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 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved. 24 * Copyright (c) 2014, Joyent, Inc. All rights reserved. 25 * Copyright (c) 2014 RackTop Systems. 26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 27 * Copyright (c) 2014 Integros [integros.com] 28 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved. 29 */ 30 31 #include <sys/dmu_objset.h> 32 #include <sys/dsl_dataset.h> 33 #include <sys/dsl_dir.h> 34 #include <sys/dsl_prop.h> 35 #include <sys/dsl_synctask.h> 36 #include <sys/dmu_traverse.h> 37 #include <sys/dmu_impl.h> 38 #include <sys/dmu_tx.h> 39 #include <sys/arc.h> 40 #include <sys/zio.h> 41 #include <sys/zap.h> 42 #include <sys/zfeature.h> 43 #include <sys/unique.h> 44 #include <sys/zfs_context.h> 45 #include <sys/zfs_ioctl.h> 46 #include <sys/spa.h> 47 #include <sys/zfs_znode.h> 48 #include <sys/zfs_onexit.h> 49 #include <sys/zvol.h> 50 #include <sys/dsl_scan.h> 51 #include <sys/dsl_deadlist.h> 52 #include <sys/dsl_destroy.h> 53 #include <sys/dsl_userhold.h> 54 #include <sys/dsl_bookmark.h> 55 #include <sys/dmu_send.h> 56 #include <sys/zio_checksum.h> 57 #include <sys/zio_compress.h> 58 #include <zfs_fletcher.h> 59 60 /* 61 * The SPA supports block sizes up to 16MB. However, very large blocks 62 * can have an impact on i/o latency (e.g. tying up a spinning disk for 63 * ~300ms), and also potentially on the memory allocator. Therefore, 64 * we do not allow the recordsize to be set larger than zfs_max_recordsize 65 * (default 1MB). Larger blocks can be created by changing this tunable, 66 * and pools with larger blocks can always be imported and used, regardless 67 * of this setting. 68 */ 69 int zfs_max_recordsize = 1 * 1024 * 1024; 70 71 #define SWITCH64(x, y) \ 72 { \ 73 uint64_t __tmp = (x); \ 74 (x) = (y); \ 75 (y) = __tmp; \ 76 } 77 78 #define DS_REF_MAX (1ULL << 62) 79 80 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds); 81 82 extern int spa_asize_inflation; 83 84 static zil_header_t zero_zil; 85 86 /* 87 * Figure out how much of this delta should be propogated to the dsl_dir 88 * layer. If there's a refreservation, that space has already been 89 * partially accounted for in our ancestors. 90 */ 91 static int64_t 92 parent_delta(dsl_dataset_t *ds, int64_t delta) 93 { 94 dsl_dataset_phys_t *ds_phys; 95 uint64_t old_bytes, new_bytes; 96 97 if (ds->ds_reserved == 0) 98 return (delta); 99 100 ds_phys = dsl_dataset_phys(ds); 101 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved); 102 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 103 104 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 105 return (new_bytes - old_bytes); 106 } 107 108 void 109 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 110 { 111 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 112 int compressed = BP_GET_PSIZE(bp); 113 int uncompressed = BP_GET_UCSIZE(bp); 114 int64_t delta; 115 116 dprintf_bp(bp, "ds=%p", ds); 117 118 ASSERT(dmu_tx_is_syncing(tx)); 119 /* It could have been compressed away to nothing */ 120 if (BP_IS_HOLE(bp)) 121 return; 122 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 123 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 124 if (ds == NULL) { 125 dsl_pool_mos_diduse_space(tx->tx_pool, 126 used, compressed, uncompressed); 127 return; 128 } 129 130 ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 131 dmu_buf_will_dirty(ds->ds_dbuf, tx); 132 mutex_enter(&ds->ds_lock); 133 delta = parent_delta(ds, used); 134 dsl_dataset_phys(ds)->ds_referenced_bytes += used; 135 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed; 136 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed; 137 dsl_dataset_phys(ds)->ds_unique_bytes += used; 138 139 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) { 140 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] = 141 B_TRUE; 142 } 143 144 spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp)); 145 if (f != SPA_FEATURE_NONE) 146 ds->ds_feature_activation_needed[f] = B_TRUE; 147 148 mutex_exit(&ds->ds_lock); 149 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 150 compressed, uncompressed, tx); 151 dsl_dir_transfer_space(ds->ds_dir, used - delta, 152 DD_USED_REFRSRV, DD_USED_HEAD, tx); 153 } 154 155 int 156 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 157 boolean_t async) 158 { 159 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 160 int compressed = BP_GET_PSIZE(bp); 161 int uncompressed = BP_GET_UCSIZE(bp); 162 163 if (BP_IS_HOLE(bp)) 164 return (0); 165 166 ASSERT(dmu_tx_is_syncing(tx)); 167 ASSERT(bp->blk_birth <= tx->tx_txg); 168 169 if (ds == NULL) { 170 dsl_free(tx->tx_pool, tx->tx_txg, bp); 171 dsl_pool_mos_diduse_space(tx->tx_pool, 172 -used, -compressed, -uncompressed); 173 return (used); 174 } 175 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 176 177 ASSERT(!ds->ds_is_snapshot); 178 dmu_buf_will_dirty(ds->ds_dbuf, tx); 179 180 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 181 int64_t delta; 182 183 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 184 dsl_free(tx->tx_pool, tx->tx_txg, bp); 185 186 mutex_enter(&ds->ds_lock); 187 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used || 188 !DS_UNIQUE_IS_ACCURATE(ds)); 189 delta = parent_delta(ds, -used); 190 dsl_dataset_phys(ds)->ds_unique_bytes -= used; 191 mutex_exit(&ds->ds_lock); 192 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 193 delta, -compressed, -uncompressed, tx); 194 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 195 DD_USED_REFRSRV, DD_USED_HEAD, tx); 196 } else { 197 dprintf_bp(bp, "putting on dead list: %s", ""); 198 if (async) { 199 /* 200 * We are here as part of zio's write done callback, 201 * which means we're a zio interrupt thread. We can't 202 * call dsl_deadlist_insert() now because it may block 203 * waiting for I/O. Instead, put bp on the deferred 204 * queue and let dsl_pool_sync() finish the job. 205 */ 206 bplist_append(&ds->ds_pending_deadlist, bp); 207 } else { 208 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); 209 } 210 ASSERT3U(ds->ds_prev->ds_object, ==, 211 dsl_dataset_phys(ds)->ds_prev_snap_obj); 212 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0); 213 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 214 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 215 ds->ds_object && bp->blk_birth > 216 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) { 217 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 218 mutex_enter(&ds->ds_prev->ds_lock); 219 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used; 220 mutex_exit(&ds->ds_prev->ds_lock); 221 } 222 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 223 dsl_dir_transfer_space(ds->ds_dir, used, 224 DD_USED_HEAD, DD_USED_SNAP, tx); 225 } 226 } 227 mutex_enter(&ds->ds_lock); 228 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used); 229 dsl_dataset_phys(ds)->ds_referenced_bytes -= used; 230 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed); 231 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed; 232 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed); 233 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed; 234 mutex_exit(&ds->ds_lock); 235 236 return (used); 237 } 238 239 uint64_t 240 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 241 { 242 uint64_t trysnap = 0; 243 244 if (ds == NULL) 245 return (0); 246 /* 247 * The snapshot creation could fail, but that would cause an 248 * incorrect FALSE return, which would only result in an 249 * overestimation of the amount of space that an operation would 250 * consume, which is OK. 251 * 252 * There's also a small window where we could miss a pending 253 * snapshot, because we could set the sync task in the quiescing 254 * phase. So this should only be used as a guess. 255 */ 256 if (ds->ds_trysnap_txg > 257 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 258 trysnap = ds->ds_trysnap_txg; 259 return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap)); 260 } 261 262 boolean_t 263 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp, 264 uint64_t blk_birth) 265 { 266 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) || 267 (bp != NULL && BP_IS_HOLE(bp))) 268 return (B_FALSE); 269 270 ddt_prefetch(dsl_dataset_get_spa(ds), bp); 271 272 return (B_TRUE); 273 } 274 275 /* 276 * We have to release the fsid syncronously or we risk that a subsequent 277 * mount of the same dataset will fail to unique_insert the fsid. This 278 * failure would manifest itself as the fsid of this dataset changing 279 * between mounts which makes NFS clients quite unhappy. 280 */ 281 static void 282 dsl_dataset_evict_sync(void *dbu) 283 { 284 dsl_dataset_t *ds = dbu; 285 286 ASSERT(ds->ds_owner == NULL); 287 288 unique_remove(ds->ds_fsid_guid); 289 } 290 291 static void 292 dsl_dataset_evict_async(void *dbu) 293 { 294 dsl_dataset_t *ds = dbu; 295 296 ASSERT(ds->ds_owner == NULL); 297 298 ds->ds_dbuf = NULL; 299 300 if (ds->ds_objset != NULL) 301 dmu_objset_evict(ds->ds_objset); 302 303 if (ds->ds_prev) { 304 dsl_dataset_rele(ds->ds_prev, ds); 305 ds->ds_prev = NULL; 306 } 307 308 bplist_destroy(&ds->ds_pending_deadlist); 309 if (ds->ds_deadlist.dl_os != NULL) 310 dsl_deadlist_close(&ds->ds_deadlist); 311 if (ds->ds_dir) 312 dsl_dir_async_rele(ds->ds_dir, ds); 313 314 ASSERT(!list_link_active(&ds->ds_synced_link)); 315 316 list_destroy(&ds->ds_prop_cbs); 317 mutex_destroy(&ds->ds_lock); 318 mutex_destroy(&ds->ds_opening_lock); 319 mutex_destroy(&ds->ds_sendstream_lock); 320 refcount_destroy(&ds->ds_longholds); 321 rrw_destroy(&ds->ds_bp_rwlock); 322 323 kmem_free(ds, sizeof (dsl_dataset_t)); 324 } 325 326 int 327 dsl_dataset_get_snapname(dsl_dataset_t *ds) 328 { 329 dsl_dataset_phys_t *headphys; 330 int err; 331 dmu_buf_t *headdbuf; 332 dsl_pool_t *dp = ds->ds_dir->dd_pool; 333 objset_t *mos = dp->dp_meta_objset; 334 335 if (ds->ds_snapname[0]) 336 return (0); 337 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) 338 return (0); 339 340 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, 341 FTAG, &headdbuf); 342 if (err != 0) 343 return (err); 344 headphys = headdbuf->db_data; 345 err = zap_value_search(dp->dp_meta_objset, 346 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 347 dmu_buf_rele(headdbuf, FTAG); 348 return (err); 349 } 350 351 int 352 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 353 { 354 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 355 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 356 matchtype_t mt; 357 int err; 358 359 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 360 mt = MT_FIRST; 361 else 362 mt = MT_EXACT; 363 364 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 365 value, mt, NULL, 0, NULL); 366 if (err == ENOTSUP && mt == MT_FIRST) 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; 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_FIRST; 384 else 385 mt = MT_EXACT; 386 387 err = zap_remove_norm(mos, snapobj, name, mt, tx); 388 if (err == ENOTSUP && mt == MT_FIRST) 389 err = zap_remove(mos, snapobj, name, tx); 390 391 if (err == 0 && adj_cnt) 392 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 393 DD_FIELD_SNAPSHOT_COUNT, tx); 394 395 return (err); 396 } 397 398 boolean_t 399 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag) 400 { 401 dmu_buf_t *dbuf = ds->ds_dbuf; 402 boolean_t result = B_FALSE; 403 404 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset, 405 ds->ds_object, DMU_BONUS_BLKID, tag)) { 406 407 if (ds == dmu_buf_get_user(dbuf)) 408 result = B_TRUE; 409 else 410 dmu_buf_rele(dbuf, tag); 411 } 412 413 return (result); 414 } 415 416 int 417 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 418 dsl_dataset_t **dsp) 419 { 420 objset_t *mos = dp->dp_meta_objset; 421 dmu_buf_t *dbuf; 422 dsl_dataset_t *ds; 423 int err; 424 dmu_object_info_t doi; 425 426 ASSERT(dsl_pool_config_held(dp)); 427 428 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 429 if (err != 0) 430 return (err); 431 432 /* Make sure dsobj has the correct object type. */ 433 dmu_object_info_from_db(dbuf, &doi); 434 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 435 dmu_buf_rele(dbuf, tag); 436 return (SET_ERROR(EINVAL)); 437 } 438 439 ds = dmu_buf_get_user(dbuf); 440 if (ds == NULL) { 441 dsl_dataset_t *winner = NULL; 442 443 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 444 ds->ds_dbuf = dbuf; 445 ds->ds_object = dsobj; 446 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0; 447 448 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 449 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 450 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 451 rrw_init(&ds->ds_bp_rwlock, B_FALSE); 452 refcount_create(&ds->ds_longholds); 453 454 bplist_create(&ds->ds_pending_deadlist); 455 dsl_deadlist_open(&ds->ds_deadlist, 456 mos, dsl_dataset_phys(ds)->ds_deadlist_obj); 457 458 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 459 offsetof(dmu_sendarg_t, dsa_link)); 460 461 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t), 462 offsetof(dsl_prop_cb_record_t, cbr_ds_node)); 463 464 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 465 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 466 if (!