1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2016 by Delphix. All rights reserved. 25 * Copyright (c) 2014, Joyent, Inc. All rights reserved. 26 * Copyright (c) 2014 RackTop Systems. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 * Copyright (c) 2014 Integros [integros.com] 29 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved. 30 * Copyright 2017 Nexenta Systems, Inc. 31 */ 32 33 #include <sys/dmu_objset.h> 34 #include <sys/dsl_dataset.h> 35 #include <sys/dsl_dir.h> 36 #include <sys/dsl_prop.h> 37 #include <sys/dsl_synctask.h> 38 #include <sys/dmu_traverse.h> 39 #include <sys/dmu_impl.h> 40 #include <sys/dmu_tx.h> 41 #include <sys/arc.h> 42 #include <sys/zio.h> 43 #include <sys/zap.h> 44 #include <sys/zfeature.h> 45 #include <sys/unique.h> 46 #include <sys/zfs_context.h> 47 #include <sys/zfs_ioctl.h> 48 #include <sys/spa.h> 49 #include <sys/zfs_znode.h> 50 #include <sys/zfs_onexit.h> 51 #include <sys/zvol.h> 52 #include <sys/dsl_scan.h> 53 #include <sys/dsl_deadlist.h> 54 #include <sys/dsl_destroy.h> 55 #include <sys/dsl_userhold.h> 56 #include <sys/dsl_bookmark.h> 57 #include <sys/dmu_send.h> 58 #include <sys/zio_checksum.h> 59 #include <sys/zio_compress.h> 60 #include <zfs_fletcher.h> 61 62 /* 63 * The SPA supports block sizes up to 16MB. However, very large blocks 64 * can have an impact on i/o latency (e.g. tying up a spinning disk for 65 * ~300ms), and also potentially on the memory allocator. Therefore, 66 * we do not allow the recordsize to be set larger than zfs_max_recordsize 67 * (default 1MB). Larger blocks can be created by changing this tunable, 68 * and pools with larger blocks can always be imported and used, regardless 69 * of this setting. 70 */ 71 int zfs_max_recordsize = 1 * 1024 * 1024; 72 73 #define SWITCH64(x, y) \ 74 { \ 75 uint64_t __tmp = (x); \ 76 (x) = (y); \ 77 (y) = __tmp; \ 78 } 79 80 #define DS_REF_MAX (1ULL << 62) 81 82 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds); 83 84 extern int spa_asize_inflation; 85 86 static zil_header_t zero_zil; 87 88 /* 89 * Figure out how much of this delta should be propogated to the dsl_dir 90 * layer. If there's a refreservation, that space has already been 91 * partially accounted for in our ancestors. 92 */ 93 static int64_t 94 parent_delta(dsl_dataset_t *ds, int64_t delta) 95 { 96 dsl_dataset_phys_t *ds_phys; 97 uint64_t old_bytes, new_bytes; 98 99 if (ds->ds_reserved == 0) 100 return (delta); 101 102 ds_phys = dsl_dataset_phys(ds); 103 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved); 104 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 105 106 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 107 return (new_bytes - old_bytes); 108 } 109 110 void 111 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 112 { 113 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 114 int compressed = BP_GET_PSIZE(bp); 115 int uncompressed = BP_GET_UCSIZE(bp); 116 int64_t delta; 117 118 dprintf_bp(bp, "ds=%p", ds); 119 120 ASSERT(dmu_tx_is_syncing(tx)); 121 /* It could have been compressed away to nothing */ 122 if (BP_IS_HOLE(bp)) 123 return; 124 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 125 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 126 if (ds == NULL) { 127 dsl_pool_mos_diduse_space(tx->tx_pool, 128 used, compressed, uncompressed); 129 return; 130 } 131 132 ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 133 dmu_buf_will_dirty(ds->ds_dbuf, tx); 134 mutex_enter(&ds->ds_lock); 135 delta = parent_delta(ds, used); 136 dsl_dataset_phys(ds)->ds_referenced_bytes += used; 137 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed; 138 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed; 139 dsl_dataset_phys(ds)->ds_unique_bytes += used; 140 141 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) { 142 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] = 143 B_TRUE; 144 } 145 146 spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp)); 147 if (f != SPA_FEATURE_NONE) 148 ds->ds_feature_activation_needed[f] = B_TRUE; 149 150 mutex_exit(&ds->ds_lock); 151 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 152 compressed, uncompressed, tx); 153 dsl_dir_transfer_space(ds->ds_dir, used - delta, 154 DD_USED_REFRSRV, DD_USED_HEAD, tx); 155 } 156 157 int 158 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 159 boolean_t async) 160 { 161 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 162 int compressed = BP_GET_PSIZE(bp); 163 int uncompressed = BP_GET_UCSIZE(bp); 164 165 if (BP_IS_HOLE(bp)) 166 return (0); 167 168 ASSERT(dmu_tx_is_syncing(tx)); 169 ASSERT(bp->blk_birth <= tx->tx_txg); 170 171 if (ds == NULL) { 172 dsl_free(tx->tx_pool, tx->tx_txg, bp); 173 dsl_pool_mos_diduse_space(tx->tx_pool, 174 -used, -compressed, -uncompressed); 175 return (used); 176 } 177 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 178 179 ASSERT(!ds->ds_is_snapshot); 180 dmu_buf_will_dirty(ds->ds_dbuf, tx); 181 182 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 183 int64_t delta; 184 185 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 186 dsl_free(tx->tx_pool, tx->tx_txg, bp); 187 188 mutex_enter(&ds->ds_lock); 189 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used || 190 !DS_UNIQUE_IS_ACCURATE(ds)); 191 delta = parent_delta(ds, -used); 192 dsl_dataset_phys(ds)->ds_unique_bytes -= used; 193 mutex_exit(&ds->ds_lock); 194 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 195 delta, -compressed, -uncompressed, tx); 196 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 197 DD_USED_REFRSRV, DD_USED_HEAD, tx); 198 } else { 199 dprintf_bp(bp, "putting on dead list: %s", ""); 200 if (async) { 201 /* 202 * We are here as part of zio's write done callback, 203 * which means we're a zio interrupt thread. We can't 204 * call dsl_deadlist_insert() now because it may block 205 * waiting for I/O. Instead, put bp on the deferred 206 * queue and let dsl_pool_sync() finish the job. 207 */ 208 bplist_append(&ds->ds_pending_deadlist, bp); 209 } else { 210 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); 211 } 212 ASSERT3U(ds->ds_prev->ds_object, ==, 213 dsl_dataset_phys(ds)->ds_prev_snap_obj); 214 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0); 215 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 216 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 217 ds->ds_object && bp->blk_birth > 218 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) { 219 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 220 mutex_enter(&ds->ds_prev->ds_lock); 221 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used; 222 mutex_exit(&ds->ds_prev->ds_lock); 223 } 224 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 225 dsl_dir_transfer_space(ds->ds_dir, used, 226 DD_USED_HEAD, DD_USED_SNAP, tx); 227 } 228 } 229 mutex_enter(&ds->ds_lock); 230 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used); 231 dsl_dataset_phys(ds)->ds_referenced_bytes -= used; 232 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed); 233 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed; 234 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed); 235 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed; 236 mutex_exit(&ds->ds_lock); 237 238 return (used); 239 } 240 241 /* 242 * We have to release the fsid syncronously or we risk that a subsequent 243 * mount of the same dataset will fail to unique_insert the fsid. This 244 * failure would manifest itself as the fsid of this dataset changing 245 * between mounts which makes NFS clients quite unhappy. 246 */ 247 static void 248 dsl_dataset_evict_sync(void *dbu) 249 { 250 dsl_dataset_t *ds = dbu; 251 252 ASSERT(ds->ds_owner == NULL); 253 254 unique_remove(ds->ds_fsid_guid); 255 } 256 257 static void 258 dsl_dataset_evict_async(void *dbu) 259 { 260 dsl_dataset_t *ds = dbu; 261 262 ASSERT(ds->ds_owner == NULL); 263 264 ds->ds_dbuf = NULL; 265 266 if (ds->ds_objset != NULL) 267 dmu_objset_evict(ds->ds_objset); 268 269 if (ds->ds_prev) { 270 dsl_dataset_rele(ds->ds_prev, ds); 271 ds->ds_prev = NULL; 272 } 273 274 bplist_destroy(&ds->ds_pending_deadlist); 275 if (ds->ds_deadlist.dl_os != NULL) 276 dsl_deadlist_close(&ds->ds_deadlist); 277 if (ds->ds_dir) 278 dsl_dir_async_rele(ds->ds_dir, ds); 279 280 ASSERT(!list_link_active(&ds->ds_synced_link)); 281 282 list_destroy(&ds->ds_prop_cbs); 283 mutex_destroy(&ds->ds_lock); 284 mutex_destroy(&ds->ds_opening_lock); 285 mutex_destroy(&ds->ds_sendstream_lock); 286 refcount_destroy(&ds->ds_longholds); 287 rrw_destroy(&ds->ds_bp_rwlock); 288 289 kmem_free(ds, sizeof (dsl_dataset_t)); 290 } 291 292 int 293 dsl_dataset_get_snapname(dsl_dataset_t *ds) 294 { 295 dsl_dataset_phys_t *headphys; 296 int err; 297 dmu_buf_t *headdbuf; 298 dsl_pool_t *dp = ds->ds_dir->dd_pool; 299 objset_t *mos = dp->dp_meta_objset; 300 301 if (ds->ds_snapname[0]) 302 return (0); 303 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) 304 return (0); 305 306 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, 307 FTAG, &headdbuf); 308 if (err != 0) 309 return (err); 310 headphys = headdbuf->db_data; 311 err = zap_value_search(dp->dp_meta_objset, 312 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 313 dmu_buf_rele(headdbuf, FTAG); 314 return (err); 315 } 316 317 int 318 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 319 { 320 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 321 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 322 matchtype_t mt = 0; 323 int err; 324 325 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 326 mt = MT_NORMALIZE; 327 328 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 329 value, mt, NULL, 0, NULL); 330 if (err == ENOTSUP && (mt & MT_NORMALIZE)) 331 err = zap_lookup(mos, snapobj, name, 8, 1, value); 332 return (err); 333 } 334 335 int 336 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 337 boolean_t adj_cnt) 338 { 339 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 340 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 341 matchtype_t mt = 0; 342 int err; 343 344 dsl_dir_snap_cmtime_update(ds->ds_dir); 345 346 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 347 mt = MT_NORMALIZE; 348 349 err = zap_remove_norm(mos, snapobj, name, mt, tx); 350 if (err == ENOTSUP && (mt & MT_NORMALIZE)) 351 err = zap_remove(mos, snapobj, name, tx); 352 353 if (err == 0 && adj_cnt) 354 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 355 DD_FIELD_SNAPSHOT_COUNT, tx); 356 357 return (err); 358 } 359 360 boolean_t 361 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag) 362 { 363 dmu_buf_t *dbuf = ds->ds_dbuf; 364 boolean_t result = B_FALSE; 365 366 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset, 367 ds->ds_object, DMU_BONUS_BLKID, tag)) { 368 369 if (ds == dmu_buf_get_user(dbuf)) 370 result = B_TRUE; 371 else 372 dmu_buf_rele(dbuf, tag); 373 } 374 375 return (result); 376 } 377 378 int 379 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 380 dsl_dataset_t **dsp) 381 { 382 objset_t *mos = dp->dp_meta_objset; 383 dmu_buf_t *dbuf; 384 dsl_dataset_t *ds; 385 int err; 386 dmu_object_info_t doi; 387 388 ASSERT(dsl_pool_config_held(dp)); 389 390 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 391 if (err != 0) 392 return (err); 393 394 /* Make sure dsobj has the correct object type. */ 395 dmu_object_info_from_db(dbuf, &doi); 396 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 397 dmu_buf_rele(dbuf, tag); 398 return (SET_ERROR(EINVAL)); 399 } 400 401 ds = dmu_buf_get_user(dbuf); 402 if (ds == NULL) { 403 dsl_dataset_t *winner = NULL; 404 405 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 406 ds->ds_dbuf = dbuf; 407 ds->ds_object = dsobj; 408 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0; 409 410 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 411 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 412 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 413 rrw_init(&ds->ds_bp_rwlock, B_FALSE); 414 refcount_create(&ds->ds_longholds); 415 416 bplist_create(&ds->ds_pending_deadlist); 417 dsl_deadlist_open(&ds->ds_deadlist, 418 mos, dsl_dataset_phys(ds)->ds_deadlist_obj); 419 420 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 421 offsetof(dmu_sendarg_t, dsa_link)); 422 423 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t), 424 offsetof(dsl_prop_cb_record_t, cbr_ds_node)); 425 426 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 427 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 428 if (!(spa_feature_table[f].fi_flags & 429 ZFEATURE_FLAG_PER_DATASET)) 430 continue; 431 err = zap_contains(mos, dsobj, 432 spa_feature_table[f].fi_guid); 433 if (err == 0) { 434 ds->ds_feature_inuse[f] = B_TRUE; 435 } else { 436 ASSERT3U(err, ==, ENOENT); 437 err = 0; 438 } 439 } 440 } 441 442 err = dsl_dir_hold_obj(dp, 443 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir); 444 if (err != 0) { 445 mutex_destroy(&ds->ds_lock); 446 mutex_destroy(&ds->ds_opening_lock); 447 mutex_destroy(&ds->ds_sendstream_lock); 448 refcount_destroy(&ds->ds_longholds); 449 bplist_destroy(&ds->ds_pending_deadlist); 450 dsl_deadlist_close(&ds->ds_deadlist); 451 kmem_free(ds, sizeof (dsl_dataset_t)); 452 dmu_buf_rele(dbuf, tag); 453 return (err); 454 } 455 456 if (!ds->ds_is_snapshot) { 457 ds->ds_snapname[0] = '\0'; 458 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 459 err = dsl_dataset_hold_obj(dp, 460 dsl_dataset_phys(ds)->ds_prev_snap_obj, 461 ds, &ds->ds_prev); 462 } 463 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 464 int zaperr = zap_lookup(mos, ds->ds_object, 465 DS_FIELD_BOOKMARK_NAMES, 466 sizeof (ds->ds_bookmarks), 1, 467 &ds->ds_bookmarks); 468 if (zaperr != ENOENT) 469 VERIFY0(zaperr); 470 } 471 } else { 472 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 473 err = dsl_dataset_get_snapname(ds); 474 if (err == 0 && 475 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { 476 err = zap_count( 477 ds->ds_dir->dd_pool->dp_meta_objset, 478 dsl_dataset_phys(ds)->ds_userrefs_obj, 479 &ds->ds_userrefs); 480 } 481 } 482 483 if (err == 0 && !ds->ds_is_snapshot) { 484 err = dsl_prop_get_int_ds(ds, 485 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 486 &ds->ds_reserved); 487 if (err == 0) { 488 err = dsl_prop_get_int_ds(ds, 489 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 490 &ds->ds_quota); 491 } 492 } else { 493 ds->ds_reserved = ds->ds_quota = 0; 494 } 495 496 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync, 497 dsl_dataset_evict_async, &ds->ds_dbuf); 498 if (err == 0) 499 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu); 500 501 if (err != 0 || winner != NULL) { 502 bplist_destroy(&ds->ds_pending_deadlist); 503 dsl_deadlist_close(&ds->ds_deadlist); 504 if (ds->ds_prev) 505 dsl_dataset_rele(ds->ds_prev, ds); 506 dsl_dir_rele(ds->ds_dir, ds); 507 mutex_destroy(&ds->ds_lock); 508 mutex_destroy(&ds->ds_opening_lock); 509 mutex_destroy(&ds->ds_sendstream_lock); 510 refcount_destroy(&ds->ds_longholds); 511 kmem_free(ds, sizeof (dsl_dataset_t)); 512 if (err != 0) { 513 dmu_buf_rele(dbuf, tag); 514 return (err); 515 } 516 ds = winner; 517 } else { 518 ds->ds_fsid_guid = 519 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); 520 if (ds->ds_fsid_guid != 521 dsl_dataset_phys(ds)->ds_fsid_guid) { 522 zfs_dbgmsg("ds_fsid_guid changed from " 523 "%llx to %llx for pool %s dataset id %llu", 524 (long long) 525 dsl_dataset_phys(ds)->ds_fsid_guid, 526 (long long)ds->ds_fsid_guid, 527 spa_name(dp->dp_spa), 528 dsobj); 529 } 530 } 531 } 532 ASSERT3P(ds->ds_dbuf, ==, dbuf); 533 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); 534 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || 535 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 536 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 537 *dsp = ds; 538 return (0); 539 } 540 541 int 542 dsl_dataset_hold(dsl_pool_t *dp, const char *name, 543 void *tag, dsl_dataset_t **dsp) 544 { 545 dsl_dir_t *dd; 546 const char *snapname; 547 uint64_t obj; 548 int err = 0; 549 dsl_dataset_t *ds; 550 551 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 552 if (err != 0) 553 return (err); 554 555 ASSERT(dsl_pool_config_held(dp)); 556 obj = dsl_dir_phys(dd)->dd_head_dataset_obj; 557 if (obj != 0) 558 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 559 else 560 err = SET_ERROR(ENOENT); 561 562 /* we may be looking for a snapshot */ 563 if (err == 0 && snapname != NULL) { 564 dsl_dataset_t *snap_ds; 565 566 if (*snapname++ != '@') { 567 dsl_dataset_rele(ds, tag); 568 dsl_dir_rele(dd, FTAG); 569 return (SET_ERROR(ENOENT)); 570 } 571 572 dprintf("looking for snapshot '%s'\n", snapname); 573 err = dsl_dataset_snap_lookup(ds, snapname, &obj); 574 if (err == 0) 575 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds); 576 dsl_dataset_rele(ds, tag); 577 578 if (err == 0) { 579 mutex_enter(&snap_ds->ds_lock); 580 if (snap_ds->ds_snapname[0] == 0) 581 (void) strlcpy(snap_ds->ds_snapname, snapname, 582 sizeof (snap_ds->ds_snapname)); 583 mutex_exit(&snap_ds->ds_lock); 584 ds = snap_ds; 585 } 586 } 587 if (err == 0) 588 *dsp = ds; 589 dsl_dir_rele(dd, FTAG); 590 return (err); 591 } 592 593 int 594 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 595 void *tag, dsl_dataset_t **dsp) 596 { 597 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 598 if (err != 0) 599 return (err); 600 if (!dsl_dataset_tryown(*dsp, tag)) { 601 dsl_dataset_rele(*dsp, tag); 602 *dsp = NULL; 603 return (SET_ERROR(EBUSY)); 604 } 605 return (0); 606 } 607 608 int 609 dsl_dataset_own(dsl_pool_t *dp, const char *name, 610 void *tag, dsl_dataset_t **dsp) 611 { 612 int err = dsl_dataset_hold(dp, name, tag, dsp); 613 if (err != 0) 614 return (err); 615 if (!dsl_dataset_tryown(*dsp, tag)) { 616 dsl_dataset_rele(*dsp, tag); 617 return (SET_ERROR(EBUSY)); 618 } 619 return (0); 620 } 621 622 /* 623 * See the comment above dsl_pool_hold() for details. In summary, a long 624 * hold is used to prevent destruction of a dataset while the pool hold 625 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 626 * 627 * The dataset and pool must be held when this function is called. After it 628 * is called, the pool hold may be released while the dataset is still held 629 * and accessed. 630 */ 631 void 632 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 633 { 634 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 635 (void) refcount_add(&ds->ds_longholds, tag); 636 } 637 638 void 639 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 640 { 641 (void) refcount_remove(&ds->ds_longholds, tag); 642 } 643 644 /* Return B_TRUE if there are any long holds on this dataset. */ 645 boolean_t 646 dsl_dataset_long_held(dsl_dataset_t *ds) 647 { 648 return (!refcount_is_zero(&ds->ds_longholds)); 649 } 650 651 void 652 dsl_dataset_name(dsl_dataset_t *ds, char *name) 653 { 654 if (ds == NULL) { 655 (void) strcpy(name, "mos"); 656 } else { 657 dsl_dir_name(ds->ds_dir, name); 658 VERIFY0(dsl_dataset_get_snapname(ds)); 659 if (ds->ds_snapname[0]) { 660 VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN), 661 <, ZFS_MAX_DATASET_NAME_LEN); 662 /* 663 * We use a "recursive" mutex so that we 664 * can call dprintf_ds() with ds_lock held. 665 */ 666 if (!MUTEX_HELD(&ds->ds_lock)) { 667 mutex_enter(&ds->ds_lock); 668 VERIFY3U(strlcat(name, ds->ds_snapname, 669 ZFS_MAX_DATASET_NAME_LEN), <, 670 ZFS_MAX_DATASET_NAME_LEN); 671 mutex_exit(&ds->ds_lock); 672 } else { 673 VERIFY3U(strlcat(name, ds->ds_snapname, 674 ZFS_MAX_DATASET_NAME_LEN), <, 675 ZFS_MAX_DATASET_NAME_LEN); 676 } 677 } 678 } 679 } 680 681 int 682 dsl_dataset_namelen(dsl_dataset_t *ds) 683 { 684 VERIFY0(dsl_dataset_get_snapname(ds)); 685 mutex_enter(&ds->ds_lock); 686 int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname); 687 mutex_exit(&ds->ds_lock); 688 return (len); 689 } 690 691 void 692 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 693 { 694 dmu_buf_rele(ds->ds_dbuf, tag); 695 } 696 697 void 698 dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 699 { 700 ASSERT3P(ds->ds_owner, ==, tag); 701 ASSERT(ds->ds_dbuf != NULL); 702 703 mutex_enter(&ds->ds_lock); 704 ds->ds_owner = NULL; 705 mutex_exit(&ds->ds_lock); 706 dsl_dataset_long_rele(ds, tag); 707 dsl_dataset_rele(ds, tag); 708 } 709 710 boolean_t 711 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 712 { 713 boolean_t gotit = FALSE; 714 715 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 716 mutex_enter(&ds->ds_lock); 717 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 718 ds->ds_owner = tag; 719 dsl_dataset_long_hold(ds, tag); 720 gotit = TRUE; 721 } 722 mutex_exit(&ds->ds_lock); 723 return (gotit); 724 } 725 726 boolean_t 727 dsl_dataset_has_owner(dsl_dataset_t *ds) 728 { 729 boolean_t rv; 730 mutex_enter(&ds->ds_lock); 731 rv = (ds->ds_owner != NULL); 732 mutex_exit(&ds->ds_lock); 733 return (rv); 734 } 735 736 static void 737 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx) 738 { 739 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 740 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 741 uint64_t zero = 0; 742 743 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 744 745 spa_feature_incr(spa, f, tx); 746 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 747 748 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, 749 sizeof (zero), 1, &zero, tx)); 750 } 751 752 void 753 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx) 754 { 755 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 756 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 757 758 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 759 760 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx)); 761 spa_feature_decr(spa, f, tx); 762 } 763 764 uint64_t 765 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 766 uint64_t flags, dmu_tx_t *tx) 767 { 768 dsl_pool_t *dp = dd->dd_pool; 769 dmu_buf_t *dbuf; 770 dsl_dataset_phys_t *dsphys; 771 uint64_t dsobj; 772 objset_t *mos = dp->dp_meta_objset; 773 774 if (origin == NULL) 775 origin = dp->dp_origin_snap; 776 777 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 778 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); 779 ASSERT(dmu_tx_is_syncing(tx)); 780 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 781 782 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 783 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 784 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 785 dmu_buf_will_dirty(dbuf, tx); 786 dsphys = dbuf->db_data; 787 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 788 dsphys->ds_dir_obj = dd->dd_object; 789 dsphys->ds_flags = flags; 790 dsphys->ds_fsid_guid = unique_create(); 791 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 792 sizeof (dsphys->ds_guid)); 793 dsphys->ds_snapnames_zapobj = 794 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 795 DMU_OT_NONE, 0, tx); 796 dsphys->ds_creation_time = gethrestime_sec(); 797 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 798 799 if (origin == NULL) { 800 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 801 } else { 802 dsl_dataset_t *ohds; /* head of the origin snapshot */ 803 804 dsphys->ds_prev_snap_obj = origin->ds_object; 805 dsphys->ds_prev_snap_txg = 806 dsl_dataset_phys(origin)->ds_creation_txg; 807 dsphys->ds_referenced_bytes = 808 dsl_dataset_phys(origin)->ds_referenced_bytes; 809 dsphys->ds_compressed_bytes = 810 dsl_dataset_phys(origin)->ds_compressed_bytes; 811 dsphys->ds_uncompressed_bytes = 812 dsl_dataset_phys(origin)->ds_uncompressed_bytes; 813 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG); 814 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; 815 rrw_exit(&origin->ds_bp_rwlock, FTAG); 816 817 /* 818 * Inherit flags that describe the dataset's contents 819 * (INCONSISTENT) or properties (Case Insensitive). 820 */ 821 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & 822 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 823 824 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 825 if (origin->ds_feature_inuse[f]) 826 dsl_dataset_activate_feature(dsobj, f, tx); 827 } 828 829 dmu_buf_will_dirty(origin->ds_dbuf, tx); 830 dsl_dataset_phys(origin)->ds_num_children++; 831 832 VERIFY0(dsl_dataset_hold_obj(dp, 833 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, 834 FTAG, &ohds)); 835 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 836 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 837 dsl_dataset_rele(ohds, FTAG); 838 839 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 840 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { 841 dsl_dataset_phys(origin)->ds_next_clones_obj = 842 zap_create(mos, 843 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 844 } 845 VERIFY0(zap_add_int(mos, 846 dsl_dataset_phys(origin)->ds_next_clones_obj, 847 dsobj, tx)); 848 } 849 850 dmu_buf_will_dirty(dd->dd_dbuf, tx); 851 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; 852 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 853 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 854 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 855 dsl_dir_phys(origin->ds_dir)->dd_clones = 856 zap_create(mos, 857 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 858 } 859 VERIFY0(zap_add_int(mos, 860 dsl_dir_phys(origin->ds_dir)->dd_clones, 861 dsobj, tx)); 862 } 863 } 864 865 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 866 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 867 868 dmu_buf_rele(dbuf, FTAG); 869 870 dmu_buf_will_dirty(dd->dd_dbuf, tx); 871 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; 872 873 return (dsobj); 874 } 875 876 static void 877 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 878 { 879 objset_t *os; 880 881 VERIFY0(dmu_objset_from_ds(ds, &os)); 882 if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) { 883 dsl_pool_t *dp = ds->ds_dir->dd_pool; 884 zio_t *zio; 885 886 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 887 888 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 889 dsl_dataset_sync(ds, zio, tx); 890 VERIFY0(zio_wait(zio)); 891 892 /* dsl_dataset_sync_done will drop this reference. */ 893 dmu_buf_add_ref(ds->ds_dbuf, ds); 894 dsl_dataset_sync_done(ds, tx); 895 } 896 } 897 898 uint64_t 899 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 900 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 901 { 902 dsl_pool_t *dp = pdd->dd_pool; 903 uint64_t dsobj, ddobj; 904 dsl_dir_t *dd; 905 906 ASSERT(dmu_tx_is_syncing(tx)); 907 ASSERT(lastname[0] != '@'); 908 909 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 910 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 911 912 dsobj = dsl_dataset_create_sync_dd(dd, origin, 913 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 914 915 dsl_deleg_set_create_perms(dd, tx, cr); 916 917 /* 918 * Since we're creating a new node we know it's a leaf, so we can 919 * initialize the counts if the limit feature is active. 920 */ 921 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 922 uint64_t cnt = 0; 923 objset_t *os = dd->dd_pool->dp_meta_objset; 924 925 dsl_dir_zapify(dd, tx); 926 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 927 sizeof (cnt), 1, &cnt, tx)); 928 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 929 sizeof (cnt), 1, &cnt, tx)); 930 } 931 932 dsl_dir_rele(dd, FTAG); 933 934 /* 935 * If we are creating a clone, make sure we zero out any stale 936 * data from the origin snapshots zil header. 937 */ 938 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 939 dsl_dataset_t *ds; 940 941 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 942 dsl_dataset_zero_zil(ds, tx); 943 dsl_dataset_rele(ds, FTAG); 944 } 945 946 return (dsobj); 947 } 948 949 /* 950 * The unique space in the head dataset can be calculated by subtracting 951 * the space used in the most recent snapshot, that is still being used 952 * in this file system, from the space currently in use. To figure out 953 * the space in the most recent snapshot still in use, we need to take 954 * the total space used in the snapshot and subtract out the space that 955 * has been freed up since the snapshot was taken. 956 */ 957 void 958 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 959 { 960 uint64_t mrs_used; 961 uint64_t dlused, dlcomp, dluncomp; 962 963 ASSERT(!ds->ds_is_snapshot); 964 965 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) 966 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; 967 else 968 mrs_used = 0; 969 970 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 971 972 ASSERT3U(dlused, <=, mrs_used); 973 dsl_dataset_phys(ds)->ds_unique_bytes = 974 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); 975 976 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 977 SPA_VERSION_UNIQUE_ACCURATE) 978 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 979 } 980 981 void 982 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 983 dmu_tx_t *tx) 984 { 985 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 986 uint64_t count; 987 int err; 988 989 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); 990 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 991 obj, tx); 992 /* 993 * The err should not be ENOENT, but a bug in a previous version 994 * of the code could cause upgrade_clones_cb() to not set 995 * ds_next_snap_obj when it should, leading to a missing entry. 