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