dsl_dataset.c revision 330590
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 * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org> 25 * Copyright (c) 2011, 2017 by Delphix. All rights reserved. 26 * Copyright (c) 2014, Joyent, Inc. All rights reserved. 27 * Copyright (c) 2014 RackTop Systems. 28 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 29 * Copyright (c) 2014 Integros [integros.com] 30 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved. 31 * Copyright 2017 Nexenta Systems, Inc. 32 */ 33 34#include <sys/dmu_objset.h> 35#include <sys/dsl_dataset.h> 36#include <sys/dsl_dir.h> 37#include <sys/dsl_prop.h> 38#include <sys/dsl_synctask.h> 39#include <sys/dmu_traverse.h> 40#include <sys/dmu_impl.h> 41#include <sys/dmu_send.h> 42#include <sys/dmu_tx.h> 43#include <sys/arc.h> 44#include <sys/zio.h> 45#include <sys/zap.h> 46#include <sys/zfeature.h> 47#include <sys/unique.h> 48#include <sys/zfs_context.h> 49#include <sys/zfs_ioctl.h> 50#include <sys/spa.h> 51#include <sys/zfs_znode.h> 52#include <sys/zfs_onexit.h> 53#include <sys/zvol.h> 54#include <sys/dsl_scan.h> 55#include <sys/dsl_deadlist.h> 56#include <sys/dsl_destroy.h> 57#include <sys/dsl_userhold.h> 58#include <sys/dsl_bookmark.h> 59#include <sys/dmu_send.h> 60#include <sys/zio_checksum.h> 61#include <sys/zio_compress.h> 62#include <zfs_fletcher.h> 63 64SYSCTL_DECL(_vfs_zfs); 65 66/* 67 * The SPA supports block sizes up to 16MB. However, very large blocks 68 * can have an impact on i/o latency (e.g. tying up a spinning disk for 69 * ~300ms), and also potentially on the memory allocator. Therefore, 70 * we do not allow the recordsize to be set larger than zfs_max_recordsize 71 * (default 1MB). Larger blocks can be created by changing this tunable, 72 * and pools with larger blocks can always be imported and used, regardless 73 * of this setting. 74 */ 75int zfs_max_recordsize = 1 * 1024 * 1024; 76SYSCTL_INT(_vfs_zfs, OID_AUTO, max_recordsize, CTLFLAG_RWTUN, 77 &zfs_max_recordsize, 0, 78 "Maximum block size. Expect dragons when tuning this."); 79 80#define SWITCH64(x, y) \ 81 { \ 82 uint64_t __tmp = (x); \ 83 (x) = (y); \ 84 (y) = __tmp; \ 85 } 86 87#define DS_REF_MAX (1ULL << 62) 88 89extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds); 90 91extern int spa_asize_inflation; 92 93static zil_header_t zero_zil; 94 95/* 96 * Figure out how much of this delta should be propogated to the dsl_dir 97 * layer. If there's a refreservation, that space has already been 98 * partially accounted for in our ancestors. 99 */ 100static int64_t 101parent_delta(dsl_dataset_t *ds, int64_t delta) 102{ 103 dsl_dataset_phys_t *ds_phys; 104 uint64_t old_bytes, new_bytes; 105 106 if (ds->ds_reserved == 0) 107 return (delta); 108 109 ds_phys = dsl_dataset_phys(ds); 110 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved); 111 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 112 113 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 114 return (new_bytes - old_bytes); 115} 116 117void 118dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 119{ 120 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 121 int compressed = BP_GET_PSIZE(bp); 122 int uncompressed = BP_GET_UCSIZE(bp); 123 int64_t delta; 124 125 dprintf_bp(bp, "ds=%p", ds); 126 127 ASSERT(dmu_tx_is_syncing(tx)); 128 /* It could have been compressed away to nothing */ 129 if (BP_IS_HOLE(bp)) 130 return; 131 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 132 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 133 if (ds == NULL) { 134 dsl_pool_mos_diduse_space(tx->tx_pool, 135 used, compressed, uncompressed); 136 return; 137 } 138 139 ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 140 dmu_buf_will_dirty(ds->ds_dbuf, tx); 141 mutex_enter(&ds->ds_lock); 142 delta = parent_delta(ds, used); 143 dsl_dataset_phys(ds)->ds_referenced_bytes += used; 144 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed; 145 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed; 146 dsl_dataset_phys(ds)->ds_unique_bytes += used; 147 148 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) { 149 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] = 150 B_TRUE; 151 } 152 153 spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp)); 154 if (f != SPA_FEATURE_NONE) 155 ds->ds_feature_activation_needed[f] = B_TRUE; 156 157 mutex_exit(&ds->ds_lock); 158 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 159 compressed, uncompressed, tx); 160 dsl_dir_transfer_space(ds->ds_dir, used - delta, 161 DD_USED_REFRSRV, DD_USED_HEAD, NULL); 162} 163 164int 165dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 166 boolean_t async) 167{ 168 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 169 int compressed = BP_GET_PSIZE(bp); 170 int uncompressed = BP_GET_UCSIZE(bp); 171 172 if (BP_IS_HOLE(bp)) 173 return (0); 174 175 ASSERT(dmu_tx_is_syncing(tx)); 176 ASSERT(bp->blk_birth <= tx->tx_txg); 177 178 if (ds == NULL) { 179 dsl_free(tx->tx_pool, tx->tx_txg, bp); 180 dsl_pool_mos_diduse_space(tx->tx_pool, 181 -used, -compressed, -uncompressed); 182 return (used); 183 } 184 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 185 186 ASSERT(!ds->ds_is_snapshot); 187 dmu_buf_will_dirty(ds->ds_dbuf, tx); 188 189 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 190 int64_t delta; 191 192 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 193 dsl_free(tx->tx_pool, tx->tx_txg, bp); 194 195 mutex_enter(&ds->ds_lock); 196 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used || 197 !DS_UNIQUE_IS_ACCURATE(ds)); 198 delta = parent_delta(ds, -used); 199 dsl_dataset_phys(ds)->ds_unique_bytes -= used; 200 mutex_exit(&ds->ds_lock); 201 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 202 delta, -compressed, -uncompressed, tx); 203 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 204 DD_USED_REFRSRV, DD_USED_HEAD, NULL); 205 } else { 206 dprintf_bp(bp, "putting on dead list: %s", ""); 207 if (async) { 208 /* 209 * We are here as part of zio's write done callback, 210 * which means we're a zio interrupt thread. We can't 211 * call dsl_deadlist_insert() now because it may block 212 * waiting for I/O. Instead, put bp on the deferred 213 * queue and let dsl_pool_sync() finish the job. 214 */ 215 bplist_append(&ds->ds_pending_deadlist, bp); 216 } else { 217 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); 218 } 219 ASSERT3U(ds->ds_prev->ds_object, ==, 220 dsl_dataset_phys(ds)->ds_prev_snap_obj); 221 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0); 222 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 223 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 224 ds->ds_object && bp->blk_birth > 225 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) { 226 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 227 mutex_enter(&ds->ds_prev->ds_lock); 228 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used; 229 mutex_exit(&ds->ds_prev->ds_lock); 230 } 231 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 232 dsl_dir_transfer_space(ds->ds_dir, used, 233 DD_USED_HEAD, DD_USED_SNAP, tx); 234 } 235 } 236 mutex_enter(&ds->ds_lock); 237 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used); 238 dsl_dataset_phys(ds)->ds_referenced_bytes -= used; 239 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed); 240 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed; 241 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed); 242 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed; 243 mutex_exit(&ds->ds_lock); 244 245 return (used); 246} 247 248/* 249 * We have to release the fsid syncronously or we risk that a subsequent 250 * mount of the same dataset will fail to unique_insert the fsid. This 251 * failure would manifest itself as the fsid of this dataset changing 252 * between mounts which makes NFS clients quite unhappy. 253 */ 254static void 255dsl_dataset_evict_sync(void *dbu) 256{ 257 dsl_dataset_t *ds = dbu; 258 259 ASSERT(ds->ds_owner == NULL); 260 261 unique_remove(ds->ds_fsid_guid); 262} 263 264static void 265dsl_dataset_evict_async(void *dbu) 266{ 267 dsl_dataset_t *ds = dbu; 268 269 ASSERT(ds->ds_owner == NULL); 270 271 ds->ds_dbuf = NULL; 272 273 if (ds->ds_objset != NULL) 274 dmu_objset_evict(ds->ds_objset); 275 276 if (ds->ds_prev) { 277 dsl_dataset_rele(ds->ds_prev, ds); 278 ds->ds_prev = NULL; 279 } 280 281 bplist_destroy(&ds->ds_pending_deadlist); 282 if (ds->ds_deadlist.dl_os != NULL) 283 dsl_deadlist_close(&ds->ds_deadlist); 284 if (ds->ds_dir) 285 dsl_dir_async_rele(ds->ds_dir, ds); 286 287 ASSERT(!list_link_active(&ds->ds_synced_link)); 288 289 list_destroy(&ds->ds_prop_cbs); 290 if (mutex_owned(&ds->ds_lock)) 291 mutex_exit(&ds->ds_lock); 292 mutex_destroy(&ds->ds_lock); 293 if (mutex_owned(&ds->ds_opening_lock)) 294 mutex_exit(&ds->ds_opening_lock); 295 mutex_destroy(&ds->ds_opening_lock); 296 mutex_destroy(&ds->ds_sendstream_lock); 297 refcount_destroy(&ds->ds_longholds); 298 rrw_destroy(&ds->ds_bp_rwlock); 299 300 kmem_free(ds, sizeof (dsl_dataset_t)); 301} 302 303int 304dsl_dataset_get_snapname(dsl_dataset_t *ds) 305{ 306 dsl_dataset_phys_t *headphys; 307 int err; 308 dmu_buf_t *headdbuf; 309 dsl_pool_t *dp = ds->ds_dir->dd_pool; 310 objset_t *mos = dp->dp_meta_objset; 311 312 if (ds->ds_snapname[0]) 313 return (0); 314 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) 315 return (0); 316 317 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, 318 FTAG, &headdbuf); 319 if (err != 0) 320 return (err); 321 headphys = headdbuf->db_data; 322 err = zap_value_search(dp->dp_meta_objset, 323 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 324 dmu_buf_rele(headdbuf, FTAG); 325 return (err); 326} 327 328int 329dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 330{ 331 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 332 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 333 matchtype_t mt = 0; 334 int err; 335 336 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 337 mt = MT_NORMALIZE; 338 339 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 340 value, mt, NULL, 0, NULL); 341 if (err == ENOTSUP && (mt & MT_NORMALIZE)) 342 err = zap_lookup(mos, snapobj, name, 8, 1, value); 343 return (err); 344} 345 346int 347dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 348 boolean_t adj_cnt) 349{ 350 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 351 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 352 matchtype_t mt = 0; 353 int err; 354 355 dsl_dir_snap_cmtime_update(ds->ds_dir); 356 357 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 358 mt = MT_NORMALIZE; 359 360 err = zap_remove_norm(mos, snapobj, name, mt, tx); 361 if (err == ENOTSUP && (mt & MT_NORMALIZE)) 362 err = zap_remove(mos, snapobj, name, tx); 363 364 if (err == 0 && adj_cnt) 365 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 366 DD_FIELD_SNAPSHOT_COUNT, tx); 367 368 return (err); 369} 370 371boolean_t 372dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag) 373{ 374 dmu_buf_t *dbuf = ds->ds_dbuf; 375 boolean_t result = B_FALSE; 376 377 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset, 378 ds->ds_object, DMU_BONUS_BLKID, tag)) { 379 380 if (ds == dmu_buf_get_user(dbuf)) 381 result = B_TRUE; 382 else 383 dmu_buf_rele(dbuf, tag); 384 } 385 386 return (result); 387} 388 389int 390dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 391 dsl_dataset_t **dsp) 392{ 393 objset_t *mos = dp->dp_meta_objset; 394 dmu_buf_t *dbuf; 395 dsl_dataset_t *ds; 396 int err; 397 dmu_object_info_t doi; 398 399 ASSERT(dsl_pool_config_held(dp)); 400 401 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 402 if (err != 0) 403 return (err); 404 405 /* Make sure dsobj has the correct object type. */ 406 dmu_object_info_from_db(dbuf, &doi); 407 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 408 dmu_buf_rele(dbuf, tag); 409 return (SET_ERROR(EINVAL)); 410 } 411 412 ds = dmu_buf_get_user(dbuf); 413 if (ds == NULL) { 414 dsl_dataset_t *winner = NULL; 415 416 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 417 ds->ds_dbuf = dbuf; 418 ds->ds_object = dsobj; 419 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0; 420 421 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 422 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 423 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 424 rrw_init(&ds->ds_bp_rwlock, B_FALSE); 425 refcount_create(&ds->ds_longholds); 426 427 bplist_create(&ds->ds_pending_deadlist); 428 dsl_deadlist_open(&ds->ds_deadlist, 429 mos, dsl_dataset_phys(ds)->ds_deadlist_obj); 430 431 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 432 offsetof(dmu_sendarg_t, dsa_link)); 433 434 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t), 435 offsetof(dsl_prop_cb_record_t, cbr_ds_node)); 436 437 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 438 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 439 if (!(spa_feature_table[f].fi_flags & 440 ZFEATURE_FLAG_PER_DATASET)) 441 continue; 442 err = zap_contains(mos, dsobj, 443 spa_feature_table[f].fi_guid); 444 if (err == 0) { 445 ds->ds_feature_inuse[f] = B_TRUE; 446 } else { 447 ASSERT3U(err, ==, ENOENT); 448 err = 0; 449 } 450 } 451 } 452 453 err = dsl_dir_hold_obj(dp, 454 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir); 455 if (err != 0) { 456 mutex_destroy(&ds->ds_lock); 457 mutex_destroy(&ds->ds_opening_lock); 458 mutex_destroy(&ds->ds_sendstream_lock); 459 refcount_destroy(&ds->ds_longholds); 460 bplist_destroy(&ds->ds_pending_deadlist); 461 dsl_deadlist_close(&ds->ds_deadlist); 462 kmem_free(ds, sizeof (dsl_dataset_t)); 463 dmu_buf_rele(dbuf, tag); 464 return (err); 465 } 466 467 if (!ds->ds_is_snapshot) { 468 ds->ds_snapname[0] = '\0'; 469 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 470 err = dsl_dataset_hold_obj(dp, 471 dsl_dataset_phys(ds)->ds_prev_snap_obj, 472 ds, &ds->ds_prev); 473 } 474 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 475 int zaperr = zap_lookup(mos, ds->ds_object, 476 DS_FIELD_BOOKMARK_NAMES, 477 sizeof (ds->ds_bookmarks), 1, 478 &ds->ds_bookmarks); 479 if (zaperr != ENOENT) 480 VERIFY0(zaperr); 481 } 482 } else { 483 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 484 err = dsl_dataset_get_snapname(ds); 485 if (err == 0 && 486 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { 487 err = zap_count( 488 ds->ds_dir->dd_pool->dp_meta_objset, 489 dsl_dataset_phys(ds)->ds_userrefs_obj, 490 &ds->ds_userrefs); 491 } 492 } 493 494 if (err == 0 && !ds->ds_is_snapshot) { 495 err = dsl_prop_get_int_ds(ds, 496 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 497 &ds->ds_reserved); 498 if (err == 0) { 499 err = dsl_prop_get_int_ds(ds, 500 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 501 &ds->ds_quota); 502 } 503 } else { 504 ds->ds_reserved = ds->ds_quota = 0; 505 } 506 507 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync, 508 dsl_dataset_evict_async, &ds->ds_dbuf); 509 if (err == 0) 510 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu); 511 512 if (err != 0 || winner != NULL) { 513 bplist_destroy(&ds->ds_pending_deadlist); 514 dsl_deadlist_close(&ds->ds_deadlist); 515 if (ds->ds_prev) 516 dsl_dataset_rele(ds->ds_prev, ds); 517 dsl_dir_rele(ds->ds_dir, ds); 518 mutex_destroy(&ds->ds_lock); 519 mutex_destroy(&ds->ds_opening_lock); 520 mutex_destroy(&ds->ds_sendstream_lock); 521 refcount_destroy(&ds->ds_longholds); 522 kmem_free(ds, sizeof (dsl_dataset_t)); 523 if (err != 0) { 524 dmu_buf_rele(dbuf, tag); 525 return (err); 526 } 527 ds = winner; 528 } else { 529 