996 * If we knew that the pool was created after 997 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 998 * ENOENT. However, at least we can check that we don't have 999 * too many entries in the next_clones_obj even after failing to 1000 * remove this one. 1001 */ 1002 if (err != ENOENT) 1003 VERIFY0(err); 1004 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1005 &count)); 1006 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); 1007 } 1008 1009 1010 blkptr_t * 1011 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1012 { 1013 return (&dsl_dataset_phys(ds)->ds_bp); 1014 } 1015 1016 spa_t * 1017 dsl_dataset_get_spa(dsl_dataset_t *ds) 1018 { 1019 return (ds->ds_dir->dd_pool->dp_spa); 1020 } 1021 1022 void 1023 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1024 { 1025 dsl_pool_t *dp; 1026 1027 if (ds == NULL) /* this is the meta-objset */ 1028 return; 1029 1030 ASSERT(ds->ds_objset != NULL); 1031 1032 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) 1033 panic("dirtying snapshot!"); 1034 1035 /* Must not dirty a dataset in the same txg where it got snapshotted. */ 1036 ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 1037 1038 dp = ds->ds_dir->dd_pool; 1039 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1040 /* up the hold count until we can be written out */ 1041 dmu_buf_add_ref(ds->ds_dbuf, ds); 1042 } 1043 } 1044 1045 boolean_t 1046 dsl_dataset_is_dirty(dsl_dataset_t *ds) 1047 { 1048 for (int t = 0; t < TXG_SIZE; t++) { 1049 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1050 ds, t)) 1051 return (B_TRUE); 1052 } 1053 return (B_FALSE); 1054 } 1055 1056 static int 1057 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1058 { 1059 uint64_t asize; 1060 1061 if (!dmu_tx_is_syncing(tx)) 1062 return (0); 1063 1064 /* 1065 * If there's an fs-only reservation, any blocks that might become 1066 * owned by the snapshot dataset must be accommodated by space 1067 * outside of the reservation. 1068 */ 1069 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1070 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); 1071 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1072 return (SET_ERROR(ENOSPC)); 1073 1074 /* 1075 * Propagate any reserved space for this snapshot to other 1076 * snapshot checks in this sync group. 1077 */ 1078 if (asize > 0) 1079 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1080 1081 return (0); 1082 } 1083 1084 typedef struct dsl_dataset_snapshot_arg { 1085 nvlist_t *ddsa_snaps; 1086 nvlist_t *ddsa_props; 1087 nvlist_t *ddsa_errors; 1088 cred_t *ddsa_cr; 1089 } dsl_dataset_snapshot_arg_t; 1090 1091 int 1092 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1093 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 1094 { 1095 int error; 1096 uint64_t value; 1097 1098 ds->ds_trysnap_txg = tx->tx_txg; 1099 1100 if (!dmu_tx_is_syncing(tx)) 1101 return (0); 1102 1103 /* 1104 * We don't allow multiple snapshots of the same txg. If there 1105 * is already one, try again. 1106 */ 1107 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) 1108 return (SET_ERROR(EAGAIN)); 1109 1110 /* 1111 * Check for conflicting snapshot name. 1112 */ 1113 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1114 if (error == 0) 1115 return (SET_ERROR(EEXIST)); 1116 if (error != ENOENT) 1117 return (error); 1118 1119 /* 1120 * We don't allow taking snapshots of inconsistent datasets, such as 1121 * those into which we are currently receiving. However, if we are 1122 * creating this snapshot as part of a receive, this check will be 1123 * executed atomically with respect to the completion of the receive 1124 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1125 * case we ignore this, knowing it will be fixed up for us shortly in 1126 * dmu_recv_end_sync(). 1127 */ 1128 if (!recv && DS_IS_INCONSISTENT(ds)) 1129 return (SET_ERROR(EBUSY)); 1130 1131 /* 1132 * Skip the check for temporary snapshots or if we have already checked 1133 * the counts in dsl_dataset_snapshot_check. This means we really only 1134 * check the count here when we're receiving a stream. 1135 */ 1136 if (cnt != 0 && cr != NULL) { 1137 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1138 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 1139 if (error != 0) 1140 return (error); 1141 } 1142 1143 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1144 if (error != 0) 1145 return (error); 1146 1147 return (0); 1148 } 1149 1150 static int 1151 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1152 { 1153 dsl_dataset_snapshot_arg_t *ddsa = arg; 1154 dsl_pool_t *dp = dmu_tx_pool(tx); 1155 nvpair_t *pair; 1156 int rv = 0; 1157 1158 /* 1159 * Pre-compute how many total new snapshots will be created for each 1160 * level in the tree and below. This is needed for validating the 1161 * snapshot limit when either taking a recursive snapshot or when 1162 * taking multiple snapshots. 1163 * 1164 * The problem is that the counts are not actually adjusted when 1165 * we are checking, only when we finally sync. For a single snapshot, 1166 * this is easy, the count will increase by 1 at each node up the tree, 1167 * but its more complicated for the recursive/multiple snapshot case. 1168 * 1169 * The dsl_fs_ss_limit_check function does recursively check the count 1170 * at each level up the tree but since it is validating each snapshot 1171 * independently we need to be sure that we are validating the complete 1172 * count for the entire set of snapshots. We do this by rolling up the 1173 * counts for each component of the name into an nvlist and then 1174 * checking each of those cases with the aggregated count. 1175 * 1176 * This approach properly handles not only the recursive snapshot 1177 * case (where we get all of those on the ddsa_snaps list) but also 1178 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1179 * validate the limit on 'a' using a count of 2). 1180 * 1181 * We validate the snapshot names in the third loop and only report 1182 * name errors once. 1183 */ 1184 if (dmu_tx_is_syncing(tx)) { 1185 nvlist_t *cnt_track = NULL; 1186 cnt_track = fnvlist_alloc(); 1187 1188 /* Rollup aggregated counts into the cnt_track list */ 1189 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1190 pair != NULL; 1191 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1192 char *pdelim; 1193 uint64_t val; 1194 char nm[MAXPATHLEN]; 1195 1196 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1197 pdelim = strchr(nm, '@'); 1198 if (pdelim == NULL) 1199 continue; 1200 *pdelim = '\0'; 1201 1202 do { 1203 if (nvlist_lookup_uint64(cnt_track, nm, 1204 &val) == 0) { 1205 /* update existing entry */ 1206 fnvlist_add_uint64(cnt_track, nm, 1207 val + 1); 1208 } else { 1209 /* add to list */ 1210 fnvlist_add_uint64(cnt_track, nm, 1); 1211 } 1212 1213 pdelim = strrchr(nm, '/'); 1214 if (pdelim != NULL) 1215 *pdelim = '\0'; 1216 } while (pdelim != NULL); 1217 } 1218 1219 /* Check aggregated counts at each level */ 1220 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1221 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1222 int error = 0; 1223 char *name; 1224 uint64_t cnt = 0; 1225 dsl_dataset_t *ds; 1226 1227 name = nvpair_name(pair); 1228 cnt = fnvpair_value_uint64(pair); 1229 ASSERT(cnt > 0); 1230 1231 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1232 if (error == 0) { 1233 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1234 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1235 ddsa->ddsa_cr); 1236 dsl_dataset_rele(ds, FTAG); 1237 } 1238 1239 if (error != 0) { 1240 if (ddsa->ddsa_errors != NULL) 1241 fnvlist_add_int32(ddsa->ddsa_errors, 1242 name, error); 1243 rv = error; 1244 /* only report one error for this check */ 1245 break; 1246 } 1247 } 1248 nvlist_free(cnt_track); 1249 } 1250 1251 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1252 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1253 int error = 0; 1254 dsl_dataset_t *ds; 1255 char *name, *atp; 1256 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1257 1258 name = nvpair_name(pair); 1259 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN) 1260 error = SET_ERROR(ENAMETOOLONG); 1261 if (error == 0) { 1262 atp = strchr(name, '@'); 1263 if (atp == NULL) 1264 error = SET_ERROR(EINVAL); 1265 if (error == 0) 1266 (void) strlcpy(dsname, name, atp - name + 1); 1267 } 1268 if (error == 0) 1269 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1270 if (error == 0) { 1271 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1272 error = dsl_dataset_snapshot_check_impl(ds, 1273 atp + 1, tx, B_FALSE, 0, NULL); 1274 dsl_dataset_rele(ds, FTAG); 1275 } 1276 1277 if (error != 0) { 1278 if (ddsa->ddsa_errors != NULL) { 1279 fnvlist_add_int32(ddsa->ddsa_errors, 1280 name, error); 1281 } 1282 rv = error; 1283 } 1284 } 1285 1286 return (rv); 1287 } 1288 1289 void 1290 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1291 dmu_tx_t *tx) 1292 { 1293 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1294 dmu_buf_t *dbuf; 1295 dsl_dataset_phys_t *dsphys; 1296 uint64_t dsobj, crtxg; 1297 objset_t *mos = dp->dp_meta_objset; 1298 objset_t *os; 1299 1300 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1301 1302 /* 1303 * If we are on an old pool, the zil must not be active, in which 1304 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1305 */ 1306 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1307 dmu_objset_from_ds(ds, &os) != 0 || 1308 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1309 sizeof (zero_zil)) == 0); 1310 1311 /* Should not snapshot a dirty dataset. */ 1312 ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1313 ds, tx->tx_txg)); 1314 1315 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1316 1317 /* 1318 * The origin's ds_creation_txg has to be < TXG_INITIAL 1319 */ 1320 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1321 crtxg = 1; 1322 else 1323 crtxg = tx->tx_txg; 1324 1325 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1326 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1327 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1328 dmu_buf_will_dirty(dbuf, tx); 1329 dsphys = dbuf->db_data; 1330 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1331 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1332 dsphys->ds_fsid_guid = unique_create(); 1333 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1334 sizeof (dsphys->ds_guid)); 1335 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1336 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 1337 dsphys->ds_next_snap_obj = ds->ds_object; 1338 dsphys->ds_num_children = 1; 1339 dsphys->ds_creation_time = gethrestime_sec(); 1340 dsphys->ds_creation_txg = crtxg; 1341 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; 1342 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; 1343 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; 1344 dsphys->ds_uncompressed_bytes = 1345 dsl_dataset_phys(ds)->ds_uncompressed_bytes; 1346 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; 1347 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1348 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; 1349 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1350 dmu_buf_rele(dbuf, FTAG); 1351 1352 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1353 if (ds->ds_feature_inuse[f]) 1354 dsl_dataset_activate_feature(dsobj, f, tx); 1355 } 1356 1357 ASSERT3U(ds->ds_prev != 0, ==, 1358 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); 1359 if (ds->ds_prev) { 1360 uint64_t next_clones_obj = 1361 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; 1362 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1363 ds->ds_object || 1364 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); 1365 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1366 ds->ds_object) { 1367 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1368 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, 1369 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); 1370 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; 1371 } else if (next_clones_obj != 0) { 1372 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1373 dsphys->ds_next_snap_obj, tx); 1374 VERIFY0(zap_add_int(mos, 1375 next_clones_obj, dsobj, tx)); 1376 } 1377 } 1378 1379 /* 1380 * If we have a reference-reservation on this dataset, we will 1381 * need to increase the amount of refreservation being charged 1382 * since our unique space is going to zero. 1383 */ 1384 if (ds->ds_reserved) { 1385 int64_t delta; 1386 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1387 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, 1388 ds->ds_reserved); 1389 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1390 delta, 0, 0, tx); 1391 } 1392 1393 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1394 dsl_dataset_phys(ds)->ds_deadlist_obj = 1395 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, 1396 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 1397 dsl_deadlist_close(&ds->ds_deadlist); 1398 dsl_deadlist_open(&ds->ds_deadlist, mos, 1399 dsl_dataset_phys(ds)->ds_deadlist_obj); 1400 dsl_deadlist_add_key(&ds->ds_deadlist, 1401 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); 1402 1403 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); 1404 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; 1405 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; 1406 dsl_dataset_phys(ds)->ds_unique_bytes = 0; 1407 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1408 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1409 1410 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1411 snapname, 8, 1, &dsobj, tx)); 1412 1413 if (ds->ds_prev) 1414 dsl_dataset_rele(ds->ds_prev, ds); 1415 VERIFY0(dsl_dataset_hold_obj(dp, 1416 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); 1417 1418 dsl_scan_ds_snapshotted(ds, tx); 1419 1420 dsl_dir_snap_cmtime_update(ds->ds_dir); 1421 1422 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1423 } 1424 1425 static void 1426 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1427 { 1428 dsl_dataset_snapshot_arg_t *ddsa = arg; 1429 dsl_pool_t *dp = dmu_tx_pool(tx); 1430 nvpair_t *pair; 1431 1432 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1433 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1434 dsl_dataset_t *ds; 1435 char *name, *atp; 1436 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1437 1438 name = nvpair_name(pair); 1439 atp = strchr(name, '@'); 1440 (void) strlcpy(dsname, name, atp - name + 1); 1441 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1442 1443 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1444 if (ddsa->ddsa_props != NULL) { 1445 dsl_props_set_sync_impl(ds->ds_prev, 1446 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1447 } 1448 dsl_dataset_rele(ds, FTAG); 1449 } 1450 } 1451 1452 /* 1453 * The snapshots must all be in the same pool. 1454 * All-or-nothing: if there are any failures, nothing will be modified. 