dd_clones = 2993 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, 2994 DMU_OT_NONE, 0, tx); 2995 } 2996 VERIFY0(zap_add_int(dp->dp_meta_objset, 2997 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); 2998 } 2999 3000 /* move snapshots to this dir */ 3001 for (snap = list_head(&ddpa->shared_snaps); snap; 3002 snap = list_next(&ddpa->shared_snaps, snap)) { 3003 dsl_dataset_t *ds = snap->ds; 3004 3005 /* 3006 * Property callbacks are registered to a particular 3007 * dsl_dir. Since ours is changing, evict the objset 3008 * so that they will be unregistered from the old dsl_dir. 3009 */ 3010 if (ds->ds_objset) { 3011 dmu_objset_evict(ds->ds_objset); 3012 ds->ds_objset = NULL; 3013 } 3014 3015 /* move snap name entry */ 3016 VERIFY0(dsl_dataset_get_snapname(ds)); 3017 VERIFY0(dsl_dataset_snap_remove(origin_head, 3018 ds->ds_snapname, tx, B_TRUE)); 3019 VERIFY0(zap_add(dp->dp_meta_objset, 3020 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 3021 8, 1, &ds->ds_object, tx)); 3022 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 3023 DD_FIELD_SNAPSHOT_COUNT, tx); 3024 3025 /* change containing dsl_dir */ 3026 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3027 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); 3028 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; 3029 ASSERT3P(ds->ds_dir, ==, odd); 3030 dsl_dir_rele(ds->ds_dir, ds); 3031 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 3032 NULL, ds, &ds->ds_dir)); 3033 3034 /* move any clone references */ 3035 if (dsl_dataset_phys(ds)->ds_next_clones_obj && 3036 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 3037 zap_cursor_t zc; 3038 zap_attribute_t za; 3039 3040 for (zap_cursor_init(&zc, dp->dp_meta_objset, 3041 dsl_dataset_phys(ds)->ds_next_clones_obj); 3042 zap_cursor_retrieve(&zc, &za) == 0; 3043 zap_cursor_advance(&zc)) { 3044 dsl_dataset_t *cnds; 3045 uint64_t o; 3046 3047 if (za.za_first_integer == oldnext_obj) { 3048 /* 3049 * We've already moved the 3050 * origin's reference. 3051 */ 3052 continue; 3053 } 3054 3055 VERIFY0(dsl_dataset_hold_obj(dp, 3056 za.za_first_integer, FTAG, &cnds)); 3057 o = dsl_dir_phys(cnds->ds_dir)-> 3058 dd_head_dataset_obj; 3059 3060 VERIFY0(zap_remove_int(dp->dp_meta_objset, 3061 dsl_dir_phys(odd)->dd_clones, o, tx)); 3062 VERIFY0(zap_add_int(dp->dp_meta_objset, 3063 dsl_dir_phys(dd)->dd_clones, o, tx)); 3064 dsl_dataset_rele(cnds, FTAG); 3065 } 3066 zap_cursor_fini(&zc); 3067 } 3068 3069 ASSERT(!dsl_prop_hascb(ds)); 3070 } 3071 3072 /* 3073 * Change space accounting. 3074 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 3075 * both be valid, or both be 0 (resulting in delta == 0). This 3076 * is true for each of {clone,origin} independently. 3077 */ 3078 3079 delta = ddpa->cloneusedsnap - 3080 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; 3081 ASSERT3S(delta, >=, 0); 3082 ASSERT3U(ddpa->used, >=, delta); 3083 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 3084 dsl_dir_diduse_space(dd, DD_USED_HEAD, 3085 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 3086 3087 delta = ddpa->originusedsnap - 3088 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; 3089 ASSERT3S(delta, <=, 0); 3090 ASSERT3U(ddpa->used, >=, -delta); 3091 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 3092 dsl_dir_diduse_space(odd, DD_USED_HEAD, 3093 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 3094 3095 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; 3096 3097 /* log history record */ 3098 spa_history_log_internal_ds(hds, "promote", tx, ""); 3099 3100 dsl_dir_rele(odd, FTAG); 3101 promote_rele(ddpa, FTAG); 3102 } 3103 3104 /* 3105 * Make a list of dsl_dataset_t's for the snapshots between first_obj 3106 * (exclusive) and last_obj (inclusive). The list will be in reverse 3107 * order (last_obj will be the list_head()). If first_obj == 0, do all 3108 * snapshots back to this dataset's origin. 