ds->ds_fsid_guid = 530 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); 531 if (ds->ds_fsid_guid != 532 dsl_dataset_phys(ds)->ds_fsid_guid) { 533 zfs_dbgmsg("ds_fsid_guid changed from " 534 "%llx to %llx for pool %s dataset id %llu", 535 (long long) 536 dsl_dataset_phys(ds)->ds_fsid_guid, 537 (long long)ds->ds_fsid_guid, 538 spa_name(dp->dp_spa), 539 dsobj); 540 } 541 } 542 } 543 ASSERT3P(ds->ds_dbuf, ==, dbuf); 544 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); 545 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || 546 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 547 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 548 *dsp = ds; 549 return (0); 550} 551 552int 553dsl_dataset_hold(dsl_pool_t *dp, const char *name, 554 void *tag, dsl_dataset_t **dsp) 555{ 556 dsl_dir_t *dd; 557 const char *snapname; 558 uint64_t obj; 559 int err = 0; 560 dsl_dataset_t *ds; 561 562 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 563 if (err != 0) 564 return (err); 565 566 ASSERT(dsl_pool_config_held(dp)); 567 obj = dsl_dir_phys(dd)->dd_head_dataset_obj; 568 if (obj != 0) 569 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 570 else 571 err = SET_ERROR(ENOENT); 572 573 /* we may be looking for a snapshot */ 574 if (err == 0 && snapname != NULL) { 575 dsl_dataset_t *snap_ds; 576 577 if (*snapname++ != '@') { 578 dsl_dataset_rele(ds, tag); 579 dsl_dir_rele(dd, FTAG); 580 return (SET_ERROR(ENOENT)); 581 } 582 583 dprintf("looking for snapshot '%s'\n", snapname); 584 err = dsl_dataset_snap_lookup(ds, snapname, &obj); 585 if (err == 0) 586 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds); 587 dsl_dataset_rele(ds, tag); 588 589 if (err == 0) { 590 mutex_enter(&snap_ds->ds_lock); 591 if (snap_ds->ds_snapname[0] == 0) 592 (void) strlcpy(snap_ds->ds_snapname, snapname, 593 sizeof (snap_ds->ds_snapname)); 594 mutex_exit(&snap_ds->ds_lock); 595 ds = snap_ds; 596 } 597 } 598 if (err == 0) 599 *dsp = ds; 600 dsl_dir_rele(dd, FTAG); 601 return (err); 602} 603 604int 605dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 606 void *tag, dsl_dataset_t **dsp) 607{ 608 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 609 if (err != 0) 610 return (err); 611 if (!dsl_dataset_tryown(*dsp, tag)) { 612 dsl_dataset_rele(*dsp, tag); 613 *dsp = NULL; 614 return (SET_ERROR(EBUSY)); 615 } 616 return (0); 617} 618 619int 620dsl_dataset_own(dsl_pool_t *dp, const char *name, 621 void *tag, dsl_dataset_t **dsp) 622{ 623 int err = dsl_dataset_hold(dp, name, tag, dsp); 624 if (err != 0) 625 return (err); 626 if (!dsl_dataset_tryown(*dsp, tag)) { 627 dsl_dataset_rele(*dsp, tag); 628 return (SET_ERROR(EBUSY)); 629 } 630 return (0); 631} 632 633/* 634 * See the comment above dsl_pool_hold() for details. In summary, a long 635 * hold is used to prevent destruction of a dataset while the pool hold 636 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 637 * 638 * The dataset and pool must be held when this function is called. After it 639 * is called, the pool hold may be released while the dataset is still held 640 * and accessed. 641 */ 642void 643dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 644{ 645 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 646 (void) refcount_add(&ds->ds_longholds, tag); 647} 648 649void 650dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 651{ 652 (void) refcount_remove(&ds->ds_longholds, tag); 653} 654 655/* Return B_TRUE if there are any long holds on this dataset. */ 656boolean_t 657dsl_dataset_long_held(dsl_dataset_t *ds) 658{ 659 return (!refcount_is_zero(&ds->ds_longholds)); 660} 661 662void 663dsl_dataset_name(dsl_dataset_t *ds, char *name) 664{ 665 if (ds == NULL) { 666 (void) strcpy(name, "mos"); 667 } else { 668 dsl_dir_name(ds->ds_dir, name); 669 VERIFY0(dsl_dataset_get_snapname(ds)); 670 if (ds->ds_snapname[0]) { 671 VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN), 672 <, ZFS_MAX_DATASET_NAME_LEN); 673 /* 674 * We use a "recursive" mutex so that we 675 * can call dprintf_ds() with ds_lock held. 676 */ 677 if (!MUTEX_HELD(&ds->ds_lock)) { 678 mutex_enter(&ds->ds_lock); 679 VERIFY3U(strlcat(name, ds->ds_snapname, 680 ZFS_MAX_DATASET_NAME_LEN), <, 681 ZFS_MAX_DATASET_NAME_LEN); 682 mutex_exit(&ds->ds_lock); 683 } else { 684 VERIFY3U(strlcat(name, ds->ds_snapname, 685 ZFS_MAX_DATASET_NAME_LEN), <, 686 ZFS_MAX_DATASET_NAME_LEN); 687 } 688 } 689 } 690} 691 692int 693dsl_dataset_namelen(dsl_dataset_t *ds) 694{ 695 VERIFY0(dsl_dataset_get_snapname(ds)); 696 mutex_enter(&ds->ds_lock); 697 int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname); 698 mutex_exit(&ds->ds_lock); 699 return (len); 700} 701 702void 703dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 704{ 705 dmu_buf_rele(ds->ds_dbuf, tag); 706} 707 708void 709dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 710{ 711 ASSERT3P(ds->ds_owner, ==, tag); 712 ASSERT(ds->ds_dbuf != NULL); 713 714 mutex_enter(&ds->ds_lock); 715 ds->ds_owner = NULL; 716 mutex_exit(&ds->ds_lock); 717 dsl_dataset_long_rele(ds, tag); 718 dsl_dataset_rele(ds, tag); 719} 720 721boolean_t 722dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 723{ 724 boolean_t gotit = FALSE; 725 726 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 727 mutex_enter(&ds->ds_lock); 728 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 729 ds->ds_owner = tag; 730 dsl_dataset_long_hold(ds, tag); 731 gotit = TRUE; 732 } 733 mutex_exit(&ds->ds_lock); 734 return (gotit); 735} 736 737boolean_t 738dsl_dataset_has_owner(dsl_dataset_t *ds) 739{ 740 boolean_t rv; 741 mutex_enter(&ds->ds_lock); 742 rv = (ds->ds_owner != NULL); 743 mutex_exit(&ds->ds_lock); 744 return (rv); 745} 746 747static void 748dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx) 749{ 750 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 751 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 752 uint64_t zero = 0; 753 754 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 755 756 spa_feature_incr(spa, f, tx); 757 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 758 759 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, 760 sizeof (zero), 1, &zero, tx)); 761} 762 763void 764dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx) 765{ 766 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 767 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 768 769 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 770 771 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx)); 772 spa_feature_decr(spa, f, tx); 773} 774 775uint64_t 776dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 777 uint64_t flags, dmu_tx_t *tx) 778{ 779 dsl_pool_t *dp = dd->dd_pool; 780 dmu_buf_t *dbuf; 781 dsl_dataset_phys_t *dsphys; 782 uint64_t dsobj; 783 objset_t *mos = dp->dp_meta_objset; 784 785 if (origin == NULL) 786 origin = dp->dp_origin_snap; 787 788 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 789 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); 790 ASSERT(dmu_tx_is_syncing(tx)); 791 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 792 793 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 794 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 795 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 796 dmu_buf_will_dirty(dbuf, tx); 797 dsphys = dbuf->db_data; 798 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 799 dsphys->ds_dir_obj = dd->dd_object; 800 dsphys->ds_flags = flags; 801 dsphys->ds_fsid_guid = unique_create(); 802 do { 803 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 804 sizeof (dsphys->ds_guid)); 805 } while (dsphys->ds_guid == 0); 806 dsphys->ds_snapnames_zapobj = 807 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 808 DMU_OT_NONE, 0, tx); 809 dsphys->ds_creation_time = gethrestime_sec(); 810 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 811 812 if (origin == NULL) { 813 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 814 } else { 815 dsl_dataset_t *ohds; /* head of the origin snapshot */ 816 817 dsphys->ds_prev_snap_obj = origin->ds_object; 818 dsphys->ds_prev_snap_txg = 819 dsl_dataset_phys(origin)->ds_creation_txg; 820 dsphys->ds_referenced_bytes = 821 dsl_dataset_phys(origin)->ds_referenced_bytes; 822 dsphys->ds_compressed_bytes = 823 dsl_dataset_phys(origin)->ds_compressed_bytes; 824 dsphys->ds_uncompressed_bytes = 825 dsl_dataset_phys(origin)->ds_uncompressed_bytes; 826 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG); 827 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; 828 rrw_exit(&origin->ds_bp_rwlock, FTAG); 829 830 /* 831 * Inherit flags that describe the dataset's contents 832 * (INCONSISTENT) or properties (Case Insensitive). 833 */ 834 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & 835 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 836 837 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 838 if (origin->ds_feature_inuse[f]) 839 dsl_dataset_activate_feature(dsobj, f, tx); 840 } 841 842 dmu_buf_will_dirty(origin->ds_dbuf, tx); 843 dsl_dataset_phys(origin)->ds_num_children++; 844 845 VERIFY0(dsl_dataset_hold_obj(dp, 846 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, 847 FTAG, &ohds)); 848 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 849 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 850 dsl_dataset_rele(ohds, FTAG); 851 852 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 853 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { 854 dsl_dataset_phys(origin)->ds_next_clones_obj = 855 zap_create(mos, 856 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 857 } 858 VERIFY0(zap_add_int(mos, 859 dsl_dataset_phys(origin)->ds_next_clones_obj, 860 dsobj, tx)); 861 } 862 863 dmu_buf_will_dirty(dd->dd_dbuf, tx); 864 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; 865 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 866 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 867 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 868 dsl_dir_phys(origin->ds_dir)->dd_clones = 869 zap_create(mos, 870 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 871 } 872 VERIFY0(zap_add_int(mos, 873 dsl_dir_phys(origin->ds_dir)->dd_clones, 874 dsobj, tx)); 875 } 876 } 877 878 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 879 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 880 881 dmu_buf_rele(dbuf, FTAG); 882 883 dmu_buf_will_dirty(dd->dd_dbuf, tx); 884 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; 885 886 return (dsobj); 887} 888 889static void 890dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 891{ 892 objset_t *os; 893 894 VERIFY0(dmu_objset_from_ds(ds, &os)); 895 if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) { 896 dsl_pool_t *dp = ds->ds_dir->dd_pool; 897 zio_t *zio; 898 899 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 900 901 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 902 dsl_dataset_sync(ds, zio, tx); 903 VERIFY0(zio_wait(zio)); 904 905 /* dsl_dataset_sync_done will drop this reference. */ 906 dmu_buf_add_ref(ds->ds_dbuf, ds); 907 dsl_dataset_sync_done(ds, tx); 908 } 909} 910 911uint64_t 912dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 913 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 914{ 915 dsl_pool_t *dp = pdd->dd_pool; 916 uint64_t dsobj, ddobj; 917 dsl_dir_t *dd; 918 919 ASSERT(dmu_tx_is_syncing(tx)); 920 ASSERT(lastname[0] != '@'); 921 922 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 923 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 924 925 dsobj = dsl_dataset_create_sync_dd(dd, origin, 926 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 927 928 dsl_deleg_set_create_perms(dd, tx, cr); 929 930 /* 931 * Since we're creating a new node we know it's a leaf, so we can 932 * initialize the counts if the limit feature is active. 933 */ 934 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 935 uint64_t cnt = 0; 936 objset_t *os = dd->dd_pool->dp_meta_objset; 937 938 dsl_dir_zapify(dd, tx); 939 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 940 sizeof (cnt), 1, &cnt, tx)); 941 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 942 sizeof (cnt), 1, &cnt, tx)); 943 } 944 945 dsl_dir_rele(dd, FTAG); 946 947 /* 948 * If we are creating a clone, make sure we zero out any stale 949 * data from the origin snapshots zil header. 950 */ 951 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 952 dsl_dataset_t *ds; 953 954 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 955 dsl_dataset_zero_zil(ds, tx); 956 dsl_dataset_rele(ds, FTAG); 957 } 958 959 return (dsobj); 960} 961 962#ifdef __FreeBSD__ 963/* FreeBSD ioctl compat begin */ 964struct destroyarg { 965 nvlist_t *nvl; 966 const char *snapname; 967}; 968 969static int 970dsl_check_snap_cb(const char *name, void *arg) 971{ 972 struct destroyarg *da = arg; 973 dsl_dataset_t *ds; 974 char *dsname; 975 976 dsname = kmem_asprintf("%s@%s", name, da->snapname); 977 fnvlist_add_boolean(da->nvl, dsname); 978 kmem_free(dsname, strlen(dsname) + 1); 979 980 return (0); 981} 982 983int 984dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname, 985 nvlist_t *snaps) 986{ 987 struct destroyarg *da; 988 int err; 989 990 da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP); 991 da->nvl = snaps; 992 da->snapname = snapname; 993 err = dmu_objset_find(fsname, dsl_check_snap_cb, da, 994 DS_FIND_CHILDREN); 995 kmem_free(da, sizeof (struct destroyarg)); 996 997 return (err); 998} 999/* FreeBSD ioctl compat end */ 1000#endif /* __FreeBSD__ */ 1001 1002/* 1003 * The unique space in the head dataset can be calculated by subtracting 1004 * the space used in the most recent snapshot, that is still being used 1005 * in this file system, from the space currently in use. To figure out 1006 * the space in the most recent snapshot still in use, we need to take 1007 * the total space used in the snapshot and subtract out the space that 1008 * has been freed up since the snapshot was taken. 1009 */ 1010void 1011dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 1012{ 1013 uint64_t mrs_used; 1014 uint64_t dlused, dlcomp, dluncomp; 1015 1016 ASSERT(!ds->ds_is_snapshot); 1017 1018 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) 1019 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; 1020 else 1021 mrs_used = 0; 1022 1023 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 1024 1025 ASSERT3U(dlused, <=, mrs_used); 1026 dsl_dataset_phys(ds)->ds_unique_bytes = 1027 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); 1028 1029 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1030 SPA_VERSION_UNIQUE_ACCURATE) 1031 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1032} 1033 1034void 1035dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 1036 dmu_tx_t *tx) 1037{ 1038 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1039 uint64_t count; 1040 int err; 1041 1042 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); 1043 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1044 obj, tx); 1045 /* 1046 * The err should not be ENOENT, but a bug in a previous version 1047 * of the code could cause upgrade_clones_cb() to not set 1048 * ds_next_snap_obj when it should, leading to a missing entry. 