dsl_dataset.c revision 288571
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 dsl_dataset_t *ds; 544 545 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 546 if (err != 0) 547 return (err); 548 549 ASSERT(dsl_pool_config_held(dp)); 550 obj = dsl_dir_phys(dd)->dd_head_dataset_obj; 551 if (obj != 0) 552 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 553 else 554 err = SET_ERROR(ENOENT); 555 556 /* we may be looking for a snapshot */ 557 if (err == 0 && snapname != NULL) { 558 dsl_dataset_t *snap_ds; 559 560 if (*snapname++ != '@') { 561 dsl_dataset_rele(ds, tag); 562 dsl_dir_rele(dd, FTAG); 563 return (SET_ERROR(ENOENT)); 564 } 565 566 dprintf("looking for snapshot '%s'\n", snapname); 567 err = dsl_dataset_snap_lookup(ds, snapname, &obj); 568 if (err == 0) 569 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds); 570 dsl_dataset_rele(ds, tag); 571 572 if (err == 0) { 573 mutex_enter(&snap_ds->ds_lock); 574 if (snap_ds->ds_snapname[0] == 0) 575 (void) strlcpy(snap_ds->ds_snapname, snapname, 576 sizeof (snap_ds->ds_snapname)); 577 mutex_exit(&snap_ds->ds_lock); 578 ds = snap_ds; 579 } 580 } 581 if (err == 0) 582 *dsp = ds; 583 dsl_dir_rele(dd, FTAG); 584 return (err); 585} 586 587int 588dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 589 void *tag, dsl_dataset_t **dsp) 590{ 591 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 592 if (err != 0) 593 return (err); 594 if (!dsl_dataset_tryown(*dsp, tag)) { 595 dsl_dataset_rele(*dsp, tag); 596 *dsp = NULL; 597 return (SET_ERROR(EBUSY)); 598 } 599 return (0); 600} 601 602int 603dsl_dataset_own(dsl_pool_t *dp, const char *name, 604 void *tag, dsl_dataset_t **dsp) 605{ 606 int err = dsl_dataset_hold(dp, name, tag, dsp); 607 if (err != 0) 608 return (err); 609 if (!dsl_dataset_tryown(*dsp, tag)) { 610 dsl_dataset_rele(*dsp, tag); 611 return (SET_ERROR(EBUSY)); 612 } 613 return (0); 614} 615 616/* 617 * See the comment above dsl_pool_hold() for details. In summary, a long 618 * hold is used to prevent destruction of a dataset while the pool hold 619 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 620 * 621 * The dataset and pool must be held when this function is called. After it 622 * is called, the pool hold may be released while the dataset is still held 623 * and accessed. 624 */ 625void 626dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 627{ 628 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 629 (void) refcount_add(&ds->ds_longholds, tag); 630} 631 632void 633dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 634{ 635 (void) refcount_remove(&ds->ds_longholds, tag); 636} 637 638/* Return B_TRUE if there are any long holds on this dataset. */ 639boolean_t 640dsl_dataset_long_held(dsl_dataset_t *ds) 641{ 642 return (!refcount_is_zero(&ds->ds_longholds)); 643} 644 645void 646dsl_dataset_name(dsl_dataset_t *ds, char *name) 647{ 648 if (ds == NULL) { 649 (void) strcpy(name, "mos"); 650 } else { 651 dsl_dir_name(ds->ds_dir, name); 652 VERIFY0(dsl_dataset_get_snapname(ds)); 653 if (ds->ds_snapname[0]) { 654 (void) strcat(name, "@"); 655 /* 656 * We use a "recursive" mutex so that we 657 * can call dprintf_ds() with ds_lock held. 658 */ 659 if (!MUTEX_HELD(&ds->ds_lock)) { 660 mutex_enter(&ds->ds_lock); 661 (void) strcat(name, ds->ds_snapname); 662 mutex_exit(&ds->ds_lock); 663 } else { 664 (void) strcat(name, ds->ds_snapname); 665 } 666 } 667 } 668} 669 670void 671dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 672{ 673 dmu_buf_rele(ds->ds_dbuf, tag); 674} 675 676void 677dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 678{ 679 ASSERT3P(ds->ds_owner, ==, tag); 680 ASSERT(ds->ds_dbuf != NULL); 681 682 mutex_enter(&ds->ds_lock); 683 ds->ds_owner = NULL; 684 mutex_exit(&ds->ds_lock); 685 dsl_dataset_long_rele(ds, tag); 686 dsl_dataset_rele(ds, tag); 687} 688 689boolean_t 690dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 691{ 692 boolean_t gotit = FALSE; 693 694 mutex_enter(&ds->ds_lock); 695 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 696 ds->ds_owner = tag; 697 dsl_dataset_long_hold(ds, tag); 698 gotit = TRUE; 699 } 700 mutex_exit(&ds->ds_lock); 701 return (gotit); 702} 703 704uint64_t 705dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 706 uint64_t flags, dmu_tx_t *tx) 707{ 708 dsl_pool_t *dp = dd->dd_pool; 709 dmu_buf_t *dbuf; 710 dsl_dataset_phys_t *dsphys; 711 uint64_t dsobj; 712 objset_t *mos = dp->dp_meta_objset; 713 714 if (origin == NULL) 715 origin = dp->dp_origin_snap; 716 717 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 718 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); 719 ASSERT(dmu_tx_is_syncing(tx)); 720 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 721 722 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 723 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 724 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 725 dmu_buf_will_dirty(dbuf, tx); 726 dsphys = dbuf->db_data; 727 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 728 dsphys->ds_dir_obj = dd->dd_object; 729 dsphys->ds_flags = flags; 730 dsphys->ds_fsid_guid = unique_create(); 731 do { 732 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 733 sizeof (dsphys->ds_guid)); 734 } while (dsphys->ds_guid == 0); 735 dsphys->ds_snapnames_zapobj = 736 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 737 DMU_OT_NONE, 0, tx); 738 dsphys->ds_creation_time = gethrestime_sec(); 739 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 740 741 if (origin == NULL) { 742 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 743 } else { 744 dsl_dataset_t *ohds; /* head of the origin snapshot */ 745 746 dsphys->ds_prev_snap_obj = origin->ds_object; 747 dsphys->ds_prev_snap_txg = 748 dsl_dataset_phys(origin)->ds_creation_txg; 749 dsphys->ds_referenced_bytes = 750 dsl_dataset_phys(origin)->ds_referenced_bytes; 751 dsphys->ds_compressed_bytes = 752 dsl_dataset_phys(origin)->ds_compressed_bytes; 753 dsphys->ds_uncompressed_bytes = 754 dsl_dataset_phys(origin)->ds_uncompressed_bytes; 755 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; 756 757 /* 758 * Inherit flags that describe the dataset's contents 759 * (INCONSISTENT) or properties (Case Insensitive). 760 */ 761 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & 762 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 763 764 if (origin->ds_large_blocks) 765 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); 766 767 dmu_buf_will_dirty(origin->ds_dbuf, tx); 768 dsl_dataset_phys(origin)->ds_num_children++; 769 770 VERIFY0(dsl_dataset_hold_obj(dp, 771 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, 772 FTAG, &ohds)); 773 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 774 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 775 dsl_dataset_rele(ohds, FTAG); 776 777 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 778 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { 779 dsl_dataset_phys(origin)->ds_next_clones_obj = 780 zap_create(mos, 781 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 782 } 783 VERIFY0(zap_add_int(mos, 784 dsl_dataset_phys(origin)->ds_next_clones_obj, 785 dsobj, tx)); 786 } 787 788 dmu_buf_will_dirty(dd->dd_dbuf, tx); 789 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; 790 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 791 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 792 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 793 dsl_dir_phys(origin->ds_dir)->dd_clones = 794 zap_create(mos, 795 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 796 } 797 VERIFY0(zap_add_int(mos, 798 dsl_dir_phys(origin->ds_dir)->dd_clones, 799 dsobj, tx)); 800 } 801 } 802 803 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 804 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 805 806 dmu_buf_rele(dbuf, FTAG); 807 808 dmu_buf_will_dirty(dd->dd_dbuf, tx); 809 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; 810 811 return (dsobj); 812} 813 814static void 815dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 816{ 817 objset_t *os; 818 819 VERIFY0(dmu_objset_from_ds(ds, &os)); 820 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 821 dsl_dataset_dirty(ds, tx); 822} 823 824uint64_t 825dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 826 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 827{ 828 dsl_pool_t *dp = pdd->dd_pool; 829 uint64_t dsobj, ddobj; 830 dsl_dir_t *dd; 831 832 ASSERT(dmu_tx_is_syncing(tx)); 833 ASSERT(lastname[0] != '@'); 834 835 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 836 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 837 838 dsobj = dsl_dataset_create_sync_dd(dd, origin, 839 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 840 841 dsl_deleg_set_create_perms(dd, tx, cr); 842 843 /* 844 * Since we're creating a new node we know it's a leaf, so we can 845 * initialize the counts if the limit feature is active. 846 */ 847 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 848 uint64_t cnt = 0; 849 objset_t *os = dd->dd_pool->dp_meta_objset; 850 851 dsl_dir_zapify(dd, tx); 852 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 853 sizeof (cnt), 1, &cnt, tx)); 854 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 855 sizeof (cnt), 1, &cnt, tx)); 856 } 857 858 dsl_dir_rele(dd, FTAG); 859 860 /* 861 * If we are creating a clone, make sure we zero out any stale 862 * data from the origin snapshots zil header. 863 */ 864 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 865 dsl_dataset_t *ds; 866 867 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 868 dsl_dataset_zero_zil(ds, tx); 869 dsl_dataset_rele(ds, FTAG); 870 } 871 872 return (dsobj); 873} 874 875#ifdef __FreeBSD__ 876/* FreeBSD ioctl compat begin */ 877struct destroyarg { 878 nvlist_t *nvl; 879 const char *snapname; 880}; 881 882static int 883dsl_check_snap_cb(const char *name, void *arg) 884{ 885 struct destroyarg *da = arg; 886 dsl_dataset_t *ds; 887 char *dsname; 888 889 dsname = kmem_asprintf("%s@%s", name, da->snapname); 890 fnvlist_add_boolean(da->nvl, dsname); 891 kmem_free(dsname, strlen(dsname) + 1); 892 893 return (0); 894} 895 896int 897dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname, 898 nvlist_t *snaps) 899{ 900 struct destroyarg *da; 901 int err; 902 903 da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP); 904 da->nvl = snaps; 905 da->snapname = snapname; 906 err = dmu_objset_find(fsname, dsl_check_snap_cb, da, 907 DS_FIND_CHILDREN); 908 kmem_free(da, sizeof (struct destroyarg)); 909 910 return (err); 911} 912/* FreeBSD ioctl compat end */ 913#endif /* __FreeBSD__ */ 914 915/* 916 * The unique space in the head dataset can be calculated by subtracting 917 * the space used in the most recent snapshot, that is still being used 918 * in this file system, from the space currently in use. To figure out 919 * the space in the most recent snapshot still in use, we need to take 920 * the total space used in the snapshot and subtract out the space that 921 * has been freed up since the snapshot was taken. 