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