dsl_dataset.c revision 286575
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org> 24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 25 * Copyright (c) 2014, Joyent, Inc. All rights reserved. 26 * Copyright (c) 2014 RackTop Systems. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 */ 29 30#include <sys/dmu_objset.h> 31#include <sys/dsl_dataset.h> 32#include <sys/dsl_dir.h> 33#include <sys/dsl_prop.h> 34#include <sys/dsl_synctask.h> 35#include <sys/dmu_traverse.h> 36#include <sys/dmu_impl.h> 37#include <sys/dmu_tx.h> 38#include <sys/arc.h> 39#include <sys/zio.h> 40#include <sys/zap.h> 41#include <sys/zfeature.h> 42#include <sys/unique.h> 43#include <sys/zfs_context.h> 44#include <sys/zfs_ioctl.h> 45#include <sys/spa.h> 46#include <sys/zfs_znode.h> 47#include <sys/zfs_onexit.h> 48#include <sys/zvol.h> 49#include <sys/dsl_scan.h> 50#include <sys/dsl_deadlist.h> 51#include <sys/dsl_destroy.h> 52#include <sys/dsl_userhold.h> 53#include <sys/dsl_bookmark.h> 54 55SYSCTL_DECL(_vfs_zfs); 56 57/* 58 * The SPA supports block sizes up to 16MB. However, very large blocks 59 * can have an impact on i/o latency (e.g. tying up a spinning disk for 60 * ~300ms), and also potentially on the memory allocator. Therefore, 61 * we do not allow the recordsize to be set larger than zfs_max_recordsize 62 * (default 1MB). Larger blocks can be created by changing this tunable, 63 * and pools with larger blocks can always be imported and used, regardless 64 * of this setting. 65 */ 66int zfs_max_recordsize = 1 * 1024 * 1024; 67SYSCTL_INT(_vfs_zfs, OID_AUTO, max_recordsize, CTLFLAG_RWTUN, 68 &zfs_max_recordsize, 0, 69 "Maximum block size. Expect dragons when tuning this."); 70 71#define SWITCH64(x, y) \ 72 { \ 73 uint64_t __tmp = (x); \ 74 (x) = (y); \ 75 (y) = __tmp; \ 76 } 77 78#define DS_REF_MAX (1ULL << 62) 79 80extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds); 81 82/* 83 * Figure out how much of this delta should be propogated to the dsl_dir 84 * layer. If there's a refreservation, that space has already been 85 * partially accounted for in our ancestors. 86 */ 87static int64_t 88parent_delta(dsl_dataset_t *ds, int64_t delta) 89{ 90 dsl_dataset_phys_t *ds_phys; 91 uint64_t old_bytes, new_bytes; 92 93 if (ds->ds_reserved == 0) 94 return (delta); 95 96 ds_phys = dsl_dataset_phys(ds); 97 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved); 98 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 99 100 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 101 return (new_bytes - old_bytes); 102} 103 104void 105dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 106{ 107 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 108 int compressed = BP_GET_PSIZE(bp); 109 int uncompressed = BP_GET_UCSIZE(bp); 110 int64_t delta; 111 112 dprintf_bp(bp, "ds=%p", ds); 113 114 ASSERT(dmu_tx_is_syncing(tx)); 115 /* It could have been compressed away to nothing */ 116 if (BP_IS_HOLE(bp)) 117 return; 118 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 119 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 120 if (ds == NULL) { 121 dsl_pool_mos_diduse_space(tx->tx_pool, 122 used, compressed, uncompressed); 123 return; 124 } 125 126 dmu_buf_will_dirty(ds->ds_dbuf, tx); 127 mutex_enter(&ds->ds_lock); 128 delta = parent_delta(ds, used); 129 dsl_dataset_phys(ds)->ds_referenced_bytes += used; 130 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed; 131 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed; 132 dsl_dataset_phys(ds)->ds_unique_bytes += used; 133 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) 134 ds->ds_need_large_blocks = B_TRUE; 135 mutex_exit(&ds->ds_lock); 136 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 137 compressed, uncompressed, tx); 138 dsl_dir_transfer_space(ds->ds_dir, used - delta, 139 DD_USED_REFRSRV, DD_USED_HEAD, NULL); 140} 141 142int 143dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 144 boolean_t async) 145{ 146 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 147 int compressed = BP_GET_PSIZE(bp); 148 int uncompressed = BP_GET_UCSIZE(bp); 149 150 if (BP_IS_HOLE(bp)) 151 return (0); 152 153 ASSERT(dmu_tx_is_syncing(tx)); 154 ASSERT(bp->blk_birth <= tx->tx_txg); 155 156 if (ds == NULL) { 157 dsl_free(tx->tx_pool, tx->tx_txg, bp); 158 dsl_pool_mos_diduse_space(tx->tx_pool, 159 -used, -compressed, -uncompressed); 160 return (used); 161 } 162 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 163 164 ASSERT(!ds->ds_is_snapshot); 165 dmu_buf_will_dirty(ds->ds_dbuf, tx); 166 167 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 168 int64_t delta; 169 170 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 171 dsl_free(tx->tx_pool, tx->tx_txg, bp); 172 173 mutex_enter(&ds->ds_lock); 174 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used || 175 !DS_UNIQUE_IS_ACCURATE(ds)); 176 delta = parent_delta(ds, -used); 177 dsl_dataset_phys(ds)->ds_unique_bytes -= used; 178 mutex_exit(&ds->ds_lock); 179 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 180 delta, -compressed, -uncompressed, tx); 181 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 182 DD_USED_REFRSRV, DD_USED_HEAD, NULL); 183 } else { 184 dprintf_bp(bp, "putting on dead list: %s", ""); 185 if (async) { 186 /* 187 * We are here as part of zio's write done callback, 188 * which means we're a zio interrupt thread. We can't 189 * call dsl_deadlist_insert() now because it may block 190 * waiting for I/O. Instead, put bp on the deferred 191 * queue and let dsl_pool_sync() finish the job. 192 */ 193 bplist_append(&ds->ds_pending_deadlist, bp); 194 } else { 195 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); 196 } 197 ASSERT3U(ds->ds_prev->ds_object, ==, 198 dsl_dataset_phys(ds)->ds_prev_snap_obj); 199 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0); 200 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 201 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 202 ds->ds_object && bp->blk_birth > 203 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) { 204 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 205 mutex_enter(&ds->ds_prev->ds_lock); 206 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used; 207 mutex_exit(&ds->ds_prev->ds_lock); 208 } 209 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 210 dsl_dir_transfer_space(ds->ds_dir, used, 211 DD_USED_HEAD, DD_USED_SNAP, tx); 212 } 213 } 214 mutex_enter(&ds->ds_lock); 215 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used); 216 dsl_dataset_phys(ds)->ds_referenced_bytes -= used; 217 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed); 218 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed; 219 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed); 220 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed; 221 mutex_exit(&ds->ds_lock); 222 223 return (used); 224} 225 226uint64_t 227dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 228{ 229 uint64_t trysnap = 0; 230 231 if (ds == NULL) 232 return (0); 233 /* 234 * The snapshot creation could fail, but that would cause an 235 * incorrect FALSE return, which would only result in an 236 * overestimation of the amount of space that an operation would 237 * consume, which is OK. 238 * 239 * There's also a small window where we could miss a pending 240 * snapshot, because we could set the sync task in the quiescing 241 * phase. So this should only be used as a guess. 242 */ 243 if (ds->ds_trysnap_txg > 244 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 245 trysnap = ds->ds_trysnap_txg; 246 return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap)); 247} 248 249boolean_t 250dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp, 251 uint64_t blk_birth) 252{ 253 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) || 254 (bp != NULL && BP_IS_HOLE(bp))) 255 return (B_FALSE); 256 257 ddt_prefetch(dsl_dataset_get_spa(ds), bp); 258 259 return (B_TRUE); 260} 261 262static void 263dsl_dataset_evict(void *dbu) 264{ 265 dsl_dataset_t *ds = dbu; 266 267 ASSERT(ds->ds_owner == NULL); 268 269 ds->ds_dbuf = NULL; 270 271 unique_remove(ds->ds_fsid_guid); 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 if (mutex_owned(&ds->ds_lock)) 290 mutex_exit(&ds->ds_lock); 291 mutex_destroy(&ds->ds_lock); 292 if (mutex_owned(&ds->ds_opening_lock)) 293 mutex_exit(&ds->ds_opening_lock); 294 mutex_destroy(&ds->ds_opening_lock); 295 mutex_destroy(&ds->ds_sendstream_lock); 296 refcount_destroy(&ds->ds_longholds); 297 298 kmem_free(ds, sizeof (dsl_dataset_t)); 299} 300 301int 302dsl_dataset_get_snapname(dsl_dataset_t *ds) 303{ 304 dsl_dataset_phys_t *headphys; 305 int err; 306 dmu_buf_t *headdbuf; 307 dsl_pool_t *dp = ds->ds_dir->dd_pool; 308 objset_t *mos = dp->dp_meta_objset; 309 310 if (ds->ds_snapname[0]) 311 return (0); 312 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) 313 return (0); 314 315 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, 316 FTAG, &headdbuf); 317 if (err != 0) 318 return (err); 319 headphys = headdbuf->db_data; 320 err = zap_value_search(dp->dp_meta_objset, 321 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 322 dmu_buf_rele(headdbuf, FTAG); 323 return (err); 324} 325 326int 327dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 328{ 329 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 330 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 331 matchtype_t mt; 332 int err; 333 334 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 335 mt = MT_FIRST; 336 else 337 mt = MT_EXACT; 338 339 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 340 value, mt, NULL, 0, NULL); 341 if (err == ENOTSUP && mt == MT_FIRST) 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; 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_FIRST; 359 else 360 mt = MT_EXACT; 361 362 err = zap_remove_norm(mos, snapobj, name, mt, tx); 363 if (err == ENOTSUP && mt == MT_FIRST) 364 err = zap_remove(mos, snapobj, name, tx); 365 366 if (err == 0 && adj_cnt) 367 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 368 DD_FIELD_SNAPSHOT_COUNT, tx); 369 370 return (err); 371} 372 373boolean_t 374dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag) 375{ 376 dmu_buf_t *dbuf = ds->ds_dbuf; 377 boolean_t result = B_FALSE; 378 379 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset, 380 ds->ds_object, DMU_BONUS_BLKID, tag)) { 381 382 if (ds == dmu_buf_get_user(dbuf)) 383 result = B_TRUE; 384 else 385 dmu_buf_rele(dbuf, tag); 386 } 387 388 return (result); 389} 390 391int 392dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 393 dsl_dataset_t **dsp) 394{ 395 objset_t *mos = dp->dp_meta_objset; 396 dmu_buf_t *dbuf; 397 dsl_dataset_t *ds; 398 int err; 399 dmu_object_info_t doi; 400 401 ASSERT(dsl_pool_config_held(dp)); 402 403 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 404 if (err != 0) 405 return (err); 406 407 /* Make sure dsobj has the correct object type. */ 408 dmu_object_info_from_db(dbuf, &doi); 409 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 410 dmu_buf_rele(dbuf, tag); 411 return (SET_ERROR(EINVAL)); 412 } 413 414 ds = dmu_buf_get_user(dbuf); 415 if (ds == NULL) { 416 dsl_dataset_t *winner = NULL; 417 418 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 419 ds->ds_dbuf = dbuf; 420 ds->ds_object = dsobj; 421 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0; 422 423 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 424 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 425 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 426 refcount_create(&ds->ds_longholds); 427 428 bplist_create(&ds->ds_pending_deadlist); 429 dsl_deadlist_open(&ds->ds_deadlist, 430 mos, dsl_dataset_phys(ds)->ds_deadlist_obj); 431 432 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 433 offsetof(dmu_sendarg_t, dsa_link)); 434 435 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 436 int zaperr = zap_contains(mos, dsobj, 437 DS_FIELD_LARGE_BLOCKS); 438 if (zaperr != ENOENT) { 439 VERIFY0(zaperr); 440 ds->ds_large_blocks = B_TRUE; 441 } 442 } 443 444 if (err == 0) { 445 err = dsl_dir_hold_obj(dp, 446 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, 447 &ds->ds_dir); 448 } 449 if (err != 0) { 450 mutex_destroy(&ds->ds_lock); 451 mutex_destroy(&ds->ds_opening_lock); 452 mutex_destroy(&ds->ds_sendstream_lock); 453 refcount_destroy(&ds->ds_longholds); 454 bplist_destroy(&ds->ds_pending_deadlist); 455 dsl_deadlist_close(&ds->ds_deadlist); 456 kmem_free(ds, sizeof (dsl_dataset_t)); 457 dmu_buf_rele(dbuf, tag); 458 return (err); 459 } 460 461 if (!ds->ds_is_snapshot) { 462 ds->ds_snapname[0] = '\0'; 463 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 464 err = dsl_dataset_hold_obj(dp, 465 dsl_dataset_phys(ds)->ds_prev_snap_obj, 466 ds, &ds->ds_prev); 467 } 468 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 469 int zaperr = zap_lookup(mos, ds->ds_object, 470 DS_FIELD_BOOKMARK_NAMES, 471 sizeof (ds->ds_bookmarks), 1, 472 &ds->ds_bookmarks); 473 if (zaperr != ENOENT) 474 VERIFY0(zaperr); 475 } 476 } else { 477 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 478 err = dsl_dataset_get_snapname(ds); 479 if (err == 0 && 480 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { 481 err = zap_count( 482 ds->ds_dir->dd_pool->dp_meta_objset, 483 dsl_dataset_phys(ds)->ds_userrefs_obj, 484 &ds->ds_userrefs); 485 } 486 } 487 488 if (err == 0 && !ds->ds_is_snapshot) { 489 err = dsl_prop_get_int_ds(ds, 490 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 491 &ds->ds_reserved); 492 if (err == 0) { 493 err = dsl_prop_get_int_ds(ds, 494 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 495 &ds->ds_quota); 496 } 497 } else { 498 ds->ds_reserved = ds->ds_quota = 0; 499 } 500 501 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf); 502 if (err == 0) 503 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu); 504 505 if (err != 0 || winner != NULL) { 506 bplist_destroy(&ds->ds_pending_deadlist); 507 dsl_deadlist_close(&ds->ds_deadlist); 508 if (ds->ds_prev) 509 dsl_dataset_rele(ds->ds_prev, ds); 510 dsl_dir_rele(ds->ds_dir, ds); 511 mutex_destroy(&ds->ds_lock); 512 mutex_destroy(&ds->ds_opening_lock); 513 mutex_destroy(&ds->ds_sendstream_lock); 514 refcount_destroy(&ds->ds_longholds); 515 kmem_free(ds, sizeof (dsl_dataset_t)); 516 if (err != 0) { 517 dmu_buf_rele(dbuf, tag); 518 return (err); 519 } 520 ds = winner; 521 } else { 522 ds->ds_fsid_guid = 523 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); 524 } 525 } 526 ASSERT3P(ds->ds_dbuf, ==, dbuf); 527 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); 528 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || 529 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 530 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 531 *dsp = ds; 532 return (0); 533} 534 535int 536dsl_dataset_hold(dsl_pool_t *dp, const char *name, 537 void *tag, dsl_dataset_t **dsp) 538{ 539 dsl_dir_t *dd; 540 const char *snapname; 541 uint64_t obj; 542 int err = 0; 543 544 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 545 if (err != 0) 546 return (err); 547 548 ASSERT(dsl_pool_config_held(dp)); 549 obj = dsl_dir_phys(dd)->dd_head_dataset_obj; 550 if (obj != 0) 551 err = dsl_dataset_hold_obj(dp, obj, tag, dsp); 552 else 553 err = SET_ERROR(ENOENT); 554 555 /* we may be looking for a snapshot */ 556 if (err == 0 && snapname != NULL) { 557 dsl_dataset_t *ds; 558 559 if (*snapname++ != '@') { 560 dsl_dataset_rele(*dsp, tag); 561 dsl_dir_rele(dd, FTAG); 562 return (SET_ERROR(ENOENT)); 563 } 564 565 dprintf("looking for snapshot '%s'\n", snapname); 566 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); 567 if (err == 0) 568 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 569 dsl_dataset_rele(*dsp, tag); 570 571 if (err == 0) { 572 mutex_enter(&ds->ds_lock); 573 if (ds->ds_snapname[0] == 0) 574 (void) strlcpy(ds->ds_snapname, snapname, 575 sizeof (ds->ds_snapname)); 576 mutex_exit(&ds->ds_lock); 577 *dsp = ds; 578 } 579 } 580 581 dsl_dir_rele(dd, FTAG); 582 return (err); 583} 584 585int 586dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 587 void *tag, dsl_dataset_t **dsp) 588{ 589 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 590 if (err != 0) 591 return (err); 592 if (!dsl_dataset_tryown(*dsp, tag)) { 593 dsl_dataset_rele(*dsp, tag); 594 *dsp = NULL; 595 return (SET_ERROR(EBUSY)); 596 } 597 return (0); 598} 599 600int 601dsl_dataset_own(dsl_pool_t *dp, const char *name, 602 void *tag, dsl_dataset_t **dsp) 603{ 604 int err = dsl_dataset_hold(dp, name, tag, dsp); 605 if (err != 0) 606 return (err); 607 if (!dsl_dataset_tryown(*dsp, tag)) { 608 dsl_dataset_rele(*dsp, tag); 609 return (SET_ERROR(EBUSY)); 610 } 611 return (0); 612} 613 614/* 615 * See the comment above dsl_pool_hold() for details. In summary, a long 616 * hold is used to prevent destruction of a dataset while the pool hold 617 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 618 * 619 * The dataset and pool must be held when this function is called. After it 620 * is called, the pool hold may be released while the dataset is still held 621 * and accessed. 622 */ 623void 624dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 625{ 626 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 627 (void) refcount_add(&ds->ds_longholds, tag); 628} 629 630void 631dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 632{ 633 (void) refcount_remove(&ds->ds_longholds, tag); 634} 635 636/* Return B_TRUE if there are any long holds on this dataset. */ 637boolean_t 638dsl_dataset_long_held(dsl_dataset_t *ds) 639{ 640 return (!refcount_is_zero(&ds->ds_longholds)); 641} 642 643void 644dsl_dataset_name(dsl_dataset_t *ds, char *name) 645{ 646 if (ds == NULL) { 647 (void) strcpy(name, "mos"); 648 } else { 649 dsl_dir_name(ds->ds_dir, name); 650 VERIFY0(dsl_dataset_get_snapname(ds)); 651 if (ds->ds_snapname[0]) { 652 (void) strcat(name, "@"); 653 /* 654 * We use a "recursive" mutex so that we 655 * can call dprintf_ds() with ds_lock held. 656 */ 657 if (!MUTEX_HELD(&ds->ds_lock)) { 658 mutex_enter(&ds->ds_lock); 659 (void) strcat(name, ds->ds_snapname); 660 mutex_exit(&ds->ds_lock); 661 } else { 662 (void) strcat(name, ds->ds_snapname); 663 } 664 } 665 } 666} 667 668void 669dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 670{ 671 dmu_buf_rele(ds->ds_dbuf, tag); 672} 673 674void 675dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 676{ 677 ASSERT3P(ds->ds_owner, ==, tag); 678 ASSERT(ds->ds_dbuf != NULL); 679 680 mutex_enter(&ds->ds_lock); 681 ds->ds_owner = NULL; 682 mutex_exit(&ds->ds_lock); 683 dsl_dataset_long_rele(ds, tag); 684 dsl_dataset_rele(ds, tag); 685} 686 687boolean_t 688dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 689{ 690 boolean_t gotit = FALSE; 691 692 mutex_enter(&ds->ds_lock); 693 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 694 ds->ds_owner = tag; 695 dsl_dataset_long_hold(ds, tag); 696 gotit = TRUE; 697 } 698 mutex_exit(&ds->ds_lock); 699 return (gotit); 700} 701 702uint64_t 703dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 704 uint64_t flags, dmu_tx_t *tx) 705{ 706 dsl_pool_t *dp = dd->dd_pool; 707 dmu_buf_t *dbuf; 708 dsl_dataset_phys_t *dsphys; 709 uint64_t dsobj; 710 objset_t *mos = dp->dp_meta_objset; 711 712 if (origin == NULL) 713 origin = dp->dp_origin_snap; 714 715 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 716 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); 717 ASSERT(dmu_tx_is_syncing(tx)); 718 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 719 720 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 721 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 722 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 723 dmu_buf_will_dirty(dbuf, tx); 724 dsphys = dbuf->db_data; 725 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 726 dsphys->ds_dir_obj = dd->dd_object; 727 dsphys->ds_flags = flags; 728 dsphys->ds_fsid_guid = unique_create(); 729 do { 730 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 731 sizeof (dsphys->ds_guid)); 732 } while (dsphys->ds_guid == 0); 733 dsphys->ds_snapnames_zapobj = 734 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 735 DMU_OT_NONE, 0, tx); 736 dsphys->ds_creation_time = gethrestime_sec(); 737 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 738 739 if (origin == NULL) { 740 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 741 } else { 742 dsl_dataset_t *ohds; /* head of the origin snapshot */ 743 744 dsphys->ds_prev_snap_obj = origin->ds_object; 745 dsphys->ds_prev_snap_txg = 746 dsl_dataset_phys(origin)->ds_creation_txg; 747 dsphys->ds_referenced_bytes = 748 dsl_dataset_phys(origin)->ds_referenced_bytes; 749 dsphys->ds_compressed_bytes = 750 dsl_dataset_phys(origin)->ds_compressed_bytes; 751 dsphys->ds_uncompressed_bytes = 752 dsl_dataset_phys(origin)->ds_uncompressed_bytes; 753 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; 754 755 /* 756 * Inherit flags that describe the dataset's contents 757 * (INCONSISTENT) or properties (Case Insensitive). 758 */ 759 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & 760 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 761 762 if (origin->ds_large_blocks) 763 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); 764 765 dmu_buf_will_dirty(origin->ds_dbuf, tx); 766 dsl_dataset_phys(origin)->ds_num_children++; 767 768 VERIFY0(dsl_dataset_hold_obj(dp, 769 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, 770 FTAG, &ohds)); 771 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 772 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 773 dsl_dataset_rele(ohds, FTAG); 774 775 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 776 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { 777 dsl_dataset_phys(origin)->ds_next_clones_obj = 778 zap_create(mos, 779 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 780 } 781 VERIFY0(zap_add_int(mos, 782 dsl_dataset_phys(origin)->ds_next_clones_obj, 783 dsobj, tx)); 784 } 785 786 dmu_buf_will_dirty(dd->dd_dbuf, tx); 787 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; 788 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 789 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 790 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 791 dsl_dir_phys(origin->ds_dir)->dd_clones = 792 zap_create(mos, 793 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 794 } 795 VERIFY0(zap_add_int(mos, 796 dsl_dir_phys(origin->ds_dir)->dd_clones, 797 dsobj, tx)); 798 } 799 } 800 801 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 802 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 803 804 dmu_buf_rele(dbuf, FTAG); 805 806 dmu_buf_will_dirty(dd->dd_dbuf, tx); 807 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; 808 809 return (dsobj); 810} 811 812static void 813dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 814{ 815 objset_t *os; 816 817 VERIFY0(dmu_objset_from_ds(ds, &os)); 818 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 819 dsl_dataset_dirty(ds, tx); 820} 821 822uint64_t 823dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 824 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 825{ 826 dsl_pool_t *dp = pdd->dd_pool; 827 uint64_t dsobj, ddobj; 828 dsl_dir_t *dd; 829 830 ASSERT(dmu_tx_is_syncing(tx)); 831 ASSERT(lastname[0] != '@'); 832 833 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 834 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 835 836 dsobj = dsl_dataset_create_sync_dd(dd, origin, 837 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 838 839 dsl_deleg_set_create_perms(dd, tx, cr); 840 841 /* 842 * Since we're creating a new node we know it's a leaf, so we can 843 * initialize the counts if the limit feature is active. 844 */ 845 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 846 uint64_t cnt = 0; 847 objset_t *os = dd->dd_pool->dp_meta_objset; 848 849 dsl_dir_zapify(dd, tx); 850 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 851 sizeof (cnt), 1, &cnt, tx)); 852 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 853 sizeof (cnt), 1, &cnt, tx)); 854 } 855 856 dsl_dir_rele(dd, FTAG); 857 858 /* 859 * If we are creating a clone, make sure we zero out any stale 860 * data from the origin snapshots zil header. 861 */ 862 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 863 dsl_dataset_t *ds; 864 865 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 866 dsl_dataset_zero_zil(ds, tx); 867 dsl_dataset_rele(ds, FTAG); 868 } 869 870 return (dsobj); 871} 872 873#ifdef __FreeBSD__ 874/* FreeBSD ioctl compat begin */ 875struct destroyarg { 876 nvlist_t *nvl; 877 const char *snapname; 878}; 879 880static int 881dsl_check_snap_cb(const char *name, void *arg) 882{ 883 struct destroyarg *da = arg; 884 dsl_dataset_t *ds; 885 char *dsname; 886 887 dsname = kmem_asprintf("%s@%s", name, da->snapname); 888 fnvlist_add_boolean(da->nvl, dsname); 889 kmem_free(dsname, strlen(dsname) + 1); 890 891 return (0); 892} 893 894int 895dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname, 896 nvlist_t *snaps) 897{ 898 struct destroyarg *da; 899 int err; 900 901 da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP); 902 da->nvl = snaps; 903 da->snapname = snapname; 904 err = dmu_objset_find(fsname, dsl_check_snap_cb, da, 905 DS_FIND_CHILDREN); 906 kmem_free(da, sizeof (struct destroyarg)); 907 908 return (err); 909} 910/* FreeBSD ioctl compat end */ 911#endif /* __FreeBSD__ */ 912 913/* 914 * The unique space in the head dataset can be calculated by subtracting 915 * the space used in the most recent snapshot, that is still being used 916 * in this file system, from the space currently in use. To figure out 917 * the space in the most recent snapshot still in use, we need to take 918 * the total space used in the snapshot and subtract out the space that 919 * has been freed up since the snapshot was taken. 