922 */ 923void 924dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 925{ 926 uint64_t mrs_used; 927 uint64_t dlused, dlcomp, dluncomp; 928 929 ASSERT(!ds->ds_is_snapshot); 930 931 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) 932 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; 933 else 934 mrs_used = 0; 935 936 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 937 938 ASSERT3U(dlused, <=, mrs_used); 939 dsl_dataset_phys(ds)->ds_unique_bytes = 940 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); 941 942 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 943 SPA_VERSION_UNIQUE_ACCURATE) 944 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 945} 946 947void 948dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 949 dmu_tx_t *tx) 950{ 951 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 952 uint64_t count; 953 int err; 954 955 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); 956 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 957 obj, tx); 958 /* 959 * The err should not be ENOENT, but a bug in a previous version 960 * of the code could cause upgrade_clones_cb() to not set 961 * ds_next_snap_obj when it should, leading to a missing entry. 962 * If we knew that the pool was created after 963 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 964 * ENOENT. However, at least we can check that we don't have 965 * too many entries in the next_clones_obj even after failing to 966 * remove this one. 967 */ 968 if (err != ENOENT) 969 VERIFY0(err); 970 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 971 &count)); 972 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); 973} 974 975 976blkptr_t * 977dsl_dataset_get_blkptr(dsl_dataset_t *ds) 978{ 979 return (&dsl_dataset_phys(ds)->ds_bp); 980} 981 982void 983dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 984{ 985 ASSERT(dmu_tx_is_syncing(tx)); 986 /* If it's the meta-objset, set dp_meta_rootbp */ 987 if (ds == NULL) { 988 tx->tx_pool->dp_meta_rootbp = *bp; 989 } else { 990 dmu_buf_will_dirty(ds->ds_dbuf, tx); 991 dsl_dataset_phys(ds)->ds_bp = *bp; 992 } 993} 994 995spa_t * 996dsl_dataset_get_spa(dsl_dataset_t *ds) 997{ 998 return (ds->ds_dir->dd_pool->dp_spa); 999} 1000 1001void 1002dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1003{ 1004 dsl_pool_t *dp; 1005 1006 if (ds == NULL) /* this is the meta-objset */ 1007 return; 1008 1009 ASSERT(ds->ds_objset != NULL); 1010 1011 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) 1012 panic("dirtying snapshot!"); 1013 1014 dp = ds->ds_dir->dd_pool; 1015 1016 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1017 /* up the hold count until we can be written out */ 1018 dmu_buf_add_ref(ds->ds_dbuf, ds); 1019 } 1020} 1021 1022boolean_t 1023dsl_dataset_is_dirty(dsl_dataset_t *ds) 1024{ 1025 for (int t = 0; t < TXG_SIZE; t++) { 1026 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1027 ds, t)) 1028 return (B_TRUE); 1029 } 1030 return (B_FALSE); 1031} 1032 1033static int 1034dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1035{ 1036 uint64_t asize; 1037 1038 if (!dmu_tx_is_syncing(tx)) 1039 return (0); 1040 1041 /* 1042 * If there's an fs-only reservation, any blocks that might become 1043 * owned by the snapshot dataset must be accommodated by space 1044 * outside of the reservation. 1045 */ 1046 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1047 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); 1048 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1049 return (SET_ERROR(ENOSPC)); 1050 1051 /* 1052 * Propagate any reserved space for this snapshot to other 1053 * snapshot checks in this sync group. 1054 */ 1055 if (asize > 0) 1056 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1057 1058 return (0); 1059} 1060 1061typedef struct dsl_dataset_snapshot_arg { 1062 nvlist_t *ddsa_snaps; 1063 nvlist_t *ddsa_props; 1064 nvlist_t *ddsa_errors; 1065 cred_t *ddsa_cr; 1066} dsl_dataset_snapshot_arg_t; 1067 1068int 1069dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1070 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 1071{ 1072 int error; 1073 uint64_t value; 1074 1075 ds->ds_trysnap_txg = tx->tx_txg; 1076 1077 if (!dmu_tx_is_syncing(tx)) 1078 return (0); 1079 1080 /* 1081 * We don't allow multiple snapshots of the same txg. If there 1082 * is already one, try again. 1083 */ 1084 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) 1085 return (SET_ERROR(EAGAIN)); 1086 1087 /* 1088 * Check for conflicting snapshot name. 1089 */ 1090 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1091 if (error == 0) 1092 return (SET_ERROR(EEXIST)); 1093 if (error != ENOENT) 1094 return (error); 1095 1096 /* 1097 * We don't allow taking snapshots of inconsistent datasets, such as 1098 * those into which we are currently receiving. However, if we are 1099 * creating this snapshot as part of a receive, this check will be 1100 * executed atomically with respect to the completion of the receive 1101 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1102 * case we ignore this, knowing it will be fixed up for us shortly in 1103 * dmu_recv_end_sync(). 1104 */ 1105 if (!recv && DS_IS_INCONSISTENT(ds)) 1106 return (SET_ERROR(EBUSY)); 1107 1108 /* 1109 * Skip the check for temporary snapshots or if we have already checked 1110 * the counts in dsl_dataset_snapshot_check. This means we really only 1111 * check the count here when we're receiving a stream. 1112 */ 1113 if (cnt != 0 && cr != NULL) { 1114 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1115 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 1116 if (error != 0) 1117 return (error); 1118 } 1119 1120 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1121 if (error != 0) 1122 return (error); 1123 1124 return (0); 1125} 1126 1127static int 1128dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1129{ 1130 dsl_dataset_snapshot_arg_t *ddsa = arg; 1131 dsl_pool_t *dp = dmu_tx_pool(tx); 1132 nvpair_t *pair; 1133 int rv = 0; 1134 1135 /* 1136 * Pre-compute how many total new snapshots will be created for each 1137 * level in the tree and below. This is needed for validating the 1138 * snapshot limit when either taking a recursive snapshot or when 1139 * taking multiple snapshots. 1140 * 1141 * The problem is that the counts are not actually adjusted when 1142 * we are checking, only when we finally sync. For a single snapshot, 1143 * this is easy, the count will increase by 1 at each node up the tree, 1144 * but its more complicated for the recursive/multiple snapshot case. 1145 * 1146 * The dsl_fs_ss_limit_check function does recursively check the count 1147 * at each level up the tree but since it is validating each snapshot 1148 * independently we need to be sure that we are validating the complete 1149 * count for the entire set of snapshots. We do this by rolling up the 1150 * counts for each component of the name into an nvlist and then 1151 * checking each of those cases with the aggregated count. 1152 * 1153 * This approach properly handles not only the recursive snapshot 1154 * case (where we get all of those on the ddsa_snaps list) but also 1155 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1156 * validate the limit on 'a' using a count of 2). 1157 * 1158 * We validate the snapshot names in the third loop and only report 1159 * name errors once. 1160 */ 1161 if (dmu_tx_is_syncing(tx)) { 1162 nvlist_t *cnt_track = NULL; 1163 cnt_track = fnvlist_alloc(); 1164 1165 /* Rollup aggregated counts into the cnt_track list */ 1166 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1167 pair != NULL; 1168 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1169 char *pdelim; 1170 uint64_t val; 1171 char nm[MAXPATHLEN]; 1172 1173 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1174 pdelim = strchr(nm, '@'); 1175 if (pdelim == NULL) 1176 continue; 1177 *pdelim = '\0'; 1178 1179 do { 1180 if (nvlist_lookup_uint64(cnt_track, nm, 1181 &val) == 0) { 1182 /* update existing entry */ 1183 fnvlist_add_uint64(cnt_track, nm, 1184 val + 1); 1185 } else { 1186 /* add to list */ 1187 fnvlist_add_uint64(cnt_track, nm, 1); 1188 } 1189 1190 pdelim = strrchr(nm, '/'); 1191 if (pdelim != NULL) 1192 *pdelim = '\0'; 1193 } while (pdelim != NULL); 1194 } 1195 1196 /* Check aggregated counts at each level */ 1197 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1198 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1199 int error = 0; 1200 char *name; 1201 uint64_t cnt = 0; 1202 dsl_dataset_t *ds; 1203 1204 name = nvpair_name(pair); 1205 cnt = fnvpair_value_uint64(pair); 1206 ASSERT(cnt > 0); 1207 1208 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1209 if (error == 0) { 1210 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1211 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1212 ddsa->ddsa_cr); 1213 dsl_dataset_rele(ds, FTAG); 1214 } 1215 1216 if (error != 0) { 1217 if (ddsa->ddsa_errors != NULL) 1218 fnvlist_add_int32(ddsa->ddsa_errors, 1219 name, error); 1220 rv = error; 1221 /* only report one error for this check */ 1222 break; 1223 } 1224 } 1225 nvlist_free(cnt_track); 1226 } 1227 1228 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1229 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1230 int error = 0; 1231 dsl_dataset_t *ds; 1232 char *name, *atp; 1233 char dsname[MAXNAMELEN]; 1234 1235 name = nvpair_name(pair); 1236 if (strlen(name) >= MAXNAMELEN) 1237 error = SET_ERROR(ENAMETOOLONG); 1238 if (error == 0) { 1239 atp = strchr(name, '@'); 1240 if (atp == NULL) 1241 error = SET_ERROR(EINVAL); 1242 if (error == 0) 1243 (void) strlcpy(dsname, name, atp - name + 1); 1244 } 1245 if (error == 0) 1246 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1247 if (error == 0) { 1248 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1249 error = dsl_dataset_snapshot_check_impl(ds, 1250 atp + 1, tx, B_FALSE, 0, NULL); 1251 dsl_dataset_rele(ds, FTAG); 1252 } 1253 1254 if (error != 0) { 1255 if (ddsa->ddsa_errors != NULL) { 1256 fnvlist_add_int32(ddsa->ddsa_errors, 1257 name, error); 1258 } 1259 rv = error; 1260 } 1261 } 1262 1263 return (rv); 1264} 1265 1266void 1267dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1268 dmu_tx_t *tx) 1269{ 1270 static zil_header_t zero_zil; 1271 1272 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1273 dmu_buf_t *dbuf; 1274 dsl_dataset_phys_t *dsphys; 1275 uint64_t dsobj, crtxg; 1276 objset_t *mos = dp->dp_meta_objset; 1277 objset_t *os; 1278 1279 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1280 1281 /* 1282 * If we are on an old pool, the zil must not be active, in which 1283 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1284 */ 1285 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1286 dmu_objset_from_ds(ds, &os) != 0 || 1287 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1288 sizeof (zero_zil)) == 0); 1289 1290 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1291 1292 /* 1293 * The origin's ds_creation_txg has to be < TXG_INITIAL 1294 */ 1295 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1296 crtxg = 1; 1297 else 1298 crtxg = tx->tx_txg; 1299 1300 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1301 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1302 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1303 dmu_buf_will_dirty(dbuf, tx); 1304 dsphys = dbuf->db_data; 1305 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1306 