(spa_feature_table[f].fi_flags & 467 ZFEATURE_FLAG_PER_DATASET)) 468 continue; 469 err = zap_contains(mos, dsobj, 470 spa_feature_table[f].fi_guid); 471 if (err == 0) { 472 ds->ds_feature_inuse[f] = B_TRUE; 473 } else { 474 ASSERT3U(err, ==, ENOENT); 475 err = 0; 476 } 477 } 478 } 479 480 err = dsl_dir_hold_obj(dp, 481 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir); 482 if (err != 0) { 483 mutex_destroy(&ds->ds_lock); 484 mutex_destroy(&ds->ds_opening_lock); 485 mutex_destroy(&ds->ds_sendstream_lock); 486 refcount_destroy(&ds->ds_longholds); 487 bplist_destroy(&ds->ds_pending_deadlist); 488 dsl_deadlist_close(&ds->ds_deadlist); 489 kmem_free(ds, sizeof (dsl_dataset_t)); 490 dmu_buf_rele(dbuf, tag); 491 return (err); 492 } 493 494 if (!ds->ds_is_snapshot) { 495 ds->ds_snapname[0] = '\0'; 496 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 497 err = dsl_dataset_hold_obj(dp, 498 dsl_dataset_phys(ds)->ds_prev_snap_obj, 499 ds, &ds->ds_prev); 500 } 501 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 502 int zaperr = zap_lookup(mos, ds->ds_object, 503 DS_FIELD_BOOKMARK_NAMES, 504 sizeof (ds->ds_bookmarks), 1, 505 &ds->ds_bookmarks); 506 if (zaperr != ENOENT) 507 VERIFY0(zaperr); 508 } 509 } else { 510 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 511 err = dsl_dataset_get_snapname(ds); 512 if (err == 0 && 513 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { 514 err = zap_count( 515 ds->ds_dir->dd_pool->dp_meta_objset, 516 dsl_dataset_phys(ds)->ds_userrefs_obj, 517 &ds->ds_userrefs); 518 } 519 } 520 521 if (err == 0 && !ds->ds_is_snapshot) { 522 err = dsl_prop_get_int_ds(ds, 523 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 524 &ds->ds_reserved); 525 if (err == 0) { 526 err = dsl_prop_get_int_ds(ds, 527 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 528 &ds->ds_quota); 529 } 530 } else { 531 ds->ds_reserved = ds->ds_quota = 0; 532 } 533 534 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync, 535 dsl_dataset_evict_async, &ds->ds_dbuf); 536 if (err == 0) 537 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu); 538 539 if (err != 0 || winner != NULL) { 540 bplist_destroy(&ds->ds_pending_deadlist); 541 dsl_deadlist_close(&ds->ds_deadlist); 542 if (ds->ds_prev) 543 dsl_dataset_rele(ds->ds_prev, ds); 544 dsl_dir_rele(ds->ds_dir, ds); 545 mutex_destroy(&ds->ds_lock); 546 mutex_destroy(&ds->ds_opening_lock); 547 mutex_destroy(&ds->ds_sendstream_lock); 548 refcount_destroy(&ds->ds_longholds); 549 kmem_free(ds, sizeof (dsl_dataset_t)); 550 if (err != 0) { 551 dmu_buf_rele(dbuf, tag); 552 return (err); 553 } 554 ds = winner; 555 } else { 556 ds->ds_fsid_guid = 557 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); 558 if (ds->ds_fsid_guid != 559 dsl_dataset_phys(ds)->ds_fsid_guid) { 560 zfs_dbgmsg("ds_fsid_guid changed from " 561 "%llx to %llx for pool %s dataset id %llu", 562 (long long) 563 dsl_dataset_phys(ds)->ds_fsid_guid, 564 (long long)ds->ds_fsid_guid, 565 spa_name(dp->dp_spa), 566 dsobj); 567 } 568 } 569 } 570 ASSERT3P(ds->ds_dbuf, ==, dbuf); 571 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); 572 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || 573 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 574 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 575 *dsp = ds; 576 return (0); 577 } 578 579 int 580 dsl_dataset_hold(dsl_pool_t *dp, const char *name, 581 void *tag, dsl_dataset_t **dsp) 582 { 583 dsl_dir_t *dd; 584 const char *snapname; 585 uint64_t obj; 586 int err = 0; 587 dsl_dataset_t *ds; 588 589 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 590 if (err != 0) 591 return (err); 592 593 ASSERT(dsl_pool_config_held(dp)); 594 obj = dsl_dir_phys(dd)->dd_head_dataset_obj; 595 if (obj != 0) 596 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 597 else 598 err = SET_ERROR(ENOENT); 599 600 /* we may be looking for a snapshot */ 601 if (err == 0 && snapname != NULL) { 602 dsl_dataset_t *snap_ds; 603 604 if (*snapname++ != '@') { 605 dsl_dataset_rele(ds, tag); 606 dsl_dir_rele(dd, FTAG); 607 return (SET_ERROR(ENOENT)); 608 } 609 610 dprintf("looking for snapshot '%s'\n", snapname); 611 err = dsl_dataset_snap_lookup(ds, snapname, &obj); 612 if (err == 0) 613 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds); 614 dsl_dataset_rele(ds, tag); 615 616 if (err == 0) { 617 mutex_enter(&snap_ds->ds_lock); 618 if (snap_ds->ds_snapname[0] == 0) 619 (void) strlcpy(snap_ds->ds_snapname, snapname, 620 sizeof (snap_ds->ds_snapname)); 621 mutex_exit(&snap_ds->ds_lock); 622 ds = snap_ds; 623 } 624 } 625 if (err == 0) 626 *dsp = ds; 627 dsl_dir_rele(dd, FTAG); 628 return (err); 629 } 630 631 int 632 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 633 void *tag, dsl_dataset_t **dsp) 634 { 635 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 636 if (err != 0) 637 return (err); 638 if (!dsl_dataset_tryown(*dsp, tag)) { 639 dsl_dataset_rele(*dsp, tag); 640 *dsp = NULL; 641 return (SET_ERROR(EBUSY)); 642 } 643 return (0); 644 } 645 646 int 647 dsl_dataset_own(dsl_pool_t *dp, const char *name, 648 void *tag, dsl_dataset_t **dsp) 649 { 650 int err = dsl_dataset_hold(dp, name, tag, dsp); 651 if (err != 0) 652 return (err); 653 if (!dsl_dataset_tryown(*dsp, tag)) { 654 dsl_dataset_rele(*dsp, tag); 655 return (SET_ERROR(EBUSY)); 656 } 657 return (0); 658 } 659 660 /* 661 * See the comment above dsl_pool_hold() for details. In summary, a long 662 * hold is used to prevent destruction of a dataset while the pool hold 663 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 664 * 665 * The dataset and pool must be held when this function is called. After it 666 * is called, the pool hold may be released while the dataset is still held 667 * and accessed. 668 */ 669 void 670 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 671 { 672 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 673 (void) refcount_add(&ds->ds_longholds, tag); 674 } 675 676 void 677 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 678 { 679 (void) refcount_remove(&ds->ds_longholds, tag); 680 } 681 682 /* Return B_TRUE if there are any long holds on this dataset. */ 683 boolean_t 684 dsl_dataset_long_held(dsl_dataset_t *ds) 685 { 686 return (!refcount_is_zero(&ds->ds_longholds)); 687 } 688 689 void 690 dsl_dataset_name(dsl_dataset_t *ds, char *name) 691 { 692 if (ds == NULL) { 693 (void) strcpy(name, "mos"); 694 } else { 695 dsl_dir_name(ds->ds_dir, name); 696 VERIFY0(dsl_dataset_get_snapname(ds)); 697 if (ds->ds_snapname[0]) { 698 VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN), 699 <, ZFS_MAX_DATASET_NAME_LEN); 700 /* 701 * We use a "recursive" mutex so that we 702 * can call dprintf_ds() with ds_lock held. 703 */ 704 if (!MUTEX_HELD(&ds->ds_lock)) { 705 mutex_enter(&ds->ds_lock); 706 VERIFY3U(strlcat(name, ds->ds_snapname, 707 ZFS_MAX_DATASET_NAME_LEN), <, 708 ZFS_MAX_DATASET_NAME_LEN); 709 mutex_exit(&ds->ds_lock); 710 } else { 711 VERIFY3U(strlcat(name, ds->ds_snapname, 712 ZFS_MAX_DATASET_NAME_LEN), <, 713 ZFS_MAX_DATASET_NAME_LEN); 714 } 715 } 716 } 717 } 718 719 int 720 dsl_dataset_namelen(dsl_dataset_t *ds) 721 { 722 VERIFY0(dsl_dataset_get_snapname(ds)); 723 mutex_enter(&ds->ds_lock); 724 int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname); 725 mutex_exit(&ds->ds_lock); 726 return (len); 727 } 728 729 void 730 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 731 { 732 dmu_buf_rele(ds->ds_dbuf, tag); 733 } 734 735 void 736 dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 737 { 738 ASSERT3P(ds->ds_owner, ==, tag); 739 ASSERT(ds->ds_dbuf != NULL); 740 741 mutex_enter(&ds->ds_lock); 742 ds->ds_owner = NULL; 743 mutex_exit(&ds->ds_lock); 744 dsl_dataset_long_rele(ds, tag); 745 dsl_dataset_rele(ds, tag); 746 } 747 748 boolean_t 749 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 750 { 751 boolean_t gotit = FALSE; 752 753 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 754 mutex_enter(&ds->ds_lock); 755 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 756 ds->ds_owner = tag; 757 dsl_dataset_long_hold(ds, tag); 758 gotit = TRUE; 759 } 760 mutex_exit(&ds->ds_lock); 761 return (gotit); 762 } 763 764 boolean_t 765 dsl_dataset_has_owner(dsl_dataset_t *ds) 766 { 767 boolean_t rv; 768 mutex_enter(&ds->ds_lock); 769 rv = (ds->ds_owner != NULL); 770 mutex_exit(&ds->ds_lock); 771 return (rv); 772 } 773 774 static void 775 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx) 776 { 777 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 778 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 779 uint64_t zero = 0; 780 781 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 782 783 spa_feature_incr(spa, f, tx); 784 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 785 786 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, 787 sizeof (zero), 1, &zero, tx)); 788 } 789 790 void 791 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx) 792 { 793 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 794 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 795 796 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 797 798 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx)); 799 spa_feature_decr(spa, f, tx); 800 } 801 802 uint64_t 803 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 804 uint64_t flags, dmu_tx_t *tx) 805 { 806 dsl_pool_t *dp = dd->dd_pool; 807 dmu_buf_t *dbuf; 808 dsl_dataset_phys_t *dsphys; 809 uint64_t dsobj; 810 objset_t *mos = dp->dp_meta_objset; 811 812 if (origin == NULL) 813 origin = dp->dp_origin_snap; 814 815 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 816 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); 817 ASSERT(dmu_tx_is_syncing(tx)); 818 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 819 820 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 821 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 822 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 823 dmu_buf_will_dirty(dbuf, tx); 824 dsphys = dbuf->db_data; 825 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 826 dsphys->ds_dir_obj = dd->dd_object; 827 dsphys->ds_flags = flags; 828 dsphys->ds_fsid_guid = unique_create(); 829 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 830 sizeof (dsphys->ds_guid)); 831 dsphys->ds_snapnames_zapobj = 832 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 833 DMU_OT_NONE, 0, tx); 834 dsphys->ds_creation_time = gethrestime_sec(); 835 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 836 837 if (origin == NULL) { 838 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 839 } else { 840 dsl_dataset_t *ohds; /* head of the origin snapshot */ 841 842 dsphys->ds_prev_snap_obj = origin->ds_object; 843 dsphys->ds_prev_snap_txg = 844 dsl_dataset_phys(origin)->ds_creation_txg; 845 dsphys->ds_referenced_bytes = 846 dsl_dataset_phys(origin)->ds_referenced_bytes; 847 dsphys->ds_compressed_bytes = 848 dsl_dataset_phys(origin)->ds_compressed_bytes; 849 dsphys->ds_uncompressed_bytes = 850 dsl_dataset_phys(origin)->ds_uncompressed_bytes; 851 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG); 852 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; 853 rrw_exit(&origin->ds_bp_rwlock, FTAG); 854 855 /* 856 * Inherit flags that describe the dataset's contents 857 * (INCONSISTENT) or properties (Case Insensitive). 858 */ 859 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & 860 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 861 862 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 863 if (origin->ds_feature_inuse[f]) 864 dsl_dataset_activate_feature(dsobj, f, tx); 865 } 866 867 dmu_buf_will_dirty(origin->ds_dbuf, tx); 868 dsl_dataset_phys(origin)->ds_num_children++; 869 870 VERIFY0(dsl_dataset_hold_obj(dp, 871 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, 872 FTAG, &ohds)); 873 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 874 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 875 dsl_dataset_rele(ohds, FTAG); 876 877 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 878 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { 879 dsl_dataset_phys(origin)->ds_next_clones_obj = 880 zap_create(mos, 881 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 882 } 883 VERIFY0(zap_add_int(mos, 884 dsl_dataset_phys(origin)->ds_next_clones_obj, 885 dsobj, tx)); 886 } 887 888 dmu_buf_will_dirty(dd->dd_dbuf, tx); 889 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; 890 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 891 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 892 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 893 dsl_dir_phys(origin->ds_dir)->dd_clones = 894 zap_create(mos, 895 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 896 } 897 VERIFY0(zap_add_int(mos, 898 dsl_dir_phys(origin->ds_dir)->dd_clones, 899 dsobj, tx)); 900 } 901 } 902 903 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 904 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 905 906 dmu_buf_rele(dbuf, FTAG); 907 908 dmu_buf_will_dirty(dd->dd_dbuf, tx); 909 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; 910 911 return (dsobj); 912 } 913 914 static void 915 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 916 { 917 objset_t *os; 918 919 VERIFY0(dmu_objset_from_ds(ds, &os)); 920 if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) { 921 dsl_pool_t *dp = ds->ds_dir->dd_pool; 922 zio_t *zio; 923 924 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 925 926 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 927 dsl_dataset_sync(ds, zio, tx); 928 VERIFY0(zio_wait(zio)); 929 930 /* dsl_dataset_sync_done will drop this reference. */ 931 dmu_buf_add_ref(ds->ds_dbuf, ds); 932 dsl_dataset_sync_done(ds, tx); 933 } 934 } 935 936 uint64_t 937 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 938 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 939 { 940 dsl_pool_t *dp = pdd->dd_pool; 941 uint64_t dsobj, ddobj; 942 dsl_dir_t *dd; 943 944 ASSERT(dmu_tx_is_syncing(tx)); 945 ASSERT(lastname[0] != '@'); 946 947 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 948 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 949 950 dsobj = dsl_dataset_create_sync_dd(dd, origin, 951 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 952 953 dsl_deleg_set_create_perms(dd, tx, cr); 954 955 /* 956 * Since we're creating a new node we know it's a leaf, so we can 957 * initialize the counts if the limit feature is active. 958 */ 959 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 960 uint64_t cnt = 0; 961 objset_t *os = dd->dd_pool->dp_meta_objset; 962 963 dsl_dir_zapify(dd, tx); 964 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 965 sizeof (cnt), 1, &cnt, tx)); 966 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 967 sizeof (cnt), 1, &cnt, tx)); 968 } 969 970 dsl_dir_rele(dd, FTAG); 971 972 /* 973 * If we are creating a clone, make sure we zero out any stale 974 * data from the origin snapshots zil header. 