920 */ 921void 922dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 923{ 924 uint64_t mrs_used; 925 uint64_t dlused, dlcomp, dluncomp; 926 927 ASSERT(!ds->ds_is_snapshot); 928 929 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) 930 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; 931 else 932 mrs_used = 0; 933 934 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 935 936 ASSERT3U(dlused, <=, mrs_used); 937 dsl_dataset_phys(ds)->ds_unique_bytes = 938 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); 939 940 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 941 SPA_VERSION_UNIQUE_ACCURATE) 942 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 943} 944 945void 946dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 947 dmu_tx_t *tx) 948{ 949 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 950 uint64_t count; 951 int err; 952 953 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); 954 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 955 obj, tx); 956 /* 957 * The err should not be ENOENT, but a bug in a previous version 958 * of the code could cause upgrade_clones_cb() to not set 959 * ds_next_snap_obj when it should, leading to a missing entry. 960 * If we knew that the pool was created after 961 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 962 * ENOENT. However, at least we can check that we don't have 963 * too many entries in the next_clones_obj even after failing to 964 * remove this one. 965 */ 966 if (err != ENOENT) 967 VERIFY0(err); 968 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 969 &count)); 970 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); 971} 972 973 974blkptr_t * 975dsl_dataset_get_blkptr(dsl_dataset_t *ds) 976{ 977 return (&dsl_dataset_phys(ds)->ds_bp); 978} 979 980void 981dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 982{ 983 ASSERT(dmu_tx_is_syncing(tx)); 984 /* If it's the meta-objset, set dp_meta_rootbp */ 985 if (ds == NULL) { 986 tx->tx_pool->dp_meta_rootbp = *bp; 987 } else { 988 dmu_buf_will_dirty(ds->ds_dbuf, tx); 989 dsl_dataset_phys(ds)->ds_bp = *bp; 990 } 991} 992 993spa_t * 994dsl_dataset_get_spa(dsl_dataset_t *ds) 995{ 996 return (ds->ds_dir->dd_pool->dp_spa); 997} 998 999void 1000dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1001{ 1002 dsl_pool_t *dp; 1003 1004 if (ds == NULL) /* this is the meta-objset */ 1005 return; 1006 1007 ASSERT(ds->ds_objset != NULL); 1008 1009 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) 1010 panic("dirtying snapshot!"); 1011 1012 dp = ds->ds_dir->dd_pool; 1013 1014 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1015 /* up the hold count until we can be written out */ 1016 dmu_buf_add_ref(ds->ds_dbuf, ds); 1017 } 1018} 1019 1020boolean_t 1021dsl_dataset_is_dirty(dsl_dataset_t *ds) 1022{ 1023 for (int t = 0; t < TXG_SIZE; t++) { 1024 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1025 ds, t)) 1026 return (B_TRUE); 1027 } 1028 return (B_FALSE); 1029} 1030 1031static int 1032dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1033{ 1034 uint64_t asize; 1035 1036 if (!dmu_tx_is_syncing(tx)) 1037 return (0); 1038 1039 /* 1040 * If there's an fs-only reservation, any blocks that might become 1041 * owned by the snapshot dataset must be accommodated by space 1042 * outside of the reservation. 1043 */ 1044 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1045 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); 1046 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1047 return (SET_ERROR(ENOSPC)); 1048 1049 /* 1050 * Propagate any reserved space for this snapshot to other 1051 * snapshot checks in this sync group. 1052 */ 1053 if (asize > 0) 1054 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1055 1056 return (0); 1057} 1058 1059typedef struct dsl_dataset_snapshot_arg { 1060 nvlist_t *ddsa_snaps; 1061 nvlist_t *ddsa_props; 1062 nvlist_t *ddsa_errors; 1063 cred_t *ddsa_cr; 1064} dsl_dataset_snapshot_arg_t; 1065 1066int 1067dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1068 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 1069{ 1070 int error; 1071 uint64_t value; 1072 1073 ds->ds_trysnap_txg = tx->tx_txg; 1074 1075 if (!dmu_tx_is_syncing(tx)) 1076 return (0); 1077 1078 /* 1079 * We don't allow multiple snapshots of the same txg. If there 1080 * is already one, try again. 1081 */ 1082 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) 1083 return (SET_ERROR(EAGAIN)); 1084 1085 /* 1086 * Check for conflicting snapshot name. 1087 */ 1088 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1089 if (error == 0) 1090 return (SET_ERROR(EEXIST)); 1091 if (error != ENOENT) 1092 return (error); 1093 1094 /* 1095 * We don't allow taking snapshots of inconsistent datasets, such as 1096 * those into which we are currently receiving. However, if we are 1097 * creating this snapshot as part of a receive, this check will be 1098 * executed atomically with respect to the completion of the receive 1099 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1100 * case we ignore this, knowing it will be fixed up for us shortly in 1101 * dmu_recv_end_sync(). 1102 */ 1103 if (!recv && DS_IS_INCONSISTENT(ds)) 1104 return (SET_ERROR(EBUSY)); 1105 1106 /* 1107 * Skip the check for temporary snapshots or if we have already checked 1108 * the counts in dsl_dataset_snapshot_check. This means we really only 1109 * check the count here when we're receiving a stream. 1110 */ 1111 if (cnt != 0 && cr != NULL) { 1112 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1113 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 1114 if (error != 0) 1115 return (error); 1116 } 1117 1118 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1119 if (error != 0) 1120 return (error); 1121 1122 return (0); 1123} 1124 1125static int 1126dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1127{ 1128 dsl_dataset_snapshot_arg_t *ddsa = arg; 1129 dsl_pool_t *dp = dmu_tx_pool(tx); 1130 nvpair_t *pair; 1131 int rv = 0; 1132 1133 /* 1134 * Pre-compute how many total new snapshots will be created for each 1135 * level in the tree and below. This is needed for validating the 1136 * snapshot limit when either taking a recursive snapshot or when 1137 * taking multiple snapshots. 1138 * 1139 * The problem is that the counts are not actually adjusted when 1140 * we are checking, only when we finally sync. For a single snapshot, 1141 * this is easy, the count will increase by 1 at each node up the tree, 1142 * but its more complicated for the recursive/multiple snapshot case. 1143 * 1144 * The dsl_fs_ss_limit_check function does recursively check the count 1145 * at each level up the tree but since it is validating each snapshot 1146 * independently we need to be sure that we are validating the complete 1147 * count for the entire set of snapshots. We do this by rolling up the 1148 * counts for each component of the name into an nvlist and then 1149 * checking each of those cases with the aggregated count. 1150 * 1151 * This approach properly handles not only the recursive snapshot 1152 * case (where we get all of those on the ddsa_snaps list) but also 1153 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1154 * validate the limit on 'a' using a count of 2). 1155 * 1156 * We validate the snapshot names in the third loop and only report 1157 * name errors once. 1158 */ 1159 if (dmu_tx_is_syncing(tx)) { 1160 nvlist_t *cnt_track = NULL; 1161 cnt_track = fnvlist_alloc(); 1162 1163 /* Rollup aggregated counts into the cnt_track list */ 1164 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1165 pair != NULL; 1166 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1167 char *pdelim; 1168 uint64_t val; 1169 char nm[MAXPATHLEN]; 1170 1171 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1172 pdelim = strchr(nm, '@'); 1173 if (pdelim == NULL) 1174 continue; 1175 *pdelim = '\0'; 1176 1177 do { 1178 if (nvlist_lookup_uint64(cnt_track, nm, 1179 &val) == 0) { 1180 /* update existing entry */ 1181 fnvlist_add_uint64(cnt_track, nm, 1182 val + 1); 1183 } else { 1184 /* add to list */ 1185 fnvlist_add_uint64(cnt_track, nm, 1); 1186 } 1187 1188 pdelim = strrchr(nm, '/'); 1189 if (pdelim != NULL) 1190 *pdelim = '\0'; 1191 } while (pdelim != NULL); 1192 } 1193 1194 /* Check aggregated counts at each level */ 1195 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1196 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1197 int error = 0; 1198 char *name; 1199 uint64_t cnt = 0; 1200 dsl_dataset_t *ds; 1201 1202 name = nvpair_name(pair); 1203 cnt = fnvpair_value_uint64(pair); 1204 ASSERT(cnt > 0); 1205 1206 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1207 if (error == 0) { 1208 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1209 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1210 ddsa->ddsa_cr); 1211 dsl_dataset_rele(ds, FTAG); 1212 } 1213 1214 if (error != 0) { 1215 if (ddsa->ddsa_errors != NULL) 1216 fnvlist_add_int32(ddsa->ddsa_errors, 1217 name, error); 1218 rv = error; 1219 /* only report one error for this check */ 1220 break; 1221 } 1222 } 1223 nvlist_free(cnt_track); 1224 } 1225 1226 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1227 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1228 int error = 0; 1229 dsl_dataset_t *ds; 1230 char *name, *atp; 1231 char dsname[MAXNAMELEN]; 1232 1233 name = nvpair_name(pair); 1234 if (strlen(name) >= MAXNAMELEN) 1235 error = SET_ERROR(ENAMETOOLONG); 1236 if (error == 0) { 1237 atp = strchr(name, '@'); 1238 if (atp == NULL) 1239 error = SET_ERROR(EINVAL); 1240 if (error == 0) 1241 (void) strlcpy(dsname, name, atp - name + 1); 1242 } 1243 if (error == 0) 1244 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1245 if (error == 0) { 1246 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1247 error = dsl_dataset_snapshot_check_impl(ds, 1248 atp + 1, tx, B_FALSE, 0, NULL); 1249 dsl_dataset_rele(ds, FTAG); 1250 } 1251 1252 if (error != 0) { 1253 if (ddsa->ddsa_errors != NULL) { 1254 fnvlist_add_int32(ddsa->ddsa_errors, 1255 name, error); 1256 } 1257 rv = error; 1258 } 1259 } 1260 1261 return (rv); 1262} 1263 1264void 1265dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1266 dmu_tx_t *tx) 1267{ 1268 static zil_header_t zero_zil; 1269 1270 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1271 dmu_buf_t *dbuf; 1272 dsl_dataset_phys_t *dsphys; 1273 uint64_t dsobj, crtxg; 1274 objset_t *mos = dp->dp_meta_objset; 1275 objset_t *os; 1276 1277 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1278 1279 /* 1280 * If we are on an old pool, the zil must not be active, in which 1281 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1282 */ 1283 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1284 dmu_objset_from_ds(ds, &os) != 0 || 1285 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1286 sizeof (zero_zil)) == 0); 1287 1288 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1289 1290 /* 1291 * The origin's ds_creation_txg has to be < TXG_INITIAL 1292 */ 1293 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1294 crtxg = 1; 1295 else 1296 crtxg = tx->tx_txg; 1297 1298 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1299 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1300 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1301 dmu_buf_will_dirty(dbuf, tx); 1302 dsphys = dbuf->db_data; 1303 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1304 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1305 dsphys->ds_fsid_guid = unique_create(); 1306 do { 1307 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1308 sizeof (dsphys->ds_guid)); 1309 } while (dsphys->ds_guid == 0); 1310 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1311 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 1312 dsphys->ds_next_snap_obj = ds->ds_object; 1313 dsphys->ds_num_children = 1; 1314 dsphys->ds_creation_time = gethrestime_sec(); 1315 dsphys->ds_creation_txg = crtxg; 1316 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; 1317 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; 1318 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; 1319 dsphys->ds_uncompressed_bytes = 1320 dsl_dataset_phys(ds)->ds_uncompressed_bytes; 1321 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; 1322 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; 1323 dmu_buf_rele(dbuf, FTAG); 1324 1325 if (ds->ds_large_blocks) 1326 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); 1327 1328 ASSERT3U(ds->ds_prev != 0, ==, 1329 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); 1330 if (ds->ds_prev) { 1331 uint64_t next_clones_obj = 1332 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; 1333 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1334 ds->ds_object || 1335 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); 1336 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1337 ds->ds_object) { 1338 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1339 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, 1340 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); 1341 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; 1342 } else if (next_clones_obj != 0) { 1343 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1344 dsphys->ds_next_snap_obj, tx); 1345 VERIFY0(zap_add_int(mos, 1346 next_clones_obj, dsobj, tx)); 1347 } 1348 } 1349 1350 /* 1351 * If we have a reference-reservation on this dataset, we