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1307 dsphys->ds_fsid_guid = unique_create(); 1308 do { 1309 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1310 sizeof (dsphys->ds_guid)); 1311 } while (dsphys->ds_guid == 0); 1312 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1313 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 1314 dsphys->ds_next_snap_obj = ds->ds_object; 1315 dsphys->ds_num_children = 1; 1316 dsphys->ds_creation_time = gethrestime_sec(); 1317 dsphys->ds_creation_txg = crtxg; 1318 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; 1319 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; 1320 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; 1321 dsphys->ds_uncompressed_bytes = 1322 dsl_dataset_phys(ds)->ds_uncompressed_bytes; 1323 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; 1324 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; 1325 dmu_buf_rele(dbuf, FTAG); 1326 1327 if (ds->ds_large_blocks) 1328 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); 1329 1330 ASSERT3U(ds->ds_prev != 0, ==, 1331 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); 1332 if (ds->ds_prev) { 1333 uint64_t next_clones_obj = 1334 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; 1335 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1336 ds->ds_object || 1337 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); 1338 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1339 ds->ds_object) { 1340 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1341 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, 1342 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); 1343 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; 1344 } else if (next_clones_obj != 0) { 1345 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1346 dsphys->ds_next_snap_obj, tx); 1347 VERIFY0(zap_add_int(mos, 1348 next_clones_obj, dsobj, tx)); 1349 } 1350 } 1351 1352 /* 1353 * If we have a reference-reservation on this dataset, we will 1354 * need to increase the amount of refreservation being charged 1355 * since our unique space is going to zero. 1356 */ 1357 if (ds->ds_reserved) { 1358 int64_t delta; 1359 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1360 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, 1361 ds->ds_reserved); 1362 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1363 delta, 0, 0, tx); 1364 } 1365 1366 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1367 dsl_dataset_phys(ds)->ds_deadlist_obj = 1368 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, 1369 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 1370 dsl_deadlist_close(&ds->ds_deadlist); 1371 dsl_deadlist_open(&ds->ds_deadlist, mos, 1372 dsl_dataset_phys(ds)->ds_deadlist_obj); 1373 dsl_deadlist_add_key(&ds->ds_deadlist, 1374 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); 1375 1376 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); 1377 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; 1378 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; 1379 dsl_dataset_phys(ds)->ds_unique_bytes = 0; 1380 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1381 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1382 1383 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1384 snapname, 8, 1, &dsobj, tx)); 1385 1386 if (ds->ds_prev) 1387 dsl_dataset_rele(ds->ds_prev, ds); 1388 VERIFY0(dsl_dataset_hold_obj(dp, 1389 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); 1390 1391 dsl_scan_ds_snapshotted(ds, tx); 1392 1393 dsl_dir_snap_cmtime_update(ds->ds_dir); 1394 1395 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1396} 1397 1398static void 1399dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1400{ 1401 dsl_dataset_snapshot_arg_t *ddsa = arg; 1402 dsl_pool_t *dp = dmu_tx_pool(tx); 1403 nvpair_t *pair; 1404 1405 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1406 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1407 dsl_dataset_t *ds; 1408 char *name, *atp; 1409 char dsname[MAXNAMELEN]; 1410 1411 name = nvpair_name(pair); 1412 atp = strchr(name, '@'); 1413 (void) strlcpy(dsname, name, atp - name + 1); 1414 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1415 1416 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1417 if (ddsa->ddsa_props != NULL) { 1418 dsl_props_set_sync_impl(ds->ds_prev, 1419 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1420 } 1421 dsl_dataset_rele(ds, FTAG); 1422 } 1423} 1424 1425/* 1426 * The snapshots must all be in the same pool. 1427 * All-or-nothing: if there are any failures, nothing will be modified. 1428 */ 1429int 1430dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1431{ 1432 dsl_dataset_snapshot_arg_t ddsa; 1433 nvpair_t *pair; 1434 boolean_t needsuspend; 1435 int error; 1436 spa_t *spa; 1437 char *firstname; 1438 nvlist_t *suspended = NULL; 1439 1440 pair = nvlist_next_nvpair(snaps, NULL); 1441 if (pair == NULL) 1442 return (0); 1443 firstname = nvpair_name(pair); 1444 1445 error = spa_open(firstname, &spa, FTAG); 1446 if (error != 0) 1447 return (error); 1448 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1449 spa_close(spa, FTAG); 1450 1451 if (needsuspend) { 1452 suspended = fnvlist_alloc(); 1453 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1454 pair = nvlist_next_nvpair(snaps, pair)) { 1455 char fsname[MAXNAMELEN]; 1456 char *snapname = nvpair_name(pair); 1457 char *atp; 1458 void *cookie; 1459 1460 atp = strchr(snapname, '@'); 1461 if (atp == NULL) { 1462 error = SET_ERROR(EINVAL); 1463 break; 1464 } 1465 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1466 1467 error = zil_suspend(fsname, &cookie); 1468 if (error != 0) 1469 break; 1470 fnvlist_add_uint64(suspended, fsname, 1471 (uintptr_t)cookie); 1472 } 1473 } 1474 1475 ddsa.ddsa_snaps = snaps; 1476 ddsa.ddsa_props = props; 1477 ddsa.ddsa_errors = errors; 1478 ddsa.ddsa_cr = CRED(); 1479 1480 if (error == 0) { 1481 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1482 dsl_dataset_snapshot_sync, &ddsa, 1483 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1484 } 1485 1486 if (suspended != NULL) { 1487 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1488 pair = nvlist_next_nvpair(suspended, pair)) { 1489 zil_resume((void *)(uintptr_t) 1490 fnvpair_value_uint64(pair)); 1491 } 1492 fnvlist_free(suspended); 1493 } 1494 1495#ifdef __FreeBSD__ 1496#ifdef _KERNEL 1497 if (error == 0) { 1498 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1499 pair = nvlist_next_nvpair(snaps, pair)) { 1500 char *snapname = nvpair_name(pair); 1501 zvol_create_minors(snapname); 1502 } 1503 } 1504#endif 1505#endif 1506 return (error); 1507} 1508 1509typedef struct dsl_dataset_snapshot_tmp_arg { 1510 const char *ddsta_fsname; 1511 const char *ddsta_snapname; 1512 minor_t ddsta_cleanup_minor; 1513 const char *ddsta_htag; 1514} dsl_dataset_snapshot_tmp_arg_t; 1515 1516static int 1517dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1518{ 1519 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1520 dsl_pool_t *dp = dmu_tx_pool(tx); 1521 dsl_dataset_t *ds; 1522 int error; 1523 1524 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1525 if (error != 0) 1526 return (error); 1527 1528 /* NULL cred means no limit check for tmp snapshot */ 1529 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1530 tx, B_FALSE, 0, NULL); 1531 if (error != 0) { 1532 dsl_dataset_rele(ds, FTAG); 1533 return (error); 1534 } 1535 1536 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1537 dsl_dataset_rele(ds, FTAG); 1538 return (SET_ERROR(ENOTSUP)); 1539 } 1540 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1541 B_TRUE, tx); 1542 if (error != 0) { 1543 dsl_dataset_rele(ds, FTAG); 1544 return (error); 1545 } 1546 1547 dsl_dataset_rele(ds, FTAG); 1548 return (0); 1549} 1550 1551static void 1552dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1553{ 1554 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1555 dsl_pool_t *dp = dmu_tx_pool(tx); 1556 dsl_dataset_t *ds; 1557 1558 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1559 1560 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1561 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1562 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1563 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1564 1565 dsl_dataset_rele(ds, FTAG); 1566} 1567 1568int 1569dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1570 minor_t cleanup_minor, const char *htag) 1571{ 1572 dsl_dataset_snapshot_tmp_arg_t ddsta; 1573 int error; 1574 spa_t *spa; 1575 boolean_t needsuspend; 1576 void *cookie; 1577 1578 ddsta.ddsta_fsname = fsname; 1579 ddsta.ddsta_snapname = snapname; 1580 ddsta.ddsta_cleanup_minor = cleanup_minor; 1581 ddsta.ddsta_htag = htag; 1582 1583 error = spa_open(fsname, &spa, FTAG); 1584 if (error != 0) 1585 return (error); 1586 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1587 spa_close(spa, FTAG); 1588 1589 if (needsuspend) { 1590 error = zil_suspend(fsname, &cookie); 1591 if (error != 0) 1592 return (error); 1593 } 1594 1595 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1596 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 1597 1598 if (needsuspend) 1599 zil_resume(cookie); 1600 return (error); 1601} 1602 1603 1604void 1605dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1606{ 1607 ASSERT(dmu_tx_is_syncing(tx)); 1608 ASSERT(ds->ds_objset != NULL); 1609 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); 1610 1611 /* 1612 * in case we had to change ds_fsid_guid when we opened it, 1613 * sync it out now. 1614 */ 1615 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1616 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; 1617 1618 dmu_objset_sync(ds->ds_objset, zio, tx); 1619 1620 if (ds->ds_need_large_blocks && !ds->ds_large_blocks) { 1621 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx); 1622 ds->ds_large_blocks = B_TRUE; 1623 } 1624} 1625 1626static void 1627get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1628{ 1629 uint64_t count = 0; 1630 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1631 zap_cursor_t zc; 1632 zap_attribute_t za; 1633 nvlist_t *propval = fnvlist_alloc(); 1634 nvlist_t *val = fnvlist_alloc(); 1635 1636 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1637 1638 /* 1639 * There may be missing entries in ds_next_clones_obj 1640 * due to a bug in a previous version of the code. 1641 * Only trust it if it has the right number of entries. 