3109 */ 3110 static int 3111 snaplist_make(dsl_pool_t *dp, 3112 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 3113 { 3114 uint64_t obj = last_obj; 3115 3116 list_create(l, sizeof (struct promotenode), 3117 offsetof(struct promotenode, link)); 3118 3119 while (obj != first_obj) { 3120 dsl_dataset_t *ds; 3121 struct promotenode *snap; 3122 int err; 3123 3124 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 3125 ASSERT(err != ENOENT); 3126 if (err != 0) 3127 return (err); 3128 3129 if (first_obj == 0) 3130 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; 3131 3132 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 3133 snap->ds = ds; 3134 list_insert_tail(l, snap); 3135 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3136 } 3137 3138 return (0); 3139 } 3140 3141 static int 3142 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 3143 { 3144 struct promotenode *snap; 3145 3146 *spacep = 0; 3147 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 3148 uint64_t used, comp, uncomp; 3149 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 3150 mintxg, UINT64_MAX, &used, &comp, &uncomp); 3151 *spacep += used; 3152 } 3153 return (0); 3154 } 3155 3156 static void 3157 snaplist_destroy(list_t *l, void *tag) 3158 { 3159 struct promotenode *snap; 3160 3161 if (l == NULL || !list_link_active(&l->list_head)) 3162 return; 3163 3164 while ((snap = list_tail(l)) != NULL) { 3165 list_remove(l, snap); 3166 dsl_dataset_rele(snap->ds, tag); 3167 kmem_free(snap, sizeof (*snap)); 3168 } 3169 list_destroy(l); 3170 } 3171 3172 static int 3173 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 3174 { 3175 int error; 3176 dsl_dir_t *dd; 3177 struct promotenode *snap; 3178 3179 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 3180 &ddpa->ddpa_clone); 3181 if (error != 0) 3182 return (error); 3183 dd = ddpa->ddpa_clone->ds_dir; 3184 3185 if (ddpa->ddpa_clone->ds_is_snapshot || 3186 !dsl_dir_is_clone(dd)) { 3187 dsl_dataset_rele(ddpa->ddpa_clone, tag); 3188 return (SET_ERROR(EINVAL)); 3189 } 3190 3191 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, 3192 &ddpa->shared_snaps, tag); 3193 if (error != 0) 3194 goto out; 3195 3196 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 3197 &ddpa->clone_snaps, tag); 3198 if (error != 0) 3199 goto out; 3200 3201 snap = list_head(&ddpa->shared_snaps); 3202 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); 3203 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, 3204 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, 3205 &ddpa->origin_snaps, tag); 3206 if (error != 0) 3207 goto out; 3208 3209 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { 3210 error = dsl_dataset_hold_obj(dp, 3211 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, 3212 tag, &ddpa->origin_origin); 3213 if (error != 0) 3214 goto out; 3215 } 3216 out: 3217 if (error != 0) 3218 promote_rele(ddpa, tag); 3219 return (error); 3220 } 3221 3222 static void 3223 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 3224 { 3225 snaplist_destroy(&ddpa->shared_snaps, tag); 3226 snaplist_destroy(&ddpa->clone_snaps, tag); 3227 snaplist_destroy(&ddpa->origin_snaps, tag); 3228 if (ddpa->origin_origin != NULL) 3229 dsl_dataset_rele(ddpa->origin_origin, tag); 3230 dsl_dataset_rele(ddpa->ddpa_clone, tag); 3231 } 3232 3233 /* 3234 * Promote a clone. 3235 * 3236 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 3237 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.) 3238 */ 3239 int 3240 dsl_dataset_promote(const char *name, char *conflsnap) 3241 { 3242 dsl_dataset_promote_arg_t ddpa = { 0 }; 3243 uint64_t numsnaps; 3244 int error; 3245 nvpair_t *snap_pair; 3246 objset_t *os; 3247 3248 /* 3249 * We will modify space proportional to the number of 3250 * snapshots. Compute numsnaps. 3251 */ 3252 error = dmu_objset_hold(name, FTAG, &os); 3253 if (error != 0) 3254 return (error); 3255 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 3256 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, 3257 &numsnaps); 3258 dmu_objset_rele(os, FTAG); 3259 if (error != 0) 3260 return (error); 3261 3262 ddpa.ddpa_clonename = name; 3263 ddpa.err_ds = fnvlist_alloc(); 3264 ddpa.cr = CRED(); 3265 3266 error = dsl_sync_task(name, dsl_dataset_promote_check, 3267 dsl_dataset_promote_sync, &ddpa, 3268 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED); 3269 3270 /* 3271 * Return the first conflicting snapshot found. 3272 */ 3273 snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL); 3274 if (snap_pair != NULL && conflsnap != NULL) 3275 (void) strcpy(conflsnap, nvpair_name(snap_pair)); 3276 3277 fnvlist_free(ddpa.err_ds); 3278 return (error); 3279 } 3280 3281 int 3282 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 3283 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 3284 { 3285 /* 3286 * "slack" factor for received datasets with refquota set on them. 3287 * See the bottom of this function for details on its use. 3288 */ 3289 uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation; 3290 int64_t unused_refres_delta; 3291 3292 /* they should both be heads */ 3293 if (clone->ds_is_snapshot || 3294 origin_head->ds_is_snapshot) 3295 return (SET_ERROR(EINVAL)); 3296 3297 /* if we are not