975 */ 976 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 977 dsl_dataset_t *ds; 978 979 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 980 dsl_dataset_zero_zil(ds, tx); 981 dsl_dataset_rele(ds, FTAG); 982 } 983 984 return (dsobj); 985 } 986 987 /* 988 * The unique space in the head dataset can be calculated by subtracting 989 * the space used in the most recent snapshot, that is still being used 990 * in this file system, from the space currently in use. To figure out 991 * the space in the most recent snapshot still in use, we need to take 992 * the total space used in the snapshot and subtract out the space that 993 * has been freed up since the snapshot was taken. 994 */ 995 void 996 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 997 { 998 uint64_t mrs_used; 999 uint64_t dlused, dlcomp, dluncomp; 1000 1001 ASSERT(!ds->ds_is_snapshot); 1002 1003 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) 1004 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; 1005 else 1006 mrs_used = 0; 1007 1008 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 1009 1010 ASSERT3U(dlused, <=, mrs_used); 1011 dsl_dataset_phys(ds)->ds_unique_bytes = 1012 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); 1013 1014 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1015 SPA_VERSION_UNIQUE_ACCURATE) 1016 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1017 } 1018 1019 void 1020 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 1021 dmu_tx_t *tx) 1022 { 1023 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1024 uint64_t count; 1025 int err; 1026 1027 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); 1028 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1029 obj, tx); 1030 /* 1031 * The err should not be ENOENT, but a bug in a previous version 1032 * of the code could cause upgrade_clones_cb() to not set 1033 * ds_next_snap_obj when it should, leading to a missing entry. 1034 * If we knew that the pool was created after 1035 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 1036 * ENOENT. However, at least we can check that we don't have 1037 * too many entries in the next_clones_obj even after failing to 1038 * remove this one. 1039 */ 1040 if (err != ENOENT) 1041 VERIFY0(err); 1042 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1043 &count)); 1044 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); 1045 } 1046 1047 1048 blkptr_t * 1049 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1050 { 1051 return (&dsl_dataset_phys(ds)->ds_bp); 1052 } 1053 1054 spa_t * 1055 dsl_dataset_get_spa(dsl_dataset_t *ds) 1056 { 1057 return (ds->ds_dir->dd_pool->dp_spa); 1058 } 1059 1060 void 1061 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1062 { 1063 dsl_pool_t *dp; 1064 1065 if (ds == NULL) /* this is the meta-objset */ 1066 return; 1067 1068 ASSERT(ds->ds_objset != NULL); 1069 1070 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) 1071 panic("dirtying snapshot!"); 1072 1073 /* Must not dirty a dataset in the same txg where it got snapshotted. */ 1074 ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 1075 1076 dp = ds->ds_dir->dd_pool; 1077 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1078 /* up the hold count until we can be written out */ 1079 dmu_buf_add_ref(ds->ds_dbuf, ds); 1080 } 1081 } 1082 1083 boolean_t 1084 dsl_dataset_is_dirty(dsl_dataset_t *ds) 1085 { 1086 for (int t = 0; t < TXG_SIZE; t++) { 1087 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1088 ds, t)) 1089 return (B_TRUE); 1090 } 1091 return (B_FALSE); 1092 } 1093 1094 static int 1095 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1096 { 1097 uint64_t asize; 1098 1099 if (!dmu_tx_is_syncing(tx)) 1100 return (0); 1101 1102 /* 1103 * If there's an fs-only reservation, any blocks that might become 1104 * owned by the snapshot dataset must be accommodated by space 1105 * outside of the reservation. 1106 */ 1107 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1108 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); 1109 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1110 return (SET_ERROR(ENOSPC)); 1111 1112 /* 1113 * Propagate any reserved space for this snapshot to other 1114 * snapshot checks in this sync group. 1115 */ 1116 if (asize > 0) 1117 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1118 1119 return (0); 1120 } 1121 1122 typedef struct dsl_dataset_snapshot_arg { 1123 nvlist_t *ddsa_snaps; 1124 nvlist_t *ddsa_props; 1125 nvlist_t *ddsa_errors; 1126 cred_t *ddsa_cr; 1127 } dsl_dataset_snapshot_arg_t; 1128 1129 int 1130 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1131 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 1132 { 1133 int error; 1134 uint64_t value; 1135 1136 ds->ds_trysnap_txg = tx->tx_txg; 1137 1138 if (!dmu_tx_is_syncing(tx)) 1139 return (0); 1140 1141 /* 1142 * We don't allow multiple snapshots of the same txg. If there 1143 * is already one, try again. 1144 */ 1145 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) 1146 return (SET_ERROR(EAGAIN)); 1147 1148 /* 1149 * Check for conflicting snapshot name. 1150 */ 1151 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1152 if (error == 0) 1153 return (SET_ERROR(EEXIST)); 1154 if (error != ENOENT) 1155 return (error); 1156 1157 /* 1158 * We don't allow taking snapshots of inconsistent datasets, such as 1159 * those into which we are currently receiving. However, if we are 1160 * creating this snapshot as part of a receive, this check will be 1161 * executed atomically with respect to the completion of the receive 1162 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1163 * case we ignore this, knowing it will be fixed up for us shortly in 1164 * dmu_recv_end_sync(). 1165 */ 1166 if (!recv && DS_IS_INCONSISTENT(ds)) 1167 return (SET_ERROR(EBUSY)); 1168 1169 /* 1170 * Skip the check for temporary snapshots or if we have already checked 1171 * the counts in dsl_dataset_snapshot_check. This means we really only 1172 * check the count here when we're receiving a stream. 1173 */ 1174 if (cnt != 0 && cr != NULL) { 1175 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1176 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 1177 if (error != 0) 1178 return (error); 1179 } 1180 1181 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1182 if (error != 0) 1183 return (error); 1184 1185 return (0); 1186 } 1187 1188 static int 1189 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1190 { 1191 dsl_dataset_snapshot_arg_t *ddsa = arg; 1192 dsl_pool_t *dp = dmu_tx_pool(tx); 1193 nvpair_t *pair; 1194 int rv = 0; 1195 1196 /* 1197 * Pre-compute how many total new snapshots will be created for each 1198 * level in the tree and below. This is needed for validating the 1199 * snapshot limit when either taking a recursive snapshot or when 1200 * taking multiple snapshots. 1201 * 1202 * The problem is that the counts are not actually adjusted when 1203 * we are checking, only when we finally sync. For a single snapshot, 1204 * this is easy, the count will increase by 1 at each node up the tree, 1205 * but its more complicated for the recursive/multiple snapshot case. 1206 * 1207 * The dsl_fs_ss_limit_check function does recursively check the count 1208 * at each level up the tree but since it is validating each snapshot 1209 * independently we need to be sure that we are validating the complete 1210 * count for the entire set of snapshots. We do this by rolling up the 1211 * counts for each component of the name into an nvlist and then 1212 * checking each of those cases with the aggregated count. 1213 * 1214 * This approach properly handles not only the recursive snapshot 1215 * case (where we get all of those on the ddsa_snaps list) but also 1216 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1217 * validate the limit on 'a' using a count of 2). 1218 * 1219 * We validate the snapshot names in the third loop and only report 1220 * name errors once. 1221 */ 1222 if (dmu_tx_is_syncing(tx)) { 1223 nvlist_t *cnt_track = NULL; 1224 cnt_track = fnvlist_alloc(); 1225 1226 /* Rollup aggregated counts into the cnt_track list */ 1227 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1228 pair != NULL; 1229 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1230 char *pdelim; 1231 uint64_t val; 1232 char nm[MAXPATHLEN]; 1233 1234 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1235 pdelim = strchr(nm, '@'); 1236 if (pdelim == NULL) 1237 continue; 1238 *pdelim = '\0'; 1239 1240 do { 1241 if (nvlist_lookup_uint64(cnt_track, nm, 1242 &val) == 0) { 1243 /* update existing entry */ 1244 fnvlist_add_uint64(cnt_track, nm, 1245 val + 1); 1246 } else { 1247 /* add to list */ 1248 fnvlist_add_uint64(cnt_track, nm, 1); 1249 } 1250 1251 pdelim = strrchr(nm, '/'); 1252 if (pdelim != NULL) 1253 *pdelim = '\0'; 1254 } while (pdelim != NULL); 1255 } 1256 1257 /* Check aggregated counts at each level */ 1258 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1259 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1260 int error = 0; 1261 char *name; 1262 uint64_t cnt = 0; 1263 dsl_dataset_t *ds; 1264 1265 name = nvpair_name(pair); 1266 cnt = fnvpair_value_uint64(pair); 1267 ASSERT(cnt > 0); 1268 1269 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1270 if (error == 0) { 1271 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1272 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1273 ddsa->ddsa_cr); 1274 dsl_dataset_rele(ds, FTAG); 1275 } 1276 1277 if (error != 0) { 1278 if (ddsa->ddsa_errors != NULL) 1279 fnvlist_add_int32(ddsa->ddsa_errors, 1280 name, error); 1281 rv = error; 1282 /* only report one error for this check */ 1283 break; 1284 } 1285 } 1286 nvlist_free(cnt_track); 1287 } 1288 1289 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1290 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1291 int error = 0; 1292 dsl_dataset_t *ds; 1293 char *name, *atp; 1294 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1295 1296 name = nvpair_name(pair); 1297 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN) 1298 error = SET_ERROR(ENAMETOOLONG); 1299 if (error == 0) { 1300 atp = strchr(name, '@'); 1301 if (atp == NULL) 1302 error = SET_ERROR(EINVAL); 1303 if (error == 0) 1304 (void) strlcpy(dsname, name, atp - name + 1); 1305 } 1306 if (error == 0) 1307 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1308 if (error == 0) { 1309 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1310 error = dsl_dataset_snapshot_check_impl(ds, 1311 atp + 1, tx, B_FALSE, 0, NULL); 1312 dsl_dataset_rele(ds, FTAG); 1313 } 1314 1315 if (error != 0) { 1316 if (ddsa->ddsa_errors != NULL) { 1317 fnvlist_add_int32(ddsa->ddsa_errors, 1318 name, error); 1319 } 1320 rv = error; 1321 } 1322 } 1323 1324 return (rv); 1325 } 1326 1327 void 1328 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1329 dmu_tx_t *tx) 1330 { 1331 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1332 dmu_buf_t *dbuf; 1333 dsl_dataset_phys_t *dsphys; 1334 uint64_t dsobj, crtxg; 1335 objset_t *mos = dp->dp_meta_objset; 1336 objset_t *os; 1337 1338 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1339 1340 /* 1341 * If we are on an old pool, the zil must not be active, in which 1342 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1343 */ 1344 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1345 dmu_objset_from_ds(ds, &os) != 0 || 1346 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1347 sizeof (zero_zil)) == 0); 1348 1349 /* Should not snapshot a dirty dataset. */ 1350 ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1351 ds, tx->tx_txg)); 1352 1353 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1354 1355 /* 1356 * The origin's ds_creation_txg has to be < TXG_INITIAL 1357 */ 1358 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1359 crtxg = 1; 1360 else 1361 crtxg = tx->tx_txg; 1362 1363 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1364 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1365 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1366 dmu_buf_will_dirty(dbuf, tx); 1367 dsphys = dbuf->db_data; 1368 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1369 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1370 dsphys->ds_fsid_guid = unique_create(); 1371 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1372 sizeof (dsphys->ds_guid)); 1373 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1374 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 1375 dsphys->ds_next_snap_obj = ds->ds_object; 1376 dsphys->ds_num_children = 1; 1377 dsphys->ds_creation_time = gethrestime_sec(); 1378 dsphys->ds_creation_txg = crtxg; 1379 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; 1380 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; 1381 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; 1382 dsphys->ds_uncompressed_bytes = 1383 dsl_dataset_phys(ds)->ds_uncompressed_bytes; 1384 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; 1385 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1386 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; 1387 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1388 dmu_buf_rele(dbuf, FTAG); 1389 1390 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1391 if (ds->ds_feature_inuse[f]) 1392 dsl_dataset_activate_feature(dsobj, f, tx); 1393 } 1394 1395 ASSERT3U(ds->ds_prev != 0, ==, 1396 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); 1397 if (ds->ds_prev) { 1398 uint64_t next_clones_obj = 1399 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; 1400 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1401 ds->ds_object || 1402 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); 1403 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1404 ds->ds_object) { 1405 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1406 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, 1407 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); 1408 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; 1409 } else if (next_clones_obj != 0) { 1410 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1411 dsphys->ds_next_snap_obj, tx); 1412 VERIFY0(zap_add_int(mos, 1413 next_clones_obj, dsobj, tx)); 1414 } 1415 } 1416 1417 /* 