1455 */ 1456 int 1457 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1458 { 1459 dsl_dataset_snapshot_arg_t ddsa; 1460 nvpair_t *pair; 1461 boolean_t needsuspend; 1462 int error; 1463 spa_t *spa; 1464 char *firstname; 1465 nvlist_t *suspended = NULL; 1466 1467 pair = nvlist_next_nvpair(snaps, NULL); 1468 if (pair == NULL) 1469 return (0); 1470 firstname = nvpair_name(pair); 1471 1472 error = spa_open(firstname, &spa, FTAG); 1473 if (error != 0) 1474 return (error); 1475 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1476 spa_close(spa, FTAG); 1477 1478 if (needsuspend) { 1479 suspended = fnvlist_alloc(); 1480 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1481 pair = nvlist_next_nvpair(snaps, pair)) { 1482 char fsname[ZFS_MAX_DATASET_NAME_LEN]; 1483 char *snapname = nvpair_name(pair); 1484 char *atp; 1485 void *cookie; 1486 1487 atp = strchr(snapname, '@'); 1488 if (atp == NULL) { 1489 error = SET_ERROR(EINVAL); 1490 break; 1491 } 1492 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1493 1494 error = zil_suspend(fsname, &cookie); 1495 if (error != 0) 1496 break; 1497 fnvlist_add_uint64(suspended, fsname, 1498 (uintptr_t)cookie); 1499 } 1500 } 1501 1502 ddsa.ddsa_snaps = snaps; 1503 ddsa.ddsa_props = props; 1504 ddsa.ddsa_errors = errors; 1505 ddsa.ddsa_cr = CRED(); 1506 1507 if (error == 0) { 1508 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1509 dsl_dataset_snapshot_sync, &ddsa, 1510 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1511 } 1512 1513 if (suspended != NULL) { 1514 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1515 pair = nvlist_next_nvpair(suspended, pair)) { 1516 zil_resume((void *)(uintptr_t) 1517 fnvpair_value_uint64(pair)); 1518 } 1519 fnvlist_free(suspended); 1520 } 1521 1522 return (error); 1523 } 1524 1525 typedef struct dsl_dataset_snapshot_tmp_arg { 1526 const char *ddsta_fsname; 1527 const char *ddsta_snapname; 1528 minor_t ddsta_cleanup_minor; 1529 const char *ddsta_htag; 1530 } dsl_dataset_snapshot_tmp_arg_t; 1531 1532 static int 1533 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1534 { 1535 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1536 dsl_pool_t *dp = dmu_tx_pool(tx); 1537 dsl_dataset_t *ds; 1538 int error; 1539 1540 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1541 if (error != 0) 1542 return (error); 1543 1544 /* NULL cred means no limit check for tmp snapshot */ 1545 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1546 tx, B_FALSE, 0, NULL); 1547 if (error != 0) { 1548 dsl_dataset_rele(ds, FTAG); 1549 return (error); 1550 } 1551 1552 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1553 dsl_dataset_rele(ds, FTAG); 1554 return (SET_ERROR(ENOTSUP)); 1555 } 1556 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1557 B_TRUE, tx); 1558 if (error != 0) { 1559 dsl_dataset_rele(ds, FTAG); 1560 return (error); 1561 } 1562 1563 dsl_dataset_rele(ds, FTAG); 1564 return (0); 1565 } 1566 1567 static void 1568 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1569 { 1570 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1571 dsl_pool_t *dp = dmu_tx_pool(tx); 1572 dsl_dataset_t *ds; 1573 1574 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1575 1576 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1577 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1578 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1579 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1580 1581 dsl_dataset_rele(ds, FTAG); 1582 } 1583 1584 int 1585 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1586 minor_t cleanup_minor, const char *htag) 1587 { 1588 dsl_dataset_snapshot_tmp_arg_t ddsta; 1589 int error; 1590 spa_t *spa; 1591 boolean_t needsuspend; 1592 void *cookie; 1593 1594 ddsta.ddsta_fsname = fsname; 1595 ddsta.ddsta_snapname = snapname; 1596 ddsta.ddsta_cleanup_minor = cleanup_minor; 1597 ddsta.ddsta_htag = htag; 1598 1599 error = spa_open(fsname, &spa, FTAG); 1600 if (error != 0) 1601 return (error); 1602 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1603 spa_close(spa, FTAG); 1604 1605 if (needsuspend) { 1606 error = zil_suspend(fsname, &cookie); 1607 if (error != 0) 1608 return (error); 1609 } 1610 1611 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1612 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 1613 1614 if (needsuspend) 1615 zil_resume(cookie); 1616 return (error); 1617 } 1618 1619 void 1620 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1621 { 1622 ASSERT(dmu_tx_is_syncing(tx)); 1623 ASSERT(ds->ds_objset != NULL); 1624 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); 1625 1626 /* 1627 * in case we had to change ds_fsid_guid when we opened it, 1628 * sync it out now. 1629 */ 1630 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1631 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; 1632 1633 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) { 1634 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1635 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1, 1636 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx)); 1637 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1638 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1, 1639 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx)); 1640 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1641 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1, 1642 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx)); 1643 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0; 1644 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0; 1645 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0; 1646 } 1647 1648 dmu_objset_sync(ds->ds_objset, zio, tx); 1649 1650 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1651 if (ds->ds_feature_activation_needed[f]) { 1652 if (ds->ds_feature_inuse[f]) 1653 continue; 1654 dsl_dataset_activate_feature(ds->ds_object, f, tx); 1655 ds->ds_feature_inuse[f] = B_TRUE; 1656 } 1657 } 1658 } 1659 1660 static int 1661 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 1662 { 1663 dsl_deadlist_t *dl = arg; 1664 dsl_deadlist_insert(dl, bp, tx); 1665 return (0); 1666 } 1667 1668 void 1669 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx) 1670 { 1671 objset_t *os = ds->ds_objset; 1672 1673 bplist_iterate(&ds->ds_pending_deadlist, 1674 deadlist_enqueue_cb, &ds->ds_deadlist, tx); 1675 1676 if (os->os_synced_dnodes != NULL) { 1677 multilist_destroy(os->os_synced_dnodes); 1678 os->os_synced_dnodes = NULL; 1679 } 1680 1681 ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx))); 1682 1683 dmu_buf_rele(ds->ds_dbuf, ds); 1684 } 1685 1686 int 1687 get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val) 1688 { 1689 uint64_t count = 0; 1690 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1691 zap_cursor_t zc; 1692 zap_attribute_t za; 1693 1694 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1695 1696 /* 1697 * There may be missing entries in ds_next_clones_obj 1698 * due to a bug in a previous version of the code. 1699 * Only trust it if it has the right number of entries. 1700 */ 1701 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { 1702 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1703 &count)); 1704 } 1705 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) { 1706 return (ENOENT); 1707 } 1708 for (zap_cursor_init(&zc, mos, 1709 dsl_dataset_phys(ds)->ds_next_clones_obj); 1710 zap_cursor_retrieve(&zc, &za) == 0; 1711 zap_cursor_advance(&zc)) { 1712 dsl_dataset_t *clone; 1713 char buf[ZFS_MAX_DATASET_NAME_LEN]; 1714 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1715 za.za_first_integer, FTAG, &clone)); 1716 dsl_dir_name(clone->ds_dir, buf); 1717 fnvlist_add_boolean(val, buf); 1718 dsl_dataset_rele(clone, FTAG); 1719 } 1720 zap_cursor_fini(&zc); 1721 return (0); 1722 } 1723 1724 void 1725 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1726 { 1727 nvlist_t *propval = fnvlist_alloc(); 1728 nvlist_t *val; 1729 1730 /* 1731 * We use nvlist_alloc() instead of fnvlist_alloc() because the 1732 * latter would allocate the list with NV_UNIQUE_NAME flag. 1733 * As a result, every time a clone name is appended to the list 1734 * it would be (linearly) searched for for a duplicate name. 1735 * We already know that all clone names must be unique and we 1736 * want avoid the quadratic complexity of double-checking that 1737 * because we can have a large number of clones. 1738 */ 1739 VERIFY0(nvlist_alloc(&val, 0, KM_SLEEP)); 1740 1741 if (get_clones_stat_impl(ds, val) == 0) { 1742 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1743 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), 1744 propval); 1745 } 1746 1747 nvlist_free(val); 1748 nvlist_free(propval); 1749 } 1750 1751 /* 1752 * Returns a string that represents the receive resume stats token. It should 1753 * be freed with strfree(). 1754 */ 1755 char * 1756 get_receive_resume_stats_impl(dsl_dataset_t *ds) 1757 { 1758 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1759 1760 if (dsl_dataset_has_resume_receive_state(ds)) { 1761 char *str; 1762 void *packed; 1763 uint8_t *compressed; 1764 uint64_t val; 1765 nvlist_t *token_nv = fnvlist_alloc(); 1766 size_t packed_size, compressed_size; 1767 1768 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1769 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) { 1770 fnvlist_add_uint64(token_nv, "fromguid", val); 1771 } 1772 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1773 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) { 1774 fnvlist_add_uint64(token_nv, "object", val); 1775 } 1776 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1777 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) { 1778 fnvlist_add_uint64(token_nv, "offset", val); 1779 } 1780 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1781 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) { 1782 fnvlist_add_uint64(token_nv, "bytes", val); 1783 } 1784 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1785 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) { 1786 fnvlist_add_uint64(token_nv, "toguid", val); 1787 } 1788 char buf[256]; 1789 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1790 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) { 1791 fnvlist_add_string(token_nv, "toname", buf); 1792 } 1793 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1794 DS_FIELD_RESUME_LARGEBLOCK) == 0) { 1795 fnvlist_add_boolean(token_nv, "largeblockok"); 1796 } 1797 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1798 DS_FIELD_RESUME_EMBEDOK) == 0) { 1799 fnvlist_add_boolean(token_nv, "embedok"); 1800 } 1801 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1802 DS_FIELD_RESUME_COMPRESSOK) == 0) { 1803 fnvlist_add_boolean(token_nv, "compressok"); 1804 } 1805 packed = fnvlist_pack(token_nv, &packed_size); 1806 fnvlist_free(token_nv); 1807 compressed = kmem_alloc(packed_size, KM_SLEEP); 1808 1809 compressed_size = gzip_compress(packed, compressed, 1810 packed_size, packed_size, 6); 1811 1812 zio_cksum_t cksum; 1813 fletcher_4_native(compressed, compressed_size, NULL, &cksum); 1814 1815 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP); 1816 for (int i = 0; i < compressed_size; i++) { 1817 (void) sprintf(str + i * 2, "%02x", compressed[i]); 1818 } 1819 str[compressed_size * 2] = '\0'; 1820 char *propval = kmem_asprintf("%u-%llx-%llx-%s", 1821 ZFS_SEND_RESUME_TOKEN_VERSION, 1822 (longlong_t)cksum.zc_word[0], 1823 (longlong_t)packed_size, str); 1824 kmem_free(packed, packed_size); 1825 kmem_free(str, compressed_size * 2 + 1); 1826 kmem_free(compressed, packed_size); 1827 return (propval); 1828 } 1829 return (strdup("")); 1830 } 1831 1832 /* 1833 * Returns a string that represents the receive resume stats token of the 1834 * dataset's child. It should be freed with strfree(). 1835 */ 1836 char * 1837 get_child_receive_stats(dsl_dataset_t *ds) 1838 { 1839 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 1840 dsl_dataset_t *recv_ds; 1841 dsl_dataset_name(ds, recvname); 1842 if (strlcat(recvname, "/", sizeof (recvname)) < 1843 sizeof (recvname) && 1844 strlcat(recvname, recv_clone_name, sizeof (recvname)) < 1845 sizeof (recvname) && 1846 dsl_dataset_hold(ds->ds_dir->dd_pool, recvname, FTAG, 1847 &recv_ds) == 0) { 1848 char *propval = get_receive_resume_stats_impl(recv_ds); 1849 dsl_dataset_rele(recv_ds, FTAG); 1850 return (propval); 1851 } 1852 return (strdup("")); 1853 } 1854 1855 static void 1856 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv) 1857 { 1858 char *propval = get_receive_resume_stats_impl(ds); 1859 if (strcmp(propval, "") != 0) { 1860 dsl_prop_nvlist_add_string(nv, 1861 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval); 1862 } else { 1863 char *childval = get_child_receive_stats(ds); 1864 if (strcmp(childval, "") != 0) { 1865 dsl_prop_nvlist_add_string(nv, 1866 ZFS_PROP_RECEIVE_RESUME_TOKEN, childval); 1867 } 1868 strfree(childval); 1869 } 1870 strfree(propval); 1871 } 1872 1873 uint64_t 1874 dsl_get_refratio(dsl_dataset_t *ds) 1875 { 1876 uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : 1877 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / 1878 dsl_dataset_phys(ds)->ds_compressed_bytes); 1879 return (ratio); 1880 } 1881 1882 uint64_t 1883 dsl_get_logicalreferenced(dsl_dataset_t *ds) 1884 { 1885 return (dsl_dataset_phys(ds)->ds_uncompressed_bytes); 1886 } 1887 1888 uint64_t 1889 dsl_get_compressratio(dsl_dataset_t *ds) 1890 { 1891 if (ds->ds_is_snapshot) { 1892 return (dsl_get_refratio(ds)); 1893 } else { 1894 dsl_dir_t *dd = ds->ds_dir; 1895 mutex_enter(&dd->dd_lock); 1896 uint64_t val = dsl_dir_get_compressratio(dd); 1897 mutex_exit(&dd->dd_lock); 1898 return (val); 1899 } 1900 } 1901 1902 uint64_t 1903 dsl_get_used(dsl_dataset_t *ds) 1904 { 1905 if (ds->ds_is_snapshot) { 1906 return (dsl_dataset_phys(ds)->ds_unique_bytes); 1907 } else { 1908 dsl_dir_t *dd = ds->ds_dir; 1909 mutex_enter(&dd->dd_lock); 1910 uint64_t val = dsl_dir_get_used(dd); 1911 mutex_exit(&dd->dd_lock); 1912 return (val); 1913 } 1914 } 1915 1916 uint64_t 1917 dsl_get_creation(dsl_dataset_t *ds) 1918 { 1919 return (dsl_dataset_phys(ds)->ds_creation_time); 1920 } 1921 1922 uint64_t 1923 dsl_get_creationtxg(dsl_dataset_t *ds) 1924 { 1925 return (dsl_dataset_phys(ds)->ds_creation_txg); 1926 } 1927 1928 uint64_t 1929 dsl_get_refquota(dsl_dataset_t *ds) 1930 { 1931 return (ds->ds_quota); 1932 } 1933 1934 uint64_t 1935 dsl_get_refreservation(dsl_dataset_t *ds) 1936 { 1937 return (ds->ds_reserved); 1938 } 1939 1940 uint64_t 1941 dsl_get_guid(dsl_dataset_t *ds) 1942 { 1943 return (dsl_dataset_phys(ds)->ds_guid); 1944 } 1945 1946 uint64_t 1947 dsl_get_unique(dsl_dataset_t *ds) 1948 { 1949 return (dsl_dataset_phys(ds)->ds_unique_bytes); 1950 } 1951 1952 uint64_t 1953 dsl_get_objsetid(dsl_dataset_t *ds) 1954 { 1955 return (ds->ds_object); 1956 } 1957 1958 uint64_t 1959 dsl_get_userrefs(dsl_dataset_t *ds) 1960 { 1961 return (ds->ds_userrefs); 1962 } 1963 1964 uint64_t 1965 dsl_get_defer_destroy(dsl_dataset_t *ds) 1966 { 1967 return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 1968 } 1969 1970 uint64_t 1971 dsl_get_referenced(dsl_dataset_t *ds) 1972 { 1973 return (dsl_dataset_phys(ds)->ds_referenced_bytes); 1974 } 1975 1976 uint64_t 1977 dsl_get_numclones(dsl_dataset_t *ds) 1978 { 1979 ASSERT(ds->ds_is_snapshot); 1980 return (dsl_dataset_phys(ds)->ds_num_children - 1); 1981 } 1982 1983 uint64_t 1984 dsl_get_inconsistent(dsl_dataset_t *ds) 1985 { 1986 return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ? 