forcing, the branch point should be just before them */ 3298 if (!force && clone->ds_prev != origin_head->ds_prev) 3299 return (SET_ERROR(EINVAL)); 3300 3301 /* clone should be the clone (unless they are unrelated) */ 3302 if (clone->ds_prev != NULL && 3303 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 3304 origin_head->ds_dir != clone->ds_prev->ds_dir) 3305 return (SET_ERROR(EINVAL)); 3306 3307 /* the clone should be a child of the origin */ 3308 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 3309 return (SET_ERROR(EINVAL)); 3310 3311 /* origin_head shouldn't be modified unless 'force' */ 3312 if (!force && 3313 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 3314 return (SET_ERROR(ETXTBSY)); 3315 3316 /* origin_head should have no long holds (e.g. is not mounted) */ 3317 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 3318 return (SET_ERROR(EBUSY)); 3319 3320 /* check amount of any unconsumed refreservation */ 3321 unused_refres_delta = 3322 (int64_t)MIN(origin_head->ds_reserved, 3323 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 3324 (int64_t)MIN(origin_head->ds_reserved, 3325 dsl_dataset_phys(clone)->ds_unique_bytes); 3326 3327 if (unused_refres_delta > 0 && 3328 unused_refres_delta > 3329 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 3330 return (SET_ERROR(ENOSPC)); 3331 3332 /* 3333 * The clone can't be too much over the head's refquota. 3334 * 3335 * To ensure that the entire refquota can be used, we allow one 3336 * transaction to exceed the the refquota. Therefore, this check 3337 * needs to also allow for the space referenced to be more than the 3338 * refquota. The maximum amount of space that one transaction can use 3339 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this 3340 * overage ensures that we are able to receive a filesystem that 3341 * exceeds the refquota on the source system. 3342 * 3343 * So that overage is the refquota_slack we use below. 3344 */ 3345 if (origin_head->ds_quota != 0 && 3346 dsl_dataset_phys(clone)->ds_referenced_bytes > 3347 origin_head->ds_quota + refquota_slack) 3348 return (SET_ERROR(EDQUOT)); 3349 3350 return (0); 3351 } 3352 3353 static void 3354 dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone, 3355 dsl_dataset_t *origin, dmu_tx_t *tx) 3356 { 3357 uint64_t clone_remap_dl_obj, origin_remap_dl_obj; 3358 dsl_pool_t *dp = dmu_tx_pool(tx); 3359 3360 ASSERT(dsl_pool_sync_context(dp)); 3361 3362 clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone); 3363 origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin); 3364 3365 if (clone_remap_dl_obj != 0) { 3366 dsl_deadlist_close(&clone->ds_remap_deadlist); 3367 dsl_dataset_unset_remap_deadlist_object(clone, tx); 3368 } 3369 if (origin_remap_dl_obj != 0) { 3370 dsl_deadlist_close(&origin->ds_remap_deadlist); 3371 dsl_dataset_unset_remap_deadlist_object(origin, tx); 3372 } 3373 3374 if (clone_remap_dl_obj != 0) { 3375 dsl_dataset_set_remap_deadlist_object(origin, 3376 clone_remap_dl_obj, tx); 3377 dsl_deadlist_open(&origin->ds_remap_deadlist, 3378 dp->dp_meta_objset, clone_remap_dl_obj); 3379 } 3380 if (origin_remap_dl_obj != 0) { 3381 dsl_dataset_set_remap_deadlist_object(clone, 3382 origin_remap_dl_obj, tx); 3383 dsl_deadlist_open(&clone->ds_remap_deadlist, 3384 dp->dp_meta_objset, origin_remap_dl_obj); 3385 } 3386 } 3387 3388 void 3389 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 3390 dsl_dataset_t *origin_head, dmu_tx_t *tx) 3391 { 3392 dsl_pool_t *dp = dmu_tx_pool(tx); 3393 int64_t unused_refres_delta; 3394 3395 ASSERT(clone->ds_reserved == 0); 3396 /* 3397 * NOTE: On DEBUG kernels there could be a race between this and 3398 * the check function if spa_asize_inflation is adjusted... 3399 */ 3400 ASSERT(origin_head->ds_quota == 0 || 3401 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota + 3402 DMU_MAX_ACCESS * spa_asize_inflation); 3403 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 3404 3405 /* 3406 * Swap per-dataset feature flags. 3407 */ 3408 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 3409 if (!