1418 * If we have a reference-reservation on this dataset, we will 1419 * need to increase the amount of refreservation being charged 1420 * since our unique space is going to zero. 1421 */ 1422 if (ds->ds_reserved) { 1423 int64_t delta; 1424 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1425 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, 1426 ds->ds_reserved); 1427 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1428 delta, 0, 0, tx); 1429 } 1430 1431 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1432 dsl_dataset_phys(ds)->ds_deadlist_obj = 1433 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, 1434 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 1435 dsl_deadlist_close(&ds->ds_deadlist); 1436 dsl_deadlist_open(&ds->ds_deadlist, mos, 1437 dsl_dataset_phys(ds)->ds_deadlist_obj); 1438 dsl_deadlist_add_key(&ds->ds_deadlist, 1439 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); 1440 1441 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); 1442 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; 1443 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; 1444 dsl_dataset_phys(ds)->ds_unique_bytes = 0; 1445 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1446 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1447 1448 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1449 snapname, 8, 1, &dsobj, tx)); 1450 1451 if (ds->ds_prev) 1452 dsl_dataset_rele(ds->ds_prev, ds); 1453 VERIFY0(dsl_dataset_hold_obj(dp, 1454 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); 1455 1456 dsl_scan_ds_snapshotted(ds, tx); 1457 1458 dsl_dir_snap_cmtime_update(ds->ds_dir); 1459 1460 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1461 } 1462 1463 static void 1464 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1465 { 1466 dsl_dataset_snapshot_arg_t *ddsa = arg; 1467 dsl_pool_t *dp = dmu_tx_pool(tx); 1468 nvpair_t *pair; 1469 1470 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1471 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1472 dsl_dataset_t *ds; 1473 char *name, *atp; 1474 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1475 1476 name = nvpair_name(pair); 1477 atp = strchr(name, '@'); 1478 (void) strlcpy(dsname, name, atp - name + 1); 1479 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1480 1481 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1482 if (ddsa->ddsa_props != NULL) { 1483 dsl_props_set_sync_impl(ds->ds_prev, 1484 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1485 } 1486 dsl_dataset_rele(ds, FTAG); 1487 } 1488 } 1489 1490 /* 1491 * The snapshots must all be in the same pool. 1492 * All-or-nothing: if there are any failures, nothing will be modified. 1493 */ 1494 int 1495 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1496 { 1497 dsl_dataset_snapshot_arg_t ddsa; 1498 nvpair_t *pair; 1499 boolean_t needsuspend; 1500 int error; 1501 spa_t *spa; 1502 char *firstname; 1503 nvlist_t *suspended = NULL; 1504 1505 pair = nvlist_next_nvpair(snaps, NULL); 1506 if (pair == NULL) 1507 return (0); 1508 firstname = nvpair_name(pair); 1509 1510 error = spa_open(firstname, &spa, FTAG); 1511 if (error != 0) 1512 return (error); 1513 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1514 spa_close(spa, FTAG); 1515 1516 if (needsuspend) { 1517 suspended = fnvlist_alloc(); 1518 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1519 pair = nvlist_next_nvpair(snaps, pair)) { 1520 char fsname[ZFS_MAX_DATASET_NAME_LEN]; 1521 char *snapname = nvpair_name(pair); 1522 char *atp; 1523 void *cookie; 1524 1525 atp = strchr(snapname, '@'); 1526 if (atp == NULL) { 1527 error = SET_ERROR(EINVAL); 1528 break; 1529 } 1530 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1531 1532 error = zil_suspend(fsname, &cookie); 1533 if (error != 0) 1534 break; 1535 fnvlist_add_uint64(suspended, fsname, 1536 (uintptr_t)cookie); 1537 } 1538 } 1539 1540 ddsa.ddsa_snaps = snaps; 1541 ddsa.ddsa_props = props; 1542 ddsa.ddsa_errors = errors; 1543 ddsa.ddsa_cr = CRED(); 1544 1545 if (error == 0) { 1546 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1547 dsl_dataset_snapshot_sync, &ddsa, 1548 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1549 } 1550 1551 if (suspended != NULL) { 1552 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1553 pair = nvlist_next_nvpair(suspended, pair)) { 1554 zil_resume((void *)(uintptr_t) 1555 fnvpair_value_uint64(pair)); 1556 } 1557 fnvlist_free(suspended); 1558 } 1559 1560 return (error); 1561 } 1562 1563 typedef struct dsl_dataset_snapshot_tmp_arg { 1564 const char *ddsta_fsname; 1565 const char *ddsta_snapname; 1566 minor_t ddsta_cleanup_minor; 1567 const char *ddsta_htag; 1568 } dsl_dataset_snapshot_tmp_arg_t; 1569 1570 static int 1571 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1572 { 1573 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1574 dsl_pool_t *dp = dmu_tx_pool(tx); 1575 dsl_dataset_t *ds; 1576 int error; 1577 1578 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1579 if (error != 0) 1580 return (error); 1581 1582 /* NULL cred means no limit check for tmp snapshot */ 1583 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1584 tx, B_FALSE, 0, NULL); 1585 if (error != 0) { 1586 dsl_dataset_rele(ds, FTAG); 1587 return (error); 1588 } 1589 1590 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1591 dsl_dataset_rele(ds, FTAG); 1592 return (SET_ERROR(ENOTSUP)); 1593 } 1594 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1595 B_TRUE, tx); 1596 if (error != 0) { 1597 dsl_dataset_rele(ds, FTAG); 1598 return (error); 1599 } 1600 1601 dsl_dataset_rele(ds, FTAG); 1602 return (0); 1603 } 1604 1605 static void 1606 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1607 { 1608 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1609 dsl_pool_t *dp = dmu_tx_pool(tx); 1610 dsl_dataset_t *ds; 1611 1612 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1613 1614 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1615 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1616 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1617 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1618 1619 dsl_dataset_rele(ds, FTAG); 1620 } 1621 1622 int 1623 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1624 minor_t cleanup_minor, const char *htag) 1625 { 1626 dsl_dataset_snapshot_tmp_arg_t ddsta; 1627 int error; 1628 spa_t *spa; 1629 boolean_t needsuspend; 1630 void *cookie; 1631 1632 ddsta.ddsta_fsname = fsname; 1633 ddsta.ddsta_snapname = snapname; 1634 ddsta.ddsta_cleanup_minor = cleanup_minor; 1635 ddsta.ddsta_htag = htag; 1636 1637 error = spa_open(fsname, &spa, FTAG); 1638 if (error != 0) 1639 return (error); 1640 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1641 spa_close(spa, FTAG); 1642 1643 if (needsuspend) { 1644 error = zil_suspend(fsname, &cookie); 1645 if (error != 0) 1646 return (error); 1647 } 1648 1649 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1650 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 1651 1652 if (needsuspend) 1653 zil_resume(cookie); 1654 return (error); 1655 } 1656 1657 1658 void 1659 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1660 { 1661 ASSERT(dmu_tx_is_syncing(tx)); 1662 ASSERT(ds->ds_objset != NULL); 1663 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); 1664 1665 /* 1666 * in case we had to change ds_fsid_guid when we opened it, 1667 * sync it out now. 1668 */ 1669 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1670 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; 1671 1672 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) { 1673 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1674 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1, 1675 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx)); 1676 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1677 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1, 1678 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx)); 1679 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1680 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1, 1681 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx)); 1682 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0; 1683 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0; 1684 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0; 1685 } 1686 1687 dmu_objset_sync(ds->ds_objset, zio, tx); 1688 1689 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1690 if (ds->ds_feature_activation_needed[f]) { 1691 if (ds->ds_feature_inuse[f]) 1692 continue; 1693 dsl_dataset_activate_feature(ds->ds_object, f, tx); 1694 ds->ds_feature_inuse[f] = B_TRUE; 1695 } 1696 } 1697 } 1698 1699 static int 1700 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 1701 { 1702 dsl_deadlist_t *dl = arg; 1703 dsl_deadlist_insert(dl, bp, tx); 1704 return (0); 1705 } 1706 1707 void 1708 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx) 1709 { 1710 objset_t *os = ds->ds_objset; 1711 1712 bplist_iterate(&ds->ds_pending_deadlist, 1713 deadlist_enqueue_cb, &ds->ds_deadlist, tx); 1714 1715 ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx))); 1716 1717 dmu_buf_rele(ds->ds_dbuf, ds); 1718 } 1719 1720 static void 1721 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1722 { 1723 uint64_t count = 0; 1724 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1725 zap_cursor_t zc; 1726 zap_attribute_t za; 1727 nvlist_t *propval = fnvlist_alloc(); 1728 nvlist_t *val = fnvlist_alloc(); 1729 1730 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1731 1732 /* 1733 * There may be missing entries in ds_next_clones_obj 1734 * due to a bug in a previous version of the code. 1735 * Only trust it if it has the right number of entries. 1736 */ 1737 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { 1738 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1739 &count)); 1740 } 1741 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) 1742 goto fail; 1743 for (zap_cursor_init(&zc, mos, 1744 dsl_dataset_phys(ds)->ds_next_clones_obj); 1745 zap_cursor_retrieve(&zc, &za) == 0; 1746 zap_cursor_advance(&zc)) { 1747 dsl_dataset_t *clone; 1748 char buf[ZFS_MAX_DATASET_NAME_LEN]; 1749 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1750 za.za_first_integer, FTAG, &clone)); 1751 dsl_dir_name(clone->ds_dir, buf); 1752 fnvlist_add_boolean(val, buf); 1753 dsl_dataset_rele(clone, FTAG); 1754 } 1755 zap_cursor_fini(&zc); 1756 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1757 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); 1758 fail: 1759 nvlist_free(val); 1760 nvlist_free(propval); 1761 } 1762 1763 static void 1764 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv) 1765 { 1766 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1767 1768 if (dsl_dataset_has_resume_receive_state(ds)) { 1769 char *str; 1770 void *packed; 1771 uint8_t *compressed; 1772 uint64_t val; 1773 nvlist_t *token_nv = fnvlist_alloc(); 1774 size_t packed_size, compressed_size; 1775 1776 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1777 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) { 1778 fnvlist_add_uint64(token_nv, "fromguid", val); 1779 } 1780 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1781 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) { 1782 fnvlist_add_uint64(token_nv, "object", val); 1783 } 1784 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1785 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) { 1786 fnvlist_add_uint64(token_nv, "offset", val); 1787 } 1788 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1789 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) { 1790 fnvlist_add_uint64(token_nv, "bytes", val); 1791 } 1792 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1793 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) { 1794 fnvlist_add_uint64(token_nv, "toguid", val); 1795 } 1796 char buf[256]; 1797 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1798 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) { 1799 fnvlist_add_string(token_nv, "toname", buf); 1800 } 1801 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1802 DS_FIELD_RESUME_LARGEBLOCK) == 0) { 1803 fnvlist_add_boolean(token_nv, "largeblockok"); 1804 } 1805 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1806 DS_FIELD_RESUME_EMBEDOK) == 0) { 1807 fnvlist_add_boolean(token_nv, "embedok"); 1808 } 1809 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1810 DS_FIELD_RESUME_COMPRESSOK) == 0) { 1811 fnvlist_add_boolean(token_nv, "compressok"); 1812 } 1813 packed = fnvlist_pack(token_nv, &packed_size); 1814 fnvlist_free(token_nv); 1815 compressed = kmem_alloc(packed_size, KM_SLEEP); 1816 1817 compressed_size = gzip_compress(packed, compressed, 1818 packed_size, packed_size, 6); 1819 1820 zio_cksum_t cksum; 1821 fletcher_4_native(compressed, compressed_size, NULL, &cksum); 1822 1823 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP); 1824 for (int i = 0; i < compressed_size; i++) { 1825 (void) sprintf(str + i * 2, "%02x", compressed[i]); 1826 } 1827 str[compressed_size * 2] = '\0'; 1828 char *propval = kmem_asprintf("%u-%llx-%llx-%s", 1829 ZFS_SEND_RESUME_TOKEN_VERSION, 1830 (longlong_t)cksum.zc_word[0], 1831 (longlong_t)packed_size, str); 1832 dsl_prop_nvlist_add_string(nv, 1833 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval); 1834 kmem_free(packed, packed_size); 1835 kmem_free(str, compressed_size * 2 + 1); 1836 kmem_free(compressed, packed_size); 1837 strfree(propval); 1838 } 1839 } 1840 1841 void 1842 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1843 { 1844 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1845 uint64_t refd, avail, uobjs, aobjs, ratio; 1846 1847 ASSERT(dsl_pool_config_held(dp)); 1848 1849 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : 1850 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / 1851 dsl_dataset_phys(ds)->ds_compressed_bytes); 1852 1853 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); 1854 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 1855 dsl_dataset_phys(ds)->ds_uncompressed_bytes); 1856 1857 if (ds->ds_is_snapshot) { 1858 