1642 */ 1643 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { 1644 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1645 &count)); 1646 } 1647 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) 1648 goto fail; 1649 for (zap_cursor_init(&zc, mos, 1650 dsl_dataset_phys(ds)->ds_next_clones_obj); 1651 zap_cursor_retrieve(&zc, &za) == 0; 1652 zap_cursor_advance(&zc)) { 1653 dsl_dataset_t *clone; 1654 char buf[ZFS_MAXNAMELEN]; 1655 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1656 za.za_first_integer, FTAG, &clone)); 1657 dsl_dir_name(clone->ds_dir, buf); 1658 fnvlist_add_boolean(val, buf); 1659 dsl_dataset_rele(clone, FTAG); 1660 } 1661 zap_cursor_fini(&zc); 1662 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1663 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); 1664fail: 1665 nvlist_free(val); 1666 nvlist_free(propval); 1667} 1668 1669void 1670dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1671{ 1672 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1673 uint64_t refd, avail, uobjs, aobjs, ratio; 1674 1675 ASSERT(dsl_pool_config_held(dp)); 1676 1677 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : 1678 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / 1679 dsl_dataset_phys(ds)->ds_compressed_bytes); 1680 1681 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); 1682 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 1683 dsl_dataset_phys(ds)->ds_uncompressed_bytes); 1684 1685 if (ds->ds_is_snapshot) { 1686 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); 1687 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 1688 dsl_dataset_phys(ds)->ds_unique_bytes); 1689 get_clones_stat(ds, nv); 1690 } else { 1691 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 1692 char buf[MAXNAMELEN]; 1693 dsl_dataset_name(ds->ds_prev, buf); 1694 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf); 1695 } 1696 1697 dsl_dir_stats(ds->ds_dir, nv); 1698 } 1699 1700 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 1701 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 1702 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 1703 1704 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 1705 dsl_dataset_phys(ds)->ds_creation_time); 1706 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 1707 dsl_dataset_phys(ds)->ds_creation_txg); 1708 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 1709 ds->ds_quota); 1710 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 1711 ds->ds_reserved); 1712 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 1713 dsl_dataset_phys(ds)->ds_guid); 1714 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 1715 dsl_dataset_phys(ds)->ds_unique_bytes); 1716 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 1717 ds->ds_object); 1718 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 1719 ds->ds_userrefs); 1720 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 1721 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 1722 1723 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 1724 uint64_t written, comp, uncomp; 1725 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1726 dsl_dataset_t *prev; 1727 1728 int err = dsl_dataset_hold_obj(dp, 1729 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 1730 if (err == 0) { 1731 err = dsl_dataset_space_written(prev, ds, &written, 1732 &comp, &uncomp); 1733 dsl_dataset_rele(prev, FTAG); 1734 if (err == 0) { 1735 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 1736 written); 1737 } 1738 } 1739 } 1740} 1741 1742void 1743dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 1744{ 1745 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1746 ASSERT(dsl_pool_config_held(dp)); 1747 1748 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg; 1749 stat->dds_inconsistent = 1750 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT; 1751 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid; 1752 stat->dds_origin[0] = '\0'; 1753 if (ds->ds_is_snapshot) { 1754 stat->dds_is_snapshot = B_TRUE; 1755 stat->dds_num_clones = 1756 dsl_dataset_phys(ds)->ds_num_children - 1; 1757 } else { 1758 stat->dds_is_snapshot = B_FALSE; 1759 stat->dds_num_clones = 0; 1760 1761 if (dsl_dir_is_clone(ds->ds_dir)) { 1762 dsl_dataset_t *ods; 1763 1764 VERIFY0(dsl_dataset_hold_obj(dp, 1765 dsl_dir_phys(ds->ds_dir)->dd_origin_obj, 1766 FTAG, &ods)); 1767 dsl_dataset_name(ods, stat->dds_origin); 1768 dsl_dataset_rele(ods, FTAG); 1769 } 1770 } 1771} 1772 1773uint64_t 1774dsl_dataset_fsid_guid(dsl_dataset_t *ds) 1775{ 1776 return (ds->ds_fsid_guid); 1777} 1778 1779void 1780dsl_dataset_space(dsl_dataset_t *ds, 1781 uint64_t *refdbytesp, uint64_t *availbytesp, 1782 uint64_t *usedobjsp, uint64_t *availobjsp) 1783{ 1784 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; 1785 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 1786 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) 1787 *availbytesp += 1788 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 1789 if (ds->ds_quota != 0) { 1790 /* 1791 * Adjust available bytes according to refquota 1792 */ 1793 if (*refdbytesp < ds->ds_quota) 1794 *availbytesp = MIN(*availbytesp, 1795 ds->ds_quota - *refdbytesp); 1796 else 1797 *availbytesp = 0; 1798 } 1799 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); 1800 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 1801} 1802 1803boolean_t 1804dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 1805{ 1806 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1807 1808 ASSERT(dsl_pool_config_held(dp)); 1809 if (snap == NULL) 1810 return (B_FALSE); 1811 if (dsl_dataset_phys(ds)->ds_bp.blk_birth > 1812 dsl_dataset_phys(snap)->ds_creation_txg) { 1813 objset_t *os, *os_snap; 1814 /* 1815 * It may be that only the ZIL differs, because it was 1816 * reset in the head. Don't count that as being 1817 * modified. 1818 */ 1819 if (dmu_objset_from_ds(ds, &os) != 0) 1820 return (B_TRUE); 1821 if (dmu_objset_from_ds(snap, &os_snap) != 0) 1822 return (B_TRUE); 1823 return (bcmp(&os->os_phys->os_meta_dnode, 1824 &os_snap->os_phys->os_meta_dnode, 1825 sizeof (os->os_phys->os_meta_dnode)) != 0); 1826 } 1827 return (B_FALSE); 1828} 1829 1830typedef struct dsl_dataset_rename_snapshot_arg { 1831 const char *ddrsa_fsname; 1832 const char *ddrsa_oldsnapname; 1833 const char *ddrsa_newsnapname; 1834 boolean_t ddrsa_recursive; 1835 dmu_tx_t *ddrsa_tx; 1836} dsl_dataset_rename_snapshot_arg_t; 1837 1838/* ARGSUSED */ 1839static int 1840dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 1841 dsl_dataset_t *hds, void *arg) 1842{ 1843 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1844 int error; 1845 uint64_t val; 1846 1847 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1848 if (error != 0) { 1849 /* ignore nonexistent snapshots */ 1850 return (error == ENOENT ? 0 : error); 1851 } 1852 1853 /* new name should not exist */ 1854 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 1855 if (error == 0) 1856 error = SET_ERROR(EEXIST); 1857 else if (error == ENOENT) 1858 error = 0; 1859 1860 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 1861 if (dsl_dir_namelen(hds->ds_dir) + 1 + 1862 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN) 1863 error = SET_ERROR(ENAMETOOLONG); 1864 1865 return (error); 1866} 1867 1868static int 1869dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 1870{ 1871 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1872 dsl_pool_t *dp = dmu_tx_pool(tx); 1873 dsl_dataset_t *hds; 1874 int error; 1875 1876 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 1877 if (error != 0) 1878 return (error); 1879 1880 if (ddrsa->ddrsa_recursive) { 1881 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1882 dsl_dataset_rename_snapshot_check_impl, ddrsa, 1883 DS_FIND_CHILDREN); 1884 } else { 1885 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 1886 } 1887 dsl_dataset_rele(hds, FTAG); 1888 return (error); 1889} 1890 1891static int 1892dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 1893 dsl_dataset_t *hds, void *arg) 1894{ 1895#ifdef __FreeBSD__ 1896#ifdef _KERNEL 1897 char *oldname, *newname; 1898#endif 1899#endif 1900 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1901 dsl_dataset_t *ds; 1902 uint64_t val; 1903 dmu_tx_t *tx = ddrsa->ddrsa_tx; 1904 int error; 1905 1906 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1907 ASSERT(error == 0 || error == ENOENT); 1908 if (error == ENOENT) { 1909 /* ignore nonexistent snapshots */ 1910 return (0); 1911 } 1912 1913 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 1914 1915 /* log before we change the name */ 1916 spa_history_log_internal_ds(ds, "rename", tx, 1917 "-> @%s", ddrsa->ddrsa_newsnapname); 1918 1919 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 1920 B_FALSE)); 1921 mutex_enter(&ds->ds_lock); 1922 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 1923 mutex_exit(&ds->ds_lock); 1924 VERIFY0(zap_add(dp->dp_meta_objset, 1925 dsl_dataset_phys(hds)->ds_snapnames_zapobj, 1926 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 1927 1928#ifdef __FreeBSD__ 1929#ifdef _KERNEL 1930 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1931 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1932 snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname, 1933 ddrsa->ddrsa_oldsnapname); 1934 snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname, 1935 ddrsa->ddrsa_newsnapname); 1936 zfsvfs_update_fromname(oldname, newname); 1937 zvol_rename_minors(oldname, newname); 1938 kmem_free(newname, MAXPATHLEN); 1939 kmem_free(oldname, MAXPATHLEN); 1940#endif 1941#endif 1942 dsl_dataset_rele(ds, FTAG); 1943 1944 return (0); 1945} 1946 1947static void 1948dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 1949{ 1950 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1951 dsl_pool_t *dp = dmu_tx_pool(tx); 1952 dsl_dataset_t *hds; 1953 1954 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 1955 ddrsa->ddrsa_tx = tx; 1956 if (ddrsa->ddrsa_recursive) { 1957 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1958 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 1959 DS_FIND_CHILDREN)); 1960 } else { 1961 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 1962 } 1963 dsl_dataset_rele(hds, FTAG); 1964} 1965 1966int 1967dsl_dataset_rename_snapshot(const char *fsname, 1968 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 1969{ 1970 dsl_dataset_rename_snapshot_arg_t ddrsa; 1971 1972 ddrsa.ddrsa_fsname = fsname; 1973 ddrsa.ddrsa_oldsnapname = oldsnapname; 1974 ddrsa.ddrsa_newsnapname = newsnapname; 1975 ddrsa.ddrsa_recursive = recursive; 1976 1977 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 1978 dsl_dataset_rename_snapshot_sync, &ddrsa, 1979 1, ZFS_SPACE_CHECK_RESERVED)); 1980} 1981 1982/* 1983 * If we're doing an ownership handoff, we need to make sure that there is 1984 * only one long hold on the dataset. We're not allowed to change anything here 1985 * so we don't permanently release the long hold or regular hold here. We want 1986 * to do this only when syncing to avoid the dataset unexpectedly going away 1987 * when we release the long hold. 1988 */ 1989static int 1990dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 1991{ 1992 boolean_t held; 1993 1994 if (!dmu_tx_is_syncing(tx)) 1995 return (0); 1996 1997 if (owner != NULL) { 1998 VERIFY3P(ds->ds_owner, ==, owner); 1999 dsl_dataset_long_rele(ds, owner); 2000 } 2001 2002 held = dsl_dataset_long_held(ds); 2003 2004 if (owner != NULL) 2005 dsl_dataset_long_hold(ds, owner); 2006 2007 if (held) 2008 return (SET_ERROR(EBUSY)); 2009 2010 return (0); 2011} 2012 2013typedef struct dsl_dataset_rollback_arg { 2014 const char *ddra_fsname; 2015 void *ddra_owner; 2016 nvlist_t *ddra_result; 2017} dsl_dataset_rollback_arg_t; 2018 2019static int 2020dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 2021{ 2022 dsl_dataset_rollback_arg_t *ddra = arg; 2023 dsl_pool_t *dp = dmu_tx_pool(tx); 2024 dsl_dataset_t *ds; 2025 int64_t unused_refres_delta; 2026 int error; 2027 2028 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 2029 if (error != 0) 2030 return (error); 2031 2032 /* must not be a snapshot */ 2033 if (ds->ds_is_snapshot) { 2034 dsl_dataset_rele(ds, FTAG); 2035 return (SET_ERROR(EINVAL)); 2036 } 2037 2038 /* must have a most recent snapshot */ 2039 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { 2040 dsl_dataset_rele(ds, FTAG); 2041 return (SET_ERROR(EINVAL)); 2042 } 2043 2044 /* must not have any bookmarks after the most recent snapshot */ 2045 nvlist_t *proprequest = fnvlist_alloc(); 2046 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2047 nvlist_t *bookmarks = fnvlist_alloc(); 2048 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 2049 fnvlist_free(proprequest); 2050 if (error != 0) 2051 return (error); 2052 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 2053 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 2054 nvlist_t *valuenv = 2055 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 2056 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2057 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 2058 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 2059 fnvlist_free(bookmarks); 2060 dsl_dataset_rele(ds, FTAG); 2061 return (SET_ERROR(EEXIST)); 2062 } 2063 } 2064 fnvlist_free(bookmarks); 2065 2066 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 2067 if (error != 0) { 2068 dsl_dataset_rele(ds, FTAG); 2069 return (error); 2070 } 2071 2072 /* 2073 * Check if the snap we are rolling back to uses more than 2074 * the refquota. 