(spa_feature_table[f].fi_flags & 3410 ZFEATURE_FLAG_PER_DATASET)) { 3411 ASSERT(!clone->ds_feature_inuse[f]); 3412 ASSERT(!origin_head->ds_feature_inuse[f]); 3413 continue; 3414 } 3415 3416 boolean_t clone_inuse = clone->ds_feature_inuse[f]; 3417 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f]; 3418 3419 if (clone_inuse) { 3420 dsl_dataset_deactivate_feature(clone->ds_object, f, tx); 3421 clone->ds_feature_inuse[f] = B_FALSE; 3422 } 3423 if (origin_head_inuse) { 3424 dsl_dataset_deactivate_feature(origin_head->ds_object, 3425 f, tx); 3426 origin_head->ds_feature_inuse[f] = B_FALSE; 3427 } 3428 if (clone_inuse) { 3429 dsl_dataset_activate_feature(origin_head->ds_object, 3430 f, tx); 3431 origin_head->ds_feature_inuse[f] = B_TRUE; 3432 } 3433 if (origin_head_inuse) { 3434 dsl_dataset_activate_feature(clone->ds_object, f, tx); 3435 clone->ds_feature_inuse[f] = B_TRUE; 3436 } 3437 } 3438 3439 dmu_buf_will_dirty(clone->ds_dbuf, tx); 3440 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 3441 3442 if (clone->ds_objset != NULL) { 3443 dmu_objset_evict(clone->ds_objset); 3444 clone->ds_objset = NULL; 3445 } 3446 3447 if (origin_head->ds_objset != NULL) { 3448 dmu_objset_evict(origin_head->ds_objset); 3449 origin_head->ds_objset = NULL; 3450 } 3451 3452 unused_refres_delta = 3453 (int64_t)MIN(origin_head->ds_reserved, 3454 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 3455 (int64_t)MIN(origin_head->ds_reserved, 3456 dsl_dataset_phys(clone)->ds_unique_bytes); 3457 3458 /* 3459 * Reset origin's unique bytes, if it exists. 3460 */ 3461 if (clone->ds_prev) { 3462 dsl_dataset_t *origin = clone->ds_prev; 3463 uint64_t comp, uncomp; 3464 3465 dmu_buf_will_dirty(origin->ds_dbuf, tx); 3466 dsl_deadlist_space_range(&clone->ds_deadlist, 3467 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, 3468 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); 3469 } 3470 3471 /* swap blkptrs */ 3472 { 3473 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG); 3474 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG); 3475 blkptr_t tmp; 3476 tmp = dsl_dataset_phys(origin_head)->ds_bp; 3477 dsl_dataset_phys(origin_head)->ds_bp = 3478 dsl_dataset_phys(clone)->ds_bp; 3479 dsl_dataset_phys(clone)->ds_bp = tmp; 3480 rrw_exit(&origin_head->ds_bp_rwlock, FTAG); 3481 rrw_exit(&clone->ds_bp_rwlock, FTAG); 3482 } 3483 3484 /* set dd_*_bytes */ 3485 { 3486 int64_t dused, dcomp, duncomp; 3487 uint64_t cdl_used, cdl_comp, cdl_uncomp; 3488 uint64_t odl_used, odl_comp, odl_uncomp; 3489 3490 ASSERT3U(dsl_dir_phys(clone->ds_dir)-> 3491 dd_used_breakdown[DD_USED_SNAP], ==, 0); 3492 3493 dsl_deadlist_space(&clone->ds_deadlist, 3494 &cdl_used, &cdl_comp, &cdl_uncomp); 3495 dsl_deadlist_space(&origin_head->ds_deadlist, 3496 &odl_used, &odl_comp, &odl_uncomp); 3497 3498 dused = dsl_dataset_phys(clone)->ds_referenced_bytes + 3499 cdl_used - 3500 (dsl_dataset_phys(origin_head)->ds_referenced_bytes + 3501 odl_used); 3502 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + 3503 cdl_comp - 3504 (dsl_dataset_phys(origin_head)->ds_compressed_bytes + 3505 odl_comp); 3506 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + 3507 cdl_uncomp - 3508 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + 3509 odl_uncomp); 3510 3511 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 3512 dused, dcomp, duncomp, tx); 3513 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 3514 -dused, -dcomp, -duncomp, tx); 3515 3516 /* 3517 * The difference in the space used by snapshots is the 3518 * difference in snapshot space due to the head's 3519 * deadlist (since that's the only thing that's 3520 * changing that affects the snapused). 3521 */ 3522 dsl_deadlist_space_range(&clone->ds_deadlist, 3523 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 3524 &cdl_used, &cdl_comp, &cdl_uncomp); 3525 dsl_deadlist_space_range(&origin_head->ds_deadlist, 3526 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 3527 &odl_used, &odl_comp, &odl_uncomp); 3528 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 3529 DD_USED_HEAD, DD_USED_SNAP, tx); 3530 } 3531 3532 /* swap ds_*_bytes */ 3533 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, 3534 dsl_dataset_phys(clone)->ds_referenced_bytes); 3535 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, 3536 dsl_dataset_phys(clone)->ds_compressed_bytes); 3537 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, 3538 dsl_dataset_phys(clone)->ds_uncompressed_bytes); 3539 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, 3540 dsl_dataset_phys(clone)->ds_unique_bytes); 3541 3542 /* apply any parent delta for change in unconsumed refreservation */ 3543 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 3544 unused_refres_delta, 0, 0, tx); 3545 3546 /* 3547 * Swap deadlists. 