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); 1859 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 1860 dsl_dataset_phys(ds)->ds_unique_bytes); 1861 get_clones_stat(ds, nv); 1862 } else { 1863 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 1864 char buf[ZFS_MAX_DATASET_NAME_LEN]; 1865 dsl_dataset_name(ds->ds_prev, buf); 1866 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf); 1867 } 1868 1869 dsl_dir_stats(ds->ds_dir, nv); 1870 } 1871 1872 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 1873 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 1874 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 1875 1876 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 1877 dsl_dataset_phys(ds)->ds_creation_time); 1878 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 1879 dsl_dataset_phys(ds)->ds_creation_txg); 1880 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 1881 ds->ds_quota); 1882 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 1883 ds->ds_reserved); 1884 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 1885 dsl_dataset_phys(ds)->ds_guid); 1886 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 1887 dsl_dataset_phys(ds)->ds_unique_bytes); 1888 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 1889 ds->ds_object); 1890 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 1891 ds->ds_userrefs); 1892 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 1893 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 1894 1895 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 1896 uint64_t written, comp, uncomp; 1897 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1898 dsl_dataset_t *prev; 1899 1900 int err = dsl_dataset_hold_obj(dp, 1901 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 1902 if (err == 0) { 1903 err = dsl_dataset_space_written(prev, ds, &written, 1904 &comp, &uncomp); 1905 dsl_dataset_rele(prev, FTAG); 1906 if (err == 0) { 1907 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 1908 written); 1909 } 1910 } 1911 } 1912 1913 if (!dsl_dataset_is_snapshot(ds)) { 1914 /* 1915 * A failed "newfs" (e.g. full) resumable receive leaves 1916 * the stats set on this dataset. Check here for the prop. 1917 */ 1918 get_receive_resume_stats(ds, nv); 1919 1920 /* 1921 * A failed incremental resumable receive leaves the 1922 * stats set on our child named "%recv". Check the child 1923 * for the prop. 1924 */ 1925 /* 6 extra bytes for /%recv */ 1926 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 1927 dsl_dataset_t *recv_ds; 1928 dsl_dataset_name(ds, recvname); 1929 if (strlcat(recvname, "/", sizeof (recvname)) < 1930 sizeof (recvname) && 1931 strlcat(recvname, recv_clone_name, sizeof (recvname)) < 1932 sizeof (recvname) && 1933 dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) { 1934 get_receive_resume_stats(recv_ds, nv); 1935 dsl_dataset_rele(recv_ds, FTAG); 1936 } 1937 } 1938 } 1939 1940 void 1941 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 1942 { 1943 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1944 ASSERT(dsl_pool_config_held(dp)); 1945 1946 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg; 1947 stat->dds_inconsistent = 1948 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT; 1949 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid; 1950 stat->dds_origin[0] = '\0'; 1951 if (ds->ds_is_snapshot) { 1952 stat->dds_is_snapshot = B_TRUE; 1953 stat->dds_num_clones = 1954 dsl_dataset_phys(ds)->ds_num_children - 1; 1955 } else { 1956 stat->dds_is_snapshot = B_FALSE; 1957 stat->dds_num_clones = 0; 1958 1959 if (dsl_dir_is_clone(ds->ds_dir)) { 1960 dsl_dataset_t *ods; 1961 1962 VERIFY0(dsl_dataset_hold_obj(dp, 1963 dsl_dir_phys(ds->ds_dir)->dd_origin_obj, 1964 FTAG, &ods)); 1965 dsl_dataset_name(ods, stat->dds_origin); 1966 dsl_dataset_rele(ods, FTAG); 1967 } 1968 } 1969 } 1970 1971 uint64_t 1972 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 1973 { 1974 return (ds->ds_fsid_guid); 1975 } 1976 1977 void 1978 dsl_dataset_space(dsl_dataset_t *ds, 1979 uint64_t *refdbytesp, uint64_t *availbytesp, 1980 uint64_t *usedobjsp, uint64_t *availobjsp) 1981 { 1982 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; 1983 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 1984 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) 1985 *availbytesp += 1986 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 1987 if (ds->ds_quota != 0) { 1988 /* 1989 * Adjust available bytes according to refquota 1990 */ 1991 if (*refdbytesp < ds->ds_quota) 1992 *availbytesp = MIN(*availbytesp, 1993 ds->ds_quota - *refdbytesp); 1994 else 1995 *availbytesp = 0; 1996 } 1997 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1998 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); 1999 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2000 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 2001 } 2002 2003 boolean_t 2004 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 2005 { 2006 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2007 uint64_t birth; 2008 2009 ASSERT(dsl_pool_config_held(dp)); 2010 if (snap == NULL) 2011 return (B_FALSE); 2012 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2013 birth = dsl_dataset_get_blkptr(ds)->blk_birth; 2014 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2015 if (birth > dsl_dataset_phys(snap)->ds_creation_txg) { 2016 objset_t *os, *os_snap; 2017 /* 2018 * It may be that only the ZIL differs, because it was 2019 * reset in the head. Don't count that as being 2020 * modified. 2021 */ 2022 if (dmu_objset_from_ds(ds, &os) != 0) 2023 return (B_TRUE); 2024 if (dmu_objset_from_ds(snap, &os_snap) != 0) 2025 return (B_TRUE); 2026 return (bcmp(&os->os_phys->os_meta_dnode, 2027 &os_snap->os_phys->os_meta_dnode, 2028 sizeof (os->os_phys->os_meta_dnode)) != 0); 2029 } 2030 return (B_FALSE); 2031 } 2032 2033 typedef struct dsl_dataset_rename_snapshot_arg { 2034 const char *ddrsa_fsname; 2035 const char *ddrsa_oldsnapname; 2036 const char *ddrsa_newsnapname; 2037 boolean_t ddrsa_recursive; 2038 dmu_tx_t *ddrsa_tx; 2039 } dsl_dataset_rename_snapshot_arg_t; 2040 2041 /* ARGSUSED */ 2042 static int 2043 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 2044 dsl_dataset_t *hds, void *arg) 2045 { 2046 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2047 int error; 2048 uint64_t val; 2049 2050 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2051 if (error != 0) { 2052 /* ignore nonexistent snapshots */ 2053 return (error == ENOENT ? 0 : error); 2054 } 2055 2056 /* new name should not exist */ 2057 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 2058 if (error == 0) 2059 error = SET_ERROR(EEXIST); 2060 else if (error == ENOENT) 2061 error = 0; 2062 2063 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2064 if (dsl_dir_namelen(hds->ds_dir) + 1 + 2065 strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN) 2066 error = SET_ERROR(ENAMETOOLONG); 2067 2068 return (error); 2069 } 2070 2071 static int 2072 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 2073 { 2074 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2075 dsl_pool_t *dp = dmu_tx_pool(tx); 2076 dsl_dataset_t *hds; 2077 int error; 2078 2079 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 2080 if (error != 0) 2081 return (error); 2082 2083 if (ddrsa->ddrsa_recursive) { 2084 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 2085 dsl_dataset_rename_snapshot_check_impl, ddrsa, 2086 DS_FIND_CHILDREN); 2087 } else { 2088 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 2089 } 2090 dsl_dataset_rele(hds, FTAG); 2091 return (error); 2092 } 2093 2094 static int 2095 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 2096 dsl_dataset_t *hds, void *arg) 2097 { 2098 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2099 dsl_dataset_t *ds; 2100 uint64_t val; 2101 dmu_tx_t *tx = ddrsa->ddrsa_tx; 2102 int error; 2103 2104 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2105 ASSERT(error == 0 || error == ENOENT); 2106 if (error == ENOENT) { 2107 /* ignore nonexistent snapshots */ 2108 return (0); 2109 } 2110 2111 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 2112 2113 /* log before we change the name */ 2114 spa_history_log_internal_ds(ds, "rename", tx, 2115 "-> @%s", ddrsa->ddrsa_newsnapname); 2116 2117 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 2118 B_FALSE)); 2119 mutex_enter(&ds->ds_lock); 2120 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 2121 mutex_exit(&ds->ds_lock); 2122 VERIFY0(zap_add(dp->dp_meta_objset, 2123 dsl_dataset_phys(hds)->ds_snapnames_zapobj, 2124 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 2125 2126 dsl_dataset_rele(ds, FTAG); 2127 return (0); 2128 } 2129 2130 static void 2131 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 2132 { 2133 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2134 dsl_pool_t *dp = dmu_tx_pool(tx); 2135 dsl_dataset_t *hds; 2136 2137 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 2138 ddrsa->ddrsa_tx = tx; 2139 if (ddrsa->ddrsa_recursive) { 2140 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 2141 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 2142 DS_FIND_CHILDREN)); 2143 } else { 2144 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 2145 } 2146 dsl_dataset_rele(hds, FTAG); 2147 } 2148 2149 int 2150 dsl_dataset_rename_snapshot(const char *fsname, 2151 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 2152 { 2153 dsl_dataset_rename_snapshot_arg_t ddrsa; 2154 2155 ddrsa.ddrsa_fsname = fsname; 2156 ddrsa.ddrsa_oldsnapname = oldsnapname; 2157 ddrsa.ddrsa_newsnapname = newsnapname; 2158 ddrsa.ddrsa_recursive = recursive; 2159 2160 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 2161 dsl_dataset_rename_snapshot_sync, &ddrsa, 2162 1, ZFS_SPACE_CHECK_RESERVED)); 2163 } 2164 2165 /* 2166 * If we're doing an ownership handoff, we need to make sure that there is 2167 * only one long hold on the dataset. We're not allowed to change anything here 2168 * so we don't permanently release the long hold or regular hold here. We want 2169 * to do this only when syncing to avoid the dataset unexpectedly going away 2170 * when we release the long hold. 2171 */ 2172 static int 2173 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 2174 { 2175 boolean_t held; 2176 2177 if (!dmu_tx_is_syncing(tx)) 2178 return (0); 2179 2180 if (owner != NULL) { 2181 VERIFY3P(ds->ds_owner, ==, owner); 2182 dsl_dataset_long_rele(ds, owner); 2183 } 2184 2185 held = dsl_dataset_long_held(ds); 2186 2187 if (owner != NULL) 2188 dsl_dataset_long_hold(ds, owner); 2189 2190 if (held) 2191 return (SET_ERROR(EBUSY)); 2192 2193 return (0); 2194 } 2195 2196 typedef struct dsl_dataset_rollback_arg { 2197 const char *ddra_fsname; 2198 void *ddra_owner; 2199 nvlist_t *ddra_result; 2200 } dsl_dataset_rollback_arg_t; 2201 2202 static int 2203 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 2204 { 2205 dsl_dataset_rollback_arg_t *ddra = arg; 2206 dsl_pool_t *dp = dmu_tx_pool(tx); 2207 dsl_dataset_t *ds; 2208 int64_t unused_refres_delta; 2209 int error; 2210 2211 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 2212 if (error != 0) 2213 return (error); 2214 2215 /* must not be a snapshot */ 2216 if (ds->ds_is_snapshot) { 2217 dsl_dataset_rele(ds, FTAG); 2218 return (SET_ERROR(EINVAL)); 2219 } 2220 2221 /* must have a most recent snapshot */ 2222 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { 2223 dsl_dataset_rele(ds, FTAG); 2224 return (SET_ERROR(EINVAL)); 2225 } 2226 2227 /* 2228 * No rollback to a snapshot created in the current txg, because 2229 * the rollback may dirty the dataset and create blocks that are 2230 * not reachable from the rootbp while having a birth txg that 2231 * falls into the snapshot's range. 2232 */ 2233 if (dmu_tx_is_syncing(tx) && 2234 dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) { 2235 dsl_dataset_rele(ds, FTAG); 2236 return (SET_ERROR(EAGAIN)); 2237 } 2238 2239 /* must not have any bookmarks after the most recent snapshot */ 2240 nvlist_t *proprequest = fnvlist_alloc(); 2241 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2242 nvlist_t *bookmarks = fnvlist_alloc(); 2243 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 2244 fnvlist_free(proprequest); 2245 if (error != 0) 2246 return (error); 2247 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 2248 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 2249 nvlist_t *valuenv = 2250 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 2251 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2252 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 2253 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 2254 fnvlist_free(bookmarks); 2255 dsl_dataset_rele(ds, FTAG); 2256 return (SET_ERROR(EEXIST)); 2257 } 2258 } 2259 fnvlist_free(bookmarks); 2260 2261 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 2262 if (error != 0) { 2263 dsl_dataset_rele(ds, FTAG); 2264 return (error); 2265 } 2266 2267 /* 2268 * Check if the snap we are rolling back to uses more than 2269 * the refquota. 2270 */ 2271 if (ds->ds_quota != 0 && 2272 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { 2273 dsl_dataset_rele(ds, FTAG); 2274 return (SET_ERROR(EDQUOT)); 2275 } 2276 2277 /* 2278 * When we do the clone swap, we will temporarily use more space 2279 * due to the refreservation (the head will no longer have any 2280 * unique space, so the entire amount of the refreservation will need 2281 * to be free). We will immediately destroy the clone, freeing 2282 * this space, but the freeing happens over many txg's. 2283 */ 2284 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 2285 dsl_dataset_phys(ds)->ds_unique_bytes); 2286 2287 if (unused_refres_delta > 0 && 2288 unused_refres_delta > 2289 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 2290 dsl_dataset_rele(ds, FTAG); 2291 return (SET_ERROR(ENOSPC)); 2292 } 2293 2294 dsl_dataset_rele(ds, FTAG); 2295 return (0); 2296 } 2297 2298 static void 2299 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 2300 { 2301 dsl_dataset_rollback_arg_t *ddra = arg; 2302 dsl_pool_t *dp = dmu_tx_pool(tx); 2303 dsl_dataset_t *ds, *clone; 2304 uint64_t cloneobj; 2305 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 2306 2307 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 2308 2309 dsl_dataset_name(ds->ds_prev, namebuf); 2310 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 2311 2312 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 2313 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 2314 2315 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 2316 2317 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 2318 dsl_dataset_zero_zil(ds, tx); 2319 2320 dsl_destroy_head_sync_impl(clone, tx); 2321 2322 dsl_dataset_rele(clone, FTAG); 2323 dsl_dataset_rele(ds, FTAG); 2324 } 2325 2326 /* 2327 * Rolls back the given filesystem or volume to the most recent snapshot. 