will 1352 * need to increase the amount of refreservation being charged 1353 * since our unique space is going to zero. 1354 */ 1355 if (ds->ds_reserved) { 1356 int64_t delta; 1357 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1358 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, 1359 ds->ds_reserved); 1360 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1361 delta, 0, 0, tx); 1362 } 1363 1364 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1365 dsl_dataset_phys(ds)->ds_deadlist_obj = 1366 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, 1367 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 1368 dsl_deadlist_close(&ds->ds_deadlist); 1369 dsl_deadlist_open(&ds->ds_deadlist, mos, 1370 dsl_dataset_phys(ds)->ds_deadlist_obj); 1371 dsl_deadlist_add_key(&ds->ds_deadlist, 1372 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); 1373 1374 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); 1375 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; 1376 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; 1377 dsl_dataset_phys(ds)->ds_unique_bytes = 0; 1378 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1379 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1380 1381 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1382 snapname, 8, 1, &dsobj, tx)); 1383 1384 if (ds->ds_prev) 1385 dsl_dataset_rele(ds->ds_prev, ds); 1386 VERIFY0(dsl_dataset_hold_obj(dp, 1387 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); 1388 1389 dsl_scan_ds_snapshotted(ds, tx); 1390 1391 dsl_dir_snap_cmtime_update(ds->ds_dir); 1392 1393 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1394} 1395 1396static void 1397dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1398{ 1399 dsl_dataset_snapshot_arg_t *ddsa = arg; 1400 dsl_pool_t *dp = dmu_tx_pool(tx); 1401 nvpair_t *pair; 1402 1403 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1404 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1405 dsl_dataset_t *ds; 1406 char *name, *atp; 1407 char dsname[MAXNAMELEN]; 1408 1409 name = nvpair_name(pair); 1410 atp = strchr(name, '@'); 1411 (void) strlcpy(dsname, name, atp - name + 1); 1412 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1413 1414 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1415 if (ddsa->ddsa_props != NULL) { 1416 dsl_props_set_sync_impl(ds->ds_prev, 1417 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1418 } 1419 dsl_dataset_rele(ds, FTAG); 1420 } 1421} 1422 1423/* 1424 * The snapshots must all be in the same pool. 1425 * All-or-nothing: if there are any failures, nothing will be modified. 1426 */ 1427int 1428dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1429{ 1430 dsl_dataset_snapshot_arg_t ddsa; 1431 nvpair_t *pair; 1432 boolean_t needsuspend; 1433 int error; 1434 spa_t *spa; 1435 char *firstname; 1436 nvlist_t *suspended = NULL; 1437 1438 pair = nvlist_next_nvpair(snaps, NULL); 1439 if (pair == NULL) 1440 return (0); 1441 firstname = nvpair_name(pair); 1442 1443 error = spa_open(firstname, &spa, FTAG); 1444 if (error != 0) 1445 return (error); 1446 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1447 spa_close(spa, FTAG); 1448 1449 if (needsuspend) { 1450 suspended = fnvlist_alloc(); 1451 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1452 pair = nvlist_next_nvpair(snaps, pair)) { 1453 char fsname[MAXNAMELEN]; 1454 char *snapname = nvpair_name(pair); 1455 char *atp; 1456 void *cookie; 1457 1458 atp = strchr(snapname, '@'); 1459 if (atp == NULL) { 1460 error = SET_ERROR(EINVAL); 1461 break; 1462 } 1463 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1464 1465 error = zil_suspend(fsname, &cookie); 1466 if (error != 0) 1467 break; 1468 fnvlist_add_uint64(suspended, fsname, 1469 (uintptr_t)cookie); 1470 } 1471 } 1472 1473 ddsa.ddsa_snaps = snaps; 1474 ddsa.ddsa_props = props; 1475 ddsa.ddsa_errors = errors; 1476 ddsa.ddsa_cr = CRED(); 1477 1478 if (error == 0) { 1479 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1480 dsl_dataset_snapshot_sync, &ddsa, 1481 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1482 } 1483 1484 if (suspended != NULL) { 1485 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1486 pair = nvlist_next_nvpair(suspended, pair)) { 1487 zil_resume((void *)(uintptr_t) 1488 fnvpair_value_uint64(pair)); 1489 } 1490 fnvlist_free(suspended); 1491 } 1492 1493#ifdef __FreeBSD__ 1494#ifdef _KERNEL 1495 if (error == 0) { 1496 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1497 pair = nvlist_next_nvpair(snaps, pair)) { 1498 char *snapname = nvpair_name(pair); 1499 zvol_create_minors(snapname); 1500 } 1501 } 1502#endif 1503#endif 1504 return (error); 1505} 1506 1507typedef struct dsl_dataset_snapshot_tmp_arg { 1508 const char *ddsta_fsname; 1509 const char *ddsta_snapname; 1510 minor_t ddsta_cleanup_minor; 1511 const char *ddsta_htag; 1512} dsl_dataset_snapshot_tmp_arg_t; 1513 1514static int 1515dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1516{ 1517 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1518 dsl_pool_t *dp = dmu_tx_pool(tx); 1519 dsl_dataset_t *ds; 1520 int error; 1521 1522 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1523 if (error != 0) 1524 return (error); 1525 1526 /* NULL cred means no limit check for tmp snapshot */ 1527 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1528 tx, B_FALSE, 0, NULL); 1529 if (error != 0) { 1530 dsl_dataset_rele(ds, FTAG); 1531 return (error); 1532 } 1533 1534 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1535 dsl_dataset_rele(ds, FTAG); 1536 return (SET_ERROR(ENOTSUP)); 1537 } 1538 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1539 B_TRUE, tx); 1540 if (error != 0) { 1541 dsl_dataset_rele(ds, FTAG); 1542 return (error); 1543 } 1544 1545 dsl_dataset_rele(ds, FTAG); 1546 return (0); 1547} 1548 1549static void 1550dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1551{ 1552 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1553 dsl_pool_t *dp = dmu_tx_pool(tx); 1554 dsl_dataset_t *ds; 1555 1556 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1557 1558 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1559 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1560 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1561 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1562 1563 dsl_dataset_rele(ds, FTAG); 1564} 1565 1566int 1567dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1568 minor_t cleanup_minor, const char *htag) 1569{ 1570 dsl_dataset_snapshot_tmp_arg_t ddsta; 1571 int error; 1572 spa_t *spa; 1573 boolean_t needsuspend; 1574 void *cookie; 1575 1576 ddsta.ddsta_fsname = fsname; 1577 ddsta.ddsta_snapname = snapname; 1578 ddsta.ddsta_cleanup_minor = cleanup_minor; 1579 ddsta.ddsta_htag = htag; 1580 1581 error = spa_open(fsname, &spa, FTAG); 1582 if (error != 0) 1583 return (error); 1584 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1585 spa_close(spa, FTAG); 1586 1587 if (needsuspend) { 1588 error = zil_suspend(fsname, &cookie); 1589 if (error != 0) 1590 return (error); 1591 } 1592 1593 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1594 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 1595 1596 if (needsuspend) 1597 zil_resume(cookie); 1598 return (error); 1599} 1600 1601 1602void 1603dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1604{ 1605 ASSERT(dmu_tx_is_syncing(tx)); 1606 ASSERT(ds->ds_objset != NULL); 1607 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); 1608 1609 /* 1610 * in case we had to change ds_fsid_guid when we opened it, 1611 * sync it out now. 1612 */ 1613 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1614 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; 1615 1616 dmu_objset_sync(ds->ds_objset, zio, tx); 1617 1618 if (ds->ds_need_large_blocks && !ds->ds_large_blocks) { 1619 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx); 1620 ds->ds_large_blocks = B_TRUE; 1621 } 1622} 1623 1624static void 1625get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1626{ 1627 uint64_t count = 0; 1628 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1629 zap_cursor_t zc; 1630 zap_attribute_t za; 1631 nvlist_t *propval = fnvlist_alloc(); 1632 nvlist_t *val = fnvlist_alloc(); 1633 1634 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1635 1636 /* 1637 * There may be missing entries in ds_next_clones_obj 1638 * due to a bug in a previous version of the code. 1639 * Only trust it if it has the right number of entries. 1640 */ 1641 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { 1642 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1643 &count)); 1644 } 1645 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) 1646 goto fail; 1647 for (zap_cursor_init(&zc, mos, 1648 dsl_dataset_phys(ds)->ds_next_clones_obj); 1649 zap_cursor_retrieve(&zc, &za) == 0; 1650 zap_cursor_advance(&zc)) { 1651 dsl_dataset_t *clone; 1652 char buf[ZFS_MAXNAMELEN]; 1653 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1654 za.za_first_integer, FTAG, &clone)); 1655 dsl_dir_name(clone->ds_dir, buf); 1656 fnvlist_add_boolean(val, buf); 1657 dsl_dataset_rele(clone, FTAG); 1658 } 1659 zap_cursor_fini(&zc); 1660 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1661 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); 1662fail: 1663 nvlist_free(val); 1664 nvlist_free(propval); 1665} 1666 1667void 1668dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1669{ 1670 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1671 uint64_t refd, avail, uobjs, aobjs, ratio; 1672 1673 ASSERT(dsl_pool_config_held(dp)); 1674 1675 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : 1676 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / 1677 dsl_dataset_phys(ds)->ds_compressed_bytes); 1678 1679 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); 1680 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 1681 dsl_dataset_phys(ds)->ds_uncompressed_bytes); 1682 1683 if (ds->ds_is_snapshot) { 1684 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); 1685 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 1686 dsl_dataset_phys(ds)->ds_unique_bytes); 1687 get_clones_stat(ds, nv); 1688 } else { 1689 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 1690 char buf[MAXNAMELEN]; 1691 dsl_dataset_name(ds->ds_prev, buf); 1692 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf); 1693 } 1694 1695 dsl_dir_stats(ds->ds_dir, nv); 1696 } 1697 1698 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 1699 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 1700 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 1701 1702 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 1703 dsl_dataset_phys(ds)->ds_creation_time); 1704 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 1705 dsl_dataset_phys(ds)->ds_creation_txg); 1706 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 1707 ds->ds_quota); 1708 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 1709 ds->ds_reserved); 1710 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 1711 dsl_dataset_phys(ds)->ds_guid); 1712 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 1713 dsl_dataset_phys(ds)->ds_unique_bytes); 1714 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 1715 ds->ds_object); 1716 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 1717 ds->ds_userrefs); 1718 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 1719 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 1720 1721 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 1722 uint64_t written, comp, uncomp; 1723 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1724 dsl_dataset_t *prev; 1725 1726 int err = dsl_dataset_hold_obj(dp, 1727 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 1728 if (err == 0) { 1729 err = dsl_dataset_space_written(prev, ds, &written, 1730 &comp, &uncomp); 1731 dsl_dataset_rele(prev, FTAG); 1732 if (err == 0) { 1733 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 1734 written); 1735 } 1736 } 1737 } 1738} 1739 1740void 1741dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 1742{ 1743 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1744 ASSERT(dsl_pool_config_held(dp)); 1745 1746 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg; 1747 stat->dds_inconsistent = 1748 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT; 1749 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid; 1750 stat->dds_origin[0] = '\0'; 1751 if (ds->ds_is_snapshot) { 1752 stat->dds_is_snapshot = B_TRUE; 1753 stat->dds_num_clones = 1754 dsl_dataset_phys(ds)->ds_num_children - 1; 1755 } else { 1756 stat->dds_is_snapshot = B_FALSE; 1757 stat->dds_num_clones = 0; 1758 1759 if (dsl_dir_is_clone(ds->ds_dir)) { 1760 dsl_dataset_t *ods; 1761 1762 VERIFY0(dsl_dataset_hold_obj(dp, 1763 dsl_dir_phys(ds->ds_dir)->dd_origin_obj, 1764 FTAG, &ods)); 1765 dsl_dataset_name(ods, stat->dds_origin); 1766 dsl_dataset_rele(ods, FTAG); 1767 } 1768 } 1769} 1770 1771uint64_t 1772dsl_dataset_fsid_guid(dsl_dataset_t *ds) 1773{ 1774 return (ds->ds_fsid_guid); 1775} 1776 1777void 1778dsl_dataset_space(dsl_dataset_t *ds, 1779 uint64_t *refdbytesp, uint64_t *availbytesp, 1780 uint64_t *usedobjsp, uint64_t *availobjsp) 1781{ 1782 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; 1783 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 1784 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) 1785 *availbytesp += 1786 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 1787 if (ds->ds_quota != 0) { 1788 /* 1789 * Adjust available bytes according to refquota 1790 */ 1791 if (*refdbytesp < ds->ds_quota) 1792 *availbytesp = MIN(*availbytesp, 1793 ds->ds_quota - *refdbytesp); 1794 else 1795 *availbytesp = 0; 1796 } 1797 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); 1798 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 1799} 1800 1801boolean_t 1802dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 1803{ 1804 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1805 1806 ASSERT(dsl_pool_config_held(dp)); 1807 if (snap == NULL) 1808 return (B_FALSE); 1809 if (dsl_dataset_phys(ds)->ds_bp.blk_birth > 1810 dsl_dataset_phys(snap)->ds_creation_txg) { 1811 objset_t *os, *os_snap; 1812 /* 1813 * It may be that only the ZIL differs, because it was 1814 * reset in the head. Don't count that as being 1815 * modified. 