3548 */ 3549 dsl_deadlist_close(&clone->ds_deadlist); 3550 dsl_deadlist_close(&origin_head->ds_deadlist); 3551 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, 3552 dsl_dataset_phys(clone)->ds_deadlist_obj); 3553 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 3554 dsl_dataset_phys(clone)->ds_deadlist_obj); 3555 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 3556 dsl_dataset_phys(origin_head)->ds_deadlist_obj); 3557 dsl_dataset_swap_remap_deadlists(clone, origin_head, tx); 3558 3559 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 3560 3561 spa_history_log_internal_ds(clone, "clone swap", tx, 3562 "parent=%s", origin_head->ds_dir->dd_myname); 3563 } 3564 3565 /* 3566 * Given a pool name and a dataset object number in that pool, 3567 * return the name of that dataset. 3568 */ 3569 int 3570 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 3571 { 3572 dsl_pool_t *dp; 3573 dsl_dataset_t *ds; 3574 int error; 3575 3576 error = dsl_pool_hold(pname, FTAG, &dp); 3577 if (error != 0) 3578 return (error); 3579 3580 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 3581 if (error == 0) { 3582 dsl_dataset_name(ds, buf); 3583 dsl_dataset_rele(ds, FTAG); 3584 } 3585 dsl_pool_rele(dp, FTAG); 3586 3587 return (error); 3588 } 3589 3590 int 3591 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 3592 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 3593 { 3594 int error = 0; 3595 3596 ASSERT3S(asize, >, 0); 3597 3598 /* 3599 * *ref_rsrv is the portion of asize that will come from any 3600 * unconsumed refreservation space. 3601 */ 3602 *ref_rsrv = 0; 3603 3604 mutex_enter(&ds->ds_lock); 3605 /* 3606 * Make a space adjustment for reserved bytes. 3607 */ 3608 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 3609 ASSERT3U(*used, >=, 3610 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 3611 *used -= 3612 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 3613 *ref_rsrv = 3614 asize - MIN(asize, parent_delta(ds, asize + inflight)); 3615 } 3616 3617 if (!check_quota || ds->ds_quota == 0) { 3618 mutex_exit(&ds->ds_lock); 3619 return (0); 3620 } 3621 /* 3622 * If they are requesting more space, and our current estimate 3623 * is over quota, they get to try again unless the actual 3624 * on-disk is over quota and there are no pending changes (which 3625 * may free up space for us). 3626 */ 3627 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= 3628 ds->ds_quota) { 3629 if (inflight > 0 || 3630 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) 3631 error = SET_ERROR(ERESTART); 3632 else 3633 error = SET_ERROR(EDQUOT); 3634 } 3635 mutex_exit(&ds->ds_lock); 3636 3637 return (error); 3638 } 3639 3640 typedef struct dsl_dataset_set_qr_arg { 3641 const char *ddsqra_name; 3642 zprop_source_t ddsqra_source; 3643 uint64_t ddsqra_value; 3644 } dsl_dataset_set_qr_arg_t; 3645 3646 3647 /* ARGSUSED */ 3648 static int 3649 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 3650 { 3651 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3652 dsl_pool_t *dp = dmu_tx_pool(tx); 3653 dsl_dataset_t *ds; 3654 int error; 3655 uint64_t newval; 3656 3657 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 3658 return (SET_ERROR(ENOTSUP)); 3659 3660 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3661 if (error != 0) 3662 return (error); 3663 3664 if (ds->ds_is_snapshot) { 3665 dsl_dataset_rele(ds, FTAG); 3666 return (SET_ERROR(EINVAL)); 3667 } 3668 3669 error = dsl_prop_predict(ds->ds_dir, 3670 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3671 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3672 if (error != 0) { 3673 dsl_dataset_rele(ds, FTAG); 3674 return (error); 3675 } 3676 3677 if (newval == 0) { 3678 dsl_dataset_rele(ds, FTAG); 3679 return (0); 3680 } 3681 3682 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || 3683 newval < ds->ds_reserved) { 3684 dsl_dataset_rele(ds, FTAG); 3685 return (SET_ERROR(ENOSPC)); 3686 } 3687 3688 dsl_dataset_rele(ds, FTAG); 3689 return (0); 3690 } 3691 3692 static void 3693 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 3694 { 3695 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3696 dsl_pool_t *dp = dmu_tx_pool(tx); 3697 dsl_dataset_t *ds; 3698 uint64_t newval; 3699 3700 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3701 3702 dsl_prop_set_sync_impl(ds, 3703 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3704 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 3705 &ddsqra->ddsqra_value, tx); 3706 3707 VERIFY0(dsl_prop_get_int_ds(ds, 3708 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 3709 3710 if (ds->ds_quota != newval) { 3711 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3712 ds->ds_quota = newval; 3713 } 3714 dsl_dataset_rele(ds, FTAG); 3715 } 3716 3717 int 3718 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 3719 uint64_t refquota) 3720 { 3721 