dsl_dataset.c revision 288538
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 */ 28 29#include <sys/dmu_objset.h> 30#include <sys/dsl_dataset.h> 31#include <sys/dsl_dir.h> 32#include <sys/dsl_prop.h> 33#include <sys/dsl_synctask.h> 34#include <sys/dmu_traverse.h> 35#include <sys/dmu_impl.h> 36#include <sys/dmu_tx.h> 37#include <sys/arc.h> 38#include <sys/zio.h> 39#include <sys/zap.h> 40#include <sys/zfeature.h> 41#include <sys/unique.h> 42#include <sys/zfs_context.h> 43#include <sys/zfs_ioctl.h> 44#include <sys/spa.h> 45#include <sys/zfs_znode.h> 46#include <sys/zfs_onexit.h> 47#include <sys/zvol.h> 48#include <sys/dsl_scan.h> 49#include <sys/dsl_deadlist.h> 50#include <sys/dsl_destroy.h> 51#include <sys/dsl_userhold.h> 52#include <sys/dsl_bookmark.h> 53 54SYSCTL_DECL(_vfs_zfs); 55 56/* 57 * The SPA supports block sizes up to 16MB. However, very large blocks 58 * can have an impact on i/o latency (e.g. tying up a spinning disk for 59 * ~300ms), and also potentially on the memory allocator. Therefore, 60 * we do not allow the recordsize to be set larger than zfs_max_recordsize 61 * (default 1MB). Larger blocks can be created by changing this tunable, 62 * and pools with larger blocks can always be imported and used, regardless 63 * of this setting. 64 */ 65int zfs_max_recordsize = 1 * 1024 * 1024; 66SYSCTL_INT(_vfs_zfs, OID_AUTO, max_recordsize, CTLFLAG_RWTUN, 67 &zfs_max_recordsize, 0, 68 "Maximum block size. Expect dragons when tuning this."); 69 70#define SWITCH64(x, y) \ 71 { \ 72 uint64_t __tmp = (x); \ 73 (x) = (y); \ 74 (y) = __tmp; \ 75 } 76 77#define DS_REF_MAX (1ULL << 62) 78 79extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds); 80extern inline boolean_t dsl_dataset_is_snapshot(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(!dsl_dataset_is_snapshot(ds)); 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} 261 262/* ARGSUSED */ 263static void 264dsl_dataset_evict(dmu_buf_t *db, void *dsv) 265{ 266 dsl_dataset_t *ds = dsv; 267 268 ASSERT(ds->ds_owner == NULL); 269 270 unique_remove(ds->ds_fsid_guid); 271 272 if (ds->ds_objset != NULL) 273 dmu_objset_evict(ds->ds_objset); 274 275 if (ds->ds_prev) { 276 dsl_dataset_rele(ds->ds_prev, ds); 277 ds->ds_prev = NULL; 278 } 279 280 bplist_destroy(&ds->ds_pending_deadlist); 281 if (dsl_dataset_phys(ds)->ds_deadlist_obj != 0) 282 dsl_deadlist_close(&ds->ds_deadlist); 283 if (ds->ds_dir) 284 dsl_dir_rele(ds->ds_dir, ds); 285 286 ASSERT(!list_link_active(&ds->ds_synced_link)); 287 288 if (mutex_owned(&ds->ds_lock)) 289 mutex_exit(&ds->ds_lock); 290 mutex_destroy(&ds->ds_lock); 291 if (mutex_owned(&ds->ds_opening_lock)) 292 mutex_exit(&ds->ds_opening_lock); 293 mutex_destroy(&ds->ds_opening_lock); 294 mutex_destroy(&ds->ds_sendstream_lock); 295 refcount_destroy(&ds->ds_longholds); 296 297 kmem_free(ds, sizeof (dsl_dataset_t)); 298} 299 300int 301dsl_dataset_get_snapname(dsl_dataset_t *ds) 302{ 303 dsl_dataset_phys_t *headphys; 304 int err; 305 dmu_buf_t *headdbuf; 306 dsl_pool_t *dp = ds->ds_dir->dd_pool; 307 objset_t *mos = dp->dp_meta_objset; 308 309 if (ds->ds_snapname[0]) 310 return (0); 311 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) 312 return (0); 313 314 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, 315 FTAG, &headdbuf); 316 if (err != 0) 317 return (err); 318 headphys = headdbuf->db_data; 319 err = zap_value_search(dp->dp_meta_objset, 320 