1049 * If we knew that the pool was created after 1050 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 1051 * ENOENT. However, at least we can check that we don't have 1052 * too many entries in the next_clones_obj even after failing to 1053 * remove this one. 1054 */ 1055 if (err != ENOENT) 1056 VERIFY0(err); 1057 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1058 &count)); 1059 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); 1060} 1061 1062 1063blkptr_t * 1064dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1065{ 1066 return (&dsl_dataset_phys(ds)->ds_bp); 1067} 1068 1069spa_t * 1070dsl_dataset_get_spa(dsl_dataset_t *ds) 1071{ 1072 return (ds->ds_dir->dd_pool->dp_spa); 1073} 1074 1075void 1076dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1077{ 1078 dsl_pool_t *dp; 1079 1080 if (ds == NULL) /* this is the meta-objset */ 1081 return; 1082 1083 ASSERT(ds->ds_objset != NULL); 1084 1085 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) 1086 panic("dirtying snapshot!"); 1087 1088 /* Must not dirty a dataset in the same txg where it got snapshotted. */ 1089 ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg); 1090 1091 dp = ds->ds_dir->dd_pool; 1092 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1093 /* up the hold count until we can be written out */ 1094 dmu_buf_add_ref(ds->ds_dbuf, ds); 1095 } 1096} 1097 1098boolean_t 1099dsl_dataset_is_dirty(dsl_dataset_t *ds) 1100{ 1101 for (int t = 0; t < TXG_SIZE; t++) { 1102 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1103 ds, t)) 1104 return (B_TRUE); 1105 } 1106 return (B_FALSE); 1107} 1108 1109static int 1110dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1111{ 1112 uint64_t asize; 1113 1114 if (!dmu_tx_is_syncing(tx)) 1115 return (0); 1116 1117 /* 1118 * If there's an fs-only reservation, any blocks that might become 1119 * owned by the snapshot dataset must be accommodated by space 1120 * outside of the reservation. 1121 */ 1122 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1123 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); 1124 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1125 return (SET_ERROR(ENOSPC)); 1126 1127 /* 1128 * Propagate any reserved space for this snapshot to other 1129 * snapshot checks in this sync group. 1130 */ 1131 if (asize > 0) 1132 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1133 1134 return (0); 1135} 1136 1137int 1138dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1139 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 1140{ 1141 int error; 1142 uint64_t value; 1143 1144 ds->ds_trysnap_txg = tx->tx_txg; 1145 1146 if (!dmu_tx_is_syncing(tx)) 1147 return (0); 1148 1149 /* 1150 * We don't allow multiple snapshots of the same txg. If there 1151 * is already one, try again. 1152 */ 1153 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) 1154 return (SET_ERROR(EAGAIN)); 1155 1156 /* 1157 * Check for conflicting snapshot name. 1158 */ 1159 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1160 if (error == 0) 1161 return (SET_ERROR(EEXIST)); 1162 if (error != ENOENT) 1163 return (error); 1164 1165 /* 1166 * We don't allow taking snapshots of inconsistent datasets, such as 1167 * those into which we are currently receiving. However, if we are 1168 * creating this snapshot as part of a receive, this check will be 1169 * executed atomically with respect to the completion of the receive 1170 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1171 * case we ignore this, knowing it will be fixed up for us shortly in 1172 * dmu_recv_end_sync(). 1173 */ 1174 if (!recv && DS_IS_INCONSISTENT(ds)) 1175 return (SET_ERROR(EBUSY)); 1176 1177 /* 1178 * Skip the check for temporary snapshots or if we have already checked 1179 * the counts in dsl_dataset_snapshot_check. This means we really only 1180 * check the count here when we're receiving a stream. 1181 */ 1182 if (cnt != 0 && cr != NULL) { 1183 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1184 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 1185 if (error != 0) 1186 return (error); 1187 } 1188 1189 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1190 if (error != 0) 1191 return (error); 1192 1193 return (0); 1194} 1195 1196int 1197dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1198{ 1199 dsl_dataset_snapshot_arg_t *ddsa = arg; 1200 dsl_pool_t *dp = dmu_tx_pool(tx); 1201 nvpair_t *pair; 1202 int rv = 0; 1203 1204 /* 1205 * Pre-compute how many total new snapshots will be created for each 1206 * level in the tree and below. This is needed for validating the 1207 * snapshot limit when either taking a recursive snapshot or when 1208 * taking multiple snapshots. 1209 * 1210 * The problem is that the counts are not actually adjusted when 1211 * we are checking, only when we finally sync. For a single snapshot, 1212 * this is easy, the count will increase by 1 at each node up the tree, 1213 * but its more complicated for the recursive/multiple snapshot case. 1214 * 1215 * The dsl_fs_ss_limit_check function does recursively check the count 1216 * at each level up the tree but since it is validating each snapshot 1217 * independently we need to be sure that we are validating the complete 1218 * count for the entire set of snapshots. We do this by rolling up the 1219 * counts for each component of the name into an nvlist and then 1220 * checking each of those cases with the aggregated count. 1221 * 1222 * This approach properly handles not only the recursive snapshot 1223 * case (where we get all of those on the ddsa_snaps list) but also 1224 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1225 * validate the limit on 'a' using a count of 2). 1226 * 1227 * We validate the snapshot names in the third loop and only report 1228 * name errors once. 1229 */ 1230 if (dmu_tx_is_syncing(tx)) { 1231 nvlist_t *cnt_track = NULL; 1232 cnt_track = fnvlist_alloc(); 1233 1234 /* Rollup aggregated counts into the cnt_track list */ 1235 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1236 pair != NULL; 1237 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1238 char *pdelim; 1239 uint64_t val; 1240 char nm[MAXPATHLEN]; 1241 1242 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1243 pdelim = strchr(nm, '@'); 1244 if (pdelim == NULL) 1245 continue; 1246 *pdelim = '\0'; 1247 1248 do { 1249 if (nvlist_lookup_uint64(cnt_track, nm, 1250 &val) == 0) { 1251 /* update existing entry */ 1252 fnvlist_add_uint64(cnt_track, nm, 1253 val + 1); 1254 } else { 1255 /* add to list */ 1256 fnvlist_add_uint64(cnt_track, nm, 1); 1257 } 1258 1259 pdelim = strrchr(nm, '/'); 1260 if (pdelim != NULL) 1261 *pdelim = '\0'; 1262 } while (pdelim != NULL); 1263 } 1264 1265 /* Check aggregated counts at each level */ 1266 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1267 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1268 int error = 0; 1269 char *name; 1270 uint64_t cnt = 0; 1271 dsl_dataset_t *ds; 1272 1273 name = nvpair_name(pair); 1274 cnt = fnvpair_value_uint64(pair); 1275 ASSERT(cnt > 0); 1276 1277 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1278 if (error == 0) { 1279 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1280 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1281 ddsa->ddsa_cr); 1282 dsl_dataset_rele(ds, FTAG); 1283 } 1284 1285 if (error != 0) { 1286 if (ddsa->ddsa_errors != NULL) 1287 fnvlist_add_int32(ddsa->ddsa_errors, 1288 name, error); 1289 rv = error; 1290 /* only report one error for this check */ 1291 break; 1292 } 1293 } 1294 nvlist_free(cnt_track); 1295 } 1296 1297 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1298 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1299 int error = 0; 1300 dsl_dataset_t *ds; 1301 char *name, *atp; 1302 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1303 1304 name = nvpair_name(pair); 1305 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN) 1306 error = SET_ERROR(ENAMETOOLONG); 1307 if (error == 0) { 1308 atp = strchr(name, '@'); 1309 if (atp == NULL) 1310 error = SET_ERROR(EINVAL); 1311 if (error == 0) 1312 (void) strlcpy(dsname, name, atp - name + 1); 1313 } 1314 if (error == 0) 1315 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1316 if (error == 0) { 1317 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1318 error = dsl_dataset_snapshot_check_impl(ds, 1319 atp + 1, tx, B_FALSE, 0, NULL); 1320 dsl_dataset_rele(ds, FTAG); 1321 } 1322 1323 if (error != 0) { 1324 if (ddsa->ddsa_errors != NULL) { 1325 fnvlist_add_int32(ddsa->ddsa_errors, 1326 name, error); 1327 } 1328 rv = error; 1329 } 1330 } 1331 1332 return (rv); 1333} 1334 1335void 1336dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1337 dmu_tx_t *tx) 1338{ 1339 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1340 dmu_buf_t *dbuf; 1341 dsl_dataset_phys_t *dsphys; 1342 uint64_t dsobj, crtxg; 1343 objset_t *mos = dp->dp_meta_objset; 1344 objset_t *os; 1345 1346 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1347 1348 /* 1349 * If we are on an old pool, the zil must not be active, in which 1350 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1351 */ 1352 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1353 dmu_objset_from_ds(ds, &os) != 0 || 1354 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1355 sizeof (zero_zil)) == 0); 1356 1357 /* Should not snapshot a dirty dataset. */ 1358 ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1359 ds, tx->tx_txg)); 1360 1361 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1362 1363 /* 1364 * The origin's ds_creation_txg has to be < TXG_INITIAL 1365 */ 1366 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1367 crtxg = 1; 1368 else 1369 crtxg = tx->tx_txg; 1370 1371 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1372 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1373 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1374 dmu_buf_will_dirty(dbuf, tx); 1375 dsphys = dbuf->db_data; 1376 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1377 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1378 dsphys->ds_fsid_guid = unique_create(); 1379 do { 1380 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1381 sizeof (dsphys->ds_guid)); 1382 } while (dsphys->ds_guid == 0); 1383 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1384 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 1385 dsphys->ds_next_snap_obj = ds->ds_object; 1386 dsphys->ds_num_children = 1; 1387 dsphys->ds_creation_time = gethrestime_sec(); 1388 dsphys->ds_creation_txg = crtxg; 1389 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; 1390 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; 1391 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; 1392 dsphys->ds_uncompressed_bytes = 1393 dsl_dataset_phys(ds)->ds_uncompressed_bytes; 1394 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; 1395 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1396 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; 1397 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1398 dmu_buf_rele(dbuf, FTAG); 1399 1400 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1401 if (ds->ds_feature_inuse[f]) 1402 dsl_dataset_activate_feature(dsobj, f, tx); 1403 } 1404 1405 ASSERT3U(ds->ds_prev != 0, ==, 1406 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); 1407 if (ds->ds_prev) { 1408 uint64_t next_clones_obj = 1409 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; 1410 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1411 ds->ds_object || 1412 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); 1413 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1414 ds->ds_object) { 1415 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1416 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, 1417 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); 1418 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; 1419 } else if (next_clones_obj != 0) { 1420 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1421 dsphys->ds_next_snap_obj, tx); 1422 VERIFY0(zap_add_int(mos, 1423 next_clones_obj, dsobj, tx)); 1424 } 1425 } 1426 1427 /* 1428 * If we have a reference-reservation on this dataset, we will 1429 * need to increase the amount of refreservation being charged 1430 * since our unique space is going to zero. 1431 */ 1432 if (ds->ds_reserved) { 1433 int64_t delta; 1434 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1435 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, 1436 ds->ds_reserved); 1437 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1438 delta, 0, 0, tx); 1439 } 1440 1441 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1442 dsl_dataset_phys(ds)->ds_deadlist_obj = 1443 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, 1444 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 1445 dsl_deadlist_close(&ds->ds_deadlist); 1446 dsl_deadlist_open(&ds->ds_deadlist, mos, 1447 dsl_dataset_phys(ds)->ds_deadlist_obj); 1448 dsl_deadlist_add_key(&ds->ds_deadlist, 1449 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); 1450 1451 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); 1452 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; 1453 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; 1454 dsl_dataset_phys(ds)->ds_unique_bytes = 0; 1455 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1456 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1457 1458 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1459 snapname, 8, 1, &dsobj, tx)); 1460 1461 if (ds->ds_prev) 1462 dsl_dataset_rele(ds->ds_prev, ds); 1463 VERIFY0(dsl_dataset_hold_obj(dp, 1464 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); 1465 1466 dsl_scan_ds_snapshotted(ds, tx); 1467 1468 dsl_dir_snap_cmtime_update(ds->ds_dir); 1469 1470 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1471} 1472 1473void 1474dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1475{ 1476 dsl_dataset_snapshot_arg_t *ddsa = arg; 1477 dsl_pool_t *dp = dmu_tx_pool(tx); 1478 nvpair_t *pair; 1479 1480 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1481 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1482 dsl_dataset_t *ds; 1483 char *name, *atp; 1484 char dsname[ZFS_MAX_DATASET_NAME_LEN]; 1485 1486 name = nvpair_name(pair); 1487 atp = strchr(name, '@'); 1488 (void) strlcpy(dsname, name, atp - name + 1); 1489 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1490 1491 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1492 if (ddsa->ddsa_props != NULL) { 1493 dsl_props_set_sync_impl(ds->ds_prev, 1494 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1495 } 1496 dsl_dataset_rele(ds, FTAG); 1497 } 1498} 1499 1500/* 1501 * The snapshots must all be in the same pool. 1502 * All-or-nothing: if there are any failures, nothing will be modified. 