dsl_dataset_set_qr_arg_t ddsqra; 3722 3723 ddsqra.ddsqra_name = dsname; 3724 ddsqra.ddsqra_source = source; 3725 ddsqra.ddsqra_value = refquota; 3726 3727 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 3728 dsl_dataset_set_refquota_sync, &ddsqra, 0, 3729 ZFS_SPACE_CHECK_EXTRA_RESERVED)); 3730 } 3731 3732 static int 3733 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 3734 { 3735 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3736 dsl_pool_t *dp = dmu_tx_pool(tx); 3737 dsl_dataset_t *ds; 3738 int error; 3739 uint64_t newval, unique; 3740 3741 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 3742 return (SET_ERROR(ENOTSUP)); 3743 3744 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3745 if (error != 0) 3746 return (error); 3747 3748 if (ds->ds_is_snapshot) { 3749 dsl_dataset_rele(ds, FTAG); 3750 return (SET_ERROR(EINVAL)); 3751 } 3752 3753 error = dsl_prop_predict(ds->ds_dir, 3754 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3755 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3756 if (error != 0) { 3757 dsl_dataset_rele(ds, FTAG); 3758 return (error); 3759 } 3760 3761 /* 3762 * If we are doing the preliminary check in open context, the 3763 * space estimates may be inaccurate. 3764 */ 3765 if (!dmu_tx_is_syncing(tx)) { 3766 dsl_dataset_rele(ds, FTAG); 3767 return (0); 3768 } 3769 3770 mutex_enter(&ds->ds_lock); 3771 if (!DS_UNIQUE_IS_ACCURATE(ds)) 3772 dsl_dataset_recalc_head_uniq(ds); 3773 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3774 mutex_exit(&ds->ds_lock); 3775 3776 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 3777 uint64_t delta = MAX(unique, newval) - 3778 MAX(unique, ds->ds_reserved); 3779 3780 if (delta > 3781 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 3782 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 3783 dsl_dataset_rele(ds, FTAG); 3784 return (SET_ERROR(ENOSPC)); 3785 } 3786 } 3787 3788 dsl_dataset_rele(ds, FTAG); 3789 return (0); 3790 } 3791 3792 void 3793 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 3794 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 3795 { 3796 uint64_t newval; 3797 uint64_t unique; 3798 int64_t delta; 3799 3800 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3801 source, sizeof (value), 1, &value, tx); 3802 3803 VERIFY0(dsl_prop_get_int_ds(ds, 3804 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 3805 3806 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3807 mutex_enter(&ds->ds_dir->dd_lock); 3808 mutex_enter(&ds->ds_lock); 3809 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3810 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3811 delta = MAX(0, (int64_t)(newval - unique)) - 3812 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3813 ds->ds_reserved = newval; 3814 mutex_exit(&ds->ds_lock); 3815 3816 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3817 mutex_exit(&ds->ds_dir->dd_lock); 3818 } 3819 3820 static void 3821 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 3822 { 3823 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3824 dsl_pool_t *dp = dmu_tx_pool(tx); 3825 dsl_dataset_t *ds; 3826 3827 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3828 dsl_dataset_set_refreservation_sync_impl(ds, 3829 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 3830 dsl_dataset_rele(ds, FTAG); 3831 } 3832 3833 int 3834 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 3835 uint64_t refreservation) 3836 { 3837 dsl_dataset_set_qr_arg_t ddsqra; 3838 3839 ddsqra.ddsqra_name = dsname; 3840 ddsqra.ddsqra_source = source; 3841 ddsqra.ddsqra_value = refreservation; 3842 3843 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 3844 dsl_dataset_set_refreservation_sync, &ddsqra, 0, 3845 ZFS_SPACE_CHECK_EXTRA_RESERVED)); 3846 } 3847 3848 /* 3849 * Return (in *usedp) the amount of space written in new that is not 3850 * present in oldsnap. New may be a snapshot or the head. Old must be 3851 * a snapshot before new, in new's filesystem (or its origin). If not then 3852 * fail and return EINVAL. 3853 * 3854 * The written space is calculated by considering two components: First, we 3855 * ignore any freed space, and calculate the written as new's used space 3856 * minus old's used space. Next, we add in the amount of space that was freed 3857 * between the two snapshots, thus reducing new's used space relative to old's. 3858 * Specifically, this is the space that was born before old->ds_creation_txg, 3859 * and freed before new (ie. on new's deadlist or a previous deadlist). 