2075 */ 2076 if (ds->ds_quota != 0 && 2077 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { 2078 dsl_dataset_rele(ds, FTAG); 2079 return (SET_ERROR(EDQUOT)); 2080 } 2081 2082 /* 2083 * When we do the clone swap, we will temporarily use more space 2084 * due to the refreservation (the head will no longer have any 2085 * unique space, so the entire amount of the refreservation will need 2086 * to be free). We will immediately destroy the clone, freeing 2087 * this space, but the freeing happens over many txg's. 2088 */ 2089 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 2090 dsl_dataset_phys(ds)->ds_unique_bytes); 2091 2092 if (unused_refres_delta > 0 && 2093 unused_refres_delta > 2094 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 2095 dsl_dataset_rele(ds, FTAG); 2096 return (SET_ERROR(ENOSPC)); 2097 } 2098 2099 dsl_dataset_rele(ds, FTAG); 2100 return (0); 2101} 2102 2103static void 2104dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 2105{ 2106 dsl_dataset_rollback_arg_t *ddra = arg; 2107 dsl_pool_t *dp = dmu_tx_pool(tx); 2108 dsl_dataset_t *ds, *clone; 2109 uint64_t cloneobj; 2110 char namebuf[ZFS_MAXNAMELEN]; 2111 2112 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 2113 2114 dsl_dataset_name(ds->ds_prev, namebuf); 2115 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 2116 2117 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 2118 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 2119 2120 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 2121 2122 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 2123 dsl_dataset_zero_zil(ds, tx); 2124 2125 dsl_destroy_head_sync_impl(clone, tx); 2126 2127 dsl_dataset_rele(clone, FTAG); 2128 dsl_dataset_rele(ds, FTAG); 2129} 2130 2131/* 2132 * Rolls back the given filesystem or volume to the most recent snapshot. 2133 * The name of the most recent snapshot will be returned under key "target" 2134 * in the result nvlist. 2135 * 2136 * If owner != NULL: 2137 * - The existing dataset MUST be owned by the specified owner at entry 2138 * - Upon return, dataset will still be held by the same owner, whether we 2139 * succeed or not. 2140 * 2141 * This mode is required any time the existing filesystem is mounted. See 2142 * notes above zfs_suspend_fs() for further details. 2143 */ 2144int 2145dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result) 2146{ 2147 dsl_dataset_rollback_arg_t ddra; 2148 2149 ddra.ddra_fsname = fsname; 2150 ddra.ddra_owner = owner; 2151 ddra.ddra_result = result; 2152 2153 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 2154 dsl_dataset_rollback_sync, &ddra, 2155 1, ZFS_SPACE_CHECK_RESERVED)); 2156} 2157 2158struct promotenode { 2159 list_node_t link; 2160 dsl_dataset_t *ds; 2161}; 2162 2163typedef struct dsl_dataset_promote_arg { 2164 const char *ddpa_clonename; 2165 dsl_dataset_t *ddpa_clone; 2166 list_t shared_snaps, origin_snaps, clone_snaps; 2167 dsl_dataset_t *origin_origin; /* origin of the origin */ 2168 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2169 char *err_ds; 2170 cred_t *cr; 2171} dsl_dataset_promote_arg_t; 2172 2173static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2174static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2175 void *tag); 2176static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2177 2178static int 2179dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2180{ 2181 dsl_dataset_promote_arg_t *ddpa = arg; 2182 dsl_pool_t *dp = dmu_tx_pool(tx); 2183 dsl_dataset_t *hds; 2184 struct promotenode *snap; 2185 dsl_dataset_t *origin_ds; 2186 int err; 2187 uint64_t unused; 2188 uint64_t ss_mv_cnt; 2189 size_t max_snap_len; 2190 2191 err = promote_hold(ddpa, dp, FTAG); 2192 if (err != 0) 2193 return (err); 2194 2195 hds = ddpa->ddpa_clone; 2196 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; 2197 2198 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { 2199 promote_rele(ddpa, FTAG); 2200 return (SET_ERROR(EXDEV)); 2201 } 2202 2203 /* 2204 * Compute and check the amount of space to transfer. Since this is 2205 * so expensive, don't do the preliminary check. 2206 */ 2207 if (!dmu_tx_is_syncing(tx)) { 2208 promote_rele(ddpa, FTAG); 2209 return (0); 2210 } 2211 2212 snap = list_head(&ddpa->shared_snaps); 2213 origin_ds = snap->ds; 2214 2215 /* compute origin's new unique space */ 2216 snap = list_tail(&ddpa->clone_snaps); 2217 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2218 origin_ds->ds_object); 2219 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2220 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, 2221 &ddpa->unique, &unused, &unused); 2222 2223 /* 2224 * Walk the snapshots that we are moving 2225 * 2226 * Compute space to transfer. Consider the incremental changes 2227 * to used by each snapshot: 2228 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2229 * So each snapshot gave birth to: 2230 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2231 * So a sequence would look like: 2232 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2233 * Which simplifies to: 2234 * uN + kN + kN-1 + ... + k1 + k0 2235 * Note however, if we stop before we reach the ORIGIN we get: 2236 * uN + kN + kN-1 + ... + kM - uM-1 2237 */ 2238 ss_mv_cnt = 0; 2239 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; 2240 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; 2241 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; 2242 for (snap = list_head(&ddpa->shared_snaps); snap; 2243 snap = list_next(&ddpa->shared_snaps, snap)) { 2244 uint64_t val, dlused, dlcomp, dluncomp; 2245 dsl_dataset_t *ds = snap->ds; 2246 2247 ss_mv_cnt++; 2248 2249 /* 2250 * If there are long holds, we won't be able to evict 2251 * the objset. 2252 */ 2253 if (dsl_dataset_long_held(ds)) { 2254 err = SET_ERROR(EBUSY); 2255 goto out; 2256 } 2257 2258 /* Check that the snapshot name does not conflict */ 2259 VERIFY0(dsl_dataset_get_snapname(ds)); 2260 if (strlen(ds->ds_snapname) >= max_snap_len) { 2261 err = SET_ERROR(ENAMETOOLONG); 2262 goto out; 2263 } 2264 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2265 if (err == 0) { 2266 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname); 2267 err = SET_ERROR(EEXIST); 2268 goto out; 2269 } 2270 if (err != ENOENT) 2271 goto out; 2272 2273 /* The very first snapshot does not have a deadlist */ 2274 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) 2275 continue; 2276 2277 dsl_deadlist_space(&ds->ds_deadlist, 2278 &dlused, &dlcomp, &dluncomp); 2279 ddpa->used += dlused; 2280 ddpa->comp += dlcomp; 2281 ddpa->uncomp += dluncomp; 2282 } 2283 2284 /* 2285 * If we are a clone of a clone then we never reached ORIGIN, 2286 * so we need to subtract out the clone origin's used space. 2287 */ 2288 if (ddpa->origin_origin) { 2289 ddpa->used -= 2290 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; 2291 ddpa->comp -= 2292 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; 2293 ddpa->uncomp -= 2294 dsl_dataset_phys(ddpa->origin_origin)-> 2295 ds_uncompressed_bytes; 2296 } 2297 2298 /* Check that there is enough space and limit headroom here */ 2299 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2300 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2301 if (err != 0) 2302 goto out; 2303 2304 /* 2305 * Compute the amounts of space that will be used by snapshots 2306 * after the promotion (for both origin and clone). For each, 2307 * it is the amount of space that will be on all of their 2308 * deadlists (that was not born before their new origin). 2309 */ 2310 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2311 uint64_t space; 2312 2313 /* 2314 * Note, typically this will not be a clone of a clone, 2315 * so dd_origin_txg will be < TXG_INITIAL, so 2316 * these snaplist_space() -> dsl_deadlist_space_range() 2317 * calls will be fast because they do not have to 2318 * iterate over all bps. 2319 */ 2320 snap = list_head(&ddpa->origin_snaps); 2321 err = snaplist_space(&ddpa->shared_snaps, 2322 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2323 if (err != 0) 2324 goto out; 2325 2326 err = snaplist_space(&ddpa->clone_snaps, 2327 snap->ds->ds_dir->dd_origin_txg, &space); 2328 if (err != 0) 2329 goto out; 2330 ddpa->cloneusedsnap += space; 2331 } 2332 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & 2333 DD_FLAG_USED_BREAKDOWN) { 2334 err = snaplist_space(&ddpa->origin_snaps, 2335 dsl_dataset_phys(origin_ds)->ds_creation_txg, 2336 &ddpa->originusedsnap); 2337 if (err != 0) 2338 goto out; 2339 } 2340 2341out: 2342 promote_rele(ddpa, FTAG); 2343 return (err); 2344} 2345 2346static void 2347dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2348{ 2349 dsl_dataset_promote_arg_t *ddpa = arg; 2350 dsl_pool_t *dp = dmu_tx_pool(tx); 2351 dsl_dataset_t *hds; 2352 struct promotenode *snap; 2353 dsl_dataset_t *origin_ds; 2354 dsl_dataset_t *origin_head; 2355 dsl_dir_t *dd; 2356 dsl_dir_t *odd = NULL; 2357 uint64_t oldnext_obj; 2358 int64_t delta; 2359#if defined(__FreeBSD__) && defined(_KERNEL) 2360 char *oldname, *newname; 2361#endif 2362 2363 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2364 hds = ddpa->ddpa_clone; 2365 2366 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); 2367 2368 snap = list_head(&ddpa->shared_snaps); 2369 origin_ds = snap->ds; 2370 dd = hds->ds_dir; 2371 2372 snap = list_head(&ddpa->origin_snaps); 2373 origin_head = snap->ds; 2374 2375 /* 2376 * We need to explicitly open odd, since origin_ds's dd will be 2377 * changing. 