1987 1 : 0); 1988 } 1989 1990 uint64_t 1991 dsl_get_available(dsl_dataset_t *ds) 1992 { 1993 uint64_t refdbytes = dsl_get_referenced(ds); 1994 uint64_t availbytes = dsl_dir_space_available(ds->ds_dir, 1995 NULL, 0, TRUE); 1996 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 1997 availbytes += 1998 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 1999 } 2000 if (ds->ds_quota != 0) { 2001 /* 2002 * Adjust available bytes according to refquota 2003 */ 2004 if (refdbytes < ds->ds_quota) { 2005 availbytes = MIN(availbytes, 2006 ds->ds_quota - refdbytes); 2007 } else { 2008 availbytes = 0; 2009 } 2010 } 2011 return (availbytes); 2012 } 2013 2014 int 2015 dsl_get_written(dsl_dataset_t *ds, uint64_t *written) 2016 { 2017 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2018 dsl_dataset_t *prev; 2019 int err = dsl_dataset_hold_obj(dp, 2020 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 2021 if (err == 0) { 2022 uint64_t comp, uncomp; 2023 err = dsl_dataset_space_written(prev, ds, written, 2024 &comp, &uncomp); 2025 dsl_dataset_rele(prev, FTAG); 2026 } 2027 return (err); 2028 } 2029 2030 /* 2031 * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN. 2032 */ 2033 int 2034 dsl_get_prev_snap(dsl_dataset_t *ds, char *snap) 2035 { 2036 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2037 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 2038 dsl_dataset_name(ds->ds_prev, snap); 2039 return (0); 2040 } else { 2041 return (ENOENT); 2042 } 2043 } 2044 2045 /* 2046 * Returns the mountpoint property and source for the given dataset in the value 2047 * and source buffers. The value buffer must be at least as large as MAXPATHLEN 2048 * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN. 2049 * Returns 0 on success and an error on failure. 2050 */ 2051 int 2052 dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value, 2053 char *source) 2054 { 2055 int error; 2056 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2057 2058 /* Retrieve the mountpoint value stored in the zap opbject */ 2059 error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1, 2060 ZAP_MAXVALUELEN, value, source); 2061 if (error != 0) { 2062 return (error); 2063 } 2064 2065 /* Process the dsname and source to find the full mountpoint string */ 2066 if (value[0] == '/') { 2067 char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP); 2068 char *root = buf; 2069 const char *relpath; 2070 2071 /* 2072 * If we inherit the mountpoint, even from a dataset 2073 * with a received value, the source will be the path of 2074 * the dataset we inherit from. If source is 2075 * ZPROP_SOURCE_VAL_RECVD, the received value is not 2076 * inherited. 2077 */ 2078 if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) { 2079 relpath = ""; 2080 } else { 2081 ASSERT0(strncmp(dsname, source, strlen(source))); 2082 relpath = dsname + strlen(source); 2083 if (relpath[0] == '/') 2084 relpath++; 2085 } 2086 2087 spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN); 2088 2089 /* 2090 * Special case an alternate root of '/'. This will 2091 * avoid having multiple leading slashes in the 2092 * mountpoint path. 2093 */ 2094 if (strcmp(root, "/") == 0) 2095 root++; 2096 2097 /* 2098 * If the mountpoint is '/' then skip over this 2099 * if we are obtaining either an alternate root or 2100 * an inherited mountpoint. 2101 */ 2102 char *mnt = value; 2103 if (value[1] == '\0' && (root[0] != '\0' || 2104 relpath[0] != '\0')) 2105 mnt = value + 1; 2106 2107 if (relpath[0] == '\0') { 2108 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s", 2109 root, mnt); 2110 } else { 2111 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s", 2112 root, mnt, relpath[0] == '@' ? "" : "/", 2113 relpath); 2114 } 2115 kmem_free(buf, ZAP_MAXVALUELEN); 2116 } else { 2117 /* 'legacy' or 'none' */ 2118 (void) snprintf(value, ZAP_MAXVALUELEN, "%s", value); 2119 } 2120 return (0); 2121 } 2122 2123 void 2124 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 2125 { 2126 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2127 2128 ASSERT(dsl_pool_config_held(dp)); 2129 2130 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, 2131 dsl_get_refratio(ds)); 2132 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 2133 dsl_get_logicalreferenced(ds)); 2134 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 2135 dsl_get_compressratio(ds)); 2136 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 2137 dsl_get_used(ds)); 2138 2139 if (ds->ds_is_snapshot) { 2140 get_clones_stat(ds, nv); 2141 } else { 2142 char buf[ZFS_MAX_DATASET_NAME_LEN]; 2143 if (dsl_get_prev_snap(ds, buf) == 0) 2144 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, 2145 buf); 2146 dsl_dir_stats(ds->ds_dir, nv); 2147 } 2148 2149 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, 2150 dsl_get_available(ds)); 2151 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, 2152 dsl_get_referenced(ds)); 2153 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 2154 dsl_get_creation(ds)); 2155 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 2156 dsl_get_creationtxg(ds)); 2157 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 2158 dsl_get_refquota(ds)); 2159 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 2160 dsl_get_refreservation(ds)); 2161 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 2162 dsl_get_guid(ds)); 2163 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 2164 dsl_get_unique(ds)); 2165 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 2166 dsl_get_objsetid(ds)); 2167 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 2168 dsl_get_userrefs(ds)); 2169 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 2170 dsl_get_defer_destroy(ds)); 2171 2172 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 2173 uint64_t written; 2174 if (dsl_get_written(ds, &written) == 0) { 2175 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 2176 written); 2177 } 2178 } 2179 2180 if (!dsl_dataset_is_snapshot(ds)) { 2181 /* 2182 * A failed "newfs" (e.g. full) resumable receive leaves 2183 * the stats set on this dataset. Check here for the prop. 2184 */ 2185 get_receive_resume_stats(ds, nv); 2186 2187 /* 2188 * A failed incremental resumable receive leaves the 2189 * stats set on our child named "%recv". Check the child 2190 * for the prop. 2191 */ 2192 /* 6 extra bytes for /%recv */ 2193 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 2194 dsl_dataset_t *recv_ds; 2195 dsl_dataset_name(ds, recvname); 2196 if (strlcat(recvname, "/", sizeof (recvname)) < 2197 sizeof (recvname) && 2198 strlcat(recvname, recv_clone_name, sizeof (recvname)) < 2199 sizeof (recvname) && 2200 dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) { 2201 get_receive_resume_stats(recv_ds, nv); 2202 dsl_dataset_rele(recv_ds, FTAG); 2203 } 2204 } 2205 } 2206 2207 void 2208 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 2209 { 2210 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2211 ASSERT(dsl_pool_config_held(dp)); 2212 2213 stat->dds_creation_txg = dsl_get_creationtxg(ds); 2214 stat->dds_inconsistent = dsl_get_inconsistent(ds); 2215 stat->dds_guid = dsl_get_guid(ds); 2216 stat->dds_origin[0] = '\0'; 2217 if (ds->ds_is_snapshot) { 2218 stat->dds_is_snapshot = B_TRUE; 2219 stat->dds_num_clones = dsl_get_numclones(ds); 2220 } else { 2221 stat->dds_is_snapshot = B_FALSE; 2222 stat->dds_num_clones = 0; 2223 2224 if (dsl_dir_is_clone(ds->ds_dir)) { 2225 dsl_dir_get_origin(ds->ds_dir, stat->dds_origin); 2226 } 2227 } 2228 } 2229 2230 uint64_t 2231 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 2232 { 2233 return (ds->ds_fsid_guid); 2234 } 2235 2236 void 2237 dsl_dataset_space(dsl_dataset_t *ds, 2238 uint64_t *refdbytesp, uint64_t *availbytesp, 2239 uint64_t *usedobjsp, uint64_t *availobjsp) 2240 { 2241 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; 2242 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 2243 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) 2244 *availbytesp += 2245 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 2246 if (ds->ds_quota != 0) { 2247 /* 2248 * Adjust available bytes according to refquota 2249 */ 2250 if (*refdbytesp < ds->ds_quota) 2251 *availbytesp = MIN(*availbytesp, 2252 ds->ds_quota - *refdbytesp); 2253 else 2254 *availbytesp = 0; 2255 } 2256 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2257 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); 2258 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2259 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 2260 } 2261 2262 boolean_t 2263 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 2264 { 2265 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2266 uint64_t birth; 2267 2268 ASSERT(dsl_pool_config_held(dp)); 2269 if (snap == NULL) 2270 return (B_FALSE); 2271 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2272 birth = dsl_dataset_get_blkptr(ds)->blk_birth; 2273 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2274 if (birth > dsl_dataset_phys(snap)->ds_creation_txg) { 2275 objset_t *os, *os_snap; 2276 /* 2277 * It may be that only the ZIL differs, because it was 2278 * reset in the head. Don't count that as being 2279 * modified. 2280 */ 2281 if (dmu_objset_from_ds(ds, &os) != 0) 2282 return (B_TRUE); 2283 if (dmu_objset_from_ds(snap, &os_snap) != 0) 2284 return (B_TRUE); 2285 return (bcmp(&os->os_phys->os_meta_dnode, 2286 &os_snap->os_phys->os_meta_dnode, 2287 sizeof (os->os_phys->os_meta_dnode)) != 0); 2288 } 2289 return (B_FALSE); 2290 } 2291 2292 typedef struct dsl_dataset_rename_snapshot_arg { 2293 const char *ddrsa_fsname; 2294 const char *ddrsa_oldsnapname; 2295 const char *ddrsa_newsnapname; 2296 boolean_t ddrsa_recursive; 2297 dmu_tx_t *ddrsa_tx; 2298 } dsl_dataset_rename_snapshot_arg_t; 2299 2300 /* ARGSUSED */ 2301 static int 2302 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 2303 dsl_dataset_t *hds, void *arg) 2304 { 2305 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2306 int error; 2307 uint64_t val; 2308 2309 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2310 if (error != 0) { 2311 /* ignore nonexistent snapshots */ 2312 return (error == ENOENT ? 0 : error); 2313 } 2314 2315 /* new name should not exist */ 2316 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 2317 if (error == 0) 2318 error = SET_ERROR(EEXIST); 2319 else if (error == ENOENT) 2320 error = 0; 2321 2322 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2323 if (dsl_dir_namelen(hds->ds_dir) + 1 + 2324 strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN) 2325 error = SET_ERROR(ENAMETOOLONG); 2326 2327 return (error); 2328 } 2329 2330 static int 2331 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 2332 { 2333 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2334 dsl_pool_t *dp = dmu_tx_pool(tx); 2335 dsl_dataset_t *hds; 2336 int error; 2337 2338 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 2339 if (error != 0) 2340 return (error); 2341 2342 if (ddrsa->ddrsa_recursive) { 2343 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 2344 dsl_dataset_rename_snapshot_check_impl, ddrsa, 2345 DS_FIND_CHILDREN); 2346 } else { 2347 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 2348 } 2349 dsl_dataset_rele(hds, FTAG); 2350 return (error); 2351 } 2352 2353 static int 2354 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 2355 dsl_dataset_t *hds, void *arg) 2356 { 2357 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2358 dsl_dataset_t *ds; 2359 uint64_t val; 2360 dmu_tx_t *tx = ddrsa->ddrsa_tx; 2361 int error; 2362 2363 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2364 ASSERT(error == 0 || error == ENOENT); 2365 if (error == ENOENT) { 2366 /* ignore nonexistent snapshots */ 2367 return (0); 2368 } 2369 2370 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 2371 2372 /* log before we change the name */ 2373 spa_history_log_internal_ds(ds, "rename", tx, 2374 "-> @%s", ddrsa->ddrsa_newsnapname); 2375 2376 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 2377 B_FALSE)); 2378 mutex_enter(&ds->ds_lock); 2379 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 2380 mutex_exit(&ds->ds_lock); 2381 VERIFY0(zap_add(dp->dp_meta_objset, 2382 dsl_dataset_phys(hds)->ds_snapnames_zapobj, 2383 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 2384 2385 dsl_dataset_rele(ds, FTAG); 2386 return (0); 2387 } 2388 2389 static void 2390 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 2391 { 2392 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2393 dsl_pool_t *dp = dmu_tx_pool(tx); 2394 dsl_dataset_t *hds; 2395 2396 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 2397 ddrsa->ddrsa_tx = tx; 2398 if (ddrsa->ddrsa_recursive) { 2399 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 2400 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 2401 DS_FIND_CHILDREN)); 2402 } else { 2403 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 2404 } 2405 dsl_dataset_rele(hds, FTAG); 2406 } 2407 2408 int 2409 dsl_dataset_rename_snapshot(const char *fsname, 2410 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 2411 { 2412 dsl_dataset_rename_snapshot_arg_t ddrsa; 2413 2414 ddrsa.ddrsa_fsname = fsname; 2415 ddrsa.ddrsa_oldsnapname = oldsnapname; 2416 ddrsa.ddrsa_newsnapname = newsnapname; 2417 ddrsa.ddrsa_recursive = recursive; 2418 2419 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 2420 dsl_dataset_rename_snapshot_sync, &ddrsa, 2421 1, ZFS_SPACE_CHECK_RESERVED)); 2422 } 2423 2424 /* 2425 * If we're doing an ownership handoff, we need to make sure that there is 2426 * only one long hold on the dataset. We're not allowed to change anything here 2427 * so we don't permanently release the long hold or regular hold here. We want 2428 * to do this only when syncing to avoid the dataset unexpectedly going away 2429 * when we release the long hold. 2430 */ 2431 static int 2432 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 2433 { 2434 boolean_t held; 2435 2436 if (!dmu_tx_is_syncing(tx)) 2437 return (0); 2438 2439 if (owner != NULL) { 2440 VERIFY3P(ds->ds_owner, ==, owner); 2441 dsl_dataset_long_rele(ds, owner); 2442 } 2443 2444 held = dsl_dataset_long_held(ds); 2445 2446 if (owner != NULL) 2447 dsl_dataset_long_hold(ds, owner); 2448 2449 if (held) 2450 return (SET_ERROR(EBUSY)); 2451 2452 return (0); 2453 } 2454 2455 typedef struct dsl_dataset_rollback_arg { 2456 const char *ddra_fsname; 2457 const char *ddra_tosnap; 2458 void *ddra_owner; 2459 nvlist_t *ddra_result; 2460 } dsl_dataset_rollback_arg_t; 2461 2462 static int 2463 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 2464 { 2465 dsl_dataset_rollback_arg_t *ddra = arg; 2466 dsl_pool_t *dp = dmu_tx_pool(tx); 2467 dsl_dataset_t *ds; 2468 int64_t unused_refres_delta; 2469 int error; 2470 2471 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 2472 if (error != 0) 2473 return (error); 2474 2475 /* must not be a snapshot */ 2476 if (ds->ds_is_snapshot) { 2477 dsl_dataset_rele(ds, FTAG); 2478 return (SET_ERROR(EINVAL)); 2479 } 2480 2481 /* must have a most recent snapshot */ 2482 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { 2483 dsl_dataset_rele(ds, FTAG); 2484 return (SET_ERROR(EINVAL)); 2485 } 2486 2487 /* 2488 * No rollback to a snapshot created in the current txg, because 2489 * the rollback may dirty the dataset and create blocks that are 2490 * not reachable from the rootbp while having a birth txg that 2491 * falls into the snapshot's range. 