1503 */ 1504int 1505dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1506{ 1507 dsl_dataset_snapshot_arg_t ddsa; 1508 nvpair_t *pair; 1509 boolean_t needsuspend; 1510 int error; 1511 spa_t *spa; 1512 char *firstname; 1513 nvlist_t *suspended = NULL; 1514 1515 pair = nvlist_next_nvpair(snaps, NULL); 1516 if (pair == NULL) 1517 return (0); 1518 firstname = nvpair_name(pair); 1519 1520 error = spa_open(firstname, &spa, FTAG); 1521 if (error != 0) 1522 return (error); 1523 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1524 spa_close(spa, FTAG); 1525 1526 if (needsuspend) { 1527 suspended = fnvlist_alloc(); 1528 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1529 pair = nvlist_next_nvpair(snaps, pair)) { 1530 char fsname[ZFS_MAX_DATASET_NAME_LEN]; 1531 char *snapname = nvpair_name(pair); 1532 char *atp; 1533 void *cookie; 1534 1535 atp = strchr(snapname, '@'); 1536 if (atp == NULL) { 1537 error = SET_ERROR(EINVAL); 1538 break; 1539 } 1540 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1541 1542 error = zil_suspend(fsname, &cookie); 1543 if (error != 0) 1544 break; 1545 fnvlist_add_uint64(suspended, fsname, 1546 (uintptr_t)cookie); 1547 } 1548 } 1549 1550 ddsa.ddsa_snaps = snaps; 1551 ddsa.ddsa_props = props; 1552 ddsa.ddsa_errors = errors; 1553 ddsa.ddsa_cr = CRED(); 1554 1555 if (error == 0) { 1556 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1557 dsl_dataset_snapshot_sync, &ddsa, 1558 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1559 } 1560 1561 if (suspended != NULL) { 1562 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1563 pair = nvlist_next_nvpair(suspended, pair)) { 1564 zil_resume((void *)(uintptr_t) 1565 fnvpair_value_uint64(pair)); 1566 } 1567 fnvlist_free(suspended); 1568 } 1569 1570#ifdef __FreeBSD__ 1571#ifdef _KERNEL 1572 if (error == 0) { 1573 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1574 pair = nvlist_next_nvpair(snaps, pair)) { 1575 char *snapname = nvpair_name(pair); 1576 zvol_create_minors(snapname); 1577 } 1578 } 1579#endif 1580#endif 1581 return (error); 1582} 1583 1584typedef struct dsl_dataset_snapshot_tmp_arg { 1585 const char *ddsta_fsname; 1586 const char *ddsta_snapname; 1587 minor_t ddsta_cleanup_minor; 1588 const char *ddsta_htag; 1589} dsl_dataset_snapshot_tmp_arg_t; 1590 1591static int 1592dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1593{ 1594 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1595 dsl_pool_t *dp = dmu_tx_pool(tx); 1596 dsl_dataset_t *ds; 1597 int error; 1598 1599 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1600 if (error != 0) 1601 return (error); 1602 1603 /* NULL cred means no limit check for tmp snapshot */ 1604 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1605 tx, B_FALSE, 0, NULL); 1606 if (error != 0) { 1607 dsl_dataset_rele(ds, FTAG); 1608 return (error); 1609 } 1610 1611 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1612 dsl_dataset_rele(ds, FTAG); 1613 return (SET_ERROR(ENOTSUP)); 1614 } 1615 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1616 B_TRUE, tx); 1617 if (error != 0) { 1618 dsl_dataset_rele(ds, FTAG); 1619 return (error); 1620 } 1621 1622 dsl_dataset_rele(ds, FTAG); 1623 return (0); 1624} 1625 1626static void 1627dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1628{ 1629 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1630 dsl_pool_t *dp = dmu_tx_pool(tx); 1631 dsl_dataset_t *ds; 1632 1633 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1634 1635 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1636 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1637 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1638 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1639 1640 dsl_dataset_rele(ds, FTAG); 1641} 1642 1643int 1644dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1645 minor_t cleanup_minor, const char *htag) 1646{ 1647 dsl_dataset_snapshot_tmp_arg_t ddsta; 1648 int error; 1649 spa_t *spa; 1650 boolean_t needsuspend; 1651 void *cookie; 1652 1653 ddsta.ddsta_fsname = fsname; 1654 ddsta.ddsta_snapname = snapname; 1655 ddsta.ddsta_cleanup_minor = cleanup_minor; 1656 ddsta.ddsta_htag = htag; 1657 1658 error = spa_open(fsname, &spa, FTAG); 1659 if (error != 0) 1660 return (error); 1661 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1662 spa_close(spa, FTAG); 1663 1664 if (needsuspend) { 1665 error = zil_suspend(fsname, &cookie); 1666 if (error != 0) 1667 return (error); 1668 } 1669 1670 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1671 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 1672 1673 if (needsuspend) 1674 zil_resume(cookie); 1675 return (error); 1676} 1677 1678void 1679dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1680{ 1681 ASSERT(dmu_tx_is_syncing(tx)); 1682 ASSERT(ds->ds_objset != NULL); 1683 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); 1684 1685 /* 1686 * in case we had to change ds_fsid_guid when we opened it, 1687 * sync it out now. 1688 */ 1689 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1690 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; 1691 1692 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) { 1693 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1694 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1, 1695 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx)); 1696 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1697 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1, 1698 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx)); 1699 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1700 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1, 1701 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx)); 1702 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0; 1703 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0; 1704 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0; 1705 } 1706 1707 dmu_objset_sync(ds->ds_objset, zio, tx); 1708 1709 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1710 if (ds->ds_feature_activation_needed[f]) { 1711 if (ds->ds_feature_inuse[f]) 1712 continue; 1713 dsl_dataset_activate_feature(ds->ds_object, f, tx); 1714 ds->ds_feature_inuse[f] = B_TRUE; 1715 } 1716 } 1717} 1718 1719static int 1720deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 1721{ 1722 dsl_deadlist_t *dl = arg; 1723 dsl_deadlist_insert(dl, bp, tx); 1724 return (0); 1725} 1726 1727void 1728dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx) 1729{ 1730 objset_t *os = ds->ds_objset; 1731 1732 bplist_iterate(&ds->ds_pending_deadlist, 1733 deadlist_enqueue_cb, &ds->ds_deadlist, tx); 1734 1735 if (os->os_synced_dnodes != NULL) { 1736 multilist_destroy(os->os_synced_dnodes); 1737 os->os_synced_dnodes = NULL; 1738 } 1739 1740 ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx))); 1741 1742 dmu_buf_rele(ds->ds_dbuf, ds); 1743} 1744 1745int 1746get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val) 1747{ 1748 uint64_t count = 0; 1749 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1750 zap_cursor_t zc; 1751 zap_attribute_t za; 1752 1753 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1754 1755 /* 1756 * There may be missing entries in ds_next_clones_obj 1757 * due to a bug in a previous version of the code. 1758 * Only trust it if it has the right number of entries. 1759 */ 1760 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { 1761 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1762 &count)); 1763 } 1764 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) { 1765 return (ENOENT); 1766 } 1767 for (zap_cursor_init(&zc, mos, 1768 dsl_dataset_phys(ds)->ds_next_clones_obj); 1769 zap_cursor_retrieve(&zc, &za) == 0; 1770 zap_cursor_advance(&zc)) { 1771 dsl_dataset_t *clone; 1772 char buf[ZFS_MAX_DATASET_NAME_LEN]; 1773 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1774 za.za_first_integer, FTAG, &clone)); 1775 dsl_dir_name(clone->ds_dir, buf); 1776 fnvlist_add_boolean(val, buf); 1777 dsl_dataset_rele(clone, FTAG); 1778 } 1779 zap_cursor_fini(&zc); 1780 return (0); 1781} 1782 1783void 1784get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1785{ 1786 nvlist_t *propval = fnvlist_alloc(); 1787 nvlist_t *val; 1788 1789 /* 1790 * We use nvlist_alloc() instead of fnvlist_alloc() because the 1791 * latter would allocate the list with NV_UNIQUE_NAME flag. 1792 * As a result, every time a clone name is appended to the list 1793 * it would be (linearly) searched for for a duplicate name. 1794 * We already know that all clone names must be unique and we 1795 * want avoid the quadratic complexity of double-checking that 1796 * because we can have a large number of clones. 1797 */ 1798 VERIFY0(nvlist_alloc(&val, 0, KM_SLEEP)); 1799 1800 if (get_clones_stat_impl(ds, val) == 0) { 1801 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1802 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), 1803 propval); 1804 } 1805 1806 nvlist_free(val); 1807 nvlist_free(propval); 1808} 1809 1810/* 1811 * Returns a string that represents the receive resume stats token. It should 1812 * be freed with strfree(). 1813 */ 1814char * 1815get_receive_resume_stats_impl(dsl_dataset_t *ds) 1816{ 1817 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1818 1819 if (dsl_dataset_has_resume_receive_state(ds)) { 1820 char *str; 1821 void *packed; 1822 uint8_t *compressed; 1823 uint64_t val; 1824 nvlist_t *token_nv = fnvlist_alloc(); 1825 size_t packed_size, compressed_size; 1826 1827 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1828 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) { 1829 fnvlist_add_uint64(token_nv, "fromguid", val); 1830 } 1831 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1832 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) { 1833 fnvlist_add_uint64(token_nv, "object", val); 1834 } 1835 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1836 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) { 1837 fnvlist_add_uint64(token_nv, "offset", val); 1838 } 1839 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1840 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) { 1841 fnvlist_add_uint64(token_nv, "bytes", val); 1842 } 1843 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1844 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) { 1845 fnvlist_add_uint64(token_nv, "toguid", val); 1846 } 1847 char buf[256]; 1848 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1849 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) { 1850 fnvlist_add_string(token_nv, "toname", buf); 1851 } 1852 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1853 DS_FIELD_RESUME_LARGEBLOCK) == 0) { 1854 fnvlist_add_boolean(token_nv, "largeblockok"); 1855 } 1856 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1857 DS_FIELD_RESUME_EMBEDOK) == 0) { 1858 fnvlist_add_boolean(token_nv, "embedok"); 1859 } 1860 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1861 DS_FIELD_RESUME_COMPRESSOK) == 0) { 1862 fnvlist_add_boolean(token_nv, "compressok"); 1863 } 1864 packed = fnvlist_pack(token_nv, &packed_size); 1865 fnvlist_free(token_nv); 1866 compressed = kmem_alloc(packed_size, KM_SLEEP); 1867 1868 compressed_size = gzip_compress(packed, compressed, 1869 packed_size, packed_size, 6); 1870 1871 zio_cksum_t cksum; 1872 fletcher_4_native(compressed, compressed_size, NULL, &cksum); 1873 1874 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP); 1875 for (int i = 0; i < compressed_size; i++) { 1876 (void) sprintf(str + i * 2, "%02x", compressed[i]); 1877 } 1878 str[compressed_size * 2] = '\0'; 1879 char *propval = kmem_asprintf("%u-%llx-%llx-%s", 1880 ZFS_SEND_RESUME_TOKEN_VERSION, 1881 (longlong_t)cksum.zc_word[0], 1882 (longlong_t)packed_size, str); 1883 kmem_free(packed, packed_size); 1884 kmem_free(str, compressed_size * 2 + 1); 1885 kmem_free(compressed, packed_size); 1886 return (propval); 1887 } 1888 return (spa_strdup("")); 1889} 1890 1891/* 1892 * Returns a string that represents the receive resume stats token of the 1893 * dataset's child. It should be freed with strfree(). 1894 */ 1895char * 1896get_child_receive_stats(dsl_dataset_t *ds) 1897{ 1898 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 1899 dsl_dataset_t *recv_ds; 1900 dsl_dataset_name(ds, recvname); 1901 if (strlcat(recvname, "/", sizeof (recvname)) < 1902 sizeof (recvname) && 1903 strlcat(recvname, recv_clone_name, sizeof (recvname)) < 1904 sizeof (recvname) && 1905 dsl_dataset_hold(ds->ds_dir->dd_pool, recvname, FTAG, 1906 &recv_ds) == 0) { 1907 char *propval = get_receive_resume_stats_impl(recv_ds); 1908 dsl_dataset_rele(recv_ds, FTAG); 1909 return (propval); 1910 } 1911 return (spa_strdup("")); 1912} 1913 1914static void 1915get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv) 1916{ 1917 char *propval = get_receive_resume_stats_impl(ds); 1918 if (strcmp(propval, "") != 0) { 1919 dsl_prop_nvlist_add_string(nv, 1920 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval); 1921 } else { 1922 char *childval = get_child_receive_stats(ds); 1923 if (strcmp(childval, "") != 0) { 1924 dsl_prop_nvlist_add_string(nv, 1925 ZFS_PROP_RECEIVE_RESUME_TOKEN, childval); 1926 } 1927 strfree(childval); 1928 } 1929 strfree(propval); 1930} 1931 1932uint64_t 1933dsl_get_refratio(dsl_dataset_t *ds) 1934{ 1935 uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : 1936 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / 1937 dsl_dataset_phys(ds)->ds_compressed_bytes); 1938 return (ratio); 1939} 1940 1941uint64_t 1942dsl_get_logicalreferenced(dsl_dataset_t *ds) 1943{ 1944 return (dsl_dataset_phys(ds)->ds_uncompressed_bytes); 1945} 1946 1947uint64_t 1948dsl_get_compressratio(dsl_dataset_t *ds) 1949{ 1950 if (ds->ds_is_snapshot) { 1951 return (dsl_get_refratio(ds)); 1952 } else { 1953 dsl_dir_t *dd = ds->ds_dir; 1954 mutex_enter(&dd->dd_lock); 1955 uint64_t val = dsl_dir_get_compressratio(dd); 1956 mutex_exit(&dd->dd_lock); 1957 return (val); 1958 } 1959} 1960 1961uint64_t 1962dsl_get_used(dsl_dataset_t *ds) 1963{ 1964 if (ds->ds_is_snapshot) { 1965 return (dsl_dataset_phys(ds)->ds_unique_bytes); 1966 } else { 1967 dsl_dir_t *dd = ds->ds_dir; 1968 mutex_enter(&dd->dd_lock); 1969 uint64_t val = dsl_dir_get_used(dd); 1970 mutex_exit(&dd->dd_lock); 1971 return (val); 1972 } 1973} 1974 1975uint64_t 1976dsl_get_creation(dsl_dataset_t *ds) 1977{ 1978 return (dsl_dataset_phys(ds)->ds_creation_time); 1979} 1980 1981uint64_t 1982dsl_get_creationtxg(dsl_dataset_t *ds) 1983{ 1984 return (dsl_dataset_phys(ds)->ds_creation_txg); 1985} 1986 1987uint64_t 1988dsl_get_refquota(dsl_dataset_t *ds) 1989{ 1990 return (ds->ds_quota); 1991} 1992 1993uint64_t 1994dsl_get_refreservation(dsl_dataset_t *ds) 1995{ 1996 return (ds->ds_reserved); 1997} 1998 1999uint64_t 2000dsl_get_guid(dsl_dataset_t *ds) 2001{ 2002 return (dsl_dataset_phys(ds)->ds_guid); 2003} 2004 2005uint64_t 2006dsl_get_unique(dsl_dataset_t *ds) 2007{ 2008 return (dsl_dataset_phys(ds)->ds_unique_bytes); 2009} 2010 2011uint64_t 2012dsl_get_objsetid(dsl_dataset_t *ds) 2013{ 2014 return (ds->ds_object); 2015} 2016 2017uint64_t 2018dsl_get_userrefs(dsl_dataset_t *ds) 2019{ 2020 return (ds->ds_userrefs); 2021} 2022 2023uint64_t 2024dsl_get_defer_destroy(dsl_dataset_t *ds) 2025{ 2026 return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 2027} 2028 2029uint64_t 2030dsl_get_referenced(dsl_dataset_t *ds) 2031{ 2032 return (dsl_dataset_phys(ds)->ds_referenced_bytes); 2033} 2034 2035uint64_t 2036dsl_get_numclones(dsl_dataset_t *ds) 2037{ 2038 ASSERT(ds->ds_is_snapshot); 2039 return (dsl_dataset_phys(ds)->ds_num_children - 1); 2040} 2041 2042uint64_t 2043dsl_get_inconsistent(dsl_dataset_t *ds) 2044{ 2045 return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ? 2046 1 : 0); 2047} 2048 2049uint64_t 2050dsl_get_available(dsl_dataset_t *ds) 2051{ 2052 uint64_t refdbytes = dsl_get_referenced(ds); 2053 uint64_t availbytes = dsl_dir_space_available(ds->ds_dir, 2054 NULL, 0, TRUE); 2055 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 2056 availbytes += 2057 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 2058 } 2059 if (ds->ds_quota != 0) { 2060 /* 2061 * Adjust available bytes according to refquota 2062 */ 2063 if (refdbytes < ds->ds_quota) { 2064 availbytes = MIN(availbytes, 2065 ds->ds_quota - refdbytes); 2066 } else { 2067 availbytes = 0; 2068 } 2069 } 2070 return (availbytes); 2071} 2072 2073int 2074dsl_get_written(dsl_dataset_t *ds, uint64_t *written) 2075{ 2076 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2077 dsl_dataset_t *prev; 2078 int err = dsl_dataset_hold_obj(dp, 2079 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 2080 if (err == 0) { 2081 uint64_t comp, uncomp; 2082 err = dsl_dataset_space_written(prev, ds, written, 2083 &comp, &uncomp); 2084 dsl_dataset_rele(prev, FTAG); 2085 } 2086 return (err); 2087} 2088 2089/* 2090 * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN. 2091 */ 2092int 2093dsl_get_prev_snap(dsl_dataset_t *ds, char *snap) 2094{ 2095 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2096 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 2097 dsl_dataset_name(ds->ds_prev, snap); 2098 return (0); 2099 } else { 2100 return (ENOENT); 2101 } 2102} 2103 2104/* 2105 * Returns the mountpoint property and source for the given dataset in the value 2106 * and source buffers. The value buffer must be at least as large as MAXPATHLEN 2107 * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN. 2108 * Returns 0 on success and an error on failure. 