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