2492 */ 2493 if (dmu_tx_is_syncing(tx) && 2494 dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) { 2495 dsl_dataset_rele(ds, FTAG); 2496 return (SET_ERROR(EAGAIN)); 2497 } 2498 2499 /* 2500 * If the expected target snapshot is specified, then check that 2501 * the latest snapshot is it. 2502 */ 2503 if (ddra->ddra_tosnap != NULL) { 2504 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 2505 2506 dsl_dataset_name(ds->ds_prev, namebuf); 2507 if (strcmp(namebuf, ddra->ddra_tosnap) != 0) 2508 return (SET_ERROR(EXDEV)); 2509 } 2510 2511 /* must not have any bookmarks after the most recent snapshot */ 2512 nvlist_t *proprequest = fnvlist_alloc(); 2513 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2514 nvlist_t *bookmarks = fnvlist_alloc(); 2515 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 2516 fnvlist_free(proprequest); 2517 if (error != 0) 2518 return (error); 2519 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 2520 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 2521 nvlist_t *valuenv = 2522 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 2523 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2524 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 2525 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 2526 fnvlist_free(bookmarks); 2527 dsl_dataset_rele(ds, FTAG); 2528 return (SET_ERROR(EEXIST)); 2529 } 2530 } 2531 fnvlist_free(bookmarks); 2532 2533 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 2534 if (error != 0) { 2535 dsl_dataset_rele(ds, FTAG); 2536 return (error); 2537 } 2538 2539 /* 2540 * Check if the snap we are rolling back to uses more than 2541 * the refquota. 2542 */ 2543 if (ds->ds_quota != 0 && 2544 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { 2545 dsl_dataset_rele(ds, FTAG); 2546 return (SET_ERROR(EDQUOT)); 2547 } 2548 2549 /* 2550 * When we do the clone swap, we will temporarily use more space 2551 * due to the refreservation (the head will no longer have any 2552 * unique space, so the entire amount of the refreservation will need 2553 * to be free). We will immediately destroy the clone, freeing 2554 * this space, but the freeing happens over many txg's. 2555 */ 2556 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 2557 dsl_dataset_phys(ds)->ds_unique_bytes); 2558 2559 if (unused_refres_delta > 0 && 2560 unused_refres_delta > 2561 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 2562 dsl_dataset_rele(ds, FTAG); 2563 return (SET_ERROR(ENOSPC)); 2564 } 2565 2566 dsl_dataset_rele(ds, FTAG); 2567 return (0); 2568 } 2569 2570 static void 2571 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 2572 { 2573 dsl_dataset_rollback_arg_t *ddra = arg; 2574 dsl_pool_t *dp = dmu_tx_pool(tx); 2575 dsl_dataset_t *ds, *clone; 2576 uint64_t cloneobj; 2577 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 2578 2579 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 2580 2581 dsl_dataset_name(ds->ds_prev, namebuf); 2582 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 2583 2584 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 2585 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 2586 2587 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 2588 2589 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 2590 dsl_dataset_zero_zil(ds, tx); 2591 2592 dsl_destroy_head_sync_impl(clone, tx); 2593 2594 dsl_dataset_rele(clone, FTAG); 2595 dsl_dataset_rele(ds, FTAG); 2596 } 2597 2598 /* 2599 * Rolls back the given filesystem or volume to the most recent snapshot. 2600 * The name of the most recent snapshot will be returned under key "target" 2601 * in the result nvlist. 2602 * 2603 * If owner != NULL: 2604 * - The existing dataset MUST be owned by the specified owner at entry 2605 * - Upon return, dataset will still be held by the same owner, whether we 2606 * succeed or not. 2607 * 2608 * This mode is required any time the existing filesystem is mounted. See 2609 * notes above zfs_suspend_fs() for further details. 2610 */ 2611 int 2612 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner, 2613 nvlist_t *result) 2614 { 2615 dsl_dataset_rollback_arg_t ddra; 2616 2617 ddra.ddra_fsname = fsname; 2618 ddra.ddra_tosnap = tosnap; 2619 ddra.ddra_owner = owner; 2620 ddra.ddra_result = result; 2621 2622 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 2623 dsl_dataset_rollback_sync, &ddra, 2624 1, ZFS_SPACE_CHECK_RESERVED)); 2625 } 2626 2627 struct promotenode { 2628 list_node_t link; 2629 dsl_dataset_t *ds; 2630 }; 2631 2632 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2633 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2634 void *tag); 2635 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2636 2637 int 2638 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2639 { 2640 dsl_dataset_promote_arg_t *ddpa = arg; 2641 dsl_pool_t *dp = dmu_tx_pool(tx); 2642 dsl_dataset_t *hds; 2643 struct promotenode *snap; 2644 dsl_dataset_t *origin_ds; 2645 int err; 2646 uint64_t unused; 2647 uint64_t ss_mv_cnt; 2648 size_t max_snap_len; 2649 boolean_t conflicting_snaps; 2650 2651 err = promote_hold(ddpa, dp, FTAG); 2652 if (err != 0) 2653 return (err); 2654 2655 hds = ddpa->ddpa_clone; 2656 snap = list_head(&ddpa->shared_snaps); 2657 origin_ds = snap->ds; 2658 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; 2659 2660 snap = list_head(&ddpa->origin_snaps); 2661 2662 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { 2663 promote_rele(ddpa, FTAG); 2664 return (SET_ERROR(EXDEV)); 2665 } 2666 2667 /* 2668 * Compute and check the amount of space to transfer. Since this is 2669 * so expensive, don't do the preliminary check. 2670 */ 2671 if (!dmu_tx_is_syncing(tx)) { 2672 promote_rele(ddpa, FTAG); 2673 return (0); 2674 } 2675 2676 /* compute origin's new unique space */ 2677 snap = list_tail(&ddpa->clone_snaps); 2678 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2679 origin_ds->ds_object); 2680 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2681 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, 2682 &ddpa->unique, &unused, &unused); 2683 2684 /* 2685 * Walk the snapshots that we are moving 2686 * 2687 * Compute space to transfer. Consider the incremental changes 2688 * to used by each snapshot: 2689 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2690 * So each snapshot gave birth to: 2691 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2692 * So a sequence would look like: 2693 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2694 * Which simplifies to: 2695 * uN + kN + kN-1 + ... + k1 + k0 2696 * Note however, if we stop before we reach the ORIGIN we get: 2697 * uN + kN + kN-1 + ... + kM - uM-1 2698 */ 2699 conflicting_snaps = B_FALSE; 2700 ss_mv_cnt = 0; 2701 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; 2702 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; 2703 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; 2704 for (snap = list_head(&ddpa->shared_snaps); snap; 2705 snap = list_next(&ddpa->shared_snaps, snap)) { 2706 uint64_t val, dlused, dlcomp, dluncomp; 2707 dsl_dataset_t *ds = snap->ds; 2708 2709 ss_mv_cnt++; 2710 2711 /* 2712 * If there are long holds, we won't be able to evict 2713 * the objset. 2714 */ 2715 if (dsl_dataset_long_held(ds)) { 2716 err = SET_ERROR(EBUSY); 2717 goto out; 2718 } 2719 2720 /* Check that the snapshot name does not conflict */ 2721 VERIFY0(dsl_dataset_get_snapname(ds)); 2722 if (strlen(ds->ds_snapname) >= max_snap_len) { 2723 err = SET_ERROR(ENAMETOOLONG); 2724 goto out; 2725 } 2726 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2727 if (err == 0) { 2728 fnvlist_add_boolean(ddpa->err_ds, 2729 snap->ds->ds_snapname); 2730 conflicting_snaps = B_TRUE; 2731 } else if (err != ENOENT) { 2732 goto out; 2733 } 2734 2735 /* The very first snapshot does not have a deadlist */ 2736 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) 2737 continue; 2738 2739 dsl_deadlist_space(&ds->ds_deadlist, 2740 &dlused, &dlcomp, &dluncomp); 2741 ddpa->used += dlused; 2742 ddpa->comp += dlcomp; 2743 ddpa->uncomp += dluncomp; 2744 } 2745 2746 /* 2747 * In order to return the full list of conflicting snapshots, we check 2748 * whether there was a conflict after traversing all of them. 2749 */ 2750 if (conflicting_snaps) { 2751 err = SET_ERROR(EEXIST); 2752 goto out; 2753 } 2754 2755 /* 2756 * If we are a clone of a clone then we never reached ORIGIN, 2757 * so we need to subtract out the clone origin's used space. 2758 */ 2759 if (ddpa->origin_origin) { 2760 ddpa->used -= 2761 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; 2762 ddpa->comp -= 2763 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; 2764 ddpa->uncomp -= 2765 dsl_dataset_phys(ddpa->origin_origin)-> 2766 ds_uncompressed_bytes; 2767 } 2768 2769 /* Check that there is enough space and limit headroom here */ 2770 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2771 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2772 if (err != 0) 2773 goto out; 2774 2775 /* 2776 * Compute the amounts of space that will be used by snapshots 2777 * after the promotion (for both origin and clone). For each, 2778 * it is the amount of space that will be on all of their 2779 * deadlists (that was not born before their new origin). 2780 */ 2781 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2782 uint64_t space; 2783 2784 /* 2785 * Note, typically this will not be a clone of a clone, 2786 * so dd_origin_txg will be < TXG_INITIAL, so 2787 * these snaplist_space() -> dsl_deadlist_space_range() 2788 * calls will be fast because they do not have to 2789 * iterate over all bps. 2790 */ 2791 snap = list_head(&ddpa->origin_snaps); 2792 err = snaplist_space(&ddpa->shared_snaps, 2793 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2794 if (err != 0) 2795 goto out; 2796 2797 err = snaplist_space(&ddpa->clone_snaps, 2798 snap->ds->ds_dir->dd_origin_txg, &space); 2799 if (err != 0) 2800 goto out; 2801 ddpa->cloneusedsnap += space; 2802 } 2803 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & 2804 DD_FLAG_USED_BREAKDOWN) { 2805 err = snaplist_space(&ddpa->origin_snaps, 2806 dsl_dataset_phys(origin_ds)->ds_creation_txg, 2807 &ddpa->originusedsnap); 2808 if (err != 0) 2809 goto out; 2810 } 2811 2812 out: 2813 promote_rele(ddpa, FTAG); 2814 return (err); 2815 } 2816 2817 void 2818 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2819 { 2820 dsl_dataset_promote_arg_t *ddpa = arg; 2821 dsl_pool_t *dp = dmu_tx_pool(tx); 2822 dsl_dataset_t *hds; 2823 struct promotenode *snap; 2824 dsl_dataset_t *origin_ds; 2825 dsl_dataset_t *origin_head; 2826 dsl_dir_t *dd; 2827 dsl_dir_t *odd = NULL; 2828 uint64_t oldnext_obj; 2829 int64_t delta; 2830 2831 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2832 hds = ddpa->ddpa_clone; 2833 2834 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); 2835 2836 snap = list_head(&ddpa->shared_snaps); 2837 origin_ds = snap->ds; 2838 dd = hds->ds_dir; 2839 2840 snap = list_head(&ddpa->origin_snaps); 2841 origin_head = snap->ds; 2842 2843 /* 2844 * We need to explicitly open odd, since origin_ds's dd will be 2845 * changing. 2846 */ 2847 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2848 NULL, FTAG, &odd)); 2849 2850 /* change origin's next snap */ 2851 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2852 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; 2853 snap = list_tail(&ddpa->clone_snaps); 2854 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2855 origin_ds->ds_object); 2856 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; 2857 2858 /* change the origin's next clone */ 2859 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { 2860 dsl_dataset_remove_from_next_clones(origin_ds, 2861 snap->ds->ds_object, tx); 2862 VERIFY0(zap_add_int(dp->dp_meta_objset, 2863 dsl_dataset_phys(origin_ds)->ds_next_clones_obj, 2864 oldnext_obj, tx)); 2865 } 2866 2867 /* change origin */ 2868 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2869 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); 2870 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; 2871 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2872 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2873 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; 2874 origin_head->ds_dir->dd_origin_txg = 2875 dsl_dataset_phys(origin_ds)->ds_creation_txg; 2876 2877 /* change dd_clone entries */ 2878 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2879 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2880 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); 2881 VERIFY0(zap_add_int(dp->dp_meta_objset, 2882 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2883 hds->ds_object, tx)); 2884 2885 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2886 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2887 origin_head->ds_object, tx)); 2888 if (dsl_dir_phys(dd)->dd_clones == 0) { 2889 dsl_dir_phys(dd)->dd_clones = 2890 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, 2891 DMU_OT_NONE, 0, tx); 2892 } 2893 VERIFY0(zap_add_int(dp->dp_meta_objset, 2894 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); 2895 } 2896 2897 /* move snapshots to this dir */ 2898 for (snap = list_head(&ddpa->shared_snaps); snap; 2899 snap = list_next(&ddpa->shared_snaps, snap)) { 2900 dsl_dataset_t *ds = snap->ds; 2901 2902 /* 2903 * Property callbacks are registered to a particular 2904 * dsl_dir. Since ours is changing, evict the objset 2905 * so that they will be unregistered from the old dsl_dir. 2906 */ 2907 if (ds->ds_objset) { 2908 dmu_objset_evict(ds->ds_objset); 2909 ds->ds_objset = NULL; 2910 } 2911 2912 /* move snap name entry */ 2913 VERIFY0(dsl_dataset_get_snapname(ds)); 2914 VERIFY0(dsl_dataset_snap_remove(origin_head, 2915 ds->ds_snapname, tx, B_TRUE)); 2916 VERIFY0(zap_add(dp->dp_meta_objset, 2917 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 2918 8, 1, &ds->ds_object, tx)); 2919 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 2920 DD_FIELD_SNAPSHOT_COUNT, tx); 2921 2922 /* change containing dsl_dir */ 2923 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2924 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); 2925 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; 2926 ASSERT3P(ds->ds_dir, ==, odd); 2927 dsl_dir_rele(ds->ds_dir, ds); 2928 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 2929 NULL, ds, &ds->ds_dir)); 2930 2931 /* move any clone references */ 2932 if (dsl_dataset_phys(ds)->ds_next_clones_obj && 2933 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2934 zap_cursor_t zc; 2935 zap_attribute_t za; 2936 2937 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2938 dsl_dataset_phys(ds)->ds_next_clones_obj); 2939 zap_cursor_retrieve(&zc, &za) == 0; 2940 zap_cursor_advance(&zc)) { 2941 dsl_dataset_t *cnds; 2942 uint64_t o; 2943 2944 if (za.za_first_integer == oldnext_obj) { 2945 /* 2946 * We've already moved the 2947 * origin's reference. 2948 */ 2949 continue; 2950 } 2951 2952 VERIFY0(dsl_dataset_hold_obj(dp, 2953 za.za_first_integer, FTAG, &cnds)); 2954 o = dsl_dir_phys(cnds->ds_dir)-> 2955 dd_head_dataset_obj; 2956 2957 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2958 dsl_dir_phys(odd)->dd_clones, o, tx)); 2959 VERIFY0(zap_add_int(dp->dp_meta_objset, 2960 dsl_dir_phys(dd)->dd_clones, o, tx)); 2961 dsl_dataset_rele(cnds, FTAG); 2962 } 2963 zap_cursor_fini(&zc); 2964 } 2965 2966 ASSERT(!dsl_prop_hascb(ds)); 2967 } 2968 2969 /* 2970 * Change space accounting. 