3860 * 3861 * space freed [---------------------] 3862 * snapshots ---O-------O--------O-------O------ 3863 * oldsnap new 3864 */ 3865 int 3866 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 3867 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3868 { 3869 int err = 0; 3870 uint64_t snapobj; 3871 dsl_pool_t *dp = new->ds_dir->dd_pool; 3872 3873 ASSERT(dsl_pool_config_held(dp)); 3874 3875 *usedp = 0; 3876 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; 3877 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes; 3878 3879 *compp = 0; 3880 *compp += dsl_dataset_phys(new)->ds_compressed_bytes; 3881 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes; 3882 3883 *uncompp = 0; 3884 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; 3885 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes; 3886 3887 snapobj = new->ds_object; 3888 while (snapobj != oldsnap->ds_object) { 3889 dsl_dataset_t *snap; 3890 uint64_t used, comp, uncomp; 3891 3892 if (snapobj == new->ds_object) { 3893 snap = new; 3894 } else { 3895 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3896 if (err != 0) 3897 break; 3898 } 3899 3900 if (dsl_dataset_phys(snap)->ds_prev_snap_txg == 3901 dsl_dataset_phys(oldsnap)->ds_creation_txg) { 3902 /* 3903 * The blocks in the deadlist can not be born after 3904 * ds_prev_snap_txg, so get the whole deadlist space, 3905 * which is more efficient (especially for old-format 3906 * deadlists). Unfortunately the deadlist code 3907 * doesn't have enough information to make this 3908 * optimization itself. 3909 */ 3910 dsl_deadlist_space(&snap->ds_deadlist, 3911 &used, &comp, &uncomp); 3912 } else { 3913 dsl_deadlist_space_range(&snap->ds_deadlist, 3914 0, dsl_dataset_phys(oldsnap)->ds_creation_txg, 3915 &used, &comp, &uncomp); 3916 } 3917 *usedp += used; 3918 *compp += comp; 3919 *uncompp += uncomp; 3920 3921 /* 3922 * If we get to the beginning of the chain of snapshots 3923 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3924 * was not a snapshot of/before new. 3925 */ 3926 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 3927 if (snap != new) 3928 dsl_dataset_rele(snap, FTAG); 3929 if (snapobj == 0) { 3930 err = SET_ERROR(EINVAL); 3931 break; 3932 } 3933 3934 } 3935 return (err); 3936 } 3937 3938 /* 3939 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3940 * lastsnap, and all snapshots in between are deleted. 3941 * 3942 * blocks that would be freed [---------------------------] 3943 * snapshots ---O-------O--------O-------O--------O 3944 * firstsnap lastsnap 3945 * 3946 * This is the set of blocks that were born after the snap before firstsnap, 3947 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3948 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3949 * We calculate this by iterating over the relevant deadlists (from the snap 3950 * after lastsnap, backward to the snap after firstsnap), summing up the 3951 * space on the deadlist that was born after the snap before firstsnap. 3952 */ 3953 int 3954 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3955 dsl_dataset_t *lastsnap, 3956 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3957 { 3958 int err = 0; 3959 uint64_t snapobj; 3960 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3961 3962 ASSERT(firstsnap->ds_is_snapshot); 3963 ASSERT(lastsnap->ds_is_snapshot); 3964 3965 /* 3966 * Check that the snapshots are in the same dsl_dir, and firstsnap 3967 * is before lastsnap. 3968 */ 3969 if (firstsnap->ds_dir != lastsnap->ds_dir || 3970 dsl_dataset_phys(firstsnap)->ds_creation_txg > 3971 dsl_dataset_phys(lastsnap)->ds_creation_txg) 3972 return (SET_ERROR(EINVAL)); 3973 3974 *usedp = *compp = *uncompp = 0; 3975 3976 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; 3977 while (snapobj != firstsnap->ds_object) { 3978 dsl_dataset_t *ds; 3979 uint64_t used, comp, uncomp; 3980 3981 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3982 if (err != 0) 3983 break; 3984 3985 dsl_deadlist_space_range(&ds->ds_deadlist, 3986 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, 3987 &used, &comp, &uncomp); 3988 *usedp += used; 3989 *compp += comp; 3990 *uncompp += uncomp; 3991 3992 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3993 ASSERT3U(snapobj, !=, 0); 3994 dsl_dataset_rele(ds, FTAG); 3995 } 3996 return (err); 3997 } 3998 3999 /* 4000 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 4001 * For example, they could both be snapshots of the same filesystem, and 4002 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 4003 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 4004 * filesystem. Or 'earlier' could be the origin's origin. 