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 321 dmu_buf_rele(headdbuf, FTAG); 322 return (err); 323} 324 325int 326dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 327{ 328 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 329 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 330 matchtype_t mt; 331 int err; 332 333 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 334 mt = MT_FIRST; 335 else 336 mt = MT_EXACT; 337 338 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 339 value, mt, NULL, 0, NULL); 340 if (err == ENOTSUP && mt == MT_FIRST) 341 err = zap_lookup(mos, snapobj, name, 8, 1, value); 342 return (err); 343} 344 345int 346dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 347 boolean_t adj_cnt) 348{ 349 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 350 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 351 matchtype_t mt; 352 int err; 353 354 dsl_dir_snap_cmtime_update(ds->ds_dir); 355 356 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 357 mt = MT_FIRST; 358 else 359 mt = MT_EXACT; 360 361 err = zap_remove_norm(mos, snapobj, name, mt, tx); 362 if (err == ENOTSUP && mt == MT_FIRST) 363 err = zap_remove(mos, snapobj, name, tx); 364 365 if (err == 0 && adj_cnt) 366 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 367 DD_FIELD_SNAPSHOT_COUNT, tx); 368 369 return (err); 370} 371 372boolean_t 373dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag) 374{ 375 return (dmu_buf_try_add_ref(ds->ds_dbuf, dp->dp_meta_objset, 376 ds->ds_object, DMU_BONUS_BLKID, tag)); 377} 378 379int 380dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 381 dsl_dataset_t **dsp) 382{ 383 objset_t *mos = dp->dp_meta_objset; 384 dmu_buf_t *dbuf; 385 dsl_dataset_t *ds; 386 int err; 387 dmu_object_info_t doi; 388 389 ASSERT(dsl_pool_config_held(dp)); 390 391 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 392 if (err != 0) 393 return (err); 394 395 /* Make sure dsobj has the correct object type. */ 396 dmu_object_info_from_db(dbuf, &doi); 397 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 398 dmu_buf_rele(dbuf, tag); 399 return (SET_ERROR(EINVAL)); 400 } 401 402 ds = dmu_buf_get_user(dbuf); 403 if (ds == NULL) { 404 dsl_dataset_t *winner = NULL; 405 406 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 407 ds->ds_dbuf = dbuf; 408 ds->ds_object = dsobj; 409 410 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 411 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 412 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 413 refcount_create(&ds->ds_longholds); 414 415 bplist_create(&ds->ds_pending_deadlist); 416 dsl_deadlist_open(&ds->ds_deadlist, 417 mos, dsl_dataset_phys(ds)->ds_deadlist_obj); 418 419 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 420 offsetof(dmu_sendarg_t, dsa_link)); 421 422 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 423 int zaperr = zap_contains(mos, dsobj, 424 DS_FIELD_LARGE_BLOCKS); 425 if (zaperr != ENOENT) { 426 VERIFY0(zaperr); 427 ds->ds_large_blocks = B_TRUE; 428 } 429 } 430 431 if (err == 0) { 432 err = dsl_dir_hold_obj(dp, 433 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, 434 &ds->ds_dir); 435 } 436 if (err != 0) { 437 mutex_destroy(&ds->ds_lock); 438 mutex_destroy(&ds->ds_opening_lock); 439 mutex_destroy(&ds->ds_sendstream_lock); 440 refcount_destroy(&ds->ds_longholds); 441 bplist_destroy(&ds->ds_pending_deadlist); 442 