1816 */ 1817 if (dmu_objset_from_ds(ds, &os) != 0) 1818 return (B_TRUE); 1819 if (dmu_objset_from_ds(snap, &os_snap) != 0) 1820 return (B_TRUE); 1821 return (bcmp(&os->os_phys->os_meta_dnode, 1822 &os_snap->os_phys->os_meta_dnode, 1823 sizeof (os->os_phys->os_meta_dnode)) != 0); 1824 } 1825 return (B_FALSE); 1826} 1827 1828typedef struct dsl_dataset_rename_snapshot_arg { 1829 const char *ddrsa_fsname; 1830 const char *ddrsa_oldsnapname; 1831 const char *ddrsa_newsnapname; 1832 boolean_t ddrsa_recursive; 1833 dmu_tx_t *ddrsa_tx; 1834} dsl_dataset_rename_snapshot_arg_t; 1835 1836/* ARGSUSED */ 1837static int 1838dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 1839 dsl_dataset_t *hds, void *arg) 1840{ 1841 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1842 int error; 1843 uint64_t val; 1844 1845 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1846 if (error != 0) { 1847 /* ignore nonexistent snapshots */ 1848 return (error == ENOENT ? 0 : error); 1849 } 1850 1851 /* new name should not exist */ 1852 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 1853 if (error == 0) 1854 error = SET_ERROR(EEXIST); 1855 else if (error == ENOENT) 1856 error = 0; 1857 1858 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 1859 if (dsl_dir_namelen(hds->ds_dir) + 1 + 1860 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN) 1861 error = SET_ERROR(ENAMETOOLONG); 1862 1863 return (error); 1864} 1865 1866static int 1867dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 1868{ 1869 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1870 dsl_pool_t *dp = dmu_tx_pool(tx); 1871 dsl_dataset_t *hds; 1872 int error; 1873 1874 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 1875 if (error != 0) 1876 return (error); 1877 1878 if (ddrsa->ddrsa_recursive) { 1879 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1880 dsl_dataset_rename_snapshot_check_impl, ddrsa, 1881 DS_FIND_CHILDREN); 1882 } else { 1883 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 1884 } 1885 dsl_dataset_rele(hds, FTAG); 1886 return (error); 1887} 1888 1889static int 1890dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 1891 dsl_dataset_t *hds, void *arg) 1892{ 1893#ifdef __FreeBSD__ 1894#ifdef _KERNEL 1895 char *oldname, *newname; 1896#endif 1897#endif 1898 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1899 dsl_dataset_t *ds; 1900 uint64_t val; 1901 dmu_tx_t *tx = ddrsa->ddrsa_tx; 1902 int error; 1903 1904 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1905 ASSERT(error == 0 || error == ENOENT); 1906 if (error == ENOENT) { 1907 /* ignore nonexistent snapshots */ 1908 return (0); 1909 } 1910 1911 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 1912 1913 /* log before we change the name */ 1914 spa_history_log_internal_ds(ds, "rename", tx, 1915 "-> @%s", ddrsa->ddrsa_newsnapname); 1916 1917 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 1918 B_FALSE)); 1919 mutex_enter(&ds->ds_lock); 1920 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 1921 mutex_exit(&ds->ds_lock); 1922 VERIFY0(zap_add(dp->dp_meta_objset, 1923 dsl_dataset_phys(hds)->ds_snapnames_zapobj, 1924 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 1925 1926#ifdef __FreeBSD__ 1927#ifdef _KERNEL 1928 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1929 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1930 snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname, 1931 ddrsa->ddrsa_oldsnapname); 1932 snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname, 1933 ddrsa->ddrsa_newsnapname); 1934 zfsvfs_update_fromname(oldname, newname); 1935 zvol_rename_minors(oldname, newname); 1936 kmem_free(newname, MAXPATHLEN); 1937 kmem_free(oldname, MAXPATHLEN); 1938#endif 1939#endif 1940 dsl_dataset_rele(ds, FTAG); 1941 1942 return (0); 1943} 1944 1945static void 1946dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 1947{ 1948 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1949 dsl_pool_t *dp = dmu_tx_pool(tx); 1950 dsl_dataset_t *hds; 1951 1952 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 1953 ddrsa->ddrsa_tx = tx; 1954 if (ddrsa->ddrsa_recursive) { 1955 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1956 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 1957 DS_FIND_CHILDREN)); 1958 } else { 1959 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 1960 } 1961 dsl_dataset_rele(hds, FTAG); 1962} 1963 1964int 1965dsl_dataset_rename_snapshot(const char *fsname, 1966 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 1967{ 1968 dsl_dataset_rename_snapshot_arg_t ddrsa; 1969 1970 ddrsa.ddrsa_fsname = fsname; 1971 ddrsa.ddrsa_oldsnapname = oldsnapname; 1972 ddrsa.ddrsa_newsnapname = newsnapname; 1973 ddrsa.ddrsa_recursive = recursive; 1974 1975 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 1976 dsl_dataset_rename_snapshot_sync, &ddrsa, 1977 1, ZFS_SPACE_CHECK_RESERVED)); 1978} 1979 1980/* 1981 * If we're doing an ownership handoff, we need to make sure that there is 1982 * only one long hold on the dataset. We're not allowed to change anything here 1983 * so we don't permanently release the long hold or regular hold here. We want 1984 * to do this only when syncing to avoid the dataset unexpectedly going away 1985 * when we release the long hold. 1986 */ 1987static int 1988dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 1989{ 1990 boolean_t held; 1991 1992 if (!dmu_tx_is_syncing(tx)) 1993 return (0); 1994 1995 if (owner != NULL) { 1996 VERIFY3P(ds->ds_owner, ==, owner); 1997 dsl_dataset_long_rele(ds, owner); 1998 } 1999 2000 held = dsl_dataset_long_held(ds); 2001 2002 if (owner != NULL) 2003 dsl_dataset_long_hold(ds, owner); 2004 2005 if (held) 2006 return (SET_ERROR(EBUSY)); 2007 2008 return (0); 2009} 2010 2011typedef struct dsl_dataset_rollback_arg { 2012 const char *ddra_fsname; 2013 void *ddra_owner; 2014 nvlist_t *ddra_result; 2015} dsl_dataset_rollback_arg_t; 2016 2017static int 2018dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 2019{ 2020 dsl_dataset_rollback_arg_t *ddra = arg; 2021 dsl_pool_t *dp = dmu_tx_pool(tx); 2022 dsl_dataset_t *ds; 2023 int64_t unused_refres_delta; 2024 int error; 2025 2026 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 2027 if (error != 0) 2028 return (error); 2029 2030 /* must not be a snapshot */ 2031 if (ds->ds_is_snapshot) { 2032 dsl_dataset_rele(ds, FTAG); 2033 return (SET_ERROR(EINVAL)); 2034 } 2035 2036 /* must have a most recent snapshot */ 2037 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { 2038 dsl_dataset_rele(ds, FTAG); 2039 return (SET_ERROR(EINVAL)); 2040 } 2041 2042 /* must not have any bookmarks after the most recent snapshot */ 2043 nvlist_t *proprequest = fnvlist_alloc(); 2044 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2045 nvlist_t *bookmarks = fnvlist_alloc(); 2046 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 2047 fnvlist_free(proprequest); 2048 if (error != 0) 2049 return (error); 2050 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 2051 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 2052 nvlist_t *valuenv = 2053 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 2054 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2055 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 2056 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 2057 fnvlist_free(bookmarks); 2058 dsl_dataset_rele(ds, FTAG); 2059 return (SET_ERROR(EEXIST)); 2060 } 2061 } 2062 fnvlist_free(bookmarks); 2063 2064 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 2065 if (error != 0) { 2066 dsl_dataset_rele(ds, FTAG); 2067 return (error); 2068 } 2069 2070 /* 2071 * Check if the snap we are rolling back to uses more than 2072 * the refquota. 2073 */ 2074 if (ds->ds_quota != 0 && 2075 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { 2076 dsl_dataset_rele(ds, FTAG); 2077 return (SET_ERROR(EDQUOT)); 2078 } 2079 2080 /* 2081 * When we do the clone swap, we will temporarily use more space 2082 * due to the refreservation (the head will no longer have any 2083 * unique space, so the entire amount of the refreservation will need 2084 * to be free). We will immediately destroy the clone, freeing 2085 * this space, but the freeing happens over many txg's. 2086 */ 2087 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 2088 dsl_dataset_phys(ds)->ds_unique_bytes); 2089 2090 if (unused_refres_delta > 0 && 2091 unused_refres_delta > 2092 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 2093 dsl_dataset_rele(ds, FTAG); 2094 return (SET_ERROR(ENOSPC)); 2095 } 2096 2097 dsl_dataset_rele(ds, FTAG); 2098 return (0); 2099} 2100 2101static void 2102dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 2103{ 2104 dsl_dataset_rollback_arg_t *ddra = arg; 2105 dsl_pool_t *dp = dmu_tx_pool(tx); 2106 dsl_dataset_t *ds, *clone; 2107 uint64_t cloneobj; 2108 char namebuf[ZFS_MAXNAMELEN]; 2109 2110 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 2111 2112 dsl_dataset_name(ds->ds_prev, namebuf); 2113 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 2114 2115 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 2116 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 2117 2118 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 2119 2120 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 2121 dsl_dataset_zero_zil(ds, tx); 2122 2123 dsl_destroy_head_sync_impl(clone, tx); 2124 2125 dsl_dataset_rele(clone, FTAG); 2126 dsl_dataset_rele(ds, FTAG); 2127} 2128 2129/* 2130 * Rolls back the given filesystem or volume to the most recent snapshot. 2131 * The name of the most recent snapshot will be returned under key "target" 2132 * in the result nvlist. 2133 * 2134 * If owner != NULL: 2135 * - The existing dataset MUST be owned by the specified owner at entry 2136 * - Upon return, dataset will still be held by the same owner, whether we 2137 * succeed or not. 2138 * 2139 * This mode is required any time the existing filesystem is mounted. See 2140 * notes above zfs_suspend_fs() for further details. 2141 */ 2142int 2143dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result) 2144{ 2145 dsl_dataset_rollback_arg_t ddra; 2146 2147 ddra.ddra_fsname = fsname; 2148 ddra.ddra_owner = owner; 2149 ddra.ddra_result = result; 2150 2151 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 2152 dsl_dataset_rollback_sync, &ddra, 2153 1, ZFS_SPACE_CHECK_RESERVED)); 2154} 2155 2156struct promotenode { 2157 list_node_t link; 2158 dsl_dataset_t *ds; 2159}; 2160 2161typedef struct dsl_dataset_promote_arg { 2162 const char *ddpa_clonename; 2163 dsl_dataset_t *ddpa_clone; 2164 list_t shared_snaps, origin_snaps, clone_snaps; 2165 dsl_dataset_t *origin_origin; /* origin of the origin */ 2166 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2167 char *err_ds; 2168 cred_t *cr; 2169} dsl_dataset_promote_arg_t; 2170 2171static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2172static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2173 void *tag); 2174static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2175 2176static int 2177dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2178{ 2179 dsl_dataset_promote_arg_t *ddpa = arg; 2180 dsl_pool_t *dp = dmu_tx_pool(tx); 2181 dsl_dataset_t *hds; 2182 struct promotenode *snap; 2183 dsl_dataset_t *origin_ds; 2184 int err; 2185 uint64_t unused; 2186 uint64_t ss_mv_cnt; 2187 size_t max_snap_len; 2188 2189 err = promote_hold(ddpa, dp, FTAG); 2190 if (err != 0) 2191 return (err); 2192 2193 hds = ddpa->ddpa_clone; 2194 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; 2195 2196 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { 2197 promote_rele(ddpa, FTAG); 2198 return (SET_ERROR(EXDEV)); 2199 } 2200 2201 /* 2202 * Compute and check the amount of space to transfer. Since this is 2203 * so expensive, don't do the preliminary check. 