2378 */ 2379 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2380 NULL, FTAG, &odd)); 2381 2382 /* change origin's next snap */ 2383 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2384 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; 2385 snap = list_tail(&ddpa->clone_snaps); 2386 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2387 origin_ds->ds_object); 2388 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; 2389 2390 /* change the origin's next clone */ 2391 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { 2392 dsl_dataset_remove_from_next_clones(origin_ds, 2393 snap->ds->ds_object, tx); 2394 VERIFY0(zap_add_int(dp->dp_meta_objset, 2395 dsl_dataset_phys(origin_ds)->ds_next_clones_obj, 2396 oldnext_obj, tx)); 2397 } 2398 2399 /* change origin */ 2400 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2401 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); 2402 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; 2403 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2404 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2405 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; 2406 origin_head->ds_dir->dd_origin_txg = 2407 dsl_dataset_phys(origin_ds)->ds_creation_txg; 2408 2409 /* change dd_clone entries */ 2410 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2411 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2412 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); 2413 VERIFY0(zap_add_int(dp->dp_meta_objset, 2414 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2415 hds->ds_object, tx)); 2416 2417 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2418 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2419 origin_head->ds_object, tx)); 2420 if (dsl_dir_phys(dd)->dd_clones == 0) { 2421 dsl_dir_phys(dd)->dd_clones = 2422 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, 2423 DMU_OT_NONE, 0, tx); 2424 } 2425 VERIFY0(zap_add_int(dp->dp_meta_objset, 2426 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); 2427 } 2428 2429#if defined(__FreeBSD__) && defined(_KERNEL) 2430 /* Take the spa_namespace_lock early so zvol renames don't deadlock. */ 2431 mutex_enter(&spa_namespace_lock); 2432 2433 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2434 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2435#endif 2436 2437 /* move snapshots to this dir */ 2438 for (snap = list_head(&ddpa->shared_snaps); snap; 2439 snap = list_next(&ddpa->shared_snaps, snap)) { 2440 dsl_dataset_t *ds = snap->ds; 2441 2442 /* 2443 * Property callbacks are registered to a particular 2444 * dsl_dir. Since ours is changing, evict the objset 2445 * so that they will be unregistered from the old dsl_dir. 2446 */ 2447 if (ds->ds_objset) { 2448 dmu_objset_evict(ds->ds_objset); 2449 ds->ds_objset = NULL; 2450 } 2451 2452 /* move snap name entry */ 2453 VERIFY0(dsl_dataset_get_snapname(ds)); 2454 VERIFY0(dsl_dataset_snap_remove(origin_head, 2455 ds->ds_snapname, tx, B_TRUE)); 2456 VERIFY0(zap_add(dp->dp_meta_objset, 2457 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 2458 8, 1, &ds->ds_object, tx)); 2459 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 2460 DD_FIELD_SNAPSHOT_COUNT, tx); 2461 2462 /* change containing dsl_dir */ 2463 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2464 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); 2465 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; 2466 ASSERT3P(ds->ds_dir, ==, odd); 2467 dsl_dir_rele(ds->ds_dir, ds); 2468 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 2469 NULL, ds, &ds->ds_dir)); 2470 2471#if defined(__FreeBSD__) && defined(_KERNEL) 2472 dsl_dataset_name(ds, newname); 2473 zfsvfs_update_fromname(oldname, newname); 2474 zvol_rename_minors(oldname, newname); 2475#endif 2476 2477 /* move any clone references */ 2478 if (dsl_dataset_phys(ds)->ds_next_clones_obj && 2479 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2480 zap_cursor_t zc; 2481 zap_attribute_t za; 2482 2483 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2484 dsl_dataset_phys(ds)->ds_next_clones_obj); 2485 zap_cursor_retrieve(&zc, &za) == 0; 2486 zap_cursor_advance(&zc)) { 2487 dsl_dataset_t *cnds; 2488 uint64_t o; 2489 2490 if (za.za_first_integer == oldnext_obj) { 2491 /* 2492 * We've already moved the 2493 * origin's reference. 2494 */ 2495 continue; 2496 } 2497 2498 VERIFY0(dsl_dataset_hold_obj(dp, 2499 za.za_first_integer, FTAG, &cnds)); 2500 o = dsl_dir_phys(cnds->ds_dir)-> 2501 dd_head_dataset_obj; 2502 2503 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2504 dsl_dir_phys(odd)->dd_clones, o, tx)); 2505 VERIFY0(zap_add_int(dp->dp_meta_objset, 2506 dsl_dir_phys(dd)->dd_clones, o, tx)); 2507 dsl_dataset_rele(cnds, FTAG); 2508 } 2509 zap_cursor_fini(&zc); 2510 } 2511 2512 ASSERT(!dsl_prop_hascb(ds)); 2513 } 2514 2515#if defined(__FreeBSD__) && defined(_KERNEL) 2516 mutex_exit(&spa_namespace_lock); 2517 2518 kmem_free(newname, MAXPATHLEN); 2519 kmem_free(oldname, MAXPATHLEN); 2520#endif 2521 /* 2522 * Change space accounting. 2523 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2524 * both be valid, or both be 0 (resulting in delta == 0). This 2525 * is true for each of {clone,origin} independently. 2526 */ 2527 2528 delta = ddpa->cloneusedsnap - 2529 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; 2530 ASSERT3S(delta, >=, 0); 2531 ASSERT3U(ddpa->used, >=, delta); 2532 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2533 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2534 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 2535 2536 delta = ddpa->originusedsnap - 2537 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; 2538 ASSERT3S(delta, <=, 0); 2539 ASSERT3U(ddpa->used, >=, -delta); 2540 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2541 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2542 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 2543 2544 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; 2545 2546 /* log history record */ 2547 spa_history_log_internal_ds(hds, "promote", tx, ""); 2548 2549 dsl_dir_rele(odd, FTAG); 2550 promote_rele(ddpa, FTAG); 2551} 2552 2553/* 2554 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2555 * (exclusive) and last_obj (inclusive). The list will be in reverse 2556 * order (last_obj will be the list_head()). If first_obj == 0, do all 2557 * snapshots back to this dataset's origin. 2558 */ 2559static int 2560snaplist_make(dsl_pool_t *dp, 2561 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 2562{ 2563 uint64_t obj = last_obj; 2564 2565 list_create(l, sizeof (struct promotenode), 2566 offsetof(struct promotenode, link)); 2567 2568 while (obj != first_obj) { 2569 dsl_dataset_t *ds; 2570 struct promotenode *snap; 2571 int err; 2572 2573 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 2574 ASSERT(err != ENOENT); 2575 if (err != 0) 2576 return (err); 2577 2578 if (first_obj == 0) 2579 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; 2580 2581 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 2582 snap->ds = ds; 2583 list_insert_tail(l, snap); 2584 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 2585 } 2586 2587 return (0); 2588} 2589 2590static int 2591snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2592{ 2593 struct promotenode *snap; 2594 2595 *spacep = 0; 2596 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2597 uint64_t used, comp, uncomp; 2598 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2599 mintxg, UINT64_MAX, &used, &comp, &uncomp); 2600 *spacep += used; 2601 } 2602 return (0); 2603} 2604 2605static void 2606snaplist_destroy(list_t *l, void *tag) 2607{ 2608 struct promotenode *snap; 2609 2610 if (l == NULL || !list_link_active(&l->list_head)) 2611 return; 2612 2613 while ((snap = list_tail(l)) != NULL) { 2614 list_remove(l, snap); 2615 dsl_dataset_rele(snap->ds, tag); 2616 kmem_free(snap, sizeof (*snap)); 2617 } 2618 list_destroy(l); 2619} 2620 2621static int 2622promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 2623{ 2624 int error; 2625 dsl_dir_t *dd; 2626 struct promotenode *snap; 2627 2628 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 2629 &ddpa->ddpa_clone); 2630 if (error != 0) 2631 return (error); 2632 dd = ddpa->ddpa_clone->ds_dir; 2633 2634 if (ddpa->ddpa_clone->ds_is_snapshot || 2635 !dsl_dir_is_clone(dd)) { 2636 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2637 return (SET_ERROR(EINVAL)); 2638 } 2639 2640 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, 2641 &ddpa->shared_snaps, tag); 2642 if (error != 0) 2643 goto out; 2644 2645 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 2646 &ddpa->clone_snaps, tag); 2647 if (error != 0) 2648 goto out; 2649 2650 snap = list_head(&ddpa->shared_snaps); 2651 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); 2652 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, 2653 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, 2654 &ddpa->origin_snaps, tag); 2655 if (error != 0) 2656 goto out; 2657 2658 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { 2659 error = dsl_dataset_hold_obj(dp, 2660 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, 2661 tag, &ddpa->origin_origin); 2662 if (error != 0) 2663 goto out; 2664 } 2665out: 2666 if (error != 0) 2667 promote_rele(ddpa, tag); 2668 return (error); 2669} 2670 2671static void 2672promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 2673{ 2674 snaplist_destroy(&ddpa->shared_snaps, tag); 2675 snaplist_destroy(&ddpa->clone_snaps, tag); 2676 snaplist_destroy(&ddpa->origin_snaps, tag); 2677 if (ddpa->origin_origin != NULL) 2678 dsl_dataset_rele(ddpa->origin_origin, tag); 2679 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2680} 2681 2682/* 2683 * Promote a clone. 2684 * 2685 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 2686 * in with the name. (It must be at least MAXNAMELEN bytes long.) 2687 */ 2688int 2689dsl_dataset_promote(const char *name, char *conflsnap) 2690{ 2691 dsl_dataset_promote_arg_t ddpa = { 0 }; 2692 uint64_t numsnaps; 2693 int error; 2694 objset_t *os; 2695 2696 /* 2697 * We will modify space proportional to the number of 2698 * snapshots. Compute numsnaps. 