4005 * 4006 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 4007 */ 4008 boolean_t 4009 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 4010 uint64_t earlier_txg) 4011 { 4012 dsl_pool_t *dp = later->ds_dir->dd_pool; 4013 int error; 4014 boolean_t ret; 4015 4016 ASSERT(dsl_pool_config_held(dp)); 4017 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); 4018 4019 if (earlier_txg == 0) 4020 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; 4021 4022 if (later->ds_is_snapshot && 4023 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) 4024 return (B_FALSE); 4025 4026 if (later->ds_dir == earlier->ds_dir) 4027 return (B_TRUE); 4028 if (!dsl_dir_is_clone(later->ds_dir)) 4029 return (B_FALSE); 4030 4031 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object) 4032 return (B_TRUE); 4033 dsl_dataset_t *origin; 4034 error = dsl_dataset_hold_obj(dp, 4035 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); 4036 if (error != 0) 4037 return (B_FALSE); 4038 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 4039 dsl_dataset_rele(origin, FTAG); 4040 return (ret); 4041 } 4042 4043 void 4044 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 4045 { 4046 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 4047 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 4048 } 4049 4050 boolean_t 4051 dsl_dataset_is_zapified(dsl_dataset_t *ds) 4052 { 4053 dmu_object_info_t doi; 4054 4055 dmu_object_info_from_db(ds->ds_dbuf, &doi); 4056 return (doi.doi_type == DMU_OTN_ZAP_METADATA); 4057 } 4058 4059 boolean_t 4060 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds) 4061 { 4062 return (dsl_dataset_is_zapified(ds) && 4063 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset, 4064 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0); 4065 } 4066 4067 uint64_t 4068 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds) 4069 { 4070 uint64_t remap_deadlist_obj; 4071 int err; 4072 4073 if (!dsl_dataset_is_zapified(ds)) 4074 return (0); 4075 4076 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, 4077 DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1, 4078 &remap_deadlist_obj); 4079 4080 if (err != 0) { 4081 VERIFY3S(err, ==, ENOENT); 4082 return (0); 4083 } 4084 4085 ASSERT(remap_deadlist_obj != 0); 4086 return (remap_deadlist_obj); 4087 } 4088 4089 boolean_t 4090 dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds) 4091 { 4092 EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist), 4093 dsl_dataset_get_remap_deadlist_object(ds) != 0); 4094 return (dsl_deadlist_is_open(&ds->ds_remap_deadlist)); 4095 } 4096 4097 static void 4098 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj, 4099 dmu_tx_t *tx) 4100 { 4101 ASSERT(obj != 0); 4102 dsl_dataset_zapify(ds, tx); 4103 VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object, 4104 DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx)); 4105 } 4106 4107 static void 4108 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx) 4109 { 4110 VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset, 4111 ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx)); 4112 } 4113 4114 void 4115 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx) 4116 { 4117 uint64_t remap_deadlist_object; 4118 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 4119 4120 ASSERT(dmu_tx_is_syncing(tx)); 4121 ASSERT(dsl_dataset_remap_deadlist_exists(ds)); 4122 4123 remap_deadlist_object = ds->ds_remap_deadlist.dl_object; 4124 dsl_deadlist_close(&ds->ds_remap_deadlist); 4125 dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx); 4126 dsl_dataset_unset_remap_deadlist_object(ds, tx); 4127 spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); 4128 } 4129 4130 void 4131 dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx) 4132 { 4133 uint64_t remap_deadlist_obj; 4134 spa_t *spa = ds->ds_dir->dd_pool->dp_spa; 4135 4136 ASSERT(dmu_tx_is_syncing(tx)); 4137 ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock)); 4138 /* 4139 * Currently we only create remap deadlists when there are indirect 4140 * vdevs with referenced mappings. 4141 */ 4142 ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL)); 4143 4144 remap_deadlist_obj = dsl_deadlist_clone( 4145 &ds->ds_deadlist, UINT64_MAX, 4146 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 4147 dsl_dataset_set_remap_deadlist_object(ds, 4148 remap_deadlist_obj, tx); 4149 dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa), 4150 remap_deadlist_obj); 4151 spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); 4152 } 4153