2204 */ 2205 if (!dmu_tx_is_syncing(tx)) { 2206 promote_rele(ddpa, FTAG); 2207 return (0); 2208 } 2209 2210 snap = list_head(&ddpa->shared_snaps); 2211 origin_ds = snap->ds; 2212 2213 /* compute origin's new unique space */ 2214 snap = list_tail(&ddpa->clone_snaps); 2215 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2216 origin_ds->ds_object); 2217 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2218 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, 2219 &ddpa->unique, &unused, &unused); 2220 2221 /* 2222 * Walk the snapshots that we are moving 2223 * 2224 * Compute space to transfer. Consider the incremental changes 2225 * to used by each snapshot: 2226 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2227 * So each snapshot gave birth to: 2228 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2229 * So a sequence would look like: 2230 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2231 * Which simplifies to: 2232 * uN + kN + kN-1 + ... + k1 + k0 2233 * Note however, if we stop before we reach the ORIGIN we get: 2234 * uN + kN + kN-1 + ... + kM - uM-1 2235 */ 2236 ss_mv_cnt = 0; 2237 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; 2238 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; 2239 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; 2240 for (snap = list_head(&ddpa->shared_snaps); snap; 2241 snap = list_next(&ddpa->shared_snaps, snap)) { 2242 uint64_t val, dlused, dlcomp, dluncomp; 2243 dsl_dataset_t *ds = snap->ds; 2244 2245 ss_mv_cnt++; 2246 2247 /* 2248 * If there are long holds, we won't be able to evict 2249 * the objset. 2250 */ 2251 if (dsl_dataset_long_held(ds)) { 2252 err = SET_ERROR(EBUSY); 2253 goto out; 2254 } 2255 2256 /* Check that the snapshot name does not conflict */ 2257 VERIFY0(dsl_dataset_get_snapname(ds)); 2258 if (strlen(ds->ds_snapname) >= max_snap_len) { 2259 err = SET_ERROR(ENAMETOOLONG); 2260 goto out; 2261 } 2262 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2263 if (err == 0) { 2264 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname); 2265 err = SET_ERROR(EEXIST); 2266 goto out; 2267 } 2268 if (err != ENOENT) 2269 goto out; 2270 2271 /* The very first snapshot does not have a deadlist */ 2272 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) 2273 continue; 2274 2275 dsl_deadlist_space(&ds->ds_deadlist, 2276 &dlused, &dlcomp, &dluncomp); 2277 ddpa->used += dlused; 2278 ddpa->comp += dlcomp; 2279 ddpa->uncomp += dluncomp; 2280 } 2281 2282 /* 2283 * If we are a clone of a clone then we never reached ORIGIN, 2284 * so we need to subtract out the clone origin's used space. 2285 */ 2286 if (ddpa->origin_origin) { 2287 ddpa->used -= 2288 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; 2289 ddpa->comp -= 2290 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; 2291 ddpa->uncomp -= 2292 dsl_dataset_phys(ddpa->origin_origin)-> 2293 ds_uncompressed_bytes; 2294 } 2295 2296 /* Check that there is enough space and limit headroom here */ 2297 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2298 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2299 if (err != 0) 2300 goto out; 2301 2302 /* 2303 * Compute the amounts of space that will be used by snapshots 2304 * after the promotion (for both origin and clone). For each, 2305 * it is the amount of space that will be on all of their 2306 * deadlists (that was not born before their new origin). 2307 */ 2308 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2309 uint64_t space; 2310 2311 /* 2312 * Note, typically this will not be a clone of a clone, 2313 * so dd_origin_txg will be < TXG_INITIAL, so 2314 * these snaplist_space() -> dsl_deadlist_space_range() 2315 * calls will be fast because they do not have to 2316 * iterate over all bps. 2317 */ 2318 snap = list_head(&ddpa->origin_snaps); 2319 err = snaplist_space(&ddpa->shared_snaps, 2320 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2321 if (err != 0) 2322 goto out; 2323 2324 err = snaplist_space(&ddpa->clone_snaps, 2325 snap->ds->ds_dir->dd_origin_txg, &space); 2326 if (err != 0) 2327 goto out; 2328 ddpa->cloneusedsnap += space; 2329 } 2330 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & 2331 DD_FLAG_USED_BREAKDOWN) { 2332 err = snaplist_space(&ddpa->origin_snaps, 2333 dsl_dataset_phys(origin_ds)->ds_creation_txg, 2334 &ddpa->originusedsnap); 2335 if (err != 0) 2336 goto out; 2337 } 2338 2339out: 2340 promote_rele(ddpa, FTAG); 2341 return (err); 2342} 2343 2344static void 2345dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2346{ 2347 dsl_dataset_promote_arg_t *ddpa = arg; 2348 dsl_pool_t *dp = dmu_tx_pool(tx); 2349 dsl_dataset_t *hds; 2350 struct promotenode *snap; 2351 dsl_dataset_t *origin_ds; 2352 dsl_dataset_t *origin_head; 2353 dsl_dir_t *dd; 2354 dsl_dir_t *odd = NULL; 2355 uint64_t oldnext_obj; 2356 int64_t delta; 2357#if defined(__FreeBSD__) && defined(_KERNEL) 2358 char *oldname, *newname; 2359#endif 2360 2361 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2362 hds = ddpa->ddpa_clone; 2363 2364 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); 2365 2366 snap = list_head(&ddpa->shared_snaps); 2367 origin_ds = snap->ds; 2368 dd = hds->ds_dir; 2369 2370 snap = list_head(&ddpa->origin_snaps); 2371 origin_head = snap->ds; 2372 2373 /* 2374 * We need to explicitly open odd, since origin_ds's dd will be 2375 * changing. 2376 */ 2377 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2378 NULL, FTAG, &odd)); 2379 2380 /* change origin's next snap */ 2381 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2382 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; 2383 snap = list_tail(&ddpa->clone_snaps); 2384 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2385 origin_ds->ds_object); 2386 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; 2387 2388 /* change the origin's next clone */ 2389 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { 2390 dsl_dataset_remove_from_next_clones(origin_ds, 2391 snap->ds->ds_object, tx); 2392 VERIFY0(zap_add_int(dp->dp_meta_objset, 2393 dsl_dataset_phys(origin_ds)->ds_next_clones_obj, 2394 oldnext_obj, tx)); 2395 } 2396 2397 /* change origin */ 2398 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2399 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); 2400 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; 2401 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2402 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2403 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; 2404 origin_head->ds_dir->dd_origin_txg = 2405 dsl_dataset_phys(origin_ds)->ds_creation_txg; 2406 2407 /* change dd_clone entries */ 2408 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2409 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2410 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); 2411 VERIFY0(zap_add_int(dp->dp_meta_objset, 2412 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2413 hds->ds_object, tx)); 2414 2415 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2416 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2417 origin_head->ds_object, tx)); 2418 if (dsl_dir_phys(dd)->dd_clones == 0) { 2419 dsl_dir_phys(dd)->dd_clones = 2420 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, 2421 DMU_OT_NONE, 0, tx); 2422 } 2423 VERIFY0(zap_add_int(dp->dp_meta_objset, 2424 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); 2425 } 2426 2427#if defined(__FreeBSD__) && defined(_KERNEL) 2428 /* Take the spa_namespace_lock early so zvol renames don't deadlock. */ 2429 mutex_enter(&spa_namespace_lock); 2430 2431 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2432 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2433#endif 2434 2435 /* move snapshots to this dir */ 2436 for (snap = list_head(&ddpa->shared_snaps); snap; 2437 snap = list_next(&ddpa->shared_snaps, snap)) { 2438 dsl_dataset_t *ds = snap->ds; 2439 2440 /* 2441 * Property callbacks are registered to a particular 2442 * dsl_dir. Since ours is changing, evict the objset 2443 * so that they will be unregistered from the old dsl_dir. 2444 */ 2445 if (ds->ds_objset) { 2446 dmu_objset_evict(ds->ds_objset); 2447 ds->ds_objset = NULL; 2448 } 2449 2450 /* move snap name entry */ 2451 VERIFY0(dsl_dataset_get_snapname(ds)); 2452 VERIFY0(dsl_dataset_snap_remove(origin_head, 2453 ds->ds_snapname, tx, B_TRUE)); 2454 VERIFY0(zap_add(dp->dp_meta_objset, 2455 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 2456 8, 1, &ds->ds_object, tx)); 2457 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 2458 DD_FIELD_SNAPSHOT_COUNT, tx); 2459 2460 /* change containing dsl_dir */ 2461 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2462 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); 2463 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; 2464 ASSERT3P(ds->ds_dir, ==, odd); 2465 dsl_dir_rele(ds->ds_dir, ds); 2466 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 2467 NULL, ds, &ds->ds_dir)); 2468 2469#if defined(__FreeBSD__) && defined(_KERNEL) 2470 dsl_dataset_name(ds, newname); 2471 zfsvfs_update_fromname(oldname, newname); 2472 zvol_rename_minors(oldname, newname); 2473#endif 2474 2475 /* move any clone references */ 2476 if (dsl_dataset_phys(ds)->ds_next_clones_obj && 2477 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2478 zap_cursor_t zc; 2479 zap_attribute_t za; 2480 2481 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2482 dsl_dataset_phys(ds)->ds_next_clones_obj); 2483 zap_cursor_retrieve(&zc, &za) == 0; 2484 zap_cursor_advance(&zc)) { 2485 dsl_dataset_t *cnds; 2486 uint64_t o; 2487 2488 if (za.za_first_integer == oldnext_obj) { 2489 /* 2490 * We've already moved the 2491 * origin's reference. 2492 */ 2493 continue; 2494 } 2495 2496 VERIFY0(dsl_dataset_hold_obj(dp, 2497 za.za_first_integer, FTAG, &cnds)); 2498 o = dsl_dir_phys(cnds->ds_dir)-> 2499 dd_head_dataset_obj; 2500 2501 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2502 dsl_dir_phys(odd)->dd_clones, o, tx)); 2503 VERIFY0(zap_add_int(dp->dp_meta_objset, 2504 dsl_dir_phys(dd)->dd_clones, o, tx)); 2505 dsl_dataset_rele(cnds, FTAG); 2506 } 2507 zap_cursor_fini(&zc); 2508 } 2509 2510 ASSERT(!dsl_prop_hascb(ds)); 2511 } 2512 2513#if defined(__FreeBSD__) && defined(_KERNEL) 2514 mutex_exit(&spa_namespace_lock); 2515 2516 kmem_free(newname, MAXPATHLEN); 2517 kmem_free(oldname, MAXPATHLEN); 2518#endif 2519 /* 2520 * Change space accounting. 2521 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2522 * both be valid, or both be 0 (resulting in delta == 0). This 2523 * is true for each of {clone,origin} independently. 2524 */ 2525 2526 delta = ddpa->cloneusedsnap - 2527 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; 2528 ASSERT3S(delta, >=, 0); 2529 ASSERT3U(ddpa->used, >=, delta); 2530 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2531 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2532 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 2533 2534 delta = ddpa->originusedsnap - 2535 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; 2536 ASSERT3S(delta, <=, 0); 2537 ASSERT3U(ddpa->used, >=, -delta); 2538 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2539 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2540 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 2541 2542 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; 2543 2544 /* log history record */ 2545 spa_history_log_internal_ds(hds, "promote", tx, ""); 2546 2547 dsl_dir_rele(odd, FTAG); 2548 promote_rele(ddpa, FTAG); 2549} 2550 2551/* 2552 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2553 * (exclusive) and last_obj (inclusive). The list will be in reverse 2554 * order (last_obj will be the list_head()). If first_obj == 0, do all 2555 * snapshots back to this dataset's origin. 