2699 */ 2700 error = dmu_objset_hold(name, FTAG, &os); 2701 if (error != 0) 2702 return (error); 2703 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 2704 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, 2705 &numsnaps); 2706 dmu_objset_rele(os, FTAG); 2707 if (error != 0) 2708 return (error); 2709 2710 ddpa.ddpa_clonename = name; 2711 ddpa.err_ds = conflsnap; 2712 ddpa.cr = CRED(); 2713 2714 return (dsl_sync_task(name, dsl_dataset_promote_check, 2715 dsl_dataset_promote_sync, &ddpa, 2716 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED)); 2717} 2718 2719int 2720dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 2721 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 2722{ 2723 int64_t unused_refres_delta; 2724 2725 /* they should both be heads */ 2726 if (clone->ds_is_snapshot || 2727 origin_head->ds_is_snapshot) 2728 return (SET_ERROR(EINVAL)); 2729 2730 /* if we are not forcing, the branch point should be just before them */ 2731 if (!force && clone->ds_prev != origin_head->ds_prev) 2732 return (SET_ERROR(EINVAL)); 2733 2734 /* clone should be the clone (unless they are unrelated) */ 2735 if (clone->ds_prev != NULL && 2736 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 2737 origin_head->ds_dir != clone->ds_prev->ds_dir) 2738 return (SET_ERROR(EINVAL)); 2739 2740 /* the clone should be a child of the origin */ 2741 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 2742 return (SET_ERROR(EINVAL)); 2743 2744 /* origin_head shouldn't be modified unless 'force' */ 2745 if (!force && 2746 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 2747 return (SET_ERROR(ETXTBSY)); 2748 2749 /* origin_head should have no long holds (e.g. is not mounted) */ 2750 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 2751 return (SET_ERROR(EBUSY)); 2752 2753 /* check amount of any unconsumed refreservation */ 2754 unused_refres_delta = 2755 (int64_t)MIN(origin_head->ds_reserved, 2756 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 2757 (int64_t)MIN(origin_head->ds_reserved, 2758 dsl_dataset_phys(clone)->ds_unique_bytes); 2759 2760 if (unused_refres_delta > 0 && 2761 unused_refres_delta > 2762 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 2763 return (SET_ERROR(ENOSPC)); 2764 2765 /* clone can't be over the head's refquota */ 2766 if (origin_head->ds_quota != 0 && 2767 dsl_dataset_phys(clone)->ds_referenced_bytes > 2768 origin_head->ds_quota) 2769 return (SET_ERROR(EDQUOT)); 2770 2771 return (0); 2772} 2773 2774void 2775dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 2776 dsl_dataset_t *origin_head, dmu_tx_t *tx) 2777{ 2778 dsl_pool_t *dp = dmu_tx_pool(tx); 2779 int64_t unused_refres_delta; 2780 2781 ASSERT(clone->ds_reserved == 0); 2782 ASSERT(origin_head->ds_quota == 0 || 2783 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota); 2784 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 2785 2786 dmu_buf_will_dirty(clone->ds_dbuf, tx); 2787 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 2788 2789 if (clone->ds_objset != NULL) { 2790 dmu_objset_evict(clone->ds_objset); 2791 clone->ds_objset = NULL; 2792 } 2793 2794 if (origin_head->ds_objset != NULL) { 2795 dmu_objset_evict(origin_head->ds_objset); 2796 origin_head->ds_objset = NULL; 2797 } 2798 2799 unused_refres_delta = 2800 (int64_t)MIN(origin_head->ds_reserved, 2801 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 2802 (int64_t)MIN(origin_head->ds_reserved, 2803 dsl_dataset_phys(clone)->ds_unique_bytes); 2804 2805 /* 2806 * Reset origin's unique bytes, if it exists. 2807 */ 2808 if (clone->ds_prev) { 2809 dsl_dataset_t *origin = clone->ds_prev; 2810 uint64_t comp, uncomp; 2811 2812 dmu_buf_will_dirty(origin->ds_dbuf, tx); 2813 dsl_deadlist_space_range(&clone->ds_deadlist, 2814 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, 2815 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); 2816 } 2817 2818 /* swap blkptrs */ 2819 { 2820 blkptr_t tmp; 2821 tmp = dsl_dataset_phys(origin_head)->ds_bp; 2822 dsl_dataset_phys(origin_head)->ds_bp = 2823 dsl_dataset_phys(clone)->ds_bp; 2824 dsl_dataset_phys(clone)->ds_bp = tmp; 2825 } 2826 2827 /* set dd_*_bytes */ 2828 { 2829 int64_t dused, dcomp, duncomp; 2830 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2831 uint64_t odl_used, odl_comp, odl_uncomp; 2832 2833 ASSERT3U(dsl_dir_phys(clone->ds_dir)-> 2834 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2835 2836 dsl_deadlist_space(&clone->ds_deadlist, 2837 &cdl_used, &cdl_comp, &cdl_uncomp); 2838 dsl_deadlist_space(&origin_head->ds_deadlist, 2839 &odl_used, &odl_comp, &odl_uncomp); 2840 2841 dused = dsl_dataset_phys(clone)->ds_referenced_bytes + 2842 cdl_used - 2843 (dsl_dataset_phys(origin_head)->ds_referenced_bytes + 2844 odl_used); 2845 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + 2846 cdl_comp - 2847 (dsl_dataset_phys(origin_head)->ds_compressed_bytes + 2848 odl_comp); 2849 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + 2850 cdl_uncomp - 2851 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + 2852 odl_uncomp); 2853 2854 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 2855 dused, dcomp, duncomp, tx); 2856 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 2857 -dused, -dcomp, -duncomp, tx); 2858 2859 /* 2860 * The difference in the space used by snapshots is the 2861 * difference in snapshot space due to the head's 2862 * deadlist (since that's the only thing that's 2863 * changing that affects the snapused). 2864 */ 2865 dsl_deadlist_space_range(&clone->ds_deadlist, 2866 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2867 &cdl_used, &cdl_comp, &cdl_uncomp); 2868 dsl_deadlist_space_range(&origin_head->ds_deadlist, 2869 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2870 &odl_used, &odl_comp, &odl_uncomp); 2871 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 2872 DD_USED_HEAD, DD_USED_SNAP, NULL); 2873 } 2874 2875 /* swap ds_*_bytes */ 2876 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, 2877 dsl_dataset_phys(clone)->ds_referenced_bytes); 2878 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, 2879 dsl_dataset_phys(clone)->ds_compressed_bytes); 2880 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, 2881 dsl_dataset_phys(clone)->ds_uncompressed_bytes); 2882 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, 2883 dsl_dataset_phys(clone)->ds_unique_bytes); 2884 2885 /* apply any parent delta for change in unconsumed refreservation */ 2886 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 2887 unused_refres_delta, 0, 0, tx); 2888 2889 /* 2890 * Swap deadlists. 2891 */ 2892 dsl_deadlist_close(&clone->ds_deadlist); 2893 dsl_deadlist_close(&origin_head->ds_deadlist); 2894 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, 2895 dsl_dataset_phys(clone)->ds_deadlist_obj); 2896 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 2897 dsl_dataset_phys(clone)->ds_deadlist_obj); 2898 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 2899 dsl_dataset_phys(origin_head)->ds_deadlist_obj); 2900 2901 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 2902 2903 spa_history_log_internal_ds(clone, "clone swap", tx, 2904 "parent=%s", origin_head->ds_dir->dd_myname); 2905} 2906 2907/* 2908 * Given a pool name and a dataset object number in that pool, 2909 * return the name of that dataset. 2910 */ 2911int 2912dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 2913{ 2914 dsl_pool_t *dp; 2915 dsl_dataset_t *ds; 2916 int error; 2917 2918 error = dsl_pool_hold(pname, FTAG, &dp); 2919 if (error != 0) 2920 return (error); 2921 2922 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 2923 if (error == 0) { 2924 dsl_dataset_name(ds, buf); 2925 dsl_dataset_rele(ds, FTAG); 2926 } 2927 dsl_pool_rele(dp, FTAG); 2928 2929 return (error); 2930} 2931 2932int 2933dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 2934 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 2935{ 2936 int error = 0; 2937 2938 ASSERT3S(asize, >, 0); 2939 2940 /* 2941 * *ref_rsrv is the portion of asize that will come from any 2942 * unconsumed refreservation space. 2943 */ 2944 *ref_rsrv = 0; 2945 2946 mutex_enter(&ds->ds_lock); 2947 /* 2948 * Make a space adjustment for reserved bytes. 2949 */ 2950 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 2951 ASSERT3U(*used, >=, 2952 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 2953 *used -= 2954 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 2955 *ref_rsrv = 2956 asize - MIN(asize, parent_delta(ds, asize + inflight)); 2957 } 2958 2959 if (!check_quota || ds->ds_quota == 0) { 2960 mutex_exit(&ds->ds_lock); 2961 return (0); 2962 } 2963 /* 2964 * If they are requesting more space, and our current estimate 2965 * is over quota, they get to try again unless the actual 2966 * on-disk is over quota and there are no pending changes (which 2967 * may free up space for us). 2968 */ 2969 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= 2970 ds->ds_quota) { 2971 if (inflight > 0 || 2972 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) 2973 error = SET_ERROR(ERESTART); 2974 else 2975 error = SET_ERROR(EDQUOT); 2976 } 2977 mutex_exit(&ds->ds_lock); 2978 2979 return (error); 2980} 2981 2982typedef struct dsl_dataset_set_qr_arg { 2983 const char *ddsqra_name; 2984 zprop_source_t ddsqra_source; 2985 uint64_t ddsqra_value; 2986} dsl_dataset_set_qr_arg_t; 2987 2988 2989/* ARGSUSED */ 2990static int 2991dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 2992{ 2993 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2994 dsl_pool_t *dp = dmu_tx_pool(tx); 2995 dsl_dataset_t *ds; 2996 int error; 2997 uint64_t newval; 2998 2999 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 3000 return (SET_ERROR(ENOTSUP)); 3001 3002 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3003 if (error != 0) 3004 return (error); 3005 3006 if (ds->ds_is_snapshot) { 3007 dsl_dataset_rele(ds, FTAG); 3008 return (SET_ERROR(EINVAL)); 3009 } 3010 3011 error = dsl_prop_predict(ds->ds_dir, 3012 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3013 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3014 if (error != 0) { 3015 dsl_dataset_rele(ds, FTAG); 3016 return (error); 3017 } 3018 3019 if (newval == 0) { 3020 dsl_dataset_rele(ds, FTAG); 3021 return (0); 3022 } 3023 3024 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || 3025 newval < ds->ds_reserved) { 3026 dsl_dataset_rele(ds, FTAG); 3027 return (SET_ERROR(ENOSPC)); 3028 } 3029 3030 dsl_dataset_rele(ds, FTAG); 3031 return (0); 3032} 3033 3034static void 3035dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 3036{ 3037 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3038 dsl_pool_t *dp = dmu_tx_pool(tx); 3039 dsl_dataset_t *ds; 3040 uint64_t newval; 3041 3042 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3043 3044 dsl_prop_set_sync_impl(ds, 3045 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3046 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 3047 &ddsqra->ddsqra_value, tx); 3048 3049 VERIFY0(dsl_prop_get_int_ds(ds, 3050 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 3051 3052 if (ds->ds_quota != newval) { 3053 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3054 ds->ds_quota = newval; 3055 } 3056 dsl_dataset_rele(ds, FTAG); 3057} 3058 3059int 3060dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 3061 uint64_t refquota) 3062{ 3063 dsl_dataset_set_qr_arg_t ddsqra; 3064 3065 ddsqra.ddsqra_name = dsname; 3066 ddsqra.ddsqra_source = source; 3067 ddsqra.ddsqra_value = refquota; 3068 3069 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 3070 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 3071} 3072 3073static int 3074dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 3075{ 3076 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3077 dsl_pool_t *dp = dmu_tx_pool(tx); 3078 dsl_dataset_t *ds; 3079 int error; 3080 uint64_t newval, unique; 3081 3082 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 3083 return (SET_ERROR(ENOTSUP)); 3084 3085 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3086 if (error != 0) 3087 return (error); 3088 3089 if (ds->ds_is_snapshot) { 3090 dsl_dataset_rele(ds, FTAG); 3091 return (SET_ERROR(EINVAL)); 3092 } 3093 3094 error = dsl_prop_predict(ds->ds_dir, 3095 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3096 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3097 if (error != 0) { 3098 dsl_dataset_rele(ds, FTAG); 3099 return (error); 3100 } 3101 3102 /* 3103 * If we are doing the preliminary check in open context, the 3104 * space estimates may be inaccurate. 