2109 */ 2110int 2111dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value, 2112 char *source) 2113{ 2114 int error; 2115 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2116 2117 /* Retrieve the mountpoint value stored in the zap opbject */ 2118 error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1, 2119 ZAP_MAXVALUELEN, value, source); 2120 if (error != 0) { 2121 return (error); 2122 } 2123 2124 /* Process the dsname and source to find the full mountpoint string */ 2125 if (value[0] == '/') { 2126 char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP); 2127 char *root = buf; 2128 const char *relpath; 2129 2130 /* 2131 * If we inherit the mountpoint, even from a dataset 2132 * with a received value, the source will be the path of 2133 * the dataset we inherit from. If source is 2134 * ZPROP_SOURCE_VAL_RECVD, the received value is not 2135 * inherited. 2136 */ 2137 if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) { 2138 relpath = ""; 2139 } else { 2140 ASSERT0(strncmp(dsname, source, strlen(source))); 2141 relpath = dsname + strlen(source); 2142 if (relpath[0] == '/') 2143 relpath++; 2144 } 2145 2146 spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN); 2147 2148 /* 2149 * Special case an alternate root of '/'. This will 2150 * avoid having multiple leading slashes in the 2151 * mountpoint path. 2152 */ 2153 if (strcmp(root, "/") == 0) 2154 root++; 2155 2156 /* 2157 * If the mountpoint is '/' then skip over this 2158 * if we are obtaining either an alternate root or 2159 * an inherited mountpoint. 2160 */ 2161 char *mnt = value; 2162 if (value[1] == '\0' && (root[0] != '\0' || 2163 relpath[0] != '\0')) 2164 mnt = value + 1; 2165 2166 if (relpath[0] == '\0') { 2167 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s", 2168 root, mnt); 2169 } else { 2170 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s", 2171 root, mnt, relpath[0] == '@' ? "" : "/", 2172 relpath); 2173 } 2174 kmem_free(buf, ZAP_MAXVALUELEN); 2175 } else { 2176 /* 'legacy' or 'none' */ 2177 (void) snprintf(value, ZAP_MAXVALUELEN, "%s", value); 2178 } 2179 return (0); 2180} 2181 2182void 2183dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 2184{ 2185 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2186 2187 ASSERT(dsl_pool_config_held(dp)); 2188 2189 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, 2190 dsl_get_refratio(ds)); 2191 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 2192 dsl_get_logicalreferenced(ds)); 2193 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 2194 dsl_get_compressratio(ds)); 2195 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 2196 dsl_get_used(ds)); 2197 2198 if (ds->ds_is_snapshot) { 2199 get_clones_stat(ds, nv); 2200 } else { 2201 char buf[ZFS_MAX_DATASET_NAME_LEN]; 2202 if (dsl_get_prev_snap(ds, buf) == 0) 2203 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, 2204 buf); 2205 dsl_dir_stats(ds->ds_dir, nv); 2206 } 2207 2208 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, 2209 dsl_get_available(ds)); 2210 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, 2211 dsl_get_referenced(ds)); 2212 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 2213 dsl_get_creation(ds)); 2214 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 2215 dsl_get_creationtxg(ds)); 2216 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 2217 dsl_get_refquota(ds)); 2218 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 2219 dsl_get_refreservation(ds)); 2220 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 2221 dsl_get_guid(ds)); 2222 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 2223 dsl_get_unique(ds)); 2224 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 2225 dsl_get_objsetid(ds)); 2226 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 2227 dsl_get_userrefs(ds)); 2228 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 2229 dsl_get_defer_destroy(ds)); 2230 2231 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 2232 uint64_t written; 2233 if (dsl_get_written(ds, &written) == 0) { 2234 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 2235 written); 2236 } 2237 } 2238 2239 if (!dsl_dataset_is_snapshot(ds)) { 2240 /* 2241 * A failed "newfs" (e.g. full) resumable receive leaves 2242 * the stats set on this dataset. Check here for the prop. 2243 */ 2244 get_receive_resume_stats(ds, nv); 2245 2246 /* 2247 * A failed incremental resumable receive leaves the 2248 * stats set on our child named "%recv". Check the child 2249 * for the prop. 2250 */ 2251 /* 6 extra bytes for /%recv */ 2252 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6]; 2253 dsl_dataset_t *recv_ds; 2254 dsl_dataset_name(ds, recvname); 2255 if (strlcat(recvname, "/", sizeof (recvname)) < 2256 sizeof (recvname) && 2257 strlcat(recvname, recv_clone_name, sizeof (recvname)) < 2258 sizeof (recvname) && 2259 dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) { 2260 get_receive_resume_stats(recv_ds, nv); 2261 dsl_dataset_rele(recv_ds, FTAG); 2262 } 2263 } 2264} 2265 2266void 2267dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 2268{ 2269 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2270 ASSERT(dsl_pool_config_held(dp)); 2271 2272 stat->dds_creation_txg = dsl_get_creationtxg(ds); 2273 stat->dds_inconsistent = dsl_get_inconsistent(ds); 2274 stat->dds_guid = dsl_get_guid(ds); 2275 stat->dds_origin[0] = '\0'; 2276 if (ds->ds_is_snapshot) { 2277 stat->dds_is_snapshot = B_TRUE; 2278 stat->dds_num_clones = dsl_get_numclones(ds); 2279 } else { 2280 stat->dds_is_snapshot = B_FALSE; 2281 stat->dds_num_clones = 0; 2282 2283 if (dsl_dir_is_clone(ds->ds_dir)) { 2284 dsl_dir_get_origin(ds->ds_dir, stat->dds_origin); 2285 } 2286 } 2287} 2288 2289uint64_t 2290dsl_dataset_fsid_guid(dsl_dataset_t *ds) 2291{ 2292 return (ds->ds_fsid_guid); 2293} 2294 2295void 2296dsl_dataset_space(dsl_dataset_t *ds, 2297 uint64_t *refdbytesp, uint64_t *availbytesp, 2298 uint64_t *usedobjsp, uint64_t *availobjsp) 2299{ 2300 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; 2301 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 2302 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) 2303 *availbytesp += 2304 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 2305 if (ds->ds_quota != 0) { 2306 /* 2307 * Adjust available bytes according to refquota 2308 */ 2309 if (*refdbytesp < ds->ds_quota) 2310 *availbytesp = MIN(*availbytesp, 2311 ds->ds_quota - *refdbytesp); 2312 else 2313 *availbytesp = 0; 2314 } 2315 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2316 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); 2317 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2318 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 2319} 2320 2321boolean_t 2322dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 2323{ 2324 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2325 uint64_t birth; 2326 2327 ASSERT(dsl_pool_config_held(dp)); 2328 if (snap == NULL) 2329 return (B_FALSE); 2330 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 2331 birth = dsl_dataset_get_blkptr(ds)->blk_birth; 2332 rrw_exit(&ds->ds_bp_rwlock, FTAG); 2333 if (birth > dsl_dataset_phys(snap)->ds_creation_txg) { 2334 objset_t *os, *os_snap; 2335 /* 2336 * It may be that only the ZIL differs, because it was 2337 * reset in the head. Don't count that as being 2338 * modified. 2339 */ 2340 if (dmu_objset_from_ds(ds, &os) != 0) 2341 return (B_TRUE); 2342 if (dmu_objset_from_ds(snap, &os_snap) != 0) 2343 return (B_TRUE); 2344 return (bcmp(&os->os_phys->os_meta_dnode, 2345 &os_snap->os_phys->os_meta_dnode, 2346 sizeof (os->os_phys->os_meta_dnode)) != 0); 2347 } 2348 return (B_FALSE); 2349} 2350 2351typedef struct dsl_dataset_rename_snapshot_arg { 2352 const char *ddrsa_fsname; 2353 const char *ddrsa_oldsnapname; 2354 const char *ddrsa_newsnapname; 2355 boolean_t ddrsa_recursive; 2356 dmu_tx_t *ddrsa_tx; 2357} dsl_dataset_rename_snapshot_arg_t; 2358 2359/* ARGSUSED */ 2360static int 2361dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 2362 dsl_dataset_t *hds, void *arg) 2363{ 2364 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2365 int error; 2366 uint64_t val; 2367 2368 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2369 if (error != 0) { 2370 /* ignore nonexistent snapshots */ 2371 return (error == ENOENT ? 0 : error); 2372 } 2373 2374 /* new name should not exist */ 2375 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 2376 if (error == 0) 2377 error = SET_ERROR(EEXIST); 2378 else if (error == ENOENT) 2379 error = 0; 2380 2381 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2382 if (dsl_dir_namelen(hds->ds_dir) + 1 + 2383 strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN) 2384 error = SET_ERROR(ENAMETOOLONG); 2385 2386 return (error); 2387} 2388 2389static int 2390dsl_dataset_rename_snapshot_check(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 int error; 2396 2397 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 2398 if (error != 0) 2399 return (error); 2400 2401 if (ddrsa->ddrsa_recursive) { 2402 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 2403 dsl_dataset_rename_snapshot_check_impl, ddrsa, 2404 DS_FIND_CHILDREN); 2405 } else { 2406 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 2407 } 2408 dsl_dataset_rele(hds, FTAG); 2409 return (error); 2410} 2411 2412static int 2413dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 2414 dsl_dataset_t *hds, void *arg) 2415{ 2416#ifdef __FreeBSD__ 2417#ifdef _KERNEL 2418 char *oldname, *newname; 2419#endif 2420#endif 2421 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2422 dsl_dataset_t *ds; 2423 uint64_t val; 2424 dmu_tx_t *tx = ddrsa->ddrsa_tx; 2425 int error; 2426 2427 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2428 ASSERT(error == 0 || error == ENOENT); 2429 if (error == ENOENT) { 2430 /* ignore nonexistent snapshots */ 2431 return (0); 2432 } 2433 2434 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 2435 2436 /* log before we change the name */ 2437 spa_history_log_internal_ds(ds, "rename", tx, 2438 "-> @%s", ddrsa->ddrsa_newsnapname); 2439 2440 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 2441 B_FALSE)); 2442 mutex_enter(&ds->ds_lock); 2443 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 2444 mutex_exit(&ds->ds_lock); 2445 VERIFY0(zap_add(dp->dp_meta_objset, 2446 dsl_dataset_phys(hds)->ds_snapnames_zapobj, 2447 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 2448 2449#ifdef __FreeBSD__ 2450#ifdef _KERNEL 2451 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2452 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2453 snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname, 2454 ddrsa->ddrsa_oldsnapname); 2455 snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname, 2456 ddrsa->ddrsa_newsnapname); 2457 zfsvfs_update_fromname(oldname, newname); 2458 zvol_rename_minors(oldname, newname); 2459 kmem_free(newname, MAXPATHLEN); 2460 kmem_free(oldname, MAXPATHLEN); 2461#endif 2462#endif 2463 dsl_dataset_rele(ds, FTAG); 2464 2465 return (0); 2466} 2467 2468static void 2469dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 2470{ 2471 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2472 dsl_pool_t *dp = dmu_tx_pool(tx); 2473 dsl_dataset_t *hds; 2474 2475 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 2476 ddrsa->ddrsa_tx = tx; 2477 if (ddrsa->ddrsa_recursive) { 2478 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 2479 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 2480 DS_FIND_CHILDREN)); 2481 } else { 2482 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 2483 } 2484 dsl_dataset_rele(hds, FTAG); 2485} 2486 2487int 2488dsl_dataset_rename_snapshot(const char *fsname, 2489 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 2490{ 2491 dsl_dataset_rename_snapshot_arg_t ddrsa; 2492 2493 ddrsa.ddrsa_fsname = fsname; 2494 ddrsa.ddrsa_oldsnapname = oldsnapname; 2495 ddrsa.ddrsa_newsnapname = newsnapname; 2496 ddrsa.ddrsa_recursive = recursive; 2497 2498 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 2499 dsl_dataset_rename_snapshot_sync, &ddrsa, 2500 1, ZFS_SPACE_CHECK_RESERVED)); 2501} 2502 2503/* 2504 * If we're doing an ownership handoff, we need to make sure that there is 2505 * only one long hold on the dataset. We're not allowed to change anything here 2506 * so we don't permanently release the long hold or regular hold here. We want 2507 * to do this only when syncing to avoid the dataset unexpectedly going away 2508 * when we release the long hold. 2509 */ 2510static int 2511dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 2512{ 2513 boolean_t held; 2514 2515 if (!dmu_tx_is_syncing(tx)) 2516 return (0); 2517 2518 if (owner != NULL) { 2519 VERIFY3P(ds->ds_owner, ==, owner); 2520 dsl_dataset_long_rele(ds, owner); 2521 } 2522 2523 held = dsl_dataset_long_held(ds); 2524 2525 if (owner != NULL) 2526 dsl_dataset_long_hold(ds, owner); 2527 2528 if (held) 2529 return (SET_ERROR(EBUSY)); 2530 2531 return (0); 2532} 2533 2534int 2535dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 2536{ 2537 dsl_dataset_rollback_arg_t *ddra = arg; 2538 dsl_pool_t *dp = dmu_tx_pool(tx); 2539 dsl_dataset_t *ds; 2540 int64_t unused_refres_delta; 2541 int error; 2542 2543 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 2544 if (error != 0) 2545 return (error); 2546 2547 /* must not be a snapshot */ 2548 if (ds->ds_is_snapshot) { 2549 dsl_dataset_rele(ds, FTAG); 2550 return (SET_ERROR(EINVAL)); 2551 } 2552 2553 /* must have a most recent snapshot */ 2554 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { 2555 dsl_dataset_rele(ds, FTAG); 2556 return (SET_ERROR(ESRCH)); 2557 } 2558 2559 /* 2560 * No rollback to a snapshot created in the current txg, because 2561 * the rollback may dirty the dataset and create blocks that are 2562 * not reachable from the rootbp while having a birth txg that 2563 * falls into the snapshot's range. 2564 */ 2565 if (dmu_tx_is_syncing(tx) && 2566 dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) { 2567 dsl_dataset_rele(ds, FTAG); 2568 return (SET_ERROR(EAGAIN)); 2569 } 2570 2571 /* 2572 * If the expected target snapshot is specified, then check that 2573 * the latest snapshot is it. 2574 */ 2575 if (ddra->ddra_tosnap != NULL) { 2576 dsl_dataset_t *snapds; 2577 2578 /* Check if the target snapshot exists at all. */ 2579 error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds); 2580 if (error != 0) { 2581 /* 2582 * ESRCH is used to signal that the target snapshot does 2583 * not exist, while ENOENT is used to report that 2584 * the rolled back dataset does not exist. 