2971 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2972 * both be valid, or both be 0 (resulting in delta == 0). This 2973 * is true for each of {clone,origin} independently. 2974 */ 2975 2976 delta = ddpa->cloneusedsnap - 2977 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; 2978 ASSERT3S(delta, >=, 0); 2979 ASSERT3U(ddpa->used, >=, delta); 2980 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2981 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2982 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 2983 2984 delta = ddpa->originusedsnap - 2985 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; 2986 ASSERT3S(delta, <=, 0); 2987 ASSERT3U(ddpa->used, >=, -delta); 2988 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2989 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2990 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 2991 2992 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; 2993 2994 /* log history record */ 2995 spa_history_log_internal_ds(hds, "promote", tx, ""); 2996 2997 dsl_dir_rele(odd, FTAG); 2998 promote_rele(ddpa, FTAG); 2999 } 3000 3001 /* 3002 * Make a list of dsl_dataset_t's for the snapshots between first_obj 3003 * (exclusive) and last_obj (inclusive). The list will be in reverse 3004 * order (last_obj will be the list_head()). If first_obj == 0, do all 3005 * snapshots back to this dataset's origin. 3006 */ 3007 static int 3008 snaplist_make(dsl_pool_t *dp, 3009 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 3010 { 3011 uint64_t obj = last_obj; 3012 3013 list_create(l, sizeof (struct promotenode), 3014 offsetof(struct promotenode, link)); 3015 3016 while (obj != first_obj) { 3017 dsl_dataset_t *ds; 3018 struct promotenode *snap; 3019 int err; 3020 3021 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 3022 ASSERT(err != ENOENT); 3023 if (err != 0) 3024 return (err); 3025 3026 if (first_obj == 0) 3027 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; 3028 3029 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 3030 snap->ds = ds; 3031 list_insert_tail(l, snap); 3032 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3033 } 3034 3035 return (0); 3036 } 3037 3038 static int 3039 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 3040 { 3041 struct promotenode *snap; 3042 3043 *spacep = 0; 3044 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 3045 uint64_t used, comp, uncomp; 3046 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 3047 mintxg, UINT64_MAX, &used, &comp, &uncomp); 3048 *spacep += used; 3049 } 3050 return (0); 3051 } 3052 3053 static void 3054 snaplist_destroy(list_t *l, void *tag) 3055 { 3056 struct promotenode *snap; 3057 3058 if (l == NULL || !list_link_active(&l->list_head)) 3059 return; 3060 3061 while ((snap = list_tail(l)) != NULL) { 3062 list_remove(l, snap); 3063 dsl_dataset_rele(snap->ds, tag); 3064 kmem_free(snap, sizeof (*snap)); 3065 } 3066 list_destroy(l); 3067 } 3068 3069 static int 3070 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 3071 { 3072 int error; 3073 dsl_dir_t *dd; 3074 struct promotenode *snap; 3075 3076 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 3077 &ddpa->ddpa_clone); 3078 if (error != 0) 3079 return (error); 3080 dd = ddpa->ddpa_clone->ds_dir; 3081 3082 if (ddpa->ddpa_clone->ds_is_snapshot || 3083 !dsl_dir_is_clone(dd)) { 3084 dsl_dataset_rele(ddpa->ddpa_clone, tag); 3085 return (SET_ERROR(EINVAL)); 3086 } 3087 3088 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, 3089 &ddpa->shared_snaps, tag); 3090 if (error != 0) 3091 goto out; 3092 3093 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 3094 &ddpa->clone_snaps, tag); 3095 if (error != 0) 3096 goto out; 3097 3098 snap = list_head(&ddpa->shared_snaps); 3099 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); 3100 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, 3101 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, 3102 &ddpa->origin_snaps, tag); 3103 if (error != 0) 3104 goto out; 3105 3106 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { 3107 error = dsl_dataset_hold_obj(dp, 3108 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, 3109 tag, &ddpa->origin_origin); 3110 if (error != 0) 3111 goto out; 3112 } 3113 out: 3114 if (error != 0) 3115 promote_rele(ddpa, tag); 3116 return (error); 3117 } 3118 3119 static void 3120 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 3121 { 3122 snaplist_destroy(&ddpa->shared_snaps, tag); 3123 snaplist_destroy(&ddpa->clone_snaps, tag); 3124 snaplist_destroy(&ddpa->origin_snaps, tag); 3125 if (ddpa->origin_origin != NULL) 3126 dsl_dataset_rele(ddpa->origin_origin, tag); 3127 dsl_dataset_rele(ddpa->ddpa_clone, tag); 3128 } 3129 3130 /* 3131 * Promote a clone. 3132 * 3133 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 3134 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.) 3135 */ 3136 int 3137 dsl_dataset_promote(const char *name, char *conflsnap) 3138 { 3139 dsl_dataset_promote_arg_t ddpa = { 0 }; 3140 uint64_t numsnaps; 3141 int error; 3142 nvpair_t *snap_pair; 3143 objset_t *os; 3144 3145 /* 3146 * We will modify space proportional to the number of 3147 * snapshots. Compute numsnaps. 3148 */ 3149 error = dmu_objset_hold(name, FTAG, &os); 3150 if (error != 0) 3151 return (error); 3152 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 3153 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, 3154 &numsnaps); 3155 dmu_objset_rele(os, FTAG); 3156 if (error != 0) 3157 return (error); 3158 3159 ddpa.ddpa_clonename = name; 3160 ddpa.err_ds = fnvlist_alloc(); 3161 ddpa.cr = CRED(); 3162 3163 error = dsl_sync_task(name, dsl_dataset_promote_check, 3164 dsl_dataset_promote_sync, &ddpa, 3165 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED); 3166 3167 /* 3168 * Return the first conflicting snapshot found. 3169 */ 3170 snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL); 3171 if (snap_pair != NULL && conflsnap != NULL) 3172 (void) strcpy(conflsnap, nvpair_name(snap_pair)); 3173 3174 fnvlist_free(ddpa.err_ds); 3175 return (error); 3176 } 3177 3178 int 3179 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 3180 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 3181 { 3182 /* 3183 * "slack" factor for received datasets with refquota set on them. 3184 * See the bottom of this function for details on its use. 3185 */ 3186 uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation; 3187 int64_t unused_refres_delta; 3188 3189 /* they should both be heads */ 3190 if (clone->ds_is_snapshot || 3191 origin_head->ds_is_snapshot) 3192 return (SET_ERROR(EINVAL)); 3193 3194 /* if we are not forcing, the branch point should be just before them */ 3195 if (!force && clone->ds_prev != origin_head->ds_prev) 3196 return (SET_ERROR(EINVAL)); 3197 3198 /* clone should be the clone (unless they are unrelated) */ 3199 if (clone->ds_prev != NULL && 3200 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 3201 origin_head->ds_dir != clone->ds_prev->ds_dir) 3202 return (SET_ERROR(EINVAL)); 3203 3204 /* the clone should be a child of the origin */ 3205 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 3206 return (SET_ERROR(EINVAL)); 3207 3208 /* origin_head shouldn't be modified unless 'force' */ 3209 if (!force && 3210 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 3211 return (SET_ERROR(ETXTBSY)); 3212 3213 /* origin_head should have no long holds (e.g. is not mounted) */ 3214 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 3215 return (SET_ERROR(EBUSY)); 3216 3217 /* check amount of any unconsumed refreservation */ 3218 unused_refres_delta = 3219 (int64_t)MIN(origin_head->ds_reserved, 3220 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 3221 (int64_t)MIN(origin_head->ds_reserved, 3222 dsl_dataset_phys(clone)->ds_unique_bytes); 3223 3224 if (unused_refres_delta > 0 && 3225 unused_refres_delta > 3226 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 3227 return (SET_ERROR(ENOSPC)); 3228 3229 /* 3230 * The clone can't be too much over the head's refquota. 3231 * 3232 * To ensure that the entire refquota can be used, we allow one 3233 * transaction to exceed the the refquota. Therefore, this check 3234 * needs to also allow for the space referenced to be more than the 3235 * refquota. The maximum amount of space that one transaction can use 3236 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this 3237 * overage ensures that we are able to receive a filesystem that 3238 * exceeds the refquota on the source system. 3239 * 3240 * So that overage is the refquota_slack we use below. 3241 */ 3242 if (origin_head->ds_quota != 0 && 3243 dsl_dataset_phys(clone)->ds_referenced_bytes > 3244 origin_head->ds_quota + refquota_slack) 3245 return (SET_ERROR(EDQUOT)); 3246 3247 return (0); 3248 } 3249 3250 void 3251 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 3252 dsl_dataset_t *origin_head, dmu_tx_t *tx) 3253 { 3254 dsl_pool_t *dp = dmu_tx_pool(tx); 3255 int64_t unused_refres_delta; 3256 3257 ASSERT(clone->ds_reserved == 0); 3258 /* 3259 * NOTE: On DEBUG kernels there could be a race between this and 3260 * the check function if spa_asize_inflation is adjusted... 3261 */ 3262 ASSERT(origin_head->ds_quota == 0 || 3263 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota + 3264 DMU_MAX_ACCESS * spa_asize_inflation); 3265 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 3266 3267 /* 3268 * Swap per-dataset feature flags. 3269 */ 3270 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 3271 if (!(spa_feature_table[f].fi_flags & 3272 ZFEATURE_FLAG_PER_DATASET)) { 3273 ASSERT(!clone->ds_feature_inuse[f]); 3274 ASSERT(!origin_head->ds_feature_inuse[f]); 3275 continue; 3276 } 3277 3278 boolean_t clone_inuse = clone->ds_feature_inuse[f]; 3279 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f]; 3280 3281 if (clone_inuse) { 3282 dsl_dataset_deactivate_feature(clone->ds_object, f, tx); 3283 clone->ds_feature_inuse[f] = B_FALSE; 3284 } 3285 if (origin_head_inuse) { 3286 dsl_dataset_deactivate_feature(origin_head->ds_object, 3287 f, tx); 3288 origin_head->ds_feature_inuse[f] = B_FALSE; 3289 } 3290 if (clone_inuse) { 3291 dsl_dataset_activate_feature(origin_head->ds_object, 3292 f, tx); 3293 origin_head->ds_feature_inuse[f] = B_TRUE; 3294 } 3295 if (origin_head_inuse) { 3296 dsl_dataset_activate_feature(clone->ds_object, f, tx); 3297 clone->ds_feature_inuse[f] = B_TRUE; 3298 } 3299 } 3300 3301 dmu_buf_will_dirty(clone->ds_dbuf, tx); 3302 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 3303 3304 if (clone->ds_objset != NULL) { 3305 dmu_objset_evict(clone->ds_objset); 3306 clone->ds_objset = NULL; 3307 } 3308 3309 if (origin_head->ds_objset != NULL) { 3310 dmu_objset_evict(origin_head->ds_objset); 3311 origin_head->ds_objset = NULL; 3312 } 3313 3314 unused_refres_delta = 3315 (int64_t)MIN(origin_head->ds_reserved, 3316 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 3317 (int64_t)MIN(origin_head->ds_reserved, 3318 dsl_dataset_phys(clone)->ds_unique_bytes); 3319 3320 /* 3321 * Reset origin's unique bytes, if it exists. 3322 */ 3323 if (clone->ds_prev) { 3324 dsl_dataset_t *origin = clone->ds_prev; 3325 uint64_t comp, uncomp; 3326 3327 dmu_buf_will_dirty(origin->ds_dbuf, tx); 3328 dsl_deadlist_space_range(&clone->ds_deadlist, 3329 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, 3330 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); 3331 } 3332 3333 /* swap blkptrs */ 3334 { 3335 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG); 3336 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG); 3337 blkptr_t tmp; 3338 tmp = dsl_dataset_phys(origin_head)->ds_bp; 3339 dsl_dataset_phys(origin_head)->ds_bp = 3340 dsl_dataset_phys(clone)->ds_bp; 3341 dsl_dataset_phys(clone)->ds_bp = tmp; 3342 rrw_exit(&origin_head->ds_bp_rwlock, FTAG); 3343 rrw_exit(&clone->ds_bp_rwlock, FTAG); 3344 } 3345 3346 /* set dd_*_bytes */ 3347 { 3348 int64_t dused, dcomp, duncomp; 3349 uint64_t cdl_used, cdl_comp, cdl_uncomp; 3350 uint64_t odl_used, odl_comp, odl_uncomp; 3351 3352 ASSERT3U(dsl_dir_phys(clone->ds_dir)-> 3353 dd_used_breakdown[DD_USED_SNAP], ==, 0); 3354 3355 dsl_deadlist_space(&clone->ds_deadlist, 3356 &cdl_used, &cdl_comp, &cdl_uncomp); 3357 dsl_deadlist_space(&origin_head->ds_deadlist, 3358 &odl_used, &odl_comp, &odl_uncomp); 3359 3360 dused = dsl_dataset_phys(clone)->ds_referenced_bytes + 3361 cdl_used - 3362 (dsl_dataset_phys(origin_head)->ds_referenced_bytes + 3363 odl_used); 3364 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + 3365 cdl_comp - 3366 (dsl_dataset_phys(origin_head)->ds_compressed_bytes + 3367 odl_comp); 3368 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + 3369 cdl_uncomp - 3370 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + 3371 odl_uncomp); 3372 3373 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 3374 dused, dcomp, duncomp, tx); 3375 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 3376 -dused, -dcomp, -duncomp, tx); 3377 3378 /* 3379 * The difference in the space used by snapshots is the 3380 * difference in snapshot space due to the head's 3381 * deadlist (since that's the only thing that's 3382 * changing that affects the snapused). 3383 */ 3384 dsl_deadlist_space_range(&clone->ds_deadlist, 3385 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 3386 &cdl_used, &cdl_comp, &cdl_uncomp); 3387 dsl_deadlist_space_range(&origin_head->ds_deadlist, 3388 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 3389 &odl_used, &odl_comp, &odl_uncomp); 3390 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 3391 DD_USED_HEAD, DD_USED_SNAP, tx); 3392 } 3393 3394 /* swap ds_*_bytes */ 3395 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, 3396 dsl_dataset_phys(clone)->ds_referenced_bytes); 3397 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, 3398 dsl_dataset_phys(clone)->ds_compressed_bytes); 3399 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, 3400 dsl_dataset_phys(clone)->ds_uncompressed_bytes); 3401 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, 3402 dsl_dataset_phys(clone)->ds_unique_bytes); 3403 3404 /* apply any parent delta for change in unconsumed refreservation */ 3405 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 3406 unused_refres_delta, 0, 0, tx); 3407 3408 /* 3409 * Swap deadlists. 3410 */ 3411 dsl_deadlist_close(&clone->ds_deadlist); 3412 dsl_deadlist_close(&origin_head->ds_deadlist); 3413 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, 3414 dsl_dataset_phys(clone)->ds_deadlist_obj); 3415 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 3416 dsl_dataset_phys(clone)->ds_deadlist_obj); 3417 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 3418 dsl_dataset_phys(origin_head)->ds_deadlist_obj); 3419 3420 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 3421 3422 spa_history_log_internal_ds(clone, "clone swap", tx, 3423 "parent=%s", origin_head->ds_dir->dd_myname); 3424 } 3425 3426 /* 3427 * Given a pool name and a dataset object number in that pool, 3428 * return the name of that dataset. 