2556 */ 2557static int 2558snaplist_make(dsl_pool_t *dp, 2559 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 2560{ 2561 uint64_t obj = last_obj; 2562 2563 list_create(l, sizeof (struct promotenode), 2564 offsetof(struct promotenode, link)); 2565 2566 while (obj != first_obj) { 2567 dsl_dataset_t *ds; 2568 struct promotenode *snap; 2569 int err; 2570 2571 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 2572 ASSERT(err != ENOENT); 2573 if (err != 0) 2574 return (err); 2575 2576 if (first_obj == 0) 2577 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; 2578 2579 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 2580 snap->ds = ds; 2581 list_insert_tail(l, snap); 2582 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 2583 } 2584 2585 return (0); 2586} 2587 2588static int 2589snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2590{ 2591 struct promotenode *snap; 2592 2593 *spacep = 0; 2594 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2595 uint64_t used, comp, uncomp; 2596 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2597 mintxg, UINT64_MAX, &used, &comp, &uncomp); 2598 *spacep += used; 2599 } 2600 return (0); 2601} 2602 2603static void 2604snaplist_destroy(list_t *l, void *tag) 2605{ 2606 struct promotenode *snap; 2607 2608 if (l == NULL || !list_link_active(&l->list_head)) 2609 return; 2610 2611 while ((snap = list_tail(l)) != NULL) { 2612 list_remove(l, snap); 2613 dsl_dataset_rele(snap->ds, tag); 2614 kmem_free(snap, sizeof (*snap)); 2615 } 2616 list_destroy(l); 2617} 2618 2619static int 2620promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 2621{ 2622 int error; 2623 dsl_dir_t *dd; 2624 struct promotenode *snap; 2625 2626 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 2627 &ddpa->ddpa_clone); 2628 if (error != 0) 2629 return (error); 2630 dd = ddpa->ddpa_clone->ds_dir; 2631 2632 if (ddpa->ddpa_clone->ds_is_snapshot || 2633 !dsl_dir_is_clone(dd)) { 2634 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2635 return (SET_ERROR(EINVAL)); 2636 } 2637 2638 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, 2639 &ddpa->shared_snaps, tag); 2640 if (error != 0) 2641 goto out; 2642 2643 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 2644 &ddpa->clone_snaps, tag); 2645 if (error != 0) 2646 goto out; 2647 2648 snap = list_head(&ddpa->shared_snaps); 2649 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); 2650 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, 2651 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, 2652 &ddpa->origin_snaps, tag); 2653 if (error != 0) 2654 goto out; 2655 2656 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { 2657 error = dsl_dataset_hold_obj(dp, 2658 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, 2659 tag, &ddpa->origin_origin); 2660 if (error != 0) 2661 goto out; 2662 } 2663out: 2664 if (error != 0) 2665 promote_rele(ddpa, tag); 2666 return (error); 2667} 2668 2669static void 2670promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 2671{ 2672 snaplist_destroy(&ddpa->shared_snaps, tag); 2673 snaplist_destroy(&ddpa->clone_snaps, tag); 2674 snaplist_destroy(&ddpa->origin_snaps, tag); 2675 if (ddpa->origin_origin != NULL) 2676 dsl_dataset_rele(ddpa->origin_origin, tag); 2677 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2678} 2679 2680/* 2681 * Promote a clone. 2682 * 2683 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 2684 * in with the name. (It must be at least MAXNAMELEN bytes long.) 2685 */ 2686int 2687dsl_dataset_promote(const char *name, char *conflsnap) 2688{ 2689 dsl_dataset_promote_arg_t ddpa = { 0 }; 2690 uint64_t numsnaps; 2691 int error; 2692 objset_t *os; 2693 2694 /* 2695 * We will modify space proportional to the number of 2696 * snapshots. Compute numsnaps. 2697 */ 2698 error = dmu_objset_hold(name, FTAG, &os); 2699 if (error != 0) 2700 return (error); 2701 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 2702 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, 2703 &numsnaps); 2704 dmu_objset_rele(os, FTAG); 2705 if (error != 0) 2706 return (error); 2707 2708 ddpa.ddpa_clonename = name; 2709 ddpa.err_ds = conflsnap; 2710 ddpa.cr = CRED(); 2711 2712 return (dsl_sync_task(name, dsl_dataset_promote_check, 2713 dsl_dataset_promote_sync, &ddpa, 2714 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED)); 2715} 2716 2717int 2718dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 2719 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 2720{ 2721 int64_t unused_refres_delta; 2722 2723 /* they should both be heads */ 2724 if (clone->ds_is_snapshot || 2725 origin_head->ds_is_snapshot) 2726 return (SET_ERROR(EINVAL)); 2727 2728 /* if we are not forcing, the branch point should be just before them */ 2729 if (!force && clone->ds_prev != origin_head->ds_prev) 2730 return (SET_ERROR(EINVAL)); 2731 2732 /* clone should be the clone (unless they are unrelated) */ 2733 if (clone->ds_prev != NULL && 2734 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 2735 origin_head->ds_dir != clone->ds_prev->ds_dir) 2736 return (SET_ERROR(EINVAL)); 2737 2738 /* the clone should be a child of the origin */ 2739 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 2740 return (SET_ERROR(EINVAL)); 2741 2742 /* origin_head shouldn't be modified unless 'force' */ 2743 if (!force && 2744 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 2745 return (SET_ERROR(ETXTBSY)); 2746 2747 /* origin_head should have no long holds (e.g. is not mounted) */ 2748 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 2749 return (SET_ERROR(EBUSY)); 2750 2751 /* check amount of any unconsumed refreservation */ 2752 unused_refres_delta = 2753 (int64_t)MIN(origin_head->ds_reserved, 2754 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 2755 (int64_t)MIN(origin_head->ds_reserved, 2756 dsl_dataset_phys(clone)->ds_unique_bytes); 2757 2758 if (unused_refres_delta > 0 && 2759 unused_refres_delta > 2760 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 2761 return (SET_ERROR(ENOSPC)); 2762 2763 /* clone can't be over the head's refquota */ 2764 if (origin_head->ds_quota != 0 && 2765 dsl_dataset_phys(clone)->ds_referenced_bytes > 2766 origin_head->ds_quota) 2767 return (SET_ERROR(EDQUOT)); 2768 2769 return (0); 2770} 2771 2772void 2773dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 2774 dsl_dataset_t *origin_head, dmu_tx_t *tx) 2775{ 2776 dsl_pool_t *dp = dmu_tx_pool(tx); 2777 int64_t unused_refres_delta; 2778 2779 ASSERT(clone->ds_reserved == 0); 2780 ASSERT(origin_head->ds_quota == 0 || 2781 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota); 2782 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 2783 2784 dmu_buf_will_dirty(clone->ds_dbuf, tx); 2785 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 2786 2787 if (clone->ds_objset != NULL) { 2788 dmu_objset_evict(clone->ds_objset); 2789 clone->ds_objset = NULL; 2790 } 2791 2792 if (origin_head->ds_objset != NULL) { 2793 dmu_objset_evict(origin_head->ds_objset); 2794 origin_head->ds_objset = NULL; 2795 } 2796 2797 unused_refres_delta = 2798 (int64_t)MIN(origin_head->ds_reserved, 2799 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 2800 (int64_t)MIN(origin_head->ds_reserved, 2801 dsl_dataset_phys(clone)->ds_unique_bytes); 2802 2803 /* 2804 * Reset origin's unique bytes, if it exists. 2805 */ 2806 if (clone->ds_prev) { 2807 dsl_dataset_t *origin = clone->ds_prev; 2808 uint64_t comp, uncomp; 2809 2810 dmu_buf_will_dirty(origin->ds_dbuf, tx); 2811 dsl_deadlist_space_range(&clone->ds_deadlist, 2812 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, 2813 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); 2814 } 2815 2816 /* swap blkptrs */ 2817 { 2818 blkptr_t tmp; 2819 tmp = dsl_dataset_phys(origin_head)->ds_bp; 2820 dsl_dataset_phys(origin_head)->ds_bp = 2821 dsl_dataset_phys(clone)->ds_bp; 2822 dsl_dataset_phys(clone)->ds_bp = tmp; 2823 } 2824 2825 /* set dd_*_bytes */ 2826 { 2827 int64_t dused, dcomp, duncomp; 2828 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2829 uint64_t odl_used, odl_comp, odl_uncomp; 2830 2831 ASSERT3U(dsl_dir_phys(clone->ds_dir)-> 2832 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2833 2834 dsl_deadlist_space(&clone->ds_deadlist, 2835 &cdl_used, &cdl_comp, &cdl_uncomp); 2836 dsl_deadlist_space(&origin_head->ds_deadlist, 2837 &odl_used, &odl_comp, &odl_uncomp); 2838 2839 dused = dsl_dataset_phys(clone)->ds_referenced_bytes + 2840 cdl_used - 2841 (dsl_dataset_phys(origin_head)->ds_referenced_bytes + 2842 odl_used); 2843 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + 2844 cdl_comp - 2845 (dsl_dataset_phys(origin_head)->ds_compressed_bytes + 2846 odl_comp); 2847 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + 2848 cdl_uncomp - 2849 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + 2850 odl_uncomp); 2851 2852 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 2853 dused, dcomp, duncomp, tx); 2854 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 2855 -dused, -dcomp, -duncomp, tx); 2856 2857 /* 2858 * The difference in the space used by snapshots is the 2859 * difference in snapshot space due to the head's 2860 * deadlist (since that's the only thing that's 2861 * changing that affects the snapused). 2862 */ 2863 dsl_deadlist_space_range(&clone->ds_deadlist, 2864 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2865 &cdl_used, &cdl_comp, &cdl_uncomp); 2866 dsl_deadlist_space_range(&origin_head->ds_deadlist, 2867 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2868 &odl_used, &odl_comp, &odl_uncomp); 2869 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 2870 DD_USED_HEAD, DD_USED_SNAP, NULL); 2871 } 2872 2873 /* swap ds_*_bytes */ 2874 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, 2875 dsl_dataset_phys(clone)->ds_referenced_bytes); 2876 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, 2877 dsl_dataset_phys(clone)->ds_compressed_bytes); 2878 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, 2879 dsl_dataset_phys(clone)->ds_uncompressed_bytes); 2880 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, 2881 dsl_dataset_phys(clone)->ds_unique_bytes); 2882 2883 /* apply any parent delta for change in unconsumed refreservation */ 2884 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 2885 unused_refres_delta, 0, 0, tx); 2886 2887 /* 2888 * Swap deadlists. 2889 */ 2890 dsl_deadlist_close(&clone->ds_deadlist); 2891 dsl_deadlist_close(&origin_head->ds_deadlist); 2892 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, 2893 dsl_dataset_phys(clone)->ds_deadlist_obj); 2894 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 2895 dsl_dataset_phys(clone)->ds_deadlist_obj); 2896 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 2897 dsl_dataset_phys(origin_head)->ds_deadlist_obj); 2898 2899 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 2900 2901 spa_history_log_internal_ds(clone, "clone swap", tx, 2902 "parent=%s", origin_head->ds_dir->dd_myname); 2903} 2904 2905/* 2906 * Given a pool name and a dataset object number in that pool, 2907 * return the name of that dataset. 2908 */ 2909int 2910dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 2911{ 2912 dsl_pool_t *dp; 2913 dsl_dataset_t *ds; 2914 int error; 2915 2916 error = dsl_pool_hold(pname, FTAG, &dp); 2917 if (error != 0) 2918 return (error); 2919 2920 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 2921 if (error == 0) { 2922 dsl_dataset_name(ds, buf); 2923 dsl_dataset_rele(ds, FTAG); 2924 } 2925 dsl_pool_rele(dp, FTAG); 2926 2927 return (error); 2928} 2929 2930int 2931dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 2932 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 2933{ 2934 int error = 0; 2935 2936 ASSERT3S(asize, >, 0); 2937 2938 /* 2939 * *ref_rsrv is the portion of asize that will come from any 2940 * unconsumed refreservation space. 2941 */ 2942 *ref_rsrv = 0; 2943 2944 mutex_enter(&ds->ds_lock); 2945 /* 2946 * Make a space adjustment for reserved bytes. 2947 */ 2948 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 2949 ASSERT3U(*used, >=, 2950 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 2951 *used -= 2952 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 2953 *ref_rsrv = 2954 asize - MIN(asize, parent_delta(ds, asize + inflight)); 2955 } 2956 2957 if (!check_quota || ds->ds_quota == 0) { 2958 mutex_exit(&ds->ds_lock); 2959 return (0); 2960 } 2961 /* 2962 * If they are requesting more space, and our current estimate 2963 * is over quota, they get to try again unless the actual 2964 * on-disk is over quota and there are no pending changes (which 2965 * may free up space for us). 