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