2328 * The name of the most recent snapshot will be returned under key "target" 2329 * in the result nvlist. 2330 * 2331 * If owner != NULL: 2332 * - The existing dataset MUST be owned by the specified owner at entry 2333 * - Upon return, dataset will still be held by the same owner, whether we 2334 * succeed or not. 2335 * 2336 * This mode is required any time the existing filesystem is mounted. See 2337 * notes above zfs_suspend_fs() for further details. 2338 */ 2339 int 2340 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result) 2341 { 2342 dsl_dataset_rollback_arg_t ddra; 2343 2344 ddra.ddra_fsname = fsname; 2345 ddra.ddra_owner = owner; 2346 ddra.ddra_result = result; 2347 2348 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 2349 dsl_dataset_rollback_sync, &ddra, 2350 1, ZFS_SPACE_CHECK_RESERVED)); 2351 } 2352 2353 struct promotenode { 2354 list_node_t link; 2355 dsl_dataset_t *ds; 2356 }; 2357 2358 typedef struct dsl_dataset_promote_arg { 2359 const char *ddpa_clonename; 2360 dsl_dataset_t *ddpa_clone; 2361 list_t shared_snaps, origin_snaps, clone_snaps; 2362 dsl_dataset_t *origin_origin; /* origin of the origin */ 2363 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2364 char *err_ds; 2365 cred_t *cr; 2366 } dsl_dataset_promote_arg_t; 2367 2368 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2369 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2370 void *tag); 2371 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2372 2373 static int 2374 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2375 { 2376 dsl_dataset_promote_arg_t *ddpa = arg; 2377 dsl_pool_t *dp = dmu_tx_pool(tx); 2378 dsl_dataset_t *hds; 2379 struct promotenode *snap; 2380 dsl_dataset_t *origin_ds; 2381 int err; 2382 uint64_t unused; 2383 uint64_t ss_mv_cnt; 2384 size_t max_snap_len; 2385 2386 err = promote_hold(ddpa, dp, FTAG); 2387 if (err != 0) 2388 return (err); 2389 2390 hds = ddpa->ddpa_clone; 2391 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; 2392 2393 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { 2394 promote_rele(ddpa, FTAG); 2395 return (SET_ERROR(EXDEV)); 2396 } 2397 2398 /* 2399 * Compute and check the amount of space to transfer. Since this is 2400 * so expensive, don't do the preliminary check. 2401 */ 2402 if (!dmu_tx_is_syncing(tx)) { 2403 promote_rele(ddpa, FTAG); 2404 return (0); 2405 } 2406 2407 snap = list_head(&ddpa->shared_snaps); 2408 origin_ds = snap->ds; 2409 2410 /* compute origin's new unique space */ 2411 snap = list_tail(&ddpa->clone_snaps); 2412 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2413 origin_ds->ds_object); 2414 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2415 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, 2416 &ddpa->unique, &unused, &unused); 2417 2418 /* 2419 * Walk the snapshots that we are moving 2420 * 2421 * Compute space to transfer. Consider the incremental changes 2422 * to used by each snapshot: 2423 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2424 * So each snapshot gave birth to: 2425 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2426 * So a sequence would look like: 2427 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2428 * Which simplifies to: 2429 * uN + kN + kN-1 + ... + k1 + k0 2430 * Note however, if we stop before we reach the ORIGIN we get: 2431 * uN + kN + kN-1 + ... + kM - uM-1 2432 */ 2433 ss_mv_cnt = 0; 2434 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; 2435 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; 2436 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; 2437 for (snap = list_head(&ddpa->shared_snaps); snap; 2438 snap = list_next(&ddpa->shared_snaps, snap)) { 2439 uint64_t val, dlused, dlcomp, dluncomp; 2440 dsl_dataset_t *ds = snap->ds; 2441 2442 ss_mv_cnt++; 2443 2444 /* 2445 * If there are long holds, we won't be able to evict 2446 * the objset. 2447 */ 2448 if (dsl_dataset_long_held(ds)) { 2449 err = SET_ERROR(EBUSY); 2450 goto out; 2451 } 2452 2453 /* Check that the snapshot name does not conflict */ 2454 VERIFY0(dsl_dataset_get_snapname(ds)); 2455 if (strlen(ds->ds_snapname) >= max_snap_len) { 2456 err = SET_ERROR(ENAMETOOLONG); 2457 goto out; 2458 } 2459 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2460 if (err == 0) { 2461 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname); 2462 err = SET_ERROR(EEXIST); 2463 goto out; 2464 } 2465 if (err != ENOENT) 2466 goto out; 2467 2468 /* The very first snapshot does not have a deadlist */ 2469 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) 2470 continue; 2471 2472 dsl_deadlist_space(&ds->ds_deadlist, 2473 &dlused, &dlcomp, &dluncomp); 2474 ddpa->used += dlused; 2475 ddpa->comp += dlcomp; 2476 ddpa->uncomp += dluncomp; 2477 } 2478 2479 /* 2480 * If we are a clone of a clone then we never reached ORIGIN, 2481 * so we need to subtract out the clone origin's used space. 2482 */ 2483 if (ddpa->origin_origin) { 2484 ddpa->used -= 2485 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; 2486 ddpa->comp -= 2487 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; 2488 ddpa->uncomp -= 2489 dsl_dataset_phys(ddpa->origin_origin)-> 2490 ds_uncompressed_bytes; 2491 } 2492 2493 /* Check that there is enough space and limit headroom here */ 2494 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2495 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2496 if (err != 0) 2497 goto out; 2498 2499 /* 2500 * Compute the amounts of space that will be used by snapshots 2501 * after the promotion (for both origin and clone). For each, 2502 * it is the amount of space that will be on all of their 2503 * deadlists (that was not born before their new origin). 2504 */ 2505 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2506 uint64_t space; 2507 2508 /* 2509 * Note, typically this will not be a clone of a clone, 2510 * so dd_origin_txg will be < TXG_INITIAL, so 2511 * these snaplist_space() -> dsl_deadlist_space_range() 2512 * calls will be fast because they do not have to 2513 * iterate over all bps. 2514 */ 2515 snap = list_head(&ddpa->origin_snaps); 2516 err = snaplist_space(&ddpa->shared_snaps, 2517 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2518 if (err != 0) 2519 goto out; 2520 2521 err = snaplist_space(&ddpa->clone_snaps, 2522 snap->ds->ds_dir->dd_origin_txg, &space); 2523 if (err != 0) 2524 goto out; 2525 ddpa->cloneusedsnap += space; 2526 } 2527 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & 2528 DD_FLAG_USED_BREAKDOWN) { 2529 err = snaplist_space(&ddpa->origin_snaps, 2530 dsl_dataset_phys(origin_ds)->ds_creation_txg, 2531 &ddpa->originusedsnap); 2532 if (err != 0) 2533 goto out; 2534 } 2535 2536 out: 2537 promote_rele(ddpa, FTAG); 2538 return (err); 2539 } 2540 2541 static void 2542 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2543 { 2544 dsl_dataset_promote_arg_t *ddpa = arg; 2545 dsl_pool_t *dp = dmu_tx_pool(tx); 2546 dsl_dataset_t *hds; 2547 struct promotenode *snap; 2548 dsl_dataset_t *origin_ds; 2549 dsl_dataset_t *origin_head; 2550 dsl_dir_t *dd; 2551 dsl_dir_t *odd = NULL; 2552 uint64_t oldnext_obj; 2553 int64_t delta; 2554 2555 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2556 hds = ddpa->ddpa_clone; 2557 2558 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); 2559 2560 snap = list_head(&ddpa->shared_snaps); 2561 origin_ds = snap->ds; 2562 dd = hds->ds_dir; 2563 2564 snap = list_head(&ddpa->origin_snaps); 2565 origin_head = snap->ds; 2566 2567 /* 2568 * We need to explicitly open odd, since origin_ds's dd will be 2569 * changing. 2570 */ 2571 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2572 NULL, FTAG, &odd)); 2573 2574 /* change origin's next snap */ 2575 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2576 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; 2577 snap = list_tail(&ddpa->clone_snaps); 2578 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2579 origin_ds->ds_object); 2580 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; 2581 2582 /* change the origin's next clone */ 2583 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { 2584 dsl_dataset_remove_from_next_clones(origin_ds, 2585 snap->ds->ds_object, tx); 2586 VERIFY0(zap_add_int(dp->dp_meta_objset, 2587 dsl_dataset_phys(origin_ds)->ds_next_clones_obj, 2588 oldnext_obj, tx)); 2589 } 2590 2591 /* change origin */ 2592 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2593 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); 2594 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; 2595 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2596 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2597 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; 2598 origin_head->ds_dir->dd_origin_txg = 2599 dsl_dataset_phys(origin_ds)->ds_creation_txg; 2600 2601 /* change dd_clone entries */ 2602 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2603 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2604 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); 2605 VERIFY0(zap_add_int(dp->dp_meta_objset, 2606 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2607 hds->ds_object, tx)); 2608 2609 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2610 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2611 origin_head->ds_object, tx)); 2612 if (dsl_dir_phys(dd)->dd_clones == 0) { 2613 dsl_dir_phys(dd)->dd_clones = 2614 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, 2615 DMU_OT_NONE, 0, tx); 2616 } 2617 VERIFY0(zap_add_int(dp->dp_meta_objset, 2618 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); 2619 } 2620 2621 /* move snapshots to this dir */ 2622 for (snap = list_head(&ddpa->shared_snaps); snap; 2623 snap = list_next(&ddpa->shared_snaps, snap)) { 2624 dsl_dataset_t *ds = snap->ds; 2625 2626 /* 2627 * Property callbacks are registered to a particular 2628 * dsl_dir. Since ours is changing, evict the objset 2629 * so that they will be unregistered from the old dsl_dir. 2630 */ 2631 if (ds->ds_objset) { 2632 dmu_objset_evict(ds->ds_objset); 2633 ds->ds_objset = NULL; 2634 } 2635 2636 /* move snap name entry */ 2637 VERIFY0(dsl_dataset_get_snapname(ds)); 2638 VERIFY0(dsl_dataset_snap_remove(origin_head, 2639 ds->ds_snapname, tx, B_TRUE)); 2640 VERIFY0(zap_add(dp->dp_meta_objset, 2641 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 2642 8, 1, &ds->ds_object, tx)); 2643 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 2644 DD_FIELD_SNAPSHOT_COUNT, tx); 2645 2646 /* change containing dsl_dir */ 2647 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2648 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); 2649 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; 2650 ASSERT3P(ds->ds_dir, ==, odd); 2651 dsl_dir_rele(ds->ds_dir, ds); 2652 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 2653 NULL, ds, &ds->ds_dir)); 2654 2655 /* move any clone references */ 2656 if (dsl_dataset_phys(ds)->ds_next_clones_obj && 2657 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2658 zap_cursor_t zc; 2659 zap_attribute_t za; 2660 2661 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2662 dsl_dataset_phys(ds)->ds_next_clones_obj); 2663 zap_cursor_retrieve(&zc, &za) == 0; 2664 zap_cursor_advance(&zc)) { 2665 dsl_dataset_t *cnds; 2666 uint64_t o; 2667 2668 if (za.za_first_integer == oldnext_obj) { 2669 /* 2670 * We've already moved the 2671 * origin's reference. 2672 */ 2673 continue; 2674 } 2675 2676 VERIFY0(dsl_dataset_hold_obj(dp, 2677 za.za_first_integer, FTAG, &cnds)); 2678 o = dsl_dir_phys(cnds->ds_dir)-> 2679 dd_head_dataset_obj; 2680 2681 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2682 dsl_dir_phys(odd)->dd_clones, o, tx)); 2683 VERIFY0(zap_add_int(dp->dp_meta_objset, 2684 dsl_dir_phys(dd)->dd_clones, o, tx)); 2685 dsl_dataset_rele(cnds, FTAG); 2686 } 2687 zap_cursor_fini(&zc); 2688 } 2689 2690 ASSERT(!dsl_prop_hascb(ds)); 2691 } 2692 2693 /* 2694 * Change space accounting. 2695 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2696 * both be valid, or both be 0 (resulting in delta == 0). This 2697 * is true for each of {clone,origin} independently. 2698 */ 2699 2700 delta = ddpa->cloneusedsnap - 2701 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; 2702 ASSERT3S(delta, >=, 0); 2703 ASSERT3U(ddpa->used, >=, delta); 2704 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2705 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2706 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 2707 2708 delta = ddpa->originusedsnap - 2709 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; 2710 ASSERT3S(delta, <=, 0); 2711 ASSERT3U(ddpa->used, >=, -delta); 2712 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2713 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2714 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 2715 2716 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; 2717 2718 /* log history record */ 2719 spa_history_log_internal_ds(hds, "promote", tx, ""); 2720 2721 dsl_dir_rele(odd, FTAG); 2722 promote_rele(ddpa, FTAG); 2723 } 2724 2725 /* 2726 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2727 * (exclusive) and last_obj (inclusive). The list will be in reverse 2728 * order (last_obj will be the list_head()). If first_obj == 0, do all 2729 * snapshots back to this dataset's origin. 