2966 */ 2967 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= 2968 ds->ds_quota) { 2969 if (inflight > 0 || 2970 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) 2971 error = SET_ERROR(ERESTART); 2972 else 2973 error = SET_ERROR(EDQUOT); 2974 } 2975 mutex_exit(&ds->ds_lock); 2976 2977 return (error); 2978} 2979 2980typedef struct dsl_dataset_set_qr_arg { 2981 const char *ddsqra_name; 2982 zprop_source_t ddsqra_source; 2983 uint64_t ddsqra_value; 2984} dsl_dataset_set_qr_arg_t; 2985 2986 2987/* ARGSUSED */ 2988static int 2989dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 2990{ 2991 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2992 dsl_pool_t *dp = dmu_tx_pool(tx); 2993 dsl_dataset_t *ds; 2994 int error; 2995 uint64_t newval; 2996 2997 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 2998 return (SET_ERROR(ENOTSUP)); 2999 3000 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3001 if (error != 0) 3002 return (error); 3003 3004 if (ds->ds_is_snapshot) { 3005 dsl_dataset_rele(ds, FTAG); 3006 return (SET_ERROR(EINVAL)); 3007 } 3008 3009 error = dsl_prop_predict(ds->ds_dir, 3010 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3011 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3012 if (error != 0) { 3013 dsl_dataset_rele(ds, FTAG); 3014 return (error); 3015 } 3016 3017 if (newval == 0) { 3018 dsl_dataset_rele(ds, FTAG); 3019 return (0); 3020 } 3021 3022 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || 3023 newval < ds->ds_reserved) { 3024 dsl_dataset_rele(ds, FTAG); 3025 return (SET_ERROR(ENOSPC)); 3026 } 3027 3028 dsl_dataset_rele(ds, FTAG); 3029 return (0); 3030} 3031 3032static void 3033dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 3034{ 3035 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3036 dsl_pool_t *dp = dmu_tx_pool(tx); 3037 dsl_dataset_t *ds; 3038 uint64_t newval; 3039 3040 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3041 3042 dsl_prop_set_sync_impl(ds, 3043 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3044 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 3045 &ddsqra->ddsqra_value, tx); 3046 3047 VERIFY0(dsl_prop_get_int_ds(ds, 3048 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 3049 3050 if (ds->ds_quota != newval) { 3051 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3052 ds->ds_quota = newval; 3053 } 3054 dsl_dataset_rele(ds, FTAG); 3055} 3056 3057int 3058dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 3059 uint64_t refquota) 3060{ 3061 dsl_dataset_set_qr_arg_t ddsqra; 3062 3063 ddsqra.ddsqra_name = dsname; 3064 ddsqra.ddsqra_source = source; 3065 ddsqra.ddsqra_value = refquota; 3066 3067 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 3068 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 3069} 3070 3071static int 3072dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 3073{ 3074 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3075 dsl_pool_t *dp = dmu_tx_pool(tx); 3076 dsl_dataset_t *ds; 3077 int error; 3078 uint64_t newval, unique; 3079 3080 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 3081 return (SET_ERROR(ENOTSUP)); 3082 3083 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3084 if (error != 0) 3085 return (error); 3086 3087 if (ds->ds_is_snapshot) { 3088 dsl_dataset_rele(ds, FTAG); 3089 return (SET_ERROR(EINVAL)); 3090 } 3091 3092 error = dsl_prop_predict(ds->ds_dir, 3093 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3094 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3095 if (error != 0) { 3096 dsl_dataset_rele(ds, FTAG); 3097 return (error); 3098 } 3099 3100 /* 3101 * If we are doing the preliminary check in open context, the 3102 * space estimates may be inaccurate. 3103 */ 3104 if (!dmu_tx_is_syncing(tx)) { 3105 dsl_dataset_rele(ds, FTAG); 3106 return (0); 3107 } 3108 3109 mutex_enter(&ds->ds_lock); 3110 if (!DS_UNIQUE_IS_ACCURATE(ds)) 3111 dsl_dataset_recalc_head_uniq(ds); 3112 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3113 mutex_exit(&ds->ds_lock); 3114 3115 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 3116 uint64_t delta = MAX(unique, newval) - 3117 MAX(unique, ds->ds_reserved); 3118 3119 if (delta > 3120 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 3121 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 3122 dsl_dataset_rele(ds, FTAG); 3123 return (SET_ERROR(ENOSPC)); 3124 } 3125 } 3126 3127 dsl_dataset_rele(ds, FTAG); 3128 return (0); 3129} 3130 3131void 3132dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 3133 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 3134{ 3135 uint64_t newval; 3136 uint64_t unique; 3137 int64_t delta; 3138 3139 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3140 source, sizeof (value), 1, &value, tx); 3141 3142 VERIFY0(dsl_prop_get_int_ds(ds, 3143 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 3144 3145 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3146 mutex_enter(&ds->ds_dir->dd_lock); 3147 mutex_enter(&ds->ds_lock); 3148 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3149 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3150 delta = MAX(0, (int64_t)(newval - unique)) - 3151 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3152 ds->ds_reserved = newval; 3153 mutex_exit(&ds->ds_lock); 3154 3155 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3156 mutex_exit(&ds->ds_dir->dd_lock); 3157} 3158 3159static void 3160dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 3161{ 3162 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3163 dsl_pool_t *dp = dmu_tx_pool(tx); 3164 dsl_dataset_t *ds; 3165 3166 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3167 dsl_dataset_set_refreservation_sync_impl(ds, 3168 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 3169 dsl_dataset_rele(ds, FTAG); 3170} 3171 3172int 3173dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 3174 uint64_t refreservation) 3175{ 3176 dsl_dataset_set_qr_arg_t ddsqra; 3177 3178 ddsqra.ddsqra_name = dsname; 3179 ddsqra.ddsqra_source = source; 3180 ddsqra.ddsqra_value = refreservation; 3181 3182 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 3183 dsl_dataset_set_refreservation_sync, &ddsqra, 3184 0, ZFS_SPACE_CHECK_NONE)); 3185} 3186 3187/* 3188 * Return (in *usedp) the amount of space written in new that is not 3189 * present in oldsnap. New may be a snapshot or the head. Old must be 3190 * a snapshot before new, in new's filesystem (or its origin). If not then 3191 * fail and return EINVAL. 3192 * 3193 * The written space is calculated by considering two components: First, we 3194 * ignore any freed space, and calculate the written as new's used space 3195 * minus old's used space. Next, we add in the amount of space that was freed 3196 * between the two snapshots, thus reducing new's used space relative to old's. 3197 * Specifically, this is the space that was born before old->ds_creation_txg, 3198 * and freed before new (ie. on new's deadlist or a previous deadlist). 3199 * 3200 * space freed [---------------------] 3201 * snapshots ---O-------O--------O-------O------ 3202 * oldsnap new 3203 */ 3204int 3205dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 3206 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3207{ 3208 int err = 0; 3209 uint64_t snapobj; 3210 dsl_pool_t *dp = new->ds_dir->dd_pool; 3211 3212 ASSERT(dsl_pool_config_held(dp)); 3213 3214 *usedp = 0; 3215 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; 3216 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes; 3217 3218 *compp = 0; 3219 *compp += dsl_dataset_phys(new)->ds_compressed_bytes; 3220 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes; 3221 3222 *uncompp = 0; 3223 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; 3224 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes; 3225 3226 snapobj = new->ds_object; 3227 while (snapobj != oldsnap->ds_object) { 3228 dsl_dataset_t *snap; 3229 uint64_t used, comp, uncomp; 3230 3231 if (snapobj == new->ds_object) { 3232 snap = new; 3233 } else { 3234 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3235 if (err != 0) 3236 break; 3237 } 3238 3239 if (dsl_dataset_phys(snap)->ds_prev_snap_txg == 3240 dsl_dataset_phys(oldsnap)->ds_creation_txg) { 3241 /* 3242 * The blocks in the deadlist can not be born after 3243 * ds_prev_snap_txg, so get the whole deadlist space, 3244 * which is more efficient (especially for old-format 3245 * deadlists). Unfortunately the deadlist code 3246 * doesn't have enough information to make this 3247 * optimization itself. 3248 */ 3249 dsl_deadlist_space(&snap->ds_deadlist, 3250 &used, &comp, &uncomp); 3251 } else { 3252 dsl_deadlist_space_range(&snap->ds_deadlist, 3253 0, dsl_dataset_phys(oldsnap)->ds_creation_txg, 3254 &used, &comp, &uncomp); 3255 } 3256 *usedp += used; 3257 *compp += comp; 3258 *uncompp += uncomp; 3259 3260 /* 3261 * If we get to the beginning of the chain of snapshots 3262 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3263 * was not a snapshot of/before new. 3264 */ 3265 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 3266 if (snap != new) 3267 dsl_dataset_rele(snap, FTAG); 3268 if (snapobj == 0) { 3269 err = SET_ERROR(EINVAL); 3270 break; 3271 } 3272 3273 } 3274 return (err); 3275} 3276 3277/* 3278 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3279 * lastsnap, and all snapshots in between are deleted. 3280 * 3281 * blocks that would be freed [---------------------------] 3282 * snapshots ---O-------O--------O-------O--------O 3283 * firstsnap lastsnap 3284 * 3285 * This is the set of blocks that were born after the snap before firstsnap, 3286 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3287 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3288 * We calculate this by iterating over the relevant deadlists (from the snap 3289 * after lastsnap, backward to the snap after firstsnap), summing up the 3290 * space on the deadlist that was born after the snap before firstsnap. 3291 */ 3292int 3293dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3294 dsl_dataset_t *lastsnap, 3295 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3296{ 3297 int err = 0; 3298 uint64_t snapobj; 3299 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3300 3301 ASSERT(firstsnap->ds_is_snapshot); 3302 ASSERT(lastsnap->ds_is_snapshot); 3303 3304 /* 3305 * Check that the snapshots are in the same dsl_dir, and firstsnap 3306 * is before lastsnap. 3307 */ 3308 if (firstsnap->ds_dir != lastsnap->ds_dir || 3309 dsl_dataset_phys(firstsnap)->ds_creation_txg > 3310 dsl_dataset_phys(lastsnap)->ds_creation_txg) 3311 return (SET_ERROR(EINVAL)); 3312 3313 *usedp = *compp = *uncompp = 0; 3314 3315 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; 3316 while (snapobj != firstsnap->ds_object) { 3317 dsl_dataset_t *ds; 3318 uint64_t used, comp, uncomp; 3319 3320 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3321 if (err != 0) 3322 break; 3323 3324 dsl_deadlist_space_range(&ds->ds_deadlist, 3325 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, 3326 &used, &comp, &uncomp); 3327 *usedp += used; 3328 *compp += comp; 3329 *uncompp += uncomp; 3330 3331 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3332 ASSERT3U(snapobj, !=, 0); 3333 dsl_dataset_rele(ds, FTAG); 3334 } 3335 return (err); 3336} 3337 3338static int 3339dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx) 3340{ 3341 const char *dsname = arg; 3342 dsl_dataset_t *ds; 3343 dsl_pool_t *dp = dmu_tx_pool(tx); 3344 int error = 0; 3345 3346 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) 3347 return (SET_ERROR(ENOTSUP)); 3348 3349 ASSERT(spa_feature_is_enabled(dp->dp_spa, 3350 SPA_FEATURE_EXTENSIBLE_DATASET)); 3351 3352 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 3353 if (error != 0) 3354 return (error); 3355 3356 if (ds->ds_large_blocks) 3357 error = EALREADY; 3358 dsl_dataset_rele(ds, FTAG); 3359 3360 return (error); 3361} 3362 3363void 3364dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx) 3365{ 3366 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 3367 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 3368 uint64_t zero = 0; 3369 3370 spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx); 3371 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 3372 3373 VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS, 3374 sizeof (zero), 1, &zero, tx)); 3375} 3376 3377static void 3378dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx) 3379{ 3380 const char *dsname = arg; 3381 dsl_dataset_t *ds; 3382 3383 VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds)); 3384 3385 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx); 3386 ASSERT(!ds->ds_large_blocks); 3387 ds->ds_large_blocks = B_TRUE; 3388 dsl_dataset_rele(ds, FTAG); 3389} 3390 3391int 3392dsl_dataset_activate_large_blocks(const char *dsname) 3393{ 3394 int error; 3395 3396 error = dsl_sync_task(dsname, 3397 dsl_dataset_activate_large_blocks_check, 3398 dsl_dataset_activate_large_blocks_sync, (void *)dsname, 3399 1, ZFS_SPACE_CHECK_RESERVED); 3400 3401 /* 3402 * EALREADY indicates that this dataset already supports large blocks. 3403 */ 3404 if (error == EALREADY) 3405 error = 0; 3406 return (error); 3407} 3408 3409/* 3410 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 3411 * For example, they could both be snapshots of the same filesystem, and 3412 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 3413 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 3414 * filesystem. Or 'earlier' could be the origin's origin. 3415 * 3416 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 3417 */ 3418boolean_t 3419dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 3420 uint64_t earlier_txg) 3421{ 3422 dsl_pool_t *dp = later->ds_dir->dd_pool; 3423 int error; 3424 boolean_t ret; 3425 3426 ASSERT(dsl_pool_config_held(dp)); 3427 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); 3428 3429 if (earlier_txg == 0) 3430 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; 3431 3432 if (later->ds_is_snapshot && 3433 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) 3434 return (B_FALSE); 3435 3436 if (later->ds_dir == earlier->ds_dir) 3437 return (B_TRUE); 3438 if (!dsl_dir_is_clone(later->ds_dir)) 3439 return (B_FALSE); 3440 3441 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object) 3442 return (B_TRUE); 3443 dsl_dataset_t *origin; 3444 error = dsl_dataset_hold_obj(dp, 3445 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); 3446 if (error != 0) 3447 return (B_FALSE); 3448 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 3449 dsl_dataset_rele(origin, FTAG); 3450 return (ret); 3451} 3452 3453 3454void 3455dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 3456{ 3457 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3458 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 3459} 3460