3429 */ 3430 int 3431 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 3432 { 3433 dsl_pool_t *dp; 3434 dsl_dataset_t *ds; 3435 int error; 3436 3437 error = dsl_pool_hold(pname, FTAG, &dp); 3438 if (error != 0) 3439 return (error); 3440 3441 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 3442 if (error == 0) { 3443 dsl_dataset_name(ds, buf); 3444 dsl_dataset_rele(ds, FTAG); 3445 } 3446 dsl_pool_rele(dp, FTAG); 3447 3448 return (error); 3449 } 3450 3451 int 3452 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 3453 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 3454 { 3455 int error = 0; 3456 3457 ASSERT3S(asize, >, 0); 3458 3459 /* 3460 * *ref_rsrv is the portion of asize that will come from any 3461 * unconsumed refreservation space. 3462 */ 3463 *ref_rsrv = 0; 3464 3465 mutex_enter(&ds->ds_lock); 3466 /* 3467 * Make a space adjustment for reserved bytes. 3468 */ 3469 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 3470 ASSERT3U(*used, >=, 3471 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 3472 *used -= 3473 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 3474 *ref_rsrv = 3475 asize - MIN(asize, parent_delta(ds, asize + inflight)); 3476 } 3477 3478 if (!check_quota || ds->ds_quota == 0) { 3479 mutex_exit(&ds->ds_lock); 3480 return (0); 3481 } 3482 /* 3483 * If they are requesting more space, and our current estimate 3484 * is over quota, they get to try again unless the actual 3485 * on-disk is over quota and there are no pending changes (which 3486 * may free up space for us). 3487 */ 3488 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= 3489 ds->ds_quota) { 3490 if (inflight > 0 || 3491 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) 3492 error = SET_ERROR(ERESTART); 3493 else 3494 error = SET_ERROR(EDQUOT); 3495 } 3496 mutex_exit(&ds->ds_lock); 3497 3498 return (error); 3499 } 3500 3501 typedef struct dsl_dataset_set_qr_arg { 3502 const char *ddsqra_name; 3503 zprop_source_t ddsqra_source; 3504 uint64_t ddsqra_value; 3505 } dsl_dataset_set_qr_arg_t; 3506 3507 3508 /* ARGSUSED */ 3509 static int 3510 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 3511 { 3512 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3513 dsl_pool_t *dp = dmu_tx_pool(tx); 3514 dsl_dataset_t *ds; 3515 int error; 3516 uint64_t newval; 3517 3518 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 3519 return (SET_ERROR(ENOTSUP)); 3520 3521 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3522 if (error != 0) 3523 return (error); 3524 3525 if (ds->ds_is_snapshot) { 3526 dsl_dataset_rele(ds, FTAG); 3527 return (SET_ERROR(EINVAL)); 3528 } 3529 3530 error = dsl_prop_predict(ds->ds_dir, 3531 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3532 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3533 if (error != 0) { 3534 dsl_dataset_rele(ds, FTAG); 3535 return (error); 3536 } 3537 3538 if (newval == 0) { 3539 dsl_dataset_rele(ds, FTAG); 3540 return (0); 3541 } 3542 3543 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || 3544 newval < ds->ds_reserved) { 3545 dsl_dataset_rele(ds, FTAG); 3546 return (SET_ERROR(ENOSPC)); 3547 } 3548 3549 dsl_dataset_rele(ds, FTAG); 3550 return (0); 3551 } 3552 3553 static void 3554 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 3555 { 3556 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3557 dsl_pool_t *dp = dmu_tx_pool(tx); 3558 dsl_dataset_t *ds; 3559 uint64_t newval; 3560 3561 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3562 3563 dsl_prop_set_sync_impl(ds, 3564 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3565 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 3566 &ddsqra->ddsqra_value, tx); 3567 3568 VERIFY0(dsl_prop_get_int_ds(ds, 3569 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 3570 3571 if (ds->ds_quota != newval) { 3572 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3573 ds->ds_quota = newval; 3574 } 3575 dsl_dataset_rele(ds, FTAG); 3576 } 3577 3578 int 3579 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 3580 uint64_t refquota) 3581 { 3582 dsl_dataset_set_qr_arg_t ddsqra; 3583 3584 ddsqra.ddsqra_name = dsname; 3585 ddsqra.ddsqra_source = source; 3586 ddsqra.ddsqra_value = refquota; 3587 3588 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 3589 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 3590 } 3591 3592 static int 3593 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 3594 { 3595 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3596 dsl_pool_t *dp = dmu_tx_pool(tx); 3597 dsl_dataset_t *ds; 3598 int error; 3599 uint64_t newval, unique; 3600 3601 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 3602 return (SET_ERROR(ENOTSUP)); 3603 3604 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3605 if (error != 0) 3606 return (error); 3607 3608 if (ds->ds_is_snapshot) { 3609 dsl_dataset_rele(ds, FTAG); 3610 return (SET_ERROR(EINVAL)); 3611 } 3612 3613 error = dsl_prop_predict(ds->ds_dir, 3614 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3615 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3616 if (error != 0) { 3617 dsl_dataset_rele(ds, FTAG); 3618 return (error); 3619 } 3620 3621 /* 3622 * If we are doing the preliminary check in open context, the 3623 * space estimates may be inaccurate. 3624 */ 3625 if (!dmu_tx_is_syncing(tx)) { 3626 dsl_dataset_rele(ds, FTAG); 3627 return (0); 3628 } 3629 3630 mutex_enter(&ds->ds_lock); 3631 if (!DS_UNIQUE_IS_ACCURATE(ds)) 3632 dsl_dataset_recalc_head_uniq(ds); 3633 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3634 mutex_exit(&ds->ds_lock); 3635 3636 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 3637 uint64_t delta = MAX(unique, newval) - 3638 MAX(unique, ds->ds_reserved); 3639 3640 if (delta > 3641 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 3642 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 3643 dsl_dataset_rele(ds, FTAG); 3644 return (SET_ERROR(ENOSPC)); 3645 } 3646 } 3647 3648 dsl_dataset_rele(ds, FTAG); 3649 return (0); 3650 } 3651 3652 void 3653 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 3654 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 3655 { 3656 uint64_t newval; 3657 uint64_t unique; 3658 int64_t delta; 3659 3660 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3661 source, sizeof (value), 1, &value, tx); 3662 3663 VERIFY0(dsl_prop_get_int_ds(ds, 3664 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 3665 3666 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3667 mutex_enter(&ds->ds_dir->dd_lock); 3668 mutex_enter(&ds->ds_lock); 3669 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3670 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3671 delta = MAX(0, (int64_t)(newval - unique)) - 3672 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3673 ds->ds_reserved = newval; 3674 mutex_exit(&ds->ds_lock); 3675 3676 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3677 mutex_exit(&ds->ds_dir->dd_lock); 3678 } 3679 3680 static void 3681 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 3682 { 3683 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3684 dsl_pool_t *dp = dmu_tx_pool(tx); 3685 dsl_dataset_t *ds; 3686 3687 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3688 dsl_dataset_set_refreservation_sync_impl(ds, 3689 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 3690 dsl_dataset_rele(ds, FTAG); 3691 } 3692 3693 int 3694 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 3695 uint64_t refreservation) 3696 { 3697 dsl_dataset_set_qr_arg_t ddsqra; 3698 3699 ddsqra.ddsqra_name = dsname; 3700 ddsqra.ddsqra_source = source; 3701 ddsqra.ddsqra_value = refreservation; 3702 3703 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 3704 dsl_dataset_set_refreservation_sync, &ddsqra, 3705 0, ZFS_SPACE_CHECK_NONE)); 3706 } 3707 3708 /* 3709 * Return (in *usedp) the amount of space written in new that is not 3710 * present in oldsnap. New may be a snapshot or the head. Old must be 3711 * a snapshot before new, in new's filesystem (or its origin). If not then 3712 * fail and return EINVAL. 3713 * 3714 * The written space is calculated by considering two components: First, we 3715 * ignore any freed space, and calculate the written as new's used space 3716 * minus old's used space. Next, we add in the amount of space that was freed 3717 * between the two snapshots, thus reducing new's used space relative to old's. 3718 * Specifically, this is the space that was born before old->ds_creation_txg, 3719 * and freed before new (ie. on new's deadlist or a previous deadlist). 3720 * 3721 * space freed [---------------------] 3722 * snapshots ---O-------O--------O-------O------ 3723 * oldsnap new 3724 */ 3725 int 3726 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 3727 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3728 { 3729 int err = 0; 3730 uint64_t snapobj; 3731 dsl_pool_t *dp = new->ds_dir->dd_pool; 3732 3733 ASSERT(dsl_pool_config_held(dp)); 3734 3735 *usedp = 0; 3736 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; 3737 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes; 3738 3739 *compp = 0; 3740 *compp += dsl_dataset_phys(new)->ds_compressed_bytes; 3741 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes; 3742 3743 *uncompp = 0; 3744 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; 3745 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes; 3746 3747 snapobj = new->ds_object; 3748 while (snapobj != oldsnap->ds_object) { 3749 dsl_dataset_t *snap; 3750 uint64_t used, comp, uncomp; 3751 3752 if (snapobj == new->ds_object) { 3753 snap = new; 3754 } else { 3755 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3756 if (err != 0) 3757 break; 3758 } 3759 3760 if (dsl_dataset_phys(snap)->ds_prev_snap_txg == 3761 dsl_dataset_phys(oldsnap)->ds_creation_txg) { 3762 /* 3763 * The blocks in the deadlist can not be born after 3764 * ds_prev_snap_txg, so get the whole deadlist space, 3765 * which is more efficient (especially for old-format 3766 * deadlists). Unfortunately the deadlist code 3767 * doesn't have enough information to make this 3768 * optimization itself. 3769 */ 3770 dsl_deadlist_space(&snap->ds_deadlist, 3771 &used, &comp, &uncomp); 3772 } else { 3773 dsl_deadlist_space_range(&snap->ds_deadlist, 3774 0, dsl_dataset_phys(oldsnap)->ds_creation_txg, 3775 &used, &comp, &uncomp); 3776 } 3777 *usedp += used; 3778 *compp += comp; 3779 *uncompp += uncomp; 3780 3781 /* 3782 * If we get to the beginning of the chain of snapshots 3783 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3784 * was not a snapshot of/before new. 3785 */ 3786 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 3787 if (snap != new) 3788 dsl_dataset_rele(snap, FTAG); 3789 if (snapobj == 0) { 3790 err = SET_ERROR(EINVAL); 3791 break; 3792 } 3793 3794 } 3795 return (err); 3796 } 3797 3798 /* 3799 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3800 * lastsnap, and all snapshots in between are deleted. 3801 * 3802 * blocks that would be freed [---------------------------] 3803 * snapshots ---O-------O--------O-------O--------O 3804 * firstsnap lastsnap 3805 * 3806 * This is the set of blocks that were born after the snap before firstsnap, 3807 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3808 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3809 * We calculate this by iterating over the relevant deadlists (from the snap 3810 * after lastsnap, backward to the snap after firstsnap), summing up the 3811 * space on the deadlist that was born after the snap before firstsnap. 3812 */ 3813 int 3814 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3815 dsl_dataset_t *lastsnap, 3816 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3817 { 3818 int err = 0; 3819 uint64_t snapobj; 3820 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3821 3822 ASSERT(firstsnap->ds_is_snapshot); 3823 ASSERT(lastsnap->ds_is_snapshot); 3824 3825 /* 3826 * Check that the snapshots are in the same dsl_dir, and firstsnap 3827 * is before lastsnap. 3828 */ 3829 if (firstsnap->ds_dir != lastsnap->ds_dir || 3830 dsl_dataset_phys(firstsnap)->ds_creation_txg > 3831 dsl_dataset_phys(lastsnap)->ds_creation_txg) 3832 return (SET_ERROR(EINVAL)); 3833 3834 *usedp = *compp = *uncompp = 0; 3835 3836 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; 3837 while (snapobj != firstsnap->ds_object) { 3838 dsl_dataset_t *ds; 3839 uint64_t used, comp, uncomp; 3840 3841 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3842 if (err != 0) 3843 break; 3844 3845 dsl_deadlist_space_range(&ds->ds_deadlist, 3846 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, 3847 &used, &comp, &uncomp); 3848 *usedp += used; 3849 *compp += comp; 3850 *uncompp += uncomp; 3851 3852 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3853 ASSERT3U(snapobj, !=, 0); 3854 dsl_dataset_rele(ds, FTAG); 3855 } 3856 return (err); 3857 } 3858 3859 /* 3860 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 3861 * For example, they could both be snapshots of the same filesystem, and 3862 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 3863 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 3864 * filesystem. Or 'earlier' could be the origin's origin. 3865 * 3866 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 3867 */ 3868 boolean_t 3869 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 3870 uint64_t earlier_txg) 3871 { 3872 dsl_pool_t *dp = later->ds_dir->dd_pool; 3873 int error; 3874 boolean_t ret; 3875 3876 ASSERT(dsl_pool_config_held(dp)); 3877 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); 3878 3879 if (earlier_txg == 0) 3880 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; 3881 3882 if (later->ds_is_snapshot && 3883 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) 3884 return (B_FALSE); 3885 3886 if (later->ds_dir == earlier->ds_dir) 3887 return (B_TRUE); 3888 if (!dsl_dir_is_clone(later->ds_dir)) 3889 return (B_FALSE); 3890 3891 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object) 3892 return (B_TRUE); 3893 dsl_dataset_t *origin; 3894 error = dsl_dataset_hold_obj(dp, 3895 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); 3896 if (error != 0) 3897 return (B_FALSE); 3898 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 3899 dsl_dataset_rele(origin, FTAG); 3900 return (ret); 3901 } 3902 3903 void 3904 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 3905 { 3906 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3907 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 3908 } 3909 3910 boolean_t 3911 dsl_dataset_is_zapified(dsl_dataset_t *ds) 3912 { 3913 dmu_object_info_t doi; 3914 3915 dmu_object_info_from_db(ds->ds_dbuf, &doi); 3916 return (doi.doi_type == DMU_OTN_ZAP_METADATA); 3917 } 3918 3919 boolean_t 3920 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds) 3921 { 3922 return (dsl_dataset_is_zapified(ds) && 3923 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset, 3924 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0); 3925 } 3926