2585 * ESRCH is also used to cover other cases where the 2586 * target snapshot is not related to the dataset being 2587 * rolled back such as being in a different pool. 2588 */ 2589 if (error == ENOENT || error == EXDEV) 2590 error = SET_ERROR(ESRCH); 2591 dsl_dataset_rele(ds, FTAG); 2592 return (error); 2593 } 2594 ASSERT(snapds->ds_is_snapshot); 2595 2596 /* Check if the snapshot is the latest snapshot indeed. */ 2597 if (snapds != ds->ds_prev) { 2598 /* 2599 * Distinguish between the case where the only problem 2600 * is intervening snapshots (EEXIST) vs the snapshot 2601 * not being a valid target for rollback (ESRCH). 2602 */ 2603 if (snapds->ds_dir == ds->ds_dir || 2604 (dsl_dir_is_clone(ds->ds_dir) && 2605 dsl_dir_phys(ds->ds_dir)->dd_origin_obj == 2606 snapds->ds_object)) { 2607 error = SET_ERROR(EEXIST); 2608 } else { 2609 error = SET_ERROR(ESRCH); 2610 } 2611 dsl_dataset_rele(snapds, FTAG); 2612 dsl_dataset_rele(ds, FTAG); 2613 return (error); 2614 } 2615 dsl_dataset_rele(snapds, FTAG); 2616 } 2617 2618 /* must not have any bookmarks after the most recent snapshot */ 2619 nvlist_t *proprequest = fnvlist_alloc(); 2620 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2621 nvlist_t *bookmarks = fnvlist_alloc(); 2622 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 2623 fnvlist_free(proprequest); 2624 if (error != 0) { 2625 dsl_dataset_rele(ds, FTAG); 2626 return (error); 2627 } 2628 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 2629 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 2630 nvlist_t *valuenv = 2631 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 2632 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2633 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 2634 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 2635 fnvlist_free(bookmarks); 2636 dsl_dataset_rele(ds, FTAG); 2637 return (SET_ERROR(EEXIST)); 2638 } 2639 } 2640 fnvlist_free(bookmarks); 2641 2642 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 2643 if (error != 0) { 2644 dsl_dataset_rele(ds, FTAG); 2645 return (error); 2646 } 2647 2648 /* 2649 * Check if the snap we are rolling back to uses more than 2650 * the refquota. 2651 */ 2652 if (ds->ds_quota != 0 && 2653 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { 2654 dsl_dataset_rele(ds, FTAG); 2655 return (SET_ERROR(EDQUOT)); 2656 } 2657 2658 /* 2659 * When we do the clone swap, we will temporarily use more space 2660 * due to the refreservation (the head will no longer have any 2661 * unique space, so the entire amount of the refreservation will need 2662 * to be free). We will immediately destroy the clone, freeing 2663 * this space, but the freeing happens over many txg's. 2664 */ 2665 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 2666 dsl_dataset_phys(ds)->ds_unique_bytes); 2667 2668 if (unused_refres_delta > 0 && 2669 unused_refres_delta > 2670 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 2671 dsl_dataset_rele(ds, FTAG); 2672 return (SET_ERROR(ENOSPC)); 2673 } 2674 2675 dsl_dataset_rele(ds, FTAG); 2676 return (0); 2677} 2678 2679void 2680dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 2681{ 2682 dsl_dataset_rollback_arg_t *ddra = arg; 2683 dsl_pool_t *dp = dmu_tx_pool(tx); 2684 dsl_dataset_t *ds, *clone; 2685 uint64_t cloneobj; 2686 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 2687 2688 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 2689 2690 dsl_dataset_name(ds->ds_prev, namebuf); 2691 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 2692 2693 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 2694 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 2695 2696 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 2697 2698 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 2699 dsl_dataset_zero_zil(ds, tx); 2700 2701 dsl_destroy_head_sync_impl(clone, tx); 2702 2703 dsl_dataset_rele(clone, FTAG); 2704 dsl_dataset_rele(ds, FTAG); 2705} 2706 2707/* 2708 * Rolls back the given filesystem or volume to the most recent snapshot. 2709 * The name of the most recent snapshot will be returned under key "target" 2710 * in the result nvlist. 2711 * 2712 * If owner != NULL: 2713 * - The existing dataset MUST be owned by the specified owner at entry 2714 * - Upon return, dataset will still be held by the same owner, whether we 2715 * succeed or not. 2716 * 2717 * This mode is required any time the existing filesystem is mounted. See 2718 * notes above zfs_suspend_fs() for further details. 2719 */ 2720int 2721dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner, 2722 nvlist_t *result) 2723{ 2724 dsl_dataset_rollback_arg_t ddra; 2725 2726 ddra.ddra_fsname = fsname; 2727 ddra.ddra_tosnap = tosnap; 2728 ddra.ddra_owner = owner; 2729 ddra.ddra_result = result; 2730 2731 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 2732 dsl_dataset_rollback_sync, &ddra, 2733 1, ZFS_SPACE_CHECK_RESERVED)); 2734} 2735 2736struct promotenode { 2737 list_node_t link; 2738 dsl_dataset_t *ds; 2739}; 2740 2741static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2742static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2743 void *tag); 2744static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2745 2746int 2747dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2748{ 2749 dsl_dataset_promote_arg_t *ddpa = arg; 2750 dsl_pool_t *dp = dmu_tx_pool(tx); 2751 dsl_dataset_t *hds; 2752 struct promotenode *snap; 2753 dsl_dataset_t *origin_ds; 2754 int err; 2755 uint64_t unused; 2756 uint64_t ss_mv_cnt; 2757 size_t max_snap_len; 2758 boolean_t conflicting_snaps; 2759 2760 err = promote_hold(ddpa, dp, FTAG); 2761 if (err != 0) 2762 return (err); 2763 2764 hds = ddpa->ddpa_clone; 2765 snap = list_head(&ddpa->shared_snaps); 2766 origin_ds = snap->ds; 2767 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; 2768 2769 snap = list_head(&ddpa->origin_snaps); 2770 2771 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { 2772 promote_rele(ddpa, FTAG); 2773 return (SET_ERROR(EXDEV)); 2774 } 2775 2776 /* 2777 * Compute and check the amount of space to transfer. Since this is 2778 * so expensive, don't do the preliminary check. 2779 */ 2780 if (!dmu_tx_is_syncing(tx)) { 2781 promote_rele(ddpa, FTAG); 2782 return (0); 2783 } 2784 2785 /* compute origin's new unique space */ 2786 snap = list_tail(&ddpa->clone_snaps); 2787 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2788 origin_ds->ds_object); 2789 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2790 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, 2791 &ddpa->unique, &unused, &unused); 2792 2793 /* 2794 * Walk the snapshots that we are moving 2795 * 2796 * Compute space to transfer. Consider the incremental changes 2797 * to used by each snapshot: 2798 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2799 * So each snapshot gave birth to: 2800 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2801 * So a sequence would look like: 2802 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2803 * Which simplifies to: 2804 * uN + kN + kN-1 + ... + k1 + k0 2805 * Note however, if we stop before we reach the ORIGIN we get: 2806 * uN + kN + kN-1 + ... + kM - uM-1 2807 */ 2808 conflicting_snaps = B_FALSE; 2809 ss_mv_cnt = 0; 2810 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; 2811 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; 2812 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; 2813 for (snap = list_head(&ddpa->shared_snaps); snap; 2814 snap = list_next(&ddpa->shared_snaps, snap)) { 2815 uint64_t val, dlused, dlcomp, dluncomp; 2816 dsl_dataset_t *ds = snap->ds; 2817 2818 ss_mv_cnt++; 2819 2820 /* 2821 * If there are long holds, we won't be able to evict 2822 * the objset. 2823 */ 2824 if (dsl_dataset_long_held(ds)) { 2825 err = SET_ERROR(EBUSY); 2826 goto out; 2827 } 2828 2829 /* Check that the snapshot name does not conflict */ 2830 VERIFY0(dsl_dataset_get_snapname(ds)); 2831 if (strlen(ds->ds_snapname) >= max_snap_len) { 2832 err = SET_ERROR(ENAMETOOLONG); 2833 goto out; 2834 } 2835 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2836 if (err == 0) { 2837 fnvlist_add_boolean(ddpa->err_ds, 2838 snap->ds->ds_snapname); 2839 conflicting_snaps = B_TRUE; 2840 } else if (err != ENOENT) { 2841 goto out; 2842 } 2843 2844 /* The very first snapshot does not have a deadlist */ 2845 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) 2846 continue; 2847 2848 dsl_deadlist_space(&ds->ds_deadlist, 2849 &dlused, &dlcomp, &dluncomp); 2850 ddpa->used += dlused; 2851 ddpa->comp += dlcomp; 2852 ddpa->uncomp += dluncomp; 2853 } 2854 2855 /* 2856 * In order to return the full list of conflicting snapshots, we check 2857 * whether there was a conflict after traversing all of them. 2858 */ 2859 if (conflicting_snaps) { 2860 err = SET_ERROR(EEXIST); 2861 goto out; 2862 } 2863 2864 /* 2865 * If we are a clone of a clone then we never reached ORIGIN, 2866 * so we need to subtract out the clone origin's used space. 2867 */ 2868 if (ddpa->origin_origin) { 2869 ddpa->used -= 2870 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; 2871 ddpa->comp -= 2872 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; 2873 ddpa->uncomp -= 2874 dsl_dataset_phys(ddpa->origin_origin)-> 2875 ds_uncompressed_bytes; 2876 } 2877 2878 /* Check that there is enough space and limit headroom here */ 2879 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2880 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2881 if (err != 0) 2882 goto out; 2883 2884 /* 2885 * Compute the amounts of space that will be used by snapshots 2886 * after the promotion (for both origin and clone). For each, 2887 * it is the amount of space that will be on all of their 2888 * deadlists (that was not born before their new origin). 2889 */ 2890 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2891 uint64_t space; 2892 2893 /* 2894 * Note, typically this will not be a clone of a clone, 2895 * so dd_origin_txg will be < TXG_INITIAL, so 2896 * these snaplist_space() -> dsl_deadlist_space_range() 2897 * calls will be fast because they do not have to 2898 * iterate over all bps. 2899 */ 2900 snap = list_head(&ddpa->origin_snaps); 2901 err = snaplist_space(&ddpa->shared_snaps, 2902 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2903 if (err != 0) 2904 goto out; 2905 2906 err = snaplist_space(&ddpa->clone_snaps, 2907 snap->ds->ds_dir->dd_origin_txg, &space); 2908 if (err != 0) 2909 goto out; 2910 ddpa->cloneusedsnap += space; 2911 } 2912 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & 2913 DD_FLAG_USED_BREAKDOWN) { 2914 err = snaplist_space(&ddpa->origin_snaps, 2915 dsl_dataset_phys(origin_ds)->ds_creation_txg, 2916 &ddpa->originusedsnap); 2917 if (err != 0) 2918 goto out; 2919 } 2920 2921out: 2922 promote_rele(ddpa, FTAG); 2923 return (err); 2924} 2925 2926void 2927dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2928{ 2929 dsl_dataset_promote_arg_t *ddpa = arg; 2930 dsl_pool_t *dp = dmu_tx_pool(tx); 2931 dsl_dataset_t *hds; 2932 struct promotenode *snap; 2933 dsl_dataset_t *origin_ds; 2934 dsl_dataset_t *origin_head; 2935 dsl_dir_t *dd; 2936 dsl_dir_t *odd = NULL; 2937 uint64_t oldnext_obj; 2938 int64_t delta; 2939#if defined(__FreeBSD__) && defined(_KERNEL) 2940 char *oldname, *newname; 2941#endif 2942 2943 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2944 hds = ddpa->ddpa_clone; 2945 2946 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); 2947 2948 snap = list_head(&ddpa->shared_snaps); 2949 origin_ds = snap->ds; 2950 dd = hds->ds_dir; 2951 2952 snap = list_head(&ddpa->origin_snaps); 2953 origin_head = snap->ds; 2954 2955 /* 2956 * We need to explicitly open odd, since origin_ds's dd will be 2957 * changing. 2958 */ 2959 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2960 NULL, FTAG, &odd)); 2961 2962 /* change origin's next snap */ 2963 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2964 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; 2965 snap = list_tail(&ddpa->clone_snaps); 2966 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2967 origin_ds->ds_object); 2968 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; 2969 2970 /* change the origin's next clone */ 2971 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { 2972 dsl_dataset_remove_from_next_clones(origin_ds, 2973 snap->ds->ds_object, tx); 2974 VERIFY0(zap_add_int(dp->dp_meta_objset, 2975 dsl_dataset_phys(origin_ds)->ds_next_clones_obj, 2976 oldnext_obj, tx)); 2977 } 2978 2979 /* change origin */ 2980 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2981 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); 2982 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; 2983 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2984 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2985 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; 2986 origin_head->ds_dir->dd_origin_txg = 2987 dsl_dataset_phys(origin_ds)->ds_creation_txg; 2988 2989 /* change dd_clone entries */ 2990 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2991 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2992 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); 2993 VERIFY0(zap_add_int(dp->dp_meta_objset, 2994 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2995 hds->ds_object, tx)); 2996 2997 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2998 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2999 origin_head->ds_object, tx)); 3000 if (dsl_dir_phys(dd)->dd_clones == 0) { 3001 dsl_dir_phys(dd)->dd_clones = 3002 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, 3003 DMU_OT_NONE, 0, tx); 3004 } 3005 VERIFY0(zap_add_int(dp->dp_meta_objset, 3006 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); 3007 } 3008 3009#if defined(__FreeBSD__) && defined(_KERNEL) 3010 /* Take the spa_namespace_lock early so zvol renames don't deadlock. */ 3011 mutex_enter(&spa_namespace_lock); 3012 3013 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3014 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3015#endif 3016 3017 /* move snapshots to this dir */ 3018 for (snap = list_head(&ddpa->shared_snaps); snap; 3019 snap = list_next(&ddpa->shared_snaps, snap)) { 3020 dsl_dataset_t *ds = snap->ds; 3021 3022 /* 3023 * Property callbacks are registered to a particular 3024 * dsl_dir. Since ours is changing, evict the objset 3025 * so that they will be unregistered from the old dsl_dir. 3026 */ 3027 if (ds->ds_objset) { 3028 dmu_objset_evict(ds->ds_objset); 3029 ds->ds_objset = NULL; 3030 } 3031 3032 /* move snap name entry */ 3033 VERIFY0(dsl_dataset_get_snapname(ds)); 3034 VERIFY0(dsl_dataset_snap_remove(origin_head, 3035 ds->ds_snapname, tx, B_TRUE)); 3036 VERIFY0(zap_add(dp->dp_meta_objset, 3037 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 3038 8, 1, &ds->ds_object, tx)); 3039 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 3040 DD_FIELD_SNAPSHOT_COUNT, tx); 3041 3042 /* change containing dsl_dir */ 3043 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3044 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); 3045 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; 3046 ASSERT3P(ds->ds_dir, ==, odd); 3047 dsl_dir_rele(ds->ds_dir, ds); 3048 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 3049 NULL, ds, &ds->ds_dir)); 3050 3051#if defined(__FreeBSD__) && defined(_KERNEL) 3052 dsl_dataset_name(ds, newname); 3053 zfsvfs_update_fromname(oldname, newname); 3054 zvol_rename_minors(oldname, newname); 3055#endif 3056 3057 /* move any clone references */ 3058 if (dsl_dataset_phys(ds)->ds_next_clones_obj && 3059 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 3060 zap_cursor_t zc; 3061 zap_attribute_t za; 3062 3063 for (zap_cursor_init(&zc, dp->dp_meta_objset, 3064 dsl_dataset_phys(ds)->ds_next_clones_obj); 3065 zap_cursor_retrieve(&zc, &za) == 0; 3066 zap_cursor_advance(&zc)) { 3067 dsl_dataset_t *cnds; 3068 uint64_t o; 3069 3070 if (za.za_first_integer == oldnext_obj) { 3071 /* 3072 * We've already moved the 3073 * origin's reference. 