3105 */ 3106 if (!dmu_tx_is_syncing(tx)) { 3107 dsl_dataset_rele(ds, FTAG); 3108 return (0); 3109 } 3110 3111 mutex_enter(&ds->ds_lock); 3112 if (!DS_UNIQUE_IS_ACCURATE(ds)) 3113 dsl_dataset_recalc_head_uniq(ds); 3114 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3115 mutex_exit(&ds->ds_lock); 3116 3117 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 3118 uint64_t delta = MAX(unique, newval) - 3119 MAX(unique, ds->ds_reserved); 3120 3121 if (delta > 3122 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 3123 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 3124 dsl_dataset_rele(ds, FTAG); 3125 return (SET_ERROR(ENOSPC)); 3126 } 3127 } 3128 3129 dsl_dataset_rele(ds, FTAG); 3130 return (0); 3131} 3132 3133void 3134dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 3135 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 3136{ 3137 uint64_t newval; 3138 uint64_t unique; 3139 int64_t delta; 3140 3141 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3142 source, sizeof (value), 1, &value, tx); 3143 3144 VERIFY0(dsl_prop_get_int_ds(ds, 3145 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 3146 3147 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3148 mutex_enter(&ds->ds_dir->dd_lock); 3149 mutex_enter(&ds->ds_lock); 3150 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3151 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3152 delta = MAX(0, (int64_t)(newval - unique)) - 3153 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3154 ds->ds_reserved = newval; 3155 mutex_exit(&ds->ds_lock); 3156 3157 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3158 mutex_exit(&ds->ds_dir->dd_lock); 3159} 3160 3161static void 3162dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 3163{ 3164 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3165 dsl_pool_t *dp = dmu_tx_pool(tx); 3166 dsl_dataset_t *ds; 3167 3168 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3169 dsl_dataset_set_refreservation_sync_impl(ds, 3170 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 3171 dsl_dataset_rele(ds, FTAG); 3172} 3173 3174int 3175dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 3176 uint64_t refreservation) 3177{ 3178 dsl_dataset_set_qr_arg_t ddsqra; 3179 3180 ddsqra.ddsqra_name = dsname; 3181 ddsqra.ddsqra_source = source; 3182 ddsqra.ddsqra_value = refreservation; 3183 3184 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 3185 dsl_dataset_set_refreservation_sync, &ddsqra, 3186 0, ZFS_SPACE_CHECK_NONE)); 3187} 3188 3189/* 3190 * Return (in *usedp) the amount of space written in new that is not 3191 * present in oldsnap. New may be a snapshot or the head. Old must be 3192 * a snapshot before new, in new's filesystem (or its origin). If not then 3193 * fail and return EINVAL. 3194 * 3195 * The written space is calculated by considering two components: First, we 3196 * ignore any freed space, and calculate the written as new's used space 3197 * minus old's used space. Next, we add in the amount of space that was freed 3198 * between the two snapshots, thus reducing new's used space relative to old's. 3199 * Specifically, this is the space that was born before old->ds_creation_txg, 3200 * and freed before new (ie. on new's deadlist or a previous deadlist). 3201 * 3202 * space freed [---------------------] 3203 * snapshots ---O-------O--------O-------O------ 3204 * oldsnap new 3205 */ 3206int 3207dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 3208 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3209{ 3210 int err = 0; 3211 uint64_t snapobj; 3212 dsl_pool_t *dp = new->ds_dir->dd_pool; 3213 3214 ASSERT(dsl_pool_config_held(dp)); 3215 3216 *usedp = 0; 3217 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; 3218 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes; 3219 3220 *compp = 0; 3221 *compp += dsl_dataset_phys(new)->ds_compressed_bytes; 3222 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes; 3223 3224 *uncompp = 0; 3225 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; 3226 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes; 3227 3228 snapobj = new->ds_object; 3229 while (snapobj != oldsnap->ds_object) { 3230 dsl_dataset_t *snap; 3231 uint64_t used, comp, uncomp; 3232 3233 if (snapobj == new->ds_object) { 3234 snap = new; 3235 } else { 3236 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3237 if (err != 0) 3238 break; 3239 } 3240 3241 if (dsl_dataset_phys(snap)->ds_prev_snap_txg == 3242 dsl_dataset_phys(oldsnap)->ds_creation_txg) { 3243 /* 3244 * The blocks in the deadlist can not be born after 3245 * ds_prev_snap_txg, so get the whole deadlist space, 3246 * which is more efficient (especially for old-format 3247 * deadlists). Unfortunately the deadlist code 3248 * doesn't have enough information to make this 3249 * optimization itself. 3250 */ 3251 dsl_deadlist_space(&snap->ds_deadlist, 3252 &used, &comp, &uncomp); 3253 } else { 3254 dsl_deadlist_space_range(&snap->ds_deadlist, 3255 0, dsl_dataset_phys(oldsnap)->ds_creation_txg, 3256 &used, &comp, &uncomp); 3257 } 3258 *usedp += used; 3259 *compp += comp; 3260 *uncompp += uncomp; 3261 3262 /* 3263 * If we get to the beginning of the chain of snapshots 3264 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3265 * was not a snapshot of/before new. 3266 */ 3267 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 3268 if (snap != new) 3269 dsl_dataset_rele(snap, FTAG); 3270 if (snapobj == 0) { 3271 err = SET_ERROR(EINVAL); 3272 break; 3273 } 3274 3275 } 3276 return (err); 3277} 3278 3279/* 3280 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3281 * lastsnap, and all snapshots in between are deleted. 3282 * 3283 * blocks that would be freed [---------------------------] 3284 * snapshots ---O-------O--------O-------O--------O 3285 * firstsnap lastsnap 3286 * 3287 * This is the set of blocks that were born after the snap before firstsnap, 3288 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3289 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3290 * We calculate this by iterating over the relevant deadlists (from the snap 3291 * after lastsnap, backward to the snap after firstsnap), summing up the 3292 * space on the deadlist that was born after the snap before firstsnap. 3293 */ 3294int 3295dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3296 dsl_dataset_t *lastsnap, 3297 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3298{ 3299 int err = 0; 3300 uint64_t snapobj; 3301 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3302 3303 ASSERT(firstsnap->ds_is_snapshot); 3304 ASSERT(lastsnap->ds_is_snapshot); 3305 3306 /* 3307 * Check that the snapshots are in the same dsl_dir, and firstsnap 3308 * is before lastsnap. 3309 */ 3310 if (firstsnap->ds_dir != lastsnap->ds_dir || 3311 dsl_dataset_phys(firstsnap)->ds_creation_txg > 3312 dsl_dataset_phys(lastsnap)->ds_creation_txg) 3313 return (SET_ERROR(EINVAL)); 3314 3315 *usedp = *compp = *uncompp = 0; 3316 3317 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; 3318 while (snapobj != firstsnap->ds_object) { 3319 dsl_dataset_t *ds; 3320 uint64_t used, comp, uncomp; 3321 3322 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3323 if (err != 0) 3324 break; 3325 3326 dsl_deadlist_space_range(&ds->ds_deadlist, 3327 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, 3328 &used, &comp, &uncomp); 3329 *usedp += used; 3330 *compp += comp; 3331 *uncompp += uncomp; 3332 3333 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3334 ASSERT3U(snapobj, !=, 0); 3335 dsl_dataset_rele(ds, FTAG); 3336 } 3337 return (err); 3338} 3339 3340static int 3341dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx) 3342{ 3343 const char *dsname = arg; 3344 dsl_dataset_t *ds; 3345 dsl_pool_t *dp = dmu_tx_pool(tx); 3346 int error = 0; 3347 3348 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) 3349 return (SET_ERROR(ENOTSUP)); 3350 3351 ASSERT(spa_feature_is_enabled(dp->dp_spa, 3352 SPA_FEATURE_EXTENSIBLE_DATASET)); 3353 3354 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 3355 if (error != 0) 3356 return (error); 3357 3358 if (ds->ds_large_blocks) 3359 error = EALREADY; 3360 dsl_dataset_rele(ds, FTAG); 3361 3362 return (error); 3363} 3364 3365void 3366dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx) 3367{ 3368 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 3369 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 3370 uint64_t zero = 0; 3371 3372 spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx); 3373 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 3374 3375 VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS, 3376 sizeof (zero), 1, &zero, tx)); 3377} 3378 3379static void 3380dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx) 3381{ 3382 const char *dsname = arg; 3383 dsl_dataset_t *ds; 3384 3385 VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds)); 3386 3387 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx); 3388 ASSERT(!ds->ds_large_blocks); 3389 ds->ds_large_blocks = B_TRUE; 3390 dsl_dataset_rele(ds, FTAG); 3391} 3392 3393int 3394dsl_dataset_activate_large_blocks(const char *dsname) 3395{ 3396 int error; 3397 3398 error = dsl_sync_task(dsname, 3399 dsl_dataset_activate_large_blocks_check, 3400 dsl_dataset_activate_large_blocks_sync, (void *)dsname, 3401 1, ZFS_SPACE_CHECK_RESERVED); 3402 3403 /* 3404 * EALREADY indicates that this dataset already supports large blocks. 3405 */ 3406 if (error == EALREADY) 3407 error = 0; 3408 return (error); 3409} 3410 3411/* 3412 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 3413 * For example, they could both be snapshots of the same filesystem, and 3414 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 3415 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 3416 * filesystem. Or 'earlier' could be the origin's origin. 3417 * 3418 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 3419 */ 3420boolean_t 3421dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 3422 uint64_t earlier_txg) 3423{ 3424 dsl_pool_t *dp = later->ds_dir->dd_pool; 3425 int error; 3426 boolean_t ret; 3427 3428 ASSERT(dsl_pool_config_held(dp)); 3429 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); 3430 3431 if (earlier_txg == 0) 3432 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; 3433 3434 if (later->ds_is_snapshot && 3435 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) 3436 return (B_FALSE); 3437 3438 if (later->ds_dir == earlier->ds_dir) 3439 return (B_TRUE); 3440 if (!dsl_dir_is_clone(later->ds_dir)) 3441 return (B_FALSE); 3442 3443 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object) 3444 return (B_TRUE); 3445 dsl_dataset_t *origin; 3446 error = dsl_dataset_hold_obj(dp, 3447 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); 3448 if (error != 0) 3449 return (B_FALSE); 3450 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 3451 dsl_dataset_rele(origin, FTAG); 3452 return (ret); 3453} 3454 3455 3456void 3457dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 3458{ 3459 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3460 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 3461} 3462