2730 */ 2731 static int 2732 snaplist_make(dsl_pool_t *dp, 2733 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 2734 { 2735 uint64_t obj = last_obj; 2736 2737 list_create(l, sizeof (struct promotenode), 2738 offsetof(struct promotenode, link)); 2739 2740 while (obj != first_obj) { 2741 dsl_dataset_t *ds; 2742 struct promotenode *snap; 2743 int err; 2744 2745 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 2746 ASSERT(err != ENOENT); 2747 if (err != 0) 2748 return (err); 2749 2750 if (first_obj == 0) 2751 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; 2752 2753 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 2754 snap->ds = ds; 2755 list_insert_tail(l, snap); 2756 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 2757 } 2758 2759 return (0); 2760 } 2761 2762 static int 2763 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2764 { 2765 struct promotenode *snap; 2766 2767 *spacep = 0; 2768 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2769 uint64_t used, comp, uncomp; 2770 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2771 mintxg, UINT64_MAX, &used, &comp, &uncomp); 2772 *spacep += used; 2773 } 2774 return (0); 2775 } 2776 2777 static void 2778 snaplist_destroy(list_t *l, void *tag) 2779 { 2780 struct promotenode *snap; 2781 2782 if (l == NULL || !list_link_active(&l->list_head)) 2783 return; 2784 2785 while ((snap = list_tail(l)) != NULL) { 2786 list_remove(l, snap); 2787 dsl_dataset_rele(snap->ds, tag); 2788 kmem_free(snap, sizeof (*snap)); 2789 } 2790 list_destroy(l); 2791 } 2792 2793 static int 2794 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 2795 { 2796 int error; 2797 dsl_dir_t *dd; 2798 struct promotenode *snap; 2799 2800 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 2801 &ddpa->ddpa_clone); 2802 if (error != 0) 2803 return (error); 2804 dd = ddpa->ddpa_clone->ds_dir; 2805 2806 if (ddpa->ddpa_clone->ds_is_snapshot || 2807 !dsl_dir_is_clone(dd)) { 2808 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2809 return (SET_ERROR(EINVAL)); 2810 } 2811 2812 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, 2813 &ddpa->shared_snaps, tag); 2814 if (error != 0) 2815 goto out; 2816 2817 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 2818 &ddpa->clone_snaps, tag); 2819 if (error != 0) 2820 goto out; 2821 2822 snap = list_head(&ddpa->shared_snaps); 2823 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); 2824 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, 2825 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, 2826 &ddpa->origin_snaps, tag); 2827 if (error != 0) 2828 goto out; 2829 2830 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { 2831 error = dsl_dataset_hold_obj(dp, 2832 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, 2833 tag, &ddpa->origin_origin); 2834 if (error != 0) 2835 goto out; 2836 } 2837 out: 2838 if (error != 0) 2839 promote_rele(ddpa, tag); 2840 return (error); 2841 } 2842 2843 static void 2844 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 2845 { 2846 snaplist_destroy(&ddpa->shared_snaps, tag); 2847 snaplist_destroy(&ddpa->clone_snaps, tag); 2848 snaplist_destroy(&ddpa->origin_snaps, tag); 2849 if (ddpa->origin_origin != NULL) 2850 dsl_dataset_rele(ddpa->origin_origin, tag); 2851 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2852 } 2853 2854 /* 2855 * Promote a clone. 2856 * 2857 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 2858 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.) 2859 */ 2860 int 2861 dsl_dataset_promote(const char *name, char *conflsnap) 2862 { 2863 dsl_dataset_promote_arg_t ddpa = { 0 }; 2864 uint64_t numsnaps; 2865 int error; 2866 objset_t *os; 2867 2868 /* 2869 * We will modify space proportional to the number of 2870 * snapshots. Compute numsnaps. 2871 */ 2872 error = dmu_objset_hold(name, FTAG, &os); 2873 if (error != 0) 2874 return (error); 2875 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 2876 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, 2877 &numsnaps); 2878 dmu_objset_rele(os, FTAG); 2879 if (error != 0) 2880 return (error); 2881 2882 ddpa.ddpa_clonename = name; 2883 ddpa.err_ds = conflsnap; 2884 ddpa.cr = CRED(); 2885 2886 return (dsl_sync_task(name, dsl_dataset_promote_check, 2887 dsl_dataset_promote_sync, &ddpa, 2888 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED)); 2889 } 2890 2891 int 2892 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 2893 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 2894 { 2895 /* 2896 * "slack" factor for received datasets with refquota set on them. 2897 * See the bottom of this function for details on its use. 2898 */ 2899 uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation; 2900 int64_t unused_refres_delta; 2901 2902 /* they should both be heads */ 2903 if (clone->ds_is_snapshot || 2904 origin_head->ds_is_snapshot) 2905 return (SET_ERROR(EINVAL)); 2906 2907 /* if we are not forcing, the branch point should be just before them */ 2908 if (!force && clone->ds_prev != origin_head->ds_prev) 2909 return (SET_ERROR(EINVAL)); 2910 2911 /* clone should be the clone (unless they are unrelated) */ 2912 if (clone->ds_prev != NULL && 2913 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 2914 origin_head->ds_dir != clone->ds_prev->ds_dir) 2915 return (SET_ERROR(EINVAL)); 2916 2917 /* the clone should be a child of the origin */ 2918 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 2919 return (SET_ERROR(EINVAL)); 2920 2921 /* origin_head shouldn't be modified unless 'force' */ 2922 if (!force && 2923 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 2924 return (SET_ERROR(ETXTBSY)); 2925 2926 /* origin_head should have no long holds (e.g. is not mounted) */ 2927 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 2928 return (SET_ERROR(EBUSY)); 2929 2930 /* check amount of any unconsumed refreservation */ 2931 unused_refres_delta = 2932 (int64_t)MIN(origin_head->ds_reserved, 2933 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 2934 (int64_t)MIN(origin_head->ds_reserved, 2935 dsl_dataset_phys(clone)->ds_unique_bytes); 2936 2937 if (unused_refres_delta > 0 && 2938 unused_refres_delta > 2939 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 2940 return (SET_ERROR(ENOSPC)); 2941 2942 /* 2943 * The clone can't be too much over the head's refquota. 2944 * 2945 * To ensure that the entire refquota can be used, we allow one 2946 * transaction to exceed the the refquota. Therefore, this check 2947 * needs to also allow for the space referenced to be more than the 2948 * refquota. The maximum amount of space that one transaction can use 2949 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this 2950 * overage ensures that we are able to receive a filesystem that 2951 * exceeds the refquota on the source system. 2952 * 2953 * So that overage is the refquota_slack we use below. 2954 */ 2955 if (origin_head->ds_quota != 0 && 2956 dsl_dataset_phys(clone)->ds_referenced_bytes > 2957 origin_head->ds_quota + refquota_slack) 2958 return (SET_ERROR(EDQUOT)); 2959 2960 return (0); 2961 } 2962 2963 void 2964 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 2965 dsl_dataset_t *origin_head, dmu_tx_t *tx) 2966 { 2967 dsl_pool_t *dp = dmu_tx_pool(tx); 2968 int64_t unused_refres_delta; 2969 2970 ASSERT(clone->ds_reserved == 0); 2971 /* 2972 * NOTE: On DEBUG kernels there could be a race between this and 2973 * the check function if spa_asize_inflation is adjusted... 2974 */ 2975 ASSERT(origin_head->ds_quota == 0 || 2976 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota + 2977 DMU_MAX_ACCESS * spa_asize_inflation); 2978 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 2979 2980 /* 2981 * Swap per-dataset feature flags. 2982 */ 2983 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 2984 if (!(spa_feature_table[f].fi_flags & 2985 ZFEATURE_FLAG_PER_DATASET)) { 2986 ASSERT(!clone->ds_feature_inuse[f]); 2987 ASSERT(!origin_head->ds_feature_inuse[f]); 2988 continue; 2989 } 2990 2991 boolean_t clone_inuse = clone->ds_feature_inuse[f]; 2992 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f]; 2993 2994 if (clone_inuse) { 2995 dsl_dataset_deactivate_feature(clone->ds_object, f, tx); 2996 clone->ds_feature_inuse[f] = B_FALSE; 2997 } 2998 if (origin_head_inuse) { 2999 dsl_dataset_deactivate_feature(origin_head->ds_object, 3000 f, tx); 3001 origin_head->ds_feature_inuse[f] = B_FALSE; 3002 } 3003 if (clone_inuse) { 3004 dsl_dataset_activate_feature(origin_head->ds_object, 3005 f, tx); 3006 origin_head->ds_feature_inuse[f] = B_TRUE; 3007 } 3008 if (origin_head_inuse) { 3009 dsl_dataset_activate_feature(clone->ds_object, f, tx); 3010 clone->ds_feature_inuse[f] = B_TRUE; 3011 } 3012 } 3013 3014 dmu_buf_will_dirty(clone->ds_dbuf, tx); 3015 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 3016 3017 if (clone->ds_objset != NULL) { 3018 dmu_objset_evict(clone->ds_objset); 3019 clone->ds_objset = NULL; 3020 } 3021 3022 if (origin_head->ds_objset != NULL) { 3023 dmu_objset_evict(origin_head->ds_objset); 3024 origin_head->ds_objset = NULL; 3025 } 3026 3027 unused_refres_delta = 3028 (int64_t)MIN(origin_head->ds_reserved, 3029 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 3030 (int64_t)MIN(origin_head->ds_reserved, 3031 dsl_dataset_phys(clone)->ds_unique_bytes); 3032 3033 /* 3034 * Reset origin's unique bytes, if it exists. 3035 */ 3036 if (clone->ds_prev) { 3037 dsl_dataset_t *origin = clone->ds_prev; 3038 uint64_t comp, uncomp; 3039 3040 dmu_buf_will_dirty(origin->ds_dbuf, tx); 3041 dsl_deadlist_space_range(&clone->ds_deadlist, 3042 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, 3043 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); 3044 } 3045 3046 /* swap blkptrs */ 3047 { 3048 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG); 3049 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG); 3050 blkptr_t tmp; 3051 tmp = dsl_dataset_phys(origin_head)->ds_bp; 3052 dsl_dataset_phys(origin_head)->ds_bp = 3053 dsl_dataset_phys(clone)->ds_bp; 3054 dsl_dataset_phys(clone)->ds_bp = tmp; 3055 rrw_exit(&origin_head->ds_bp_rwlock, FTAG); 3056 rrw_exit(&clone->ds_bp_rwlock, FTAG); 3057 } 3058 3059 /* set dd_*_bytes */ 3060 { 3061 int64_t dused, dcomp, duncomp; 3062 uint64_t cdl_used, cdl_comp, cdl_uncomp; 3063 uint64_t odl_used, odl_comp, odl_uncomp; 3064 3065 ASSERT3U(dsl_dir_phys(clone->ds_dir)-> 3066 dd_used_breakdown[DD_USED_SNAP], ==, 0); 3067 3068 dsl_deadlist_space(&clone->ds_deadlist, 3069 &cdl_used, &cdl_comp, &cdl_uncomp); 3070 dsl_deadlist_space(&origin_head->ds_deadlist, 3071 &odl_used, &odl_comp, &odl_uncomp); 3072 3073 dused = dsl_dataset_phys(clone)->ds_referenced_bytes + 3074 cdl_used - 3075 (dsl_dataset_phys(origin_head)->ds_referenced_bytes + 3076 odl_used); 3077 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + 3078 cdl_comp - 3079 (dsl_dataset_phys(origin_head)->ds_compressed_bytes + 3080 odl_comp); 3081 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + 3082 cdl_uncomp - 3083 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + 3084 odl_uncomp); 3085 3086 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 3087 dused, dcomp, duncomp, tx); 3088 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 3089 -dused, -dcomp, -duncomp, tx); 3090 3091 /* 3092 * The difference in the space used by snapshots is the 3093 * difference in snapshot space due to the head's 3094 * deadlist (since that's the only thing that's 3095 * changing that affects the snapused). 3096 */ 3097 dsl_deadlist_space_range(&clone->ds_deadlist, 3098 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 3099 &cdl_used, &cdl_comp, &cdl_uncomp); 3100 dsl_deadlist_space_range(&origin_head->ds_deadlist, 3101 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 3102 &odl_used, &odl_comp, &odl_uncomp); 3103 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 3104 DD_USED_HEAD, DD_USED_SNAP, tx); 3105 } 3106 3107 /* swap ds_*_bytes */ 3108 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, 3109 dsl_dataset_phys(clone)->ds_referenced_bytes); 3110 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, 3111 dsl_dataset_phys(clone)->ds_compressed_bytes); 3112 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, 3113 dsl_dataset_phys(clone)->ds_uncompressed_bytes); 3114 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, 3115 dsl_dataset_phys(clone)->ds_unique_bytes); 3116 3117 /* apply any parent delta for change in unconsumed refreservation */ 3118 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 3119 unused_refres_delta, 0, 0, tx); 3120 3121 /* 3122 * Swap deadlists. 3123 */ 3124 dsl_deadlist_close(&clone->ds_deadlist); 3125 dsl_deadlist_close(&origin_head->ds_deadlist); 3126 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, 3127 dsl_dataset_phys(clone)->ds_deadlist_obj); 3128 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 3129 dsl_dataset_phys(clone)->ds_deadlist_obj); 3130 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 3131 dsl_dataset_phys(origin_head)->ds_deadlist_obj); 3132 3133 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 3134 3135 spa_history_log_internal_ds(clone, "clone swap", tx, 3136 "parent=%s", origin_head->ds_dir->dd_myname); 3137 } 3138 3139 /* 3140 * Given a pool name and a dataset object number in that pool, 3141 * return the name of that dataset. 3142 */ 3143 int 3144 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 3145 { 3146 dsl_pool_t *dp; 3147 dsl_dataset_t *ds; 3148 int error; 3149 3150 error = dsl_pool_hold(pname, FTAG, &dp); 3151 if (error != 0) 3152 return (error); 3153 3154 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 3155 if (error == 0) { 3156 dsl_dataset_name(ds, buf); 3157 dsl_dataset_rele(ds, FTAG); 3158 } 3159 dsl_pool_rele(dp, FTAG); 3160 3161 return (error); 3162 } 3163 3164 int 3165 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 3166 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 3167 { 3168 int error = 0; 3169 3170 ASSERT3S(asize, >, 0); 3171 3172 /* 3173 * *ref_rsrv is the portion of asize that will come from any 3174 * unconsumed refreservation space. 3175 */ 3176 *ref_rsrv = 0; 3177 3178 mutex_enter(&ds->ds_lock); 3179 /* 3180 * Make a space adjustment for reserved bytes. 