3074 */ 3075 continue; 3076 } 3077 3078 VERIFY0(dsl_dataset_hold_obj(dp, 3079 za.za_first_integer, FTAG, &cnds)); 3080 o = dsl_dir_phys(cnds->ds_dir)-> 3081 dd_head_dataset_obj; 3082 3083 VERIFY0(zap_remove_int(dp->dp_meta_objset, 3084 dsl_dir_phys(odd)->dd_clones, o, tx)); 3085 VERIFY0(zap_add_int(dp->dp_meta_objset, 3086 dsl_dir_phys(dd)->dd_clones, o, tx)); 3087 dsl_dataset_rele(cnds, FTAG); 3088 } 3089 zap_cursor_fini(&zc); 3090 } 3091 3092 ASSERT(!dsl_prop_hascb(ds)); 3093 } 3094 3095#if defined(__FreeBSD__) && defined(_KERNEL) 3096 mutex_exit(&spa_namespace_lock); 3097 3098 kmem_free(newname, MAXPATHLEN); 3099 kmem_free(oldname, MAXPATHLEN); 3100#endif 3101 /* 3102 * Change space accounting. 3103 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 3104 * both be valid, or both be 0 (resulting in delta == 0). This 3105 * is true for each of {clone,origin} independently. 3106 */ 3107 3108 delta = ddpa->cloneusedsnap - 3109 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; 3110 ASSERT3S(delta, >=, 0); 3111 ASSERT3U(ddpa->used, >=, delta); 3112 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 3113 dsl_dir_diduse_space(dd, DD_USED_HEAD, 3114 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 3115 3116 delta = ddpa->originusedsnap - 3117 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; 3118 ASSERT3S(delta, <=, 0); 3119 ASSERT3U(ddpa->used, >=, -delta); 3120 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 3121 dsl_dir_diduse_space(odd, DD_USED_HEAD, 3122 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 3123 3124 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; 3125 3126 /* log history record */ 3127 spa_history_log_internal_ds(hds, "promote", tx, ""); 3128 3129 dsl_dir_rele(odd, FTAG); 3130 promote_rele(ddpa, FTAG); 3131} 3132 3133/* 3134 * Make a list of dsl_dataset_t's for the snapshots between first_obj 3135 * (exclusive) and last_obj (inclusive). The list will be in reverse 3136 * order (last_obj will be the list_head()). If first_obj == 0, do all 3137 * snapshots back to this dataset's origin. 3138 */ 3139static int 3140snaplist_make(dsl_pool_t *dp, 3141 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 3142{ 3143 uint64_t obj = last_obj; 3144 3145 list_create(l, sizeof (struct promotenode), 3146 offsetof(struct promotenode, link)); 3147 3148 while (obj != first_obj) { 3149 dsl_dataset_t *ds; 3150 struct promotenode *snap; 3151 int err; 3152 3153 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 3154 ASSERT(err != ENOENT); 3155 if (err != 0) 3156 return (err); 3157 3158 if (first_obj == 0) 3159 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; 3160 3161 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 3162 snap->ds = ds; 3163 list_insert_tail(l, snap); 3164 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3165 } 3166 3167 return (0); 3168} 3169 3170static int 3171snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 3172{ 3173 struct promotenode *snap; 3174 3175 *spacep = 0; 3176 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 3177 uint64_t used, comp, uncomp; 3178 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 3179 mintxg, UINT64_MAX, &used, &comp, &uncomp); 3180 *spacep += used; 3181 } 3182 return (0); 3183} 3184 3185static void 3186snaplist_destroy(list_t *l, void *tag) 3187{ 3188 struct promotenode *snap; 3189 3190 if (l == NULL || !list_link_active(&l->list_head)) 3191 return; 3192 3193 while ((snap = list_tail(l)) != NULL) { 3194 list_remove(l, snap); 3195 dsl_dataset_rele(snap->ds, tag); 3196 kmem_free(snap, sizeof (*snap)); 3197 } 3198 list_destroy(l); 3199} 3200 3201static int 3202promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 3203{ 3204 int error; 3205 dsl_dir_t *dd; 3206 struct promotenode *snap; 3207 3208 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 3209 &ddpa->ddpa_clone); 3210 if (error != 0) 3211 return (error); 3212 dd = ddpa->ddpa_clone->ds_dir; 3213 3214 if (ddpa->ddpa_clone->ds_is_snapshot || 3215 !dsl_dir_is_clone(dd)) { 3216 dsl_dataset_rele(ddpa->ddpa_clone, tag); 3217 return (SET_ERROR(EINVAL)); 3218 } 3219 3220 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, 3221 &ddpa->shared_snaps, tag); 3222 if (error != 0) 3223 goto out; 3224 3225 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 3226 &ddpa->clone_snaps, tag); 3227 if (error != 0) 3228 goto out; 3229 3230 snap = list_head(&ddpa->shared_snaps); 3231 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); 3232 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, 3233 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, 3234 &ddpa->origin_snaps, tag); 3235 if (error != 0) 3236 goto out; 3237 3238 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { 3239 error = dsl_dataset_hold_obj(dp, 3240 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, 3241 tag, &ddpa->origin_origin); 3242 if (error != 0) 3243 goto out; 3244 } 3245out: 3246 if (error != 0) 3247 promote_rele(ddpa, tag); 3248 return (error); 3249} 3250 3251static void 3252promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 3253{ 3254 snaplist_destroy(&ddpa->shared_snaps, tag); 3255 snaplist_destroy(&ddpa->clone_snaps, tag); 3256 snaplist_destroy(&ddpa->origin_snaps, tag); 3257 if (ddpa->origin_origin != NULL) 3258 dsl_dataset_rele(ddpa->origin_origin, tag); 3259 dsl_dataset_rele(ddpa->ddpa_clone, tag); 3260} 3261 3262/* 3263 * Promote a clone. 3264 * 3265 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 3266 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.) 3267 */ 3268int 3269dsl_dataset_promote(const char *name, char *conflsnap) 3270{ 3271 dsl_dataset_promote_arg_t ddpa = { 0 }; 3272 uint64_t numsnaps; 3273 int error; 3274 nvpair_t *snap_pair; 3275 objset_t *os; 3276 3277 /* 3278 * We will modify space proportional to the number of 3279 * snapshots. Compute numsnaps. 3280 */ 3281 error = dmu_objset_hold(name, FTAG, &os); 3282 if (error != 0) 3283 return (error); 3284 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 3285 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, 3286 &numsnaps); 3287 dmu_objset_rele(os, FTAG); 3288 if (error != 0) 3289 return (error); 3290 3291 ddpa.ddpa_clonename = name; 3292 ddpa.err_ds = fnvlist_alloc(); 3293 ddpa.cr = CRED(); 3294 3295 error = dsl_sync_task(name, dsl_dataset_promote_check, 3296 dsl_dataset_promote_sync, &ddpa, 3297 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED); 3298 3299 /* 3300 * Return the first conflicting snapshot found. 3301 */ 3302 snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL); 3303 if (snap_pair != NULL && conflsnap != NULL) 3304 (void) strcpy(conflsnap, nvpair_name(snap_pair)); 3305 3306 fnvlist_free(ddpa.err_ds); 3307 return (error); 3308} 3309 3310int 3311dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 3312 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 3313{ 3314 /* 3315 * "slack" factor for received datasets with refquota set on them. 3316 * See the bottom of this function for details on its use. 3317 */ 3318 uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation; 3319 int64_t unused_refres_delta; 3320 3321 /* they should both be heads */ 3322 if (clone->ds_is_snapshot || 3323 origin_head->ds_is_snapshot) 3324 return (SET_ERROR(EINVAL)); 3325 3326 /* if we are not forcing, the branch point should be just before them */ 3327 if (!force && clone->ds_prev != origin_head->ds_prev) 3328 return (SET_ERROR(EINVAL)); 3329 3330 /* clone should be the clone (unless they are unrelated) */ 3331 if (clone->ds_prev != NULL && 3332 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 3333 origin_head->ds_dir != clone->ds_prev->ds_dir) 3334 return (SET_ERROR(EINVAL)); 3335 3336 /* the clone should be a child of the origin */ 3337 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 3338 return (SET_ERROR(EINVAL)); 3339 3340 /* origin_head shouldn't be modified unless 'force' */ 3341 if (!force && 3342 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 3343 return (SET_ERROR(ETXTBSY)); 3344 3345 /* origin_head should have no long holds (e.g. is not mounted) */ 3346 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 3347 return (SET_ERROR(EBUSY)); 3348 3349 /* check amount of any unconsumed refreservation */ 3350 unused_refres_delta = 3351 (int64_t)MIN(origin_head->ds_reserved, 3352 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 3353 (int64_t)MIN(origin_head->ds_reserved, 3354 dsl_dataset_phys(clone)->ds_unique_bytes); 3355 3356 if (unused_refres_delta > 0 && 3357 unused_refres_delta > 3358 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 3359 return (SET_ERROR(ENOSPC)); 3360 3361 /* 3362 * The clone can't be too much over the head's refquota. 3363 * 3364 * To ensure that the entire refquota can be used, we allow one 3365 * transaction to exceed the the refquota. Therefore, this check 3366 * needs to also allow for the space referenced to be more than the 3367 * refquota. The maximum amount of space that one transaction can use 3368 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this 3369 * overage ensures that we are able to receive a filesystem that 3370 * exceeds the refquota on the source system. 3371 * 3372 * So that overage is the refquota_slack we use below. 3373 */ 3374 if (origin_head->ds_quota != 0 && 3375 dsl_dataset_phys(clone)->ds_referenced_bytes > 3376 origin_head->ds_quota + refquota_slack) 3377 return (SET_ERROR(EDQUOT)); 3378 3379 return (0); 3380} 3381 3382void 3383dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 3384 dsl_dataset_t *origin_head, dmu_tx_t *tx) 3385{ 3386 dsl_pool_t *dp = dmu_tx_pool(tx); 3387 int64_t unused_refres_delta; 3388 3389 ASSERT(clone->ds_reserved == 0); 3390 /* 3391 * NOTE: On DEBUG kernels there could be a race between this and 3392 * the check function if spa_asize_inflation is adjusted... 3393 */ 3394 ASSERT(origin_head->ds_quota == 0 || 3395 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota + 3396 DMU_MAX_ACCESS * spa_asize_inflation); 3397 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 3398 3399 /* 3400 * Swap per-dataset feature flags. 3401 */ 3402 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 3403 if (!(spa_feature_table[f].fi_flags & 3404 ZFEATURE_FLAG_PER_DATASET)) { 3405 ASSERT(!clone->ds_feature_inuse[f]); 3406 ASSERT(!origin_head->ds_feature_inuse[f]); 3407 continue; 3408 } 3409 3410 boolean_t clone_inuse = clone->ds_feature_inuse[f]; 3411 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f]; 3412 3413 if (clone_inuse) { 3414 dsl_dataset_deactivate_feature(clone->ds_object, f, tx); 3415 clone->ds_feature_inuse[f] = B_FALSE; 3416 } 3417 if (origin_head_inuse) { 3418 dsl_dataset_deactivate_feature(origin_head->ds_object, 3419 f, tx); 3420 origin_head->ds_feature_inuse[f] = B_FALSE; 3421 } 3422 if (clone_inuse) { 3423 dsl_dataset_activate_feature(origin_head->ds_object, 3424 f, tx); 3425 origin_head->ds_feature_inuse[f] = B_TRUE; 3426 } 3427 if (origin_head_inuse) { 3428 dsl_dataset_activate_feature(clone->ds_object, f, tx); 3429 clone->ds_feature_inuse[f] = B_TRUE; 3430 } 3431 } 3432 3433 dmu_buf_will_dirty(clone->ds_dbuf, tx); 3434 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 3435 3436 if (clone->ds_objset != NULL) { 3437 dmu_objset_evict(clone->ds_objset); 3438 clone->ds_objset = NULL; 3439 } 3440 3441 if (origin_head->ds_objset != NULL) { 3442 dmu_objset_evict(origin_head->ds_objset); 3443 origin_head->ds_objset = NULL; 3444 } 3445 3446 unused_refres_delta = 3447 (int64_t)MIN(origin_head->ds_reserved, 3448 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 3449 (int64_t)MIN(origin_head->ds_reserved, 3450 dsl_dataset_phys(clone)->ds_unique_bytes); 3451 3452 /* 3453 * Reset origin's unique bytes, if it exists. 3454 */ 3455 if (clone->ds_prev) { 3456 dsl_dataset_t *origin = clone->ds_prev; 3457 uint64_t comp, uncomp; 3458 3459 dmu_buf_will_dirty(origin->ds_dbuf, tx); 3460 dsl_deadlist_space_range(&clone->ds_deadlist, 3461 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, 3462 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); 3463 } 3464 3465 /* swap blkptrs */ 3466 { 3467 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG); 3468 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG); 3469 blkptr_t tmp; 3470 tmp = dsl_dataset_phys(origin_head)->ds_bp; 3471 dsl_dataset_phys(origin_head)->ds_bp = 3472 dsl_dataset_phys(clone)->ds_bp; 3473 dsl_dataset_phys(clone)->ds_bp = tmp; 3474 rrw_exit(&origin_head->ds_bp_rwlock, FTAG); 3475 rrw_exit(&clone->ds_bp_rwlock, FTAG); 3476 } 3477 3478 /* set dd_*_bytes */ 3479 { 3480 int64_t dused, dcomp, duncomp; 3481 uint64_t cdl_used, cdl_comp, cdl_uncomp; 3482 uint64_t odl_used, odl_comp, odl_uncomp; 3483 3484 ASSERT3U(dsl_dir_phys(clone->ds_dir)-> 3485 dd_used_breakdown[DD_USED_SNAP], ==, 0); 3486 3487 dsl_deadlist_space(&clone->ds_deadlist, 3488 &cdl_used, &cdl_comp, &cdl_uncomp); 3489 dsl_deadlist_space(&origin_head->ds_deadlist, 3490 &odl_used, &odl_comp, &odl_uncomp); 3491 3492 dused = dsl_dataset_phys(clone)->ds_referenced_bytes + 3493 cdl_used - 3494 (dsl_dataset_phys(origin_head)->ds_referenced_bytes + 3495 odl_used); 3496 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + 3497 cdl_comp - 3498 (dsl_dataset_phys(origin_head)->ds_compressed_bytes + 3499 odl_comp); 3500 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + 3501 cdl_uncomp - 3502 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + 3503 odl_uncomp); 3504 3505 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 3506 dused, dcomp, duncomp, tx); 3507 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 3508 -dused, -dcomp, -duncomp, tx); 3509 3510 /* 3511 * The difference in the space used by snapshots is the 3512 * difference in snapshot space due to the head's 3513 * deadlist (since that's the only thing that's 3514 * changing that affects the snapused). 3515 */ 3516 dsl_deadlist_space_range(&clone->ds_deadlist, 3517 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 3518 &cdl_used, &cdl_comp, &cdl_uncomp); 3519 dsl_deadlist_space_range(&origin_head->ds_deadlist, 3520 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 3521 &odl_used, &odl_comp, &odl_uncomp); 3522 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 3523 DD_USED_HEAD, DD_USED_SNAP, NULL); 3524 } 3525 3526 /* swap ds_*_bytes */ 3527 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, 3528 dsl_dataset_phys(clone)->ds_referenced_bytes); 3529 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, 3530 dsl_dataset_phys(clone)->ds_compressed_bytes); 3531 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, 3532 dsl_dataset_phys(clone)->ds_uncompressed_bytes); 3533 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, 3534 dsl_dataset_phys(clone)->ds_unique_bytes); 3535 3536 /* apply any parent delta for change in unconsumed refreservation */ 3537 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 3538 unused_refres_delta, 0, 0, tx); 3539 3540 /* 3541 * Swap deadlists. 