3181 */ 3182 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 3183 ASSERT3U(*used, >=, 3184 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 3185 *used -= 3186 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 3187 *ref_rsrv = 3188 asize - MIN(asize, parent_delta(ds, asize + inflight)); 3189 } 3190 3191 if (!check_quota || ds->ds_quota == 0) { 3192 mutex_exit(&ds->ds_lock); 3193 return (0); 3194 } 3195 /* 3196 * If they are requesting more space, and our current estimate 3197 * is over quota, they get to try again unless the actual 3198 * on-disk is over quota and there are no pending changes (which 3199 * may free up space for us). 3200 */ 3201 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= 3202 ds->ds_quota) { 3203 if (inflight > 0 || 3204 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) 3205 error = SET_ERROR(ERESTART); 3206 else 3207 error = SET_ERROR(EDQUOT); 3208 } 3209 mutex_exit(&ds->ds_lock); 3210 3211 return (error); 3212 } 3213 3214 typedef struct dsl_dataset_set_qr_arg { 3215 const char *ddsqra_name; 3216 zprop_source_t ddsqra_source; 3217 uint64_t ddsqra_value; 3218 } dsl_dataset_set_qr_arg_t; 3219 3220 3221 /* ARGSUSED */ 3222 static int 3223 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 3224 { 3225 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3226 dsl_pool_t *dp = dmu_tx_pool(tx); 3227 dsl_dataset_t *ds; 3228 int error; 3229 uint64_t newval; 3230 3231 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 3232 return (SET_ERROR(ENOTSUP)); 3233 3234 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3235 if (error != 0) 3236 return (error); 3237 3238 if (ds->ds_is_snapshot) { 3239 dsl_dataset_rele(ds, FTAG); 3240 return (SET_ERROR(EINVAL)); 3241 } 3242 3243 error = dsl_prop_predict(ds->ds_dir, 3244 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3245 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3246 if (error != 0) { 3247 dsl_dataset_rele(ds, FTAG); 3248 return (error); 3249 } 3250 3251 if (newval == 0) { 3252 dsl_dataset_rele(ds, FTAG); 3253 return (0); 3254 } 3255 3256 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || 3257 newval < ds->ds_reserved) { 3258 dsl_dataset_rele(ds, FTAG); 3259 return (SET_ERROR(ENOSPC)); 3260 } 3261 3262 dsl_dataset_rele(ds, FTAG); 3263 return (0); 3264 } 3265 3266 static void 3267 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 3268 { 3269 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3270 dsl_pool_t *dp = dmu_tx_pool(tx); 3271 dsl_dataset_t *ds; 3272 uint64_t newval; 3273 3274 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3275 3276 dsl_prop_set_sync_impl(ds, 3277 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3278 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 3279 &ddsqra->ddsqra_value, tx); 3280 3281 VERIFY0(dsl_prop_get_int_ds(ds, 3282 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 3283 3284 if (ds->ds_quota != newval) { 3285 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3286 ds->ds_quota = newval; 3287 } 3288 dsl_dataset_rele(ds, FTAG); 3289 } 3290 3291 int 3292 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 3293 uint64_t refquota) 3294 { 3295 dsl_dataset_set_qr_arg_t ddsqra; 3296 3297 ddsqra.ddsqra_name = dsname; 3298 ddsqra.ddsqra_source = source; 3299 ddsqra.ddsqra_value = refquota; 3300 3301 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 3302 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 3303 } 3304 3305 static int 3306 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 3307 { 3308 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3309 dsl_pool_t *dp = dmu_tx_pool(tx); 3310 dsl_dataset_t *ds; 3311 int error; 3312 uint64_t newval, unique; 3313 3314 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 3315 return (SET_ERROR(ENOTSUP)); 3316 3317 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3318 if (error != 0) 3319 return (error); 3320 3321 if (ds->ds_is_snapshot) { 3322 dsl_dataset_rele(ds, FTAG); 3323 return (SET_ERROR(EINVAL)); 3324 } 3325 3326 error = dsl_prop_predict(ds->ds_dir, 3327 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3328 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3329 if (error != 0) { 3330 dsl_dataset_rele(ds, FTAG); 3331 return (error); 3332 } 3333 3334 /* 3335 * If we are doing the preliminary check in open context, the 3336 * space estimates may be inaccurate. 3337 */ 3338 if (!dmu_tx_is_syncing(tx)) { 3339 dsl_dataset_rele(ds, FTAG); 3340 return (0); 3341 } 3342 3343 mutex_enter(&ds->ds_lock); 3344 if (!DS_UNIQUE_IS_ACCURATE(ds)) 3345 dsl_dataset_recalc_head_uniq(ds); 3346 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3347 mutex_exit(&ds->ds_lock); 3348 3349 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 3350 uint64_t delta = MAX(unique, newval) - 3351 MAX(unique, ds->ds_reserved); 3352 3353 if (delta > 3354 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 3355 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 3356 dsl_dataset_rele(ds, FTAG); 3357 return (SET_ERROR(ENOSPC)); 3358 } 3359 } 3360 3361 dsl_dataset_rele(ds, FTAG); 3362 return (0); 3363 } 3364 3365 void 3366 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 3367 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 3368 { 3369 uint64_t newval; 3370 uint64_t unique; 3371 int64_t delta; 3372 3373 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3374 source, sizeof (value), 1, &value, tx); 3375 3376 VERIFY0(dsl_prop_get_int_ds(ds, 3377 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 3378 3379 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3380 mutex_enter(&ds->ds_dir->dd_lock); 3381 mutex_enter(&ds->ds_lock); 3382 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3383 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3384 delta = MAX(0, (int64_t)(newval - unique)) - 3385 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3386 ds->ds_reserved = newval; 3387 mutex_exit(&ds->ds_lock); 3388 3389 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3390 mutex_exit(&ds->ds_dir->dd_lock); 3391 } 3392 3393 static void 3394 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 3395 { 3396 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3397 dsl_pool_t *dp = dmu_tx_pool(tx); 3398 dsl_dataset_t *ds; 3399 3400 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3401 dsl_dataset_set_refreservation_sync_impl(ds, 3402 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 3403 dsl_dataset_rele(ds, FTAG); 3404 } 3405 3406 int 3407 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 3408 uint64_t refreservation) 3409 { 3410 dsl_dataset_set_qr_arg_t ddsqra; 3411 3412 ddsqra.ddsqra_name = dsname; 3413 ddsqra.ddsqra_source = source; 3414 ddsqra.ddsqra_value = refreservation; 3415 3416 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 3417 dsl_dataset_set_refreservation_sync, &ddsqra, 3418 0, ZFS_SPACE_CHECK_NONE)); 3419 } 3420 3421 /* 3422 * Return (in *usedp) the amount of space written in new that is not 3423 * present in oldsnap. New may be a snapshot or the head. Old must be 3424 * a snapshot before new, in new's filesystem (or its origin). If not then 3425 * fail and return EINVAL. 3426 * 3427 * The written space is calculated by considering two components: First, we 3428 * ignore any freed space, and calculate the written as new's used space 3429 * minus old's used space. Next, we add in the amount of space that was freed 3430 * between the two snapshots, thus reducing new's used space relative to old's. 3431 * Specifically, this is the space that was born before old->ds_creation_txg, 3432 * and freed before new (ie. on new's deadlist or a previous deadlist). 3433 * 3434 * space freed [---------------------] 3435 * snapshots ---O-------O--------O-------O------ 3436 * oldsnap new 3437 */ 3438 int 3439 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 3440 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3441 { 3442 int err = 0; 3443 uint64_t snapobj; 3444 dsl_pool_t *dp = new->ds_dir->dd_pool; 3445 3446 ASSERT(dsl_pool_config_held(dp)); 3447 3448 *usedp = 0; 3449 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; 3450 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes; 3451 3452 *compp = 0; 3453 *compp += dsl_dataset_phys(new)->ds_compressed_bytes; 3454 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes; 3455 3456 *uncompp = 0; 3457 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; 3458 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes; 3459 3460 snapobj = new->ds_object; 3461 while (snapobj != oldsnap->ds_object) { 3462 dsl_dataset_t *snap; 3463 uint64_t used, comp, uncomp; 3464 3465 if (snapobj == new->ds_object) { 3466 snap = new; 3467 } else { 3468 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3469 if (err != 0) 3470 break; 3471 } 3472 3473 if (dsl_dataset_phys(snap)->ds_prev_snap_txg == 3474 dsl_dataset_phys(oldsnap)->ds_creation_txg) { 3475 /* 3476 * The blocks in the deadlist can not be born after 3477 * ds_prev_snap_txg, so get the whole deadlist space, 3478 * which is more efficient (especially for old-format 3479 * deadlists). Unfortunately the deadlist code 3480 * doesn't have enough information to make this 3481 * optimization itself. 3482 */ 3483 dsl_deadlist_space(&snap->ds_deadlist, 3484 &used, &comp, &uncomp); 3485 } else { 3486 dsl_deadlist_space_range(&snap->ds_deadlist, 3487 0, dsl_dataset_phys(oldsnap)->ds_creation_txg, 3488 &used, &comp, &uncomp); 3489 } 3490 *usedp += used; 3491 *compp += comp; 3492 *uncompp += uncomp; 3493 3494 /* 3495 * If we get to the beginning of the chain of snapshots 3496 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3497 * was not a snapshot of/before new. 3498 */ 3499 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 3500 if (snap != new) 3501 dsl_dataset_rele(snap, FTAG); 3502 if (snapobj == 0) { 3503 err = SET_ERROR(EINVAL); 3504 break; 3505 } 3506 3507 } 3508 return (err); 3509 } 3510 3511 /* 3512 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3513 * lastsnap, and all snapshots in between are deleted. 3514 * 3515 * blocks that would be freed [---------------------------] 3516 * snapshots ---O-------O--------O-------O--------O 3517 * firstsnap lastsnap 3518 * 3519 * This is the set of blocks that were born after the snap before firstsnap, 3520 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3521 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3522 * We calculate this by iterating over the relevant deadlists (from the snap 3523 * after lastsnap, backward to the snap after firstsnap), summing up the 3524 * space on the deadlist that was born after the snap before firstsnap. 3525 */ 3526 int 3527 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3528 dsl_dataset_t *lastsnap, 3529 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3530 { 3531 int err = 0; 3532 uint64_t snapobj; 3533 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3534 3535 ASSERT(firstsnap->ds_is_snapshot); 3536 ASSERT(lastsnap->ds_is_snapshot); 3537 3538 /* 3539 * Check that the snapshots are in the same dsl_dir, and firstsnap 3540 * is before lastsnap. 3541 */ 3542 if (firstsnap->ds_dir != lastsnap->ds_dir || 3543 dsl_dataset_phys(firstsnap)->ds_creation_txg > 3544 dsl_dataset_phys(lastsnap)->ds_creation_txg) 3545 return (SET_ERROR(EINVAL)); 3546 3547 *usedp = *compp = *uncompp = 0; 3548 3549 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; 3550 while (snapobj != firstsnap->ds_object) { 3551 dsl_dataset_t *ds; 3552 uint64_t used, comp, uncomp; 3553 3554 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3555 if (err != 0) 3556 break; 3557 3558 dsl_deadlist_space_range(&ds->ds_deadlist, 3559 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, 3560 &used, &comp, &uncomp); 3561 *usedp += used; 3562 *compp += comp; 3563 *uncompp += uncomp; 3564 3565 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3566 ASSERT3U(snapobj, !=, 0); 3567 dsl_dataset_rele(ds, FTAG); 3568 } 3569 return (err); 3570 } 3571 3572 /* 3573 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 3574 * For example, they could both be snapshots of the same filesystem, and 3575 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 3576 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 3577 * filesystem. Or 'earlier' could be the origin's origin. 3578 * 3579 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 3580 */ 3581 boolean_t 3582 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 3583 uint64_t earlier_txg) 3584 { 3585 dsl_pool_t *dp = later->ds_dir->dd_pool; 3586 int error; 3587 boolean_t ret; 3588 3589 ASSERT(dsl_pool_config_held(dp)); 3590 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); 3591 3592 if (earlier_txg == 0) 3593 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; 3594 3595 if (later->ds_is_snapshot && 3596 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) 3597 return (B_FALSE); 3598 3599 if (later->ds_dir == earlier->ds_dir) 3600 return (B_TRUE); 3601 if (!dsl_dir_is_clone(later->ds_dir)) 3602 return (B_FALSE); 3603 3604 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object) 3605 return (B_TRUE); 3606 dsl_dataset_t *origin; 3607 error = dsl_dataset_hold_obj(dp, 3608 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); 3609 if (error != 0) 3610 return (B_FALSE); 3611 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 3612 dsl_dataset_rele(origin, FTAG); 3613 return (ret); 3614 } 3615 3616 void 3617 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 3618 { 3619 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3620 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 3621 } 3622 3623 boolean_t 3624 dsl_dataset_is_zapified(dsl_dataset_t *ds) 3625 { 3626 dmu_object_info_t doi; 3627 3628 dmu_object_info_from_db(ds->ds_dbuf, &doi); 3629 return (doi.doi_type == DMU_OTN_ZAP_METADATA); 3630 } 3631 3632 boolean_t 3633 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds) 3634 { 3635 return (dsl_dataset_is_zapified(ds) && 3636 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset, 3637 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0); 3638 } 3639