3542 */ 3543 dsl_deadlist_close(&clone->ds_deadlist); 3544 dsl_deadlist_close(&origin_head->ds_deadlist); 3545 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, 3546 dsl_dataset_phys(clone)->ds_deadlist_obj); 3547 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 3548 dsl_dataset_phys(clone)->ds_deadlist_obj); 3549 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 3550 dsl_dataset_phys(origin_head)->ds_deadlist_obj); 3551 3552 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 3553 3554 spa_history_log_internal_ds(clone, "clone swap", tx, 3555 "parent=%s", origin_head->ds_dir->dd_myname); 3556} 3557 3558/* 3559 * Given a pool name and a dataset object number in that pool, 3560 * return the name of that dataset. 3561 */ 3562int 3563dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 3564{ 3565 dsl_pool_t *dp; 3566 dsl_dataset_t *ds; 3567 int error; 3568 3569 error = dsl_pool_hold(pname, FTAG, &dp); 3570 if (error != 0) 3571 return (error); 3572 3573 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 3574 if (error == 0) { 3575 dsl_dataset_name(ds, buf); 3576 dsl_dataset_rele(ds, FTAG); 3577 } 3578 dsl_pool_rele(dp, FTAG); 3579 3580 return (error); 3581} 3582 3583int 3584dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 3585 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 3586{ 3587 int error = 0; 3588 3589 ASSERT3S(asize, >, 0); 3590 3591 /* 3592 * *ref_rsrv is the portion of asize that will come from any 3593 * unconsumed refreservation space. 3594 */ 3595 *ref_rsrv = 0; 3596 3597 mutex_enter(&ds->ds_lock); 3598 /* 3599 * Make a space adjustment for reserved bytes. 3600 */ 3601 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 3602 ASSERT3U(*used, >=, 3603 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 3604 *used -= 3605 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 3606 *ref_rsrv = 3607 asize - MIN(asize, parent_delta(ds, asize + inflight)); 3608 } 3609 3610 if (!check_quota || ds->ds_quota == 0) { 3611 mutex_exit(&ds->ds_lock); 3612 return (0); 3613 } 3614 /* 3615 * If they are requesting more space, and our current estimate 3616 * is over quota, they get to try again unless the actual 3617 * on-disk is over quota and there are no pending changes (which 3618 * may free up space for us). 3619 */ 3620 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= 3621 ds->ds_quota) { 3622 if (inflight > 0 || 3623 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) 3624 error = SET_ERROR(ERESTART); 3625 else 3626 error = SET_ERROR(EDQUOT); 3627 } 3628 mutex_exit(&ds->ds_lock); 3629 3630 return (error); 3631} 3632 3633typedef struct dsl_dataset_set_qr_arg { 3634 const char *ddsqra_name; 3635 zprop_source_t ddsqra_source; 3636 uint64_t ddsqra_value; 3637} dsl_dataset_set_qr_arg_t; 3638 3639 3640/* ARGSUSED */ 3641static int 3642dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 3643{ 3644 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3645 dsl_pool_t *dp = dmu_tx_pool(tx); 3646 dsl_dataset_t *ds; 3647 int error; 3648 uint64_t newval; 3649 3650 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 3651 return (SET_ERROR(ENOTSUP)); 3652 3653 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3654 if (error != 0) 3655 return (error); 3656 3657 if (ds->ds_is_snapshot) { 3658 dsl_dataset_rele(ds, FTAG); 3659 return (SET_ERROR(EINVAL)); 3660 } 3661 3662 error = dsl_prop_predict(ds->ds_dir, 3663 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3664 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3665 if (error != 0) { 3666 dsl_dataset_rele(ds, FTAG); 3667 return (error); 3668 } 3669 3670 if (newval == 0) { 3671 dsl_dataset_rele(ds, FTAG); 3672 return (0); 3673 } 3674 3675 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || 3676 newval < ds->ds_reserved) { 3677 dsl_dataset_rele(ds, FTAG); 3678 return (SET_ERROR(ENOSPC)); 3679 } 3680 3681 dsl_dataset_rele(ds, FTAG); 3682 return (0); 3683} 3684 3685static void 3686dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 3687{ 3688 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3689 dsl_pool_t *dp = dmu_tx_pool(tx); 3690 dsl_dataset_t *ds; 3691 uint64_t newval; 3692 3693 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3694 3695 dsl_prop_set_sync_impl(ds, 3696 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3697 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 3698 &ddsqra->ddsqra_value, tx); 3699 3700 VERIFY0(dsl_prop_get_int_ds(ds, 3701 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 3702 3703 if (ds->ds_quota != newval) { 3704 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3705 ds->ds_quota = newval; 3706 } 3707 dsl_dataset_rele(ds, FTAG); 3708} 3709 3710int 3711dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 3712 uint64_t refquota) 3713{ 3714 dsl_dataset_set_qr_arg_t ddsqra; 3715 3716 ddsqra.ddsqra_name = dsname; 3717 ddsqra.ddsqra_source = source; 3718 ddsqra.ddsqra_value = refquota; 3719 3720 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 3721 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 3722} 3723 3724static int 3725dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 3726{ 3727 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3728 dsl_pool_t *dp = dmu_tx_pool(tx); 3729 dsl_dataset_t *ds; 3730 int error; 3731 uint64_t newval, unique; 3732 3733 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 3734 return (SET_ERROR(ENOTSUP)); 3735 3736 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3737 if (error != 0) 3738 return (error); 3739 3740 if (ds->ds_is_snapshot) { 3741 dsl_dataset_rele(ds, FTAG); 3742 return (SET_ERROR(EINVAL)); 3743 } 3744 3745 error = dsl_prop_predict(ds->ds_dir, 3746 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3747 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3748 if (error != 0) { 3749 dsl_dataset_rele(ds, FTAG); 3750 return (error); 3751 } 3752 3753 /* 3754 * If we are doing the preliminary check in open context, the 3755 * space estimates may be inaccurate. 3756 */ 3757 if (!dmu_tx_is_syncing(tx)) { 3758 dsl_dataset_rele(ds, FTAG); 3759 return (0); 3760 } 3761 3762 mutex_enter(&ds->ds_lock); 3763 if (!DS_UNIQUE_IS_ACCURATE(ds)) 3764 dsl_dataset_recalc_head_uniq(ds); 3765 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3766 mutex_exit(&ds->ds_lock); 3767 3768 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 3769 uint64_t delta = MAX(unique, newval) - 3770 MAX(unique, ds->ds_reserved); 3771 3772 if (delta > 3773 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 3774 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 3775 dsl_dataset_rele(ds, FTAG); 3776 return (SET_ERROR(ENOSPC)); 3777 } 3778 } 3779 3780 dsl_dataset_rele(ds, FTAG); 3781 return (0); 3782} 3783 3784void 3785dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 3786 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 3787{ 3788 uint64_t newval; 3789 uint64_t unique; 3790 int64_t delta; 3791 3792 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3793 source, sizeof (value), 1, &value, tx); 3794 3795 VERIFY0(dsl_prop_get_int_ds(ds, 3796 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 3797 3798 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3799 mutex_enter(&ds->ds_dir->dd_lock); 3800 mutex_enter(&ds->ds_lock); 3801 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3802 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3803 delta = MAX(0, (int64_t)(newval - unique)) - 3804 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3805 ds->ds_reserved = newval; 3806 mutex_exit(&ds->ds_lock); 3807 3808 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3809 mutex_exit(&ds->ds_dir->dd_lock); 3810} 3811 3812static void 3813dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 3814{ 3815 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3816 dsl_pool_t *dp = dmu_tx_pool(tx); 3817 dsl_dataset_t *ds; 3818 3819 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3820 dsl_dataset_set_refreservation_sync_impl(ds, 3821 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 3822 dsl_dataset_rele(ds, FTAG); 3823} 3824 3825int 3826dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 3827 uint64_t refreservation) 3828{ 3829 dsl_dataset_set_qr_arg_t ddsqra; 3830 3831 ddsqra.ddsqra_name = dsname; 3832 ddsqra.ddsqra_source = source; 3833 ddsqra.ddsqra_value = refreservation; 3834 3835 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 3836 dsl_dataset_set_refreservation_sync, &ddsqra, 3837 0, ZFS_SPACE_CHECK_NONE)); 3838} 3839 3840/* 3841 * Return (in *usedp) the amount of space written in new that is not 3842 * present in oldsnap. New may be a snapshot or the head. Old must be 3843 * a snapshot before new, in new's filesystem (or its origin). If not then 3844 * fail and return EINVAL. 3845 * 3846 * The written space is calculated by considering two components: First, we 3847 * ignore any freed space, and calculate the written as new's used space 3848 * minus old's used space. Next, we add in the amount of space that was freed 3849 * between the two snapshots, thus reducing new's used space relative to old's. 3850 * Specifically, this is the space that was born before old->ds_creation_txg, 3851 * and freed before new (ie. on new's deadlist or a previous deadlist). 3852 * 3853 * space freed [---------------------] 3854 * snapshots ---O-------O--------O-------O------ 3855 * oldsnap new 3856 */ 3857int 3858dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 3859 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3860{ 3861 int err = 0; 3862 uint64_t snapobj; 3863 dsl_pool_t *dp = new->ds_dir->dd_pool; 3864 3865 ASSERT(dsl_pool_config_held(dp)); 3866 3867 *usedp = 0; 3868 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; 3869 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes; 3870 3871 *compp = 0; 3872 *compp += dsl_dataset_phys(new)->ds_compressed_bytes; 3873 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes; 3874 3875 *uncompp = 0; 3876 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; 3877 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes; 3878 3879 snapobj = new->ds_object; 3880 while (snapobj != oldsnap->ds_object) { 3881 dsl_dataset_t *snap; 3882 uint64_t used, comp, uncomp; 3883 3884 if (snapobj == new->ds_object) { 3885 snap = new; 3886 } else { 3887 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3888 if (err != 0) 3889 break; 3890 } 3891 3892 if (dsl_dataset_phys(snap)->ds_prev_snap_txg == 3893 dsl_dataset_phys(oldsnap)->ds_creation_txg) { 3894 /* 3895 * The blocks in the deadlist can not be born after 3896 * ds_prev_snap_txg, so get the whole deadlist space, 3897 * which is more efficient (especially for old-format 3898 * deadlists). Unfortunately the deadlist code 3899 * doesn't have enough information to make this 3900 * optimization itself. 3901 */ 3902 dsl_deadlist_space(&snap->ds_deadlist, 3903 &used, &comp, &uncomp); 3904 } else { 3905 dsl_deadlist_space_range(&snap->ds_deadlist, 3906 0, dsl_dataset_phys(oldsnap)->ds_creation_txg, 3907 &used, &comp, &uncomp); 3908 } 3909 *usedp += used; 3910 *compp += comp; 3911 *uncompp += uncomp; 3912 3913 /* 3914 * If we get to the beginning of the chain of snapshots 3915 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3916 * was not a snapshot of/before new. 3917 */ 3918 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 3919 if (snap != new) 3920 dsl_dataset_rele(snap, FTAG); 3921 if (snapobj == 0) { 3922 err = SET_ERROR(EINVAL); 3923 break; 3924 } 3925 3926 } 3927 return (err); 3928} 3929 3930/* 3931 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3932 * lastsnap, and all snapshots in between are deleted. 3933 * 3934 * blocks that would be freed [---------------------------] 3935 * snapshots ---O-------O--------O-------O--------O 3936 * firstsnap lastsnap 3937 * 3938 * This is the set of blocks that were born after the snap before firstsnap, 3939 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3940 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3941 * We calculate this by iterating over the relevant deadlists (from the snap 3942 * after lastsnap, backward to the snap after firstsnap), summing up the 3943 * space on the deadlist that was born after the snap before firstsnap. 3944 */ 3945int 3946dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3947 dsl_dataset_t *lastsnap, 3948 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3949{ 3950 int err = 0; 3951 uint64_t snapobj; 3952 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3953 3954 ASSERT(firstsnap->ds_is_snapshot); 3955 ASSERT(lastsnap->ds_is_snapshot); 3956 3957 /* 3958 * Check that the snapshots are in the same dsl_dir, and firstsnap 3959 * is before lastsnap. 3960 */ 3961 if (firstsnap->ds_dir != lastsnap->ds_dir || 3962 dsl_dataset_phys(firstsnap)->ds_creation_txg > 3963 dsl_dataset_phys(lastsnap)->ds_creation_txg) 3964 return (SET_ERROR(EINVAL)); 3965 3966 *usedp = *compp = *uncompp = 0; 3967 3968 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; 3969 while (snapobj != firstsnap->ds_object) { 3970 dsl_dataset_t *ds; 3971 uint64_t used, comp, uncomp; 3972 3973 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3974 if (err != 0) 3975 break; 3976 3977 dsl_deadlist_space_range(&ds->ds_deadlist, 3978 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, 3979 &used, &comp, &uncomp); 3980 *usedp += used; 3981 *compp += comp; 3982 *uncompp += uncomp; 3983 3984 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3985 ASSERT3U(snapobj, !=, 0); 3986 dsl_dataset_rele(ds, FTAG); 3987 } 3988 return (err); 3989} 3990 3991/* 3992 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 3993 * For example, they could both be snapshots of the same filesystem, and 3994 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 3995 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 3996 * filesystem. Or 'earlier' could be the origin's origin. 3997 * 3998 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 3999 */ 4000boolean_t 4001dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 4002 uint64_t earlier_txg) 4003{ 4004 dsl_pool_t *dp = later->ds_dir->dd_pool; 4005 int error; 4006 boolean_t ret; 4007 4008 ASSERT(dsl_pool_config_held(dp)); 4009 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); 4010 4011 if (earlier_txg == 0) 4012 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; 4013 4014 if (later->ds_is_snapshot && 4015 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) 4016 return (B_FALSE); 4017 4018 if (later->ds_dir == earlier->ds_dir) 4019 return (B_TRUE); 4020 if (!dsl_dir_is_clone(later->ds_dir)) 4021 return (B_FALSE); 4022 4023 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object) 4024 return (B_TRUE); 4025 dsl_dataset_t *origin; 4026 error = dsl_dataset_hold_obj(dp, 4027 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); 4028 if (error != 0) 4029 return (B_FALSE); 4030 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 4031 dsl_dataset_rele(origin, FTAG); 4032 return (ret); 4033} 4034 4035void 4036dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 4037{ 4038 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 4039 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 4040} 4041 4042boolean_t 4043dsl_dataset_is_zapified(dsl_dataset_t *ds) 4044{ 4045 dmu_object_info_t doi; 4046 4047 dmu_object_info_from_db(ds->ds_dbuf, &doi); 4048 return (doi.doi_type == DMU_OTN_ZAP_METADATA); 4049} 4050 4051boolean_t 4052dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds) 4053{ 4054 return (dsl_dataset_is_zapified(ds) && 4055 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset, 4056 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0); 4057} 4058