dsl_deadlist_close(&ds->ds_deadlist); 443 kmem_free(ds, sizeof (dsl_dataset_t)); 444 dmu_buf_rele(dbuf, tag); 445 return (err); 446 } 447 448 if (!dsl_dataset_is_snapshot(ds)) { 449 ds->ds_snapname[0] = '\0'; 450 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 451 err = dsl_dataset_hold_obj(dp, 452 dsl_dataset_phys(ds)->ds_prev_snap_obj, 453 ds, &ds->ds_prev); 454 } 455 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 456 int zaperr = zap_lookup(mos, ds->ds_object, 457 DS_FIELD_BOOKMARK_NAMES, 458 sizeof (ds->ds_bookmarks), 1, 459 &ds->ds_bookmarks); 460 if (zaperr != ENOENT) 461 VERIFY0(zaperr); 462 } 463 } else { 464 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 465 err = dsl_dataset_get_snapname(ds); 466 if (err == 0 && 467 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { 468 err = zap_count( 469 ds->ds_dir->dd_pool->dp_meta_objset, 470 dsl_dataset_phys(ds)->ds_userrefs_obj, 471 &ds->ds_userrefs); 472 } 473 } 474 475 if (err == 0 && !dsl_dataset_is_snapshot(ds)) { 476 err = dsl_prop_get_int_ds(ds, 477 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 478 &ds->ds_reserved); 479 if (err == 0) { 480 err = dsl_prop_get_int_ds(ds, 481 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 482 &ds->ds_quota); 483 } 484 } else { 485 ds->ds_reserved = ds->ds_quota = 0; 486 } 487 488 if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds, 489 dsl_dataset_evict)) != NULL) { 490 bplist_destroy(&ds->ds_pending_deadlist); 491 dsl_deadlist_close(&ds->ds_deadlist); 492 if (ds->ds_prev) 493 dsl_dataset_rele(ds->ds_prev, ds); 494 dsl_dir_rele(ds->ds_dir, ds); 495 mutex_destroy(&ds->ds_lock); 496 mutex_destroy(&ds->ds_opening_lock); 497 mutex_destroy(&ds->ds_sendstream_lock); 498 refcount_destroy(&ds->ds_longholds); 499 kmem_free(ds, sizeof (dsl_dataset_t)); 500 if (err != 0) { 501 dmu_buf_rele(dbuf, tag); 502 return (err); 503 } 504 ds = winner; 505 } else { 506 ds->ds_fsid_guid = 507 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); 508 } 509 } 510 ASSERT3P(ds->ds_dbuf, ==, dbuf); 511 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); 512 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || 513 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 514 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 515 *dsp = ds; 516 return (0); 517} 518 519int 520dsl_dataset_hold(dsl_pool_t *dp, const char *name, 521 void *tag, dsl_dataset_t **dsp) 522{ 523 dsl_dir_t *dd; 524 const char *snapname; 525 uint64_t obj; 526 int err = 0; 527 528 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 529 if (err != 0) 530 return (err); 531 532 ASSERT(dsl_pool_config_held(dp)); 533 obj = dsl_dir_phys(dd)->dd_head_dataset_obj; 534 if (obj != 0) 535 err = dsl_dataset_hold_obj(dp, obj, tag, dsp); 536 else 537 err = SET_ERROR(ENOENT); 538 539 /* we may be looking for a snapshot */ 540 if (err == 0 && snapname != NULL) { 541 dsl_dataset_t *ds; 542 543 if (*snapname++ != '@') { 544 dsl_dataset_rele(*dsp, tag); 545 dsl_dir_rele(dd, FTAG); 546 return (SET_ERROR(ENOENT)); 547 } 548 549 dprintf("looking for snapshot '%s'\n", snapname); 550 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); 551 if (err == 0) 552 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 553 dsl_dataset_rele(*dsp, tag); 554 555 if (err == 0) { 556 mutex_enter(&ds->ds_lock); 557 if (ds->ds_snapname[0] == 0) 558 (void) strlcpy(ds->ds_snapname, snapname, 559 sizeof (ds->ds_snapname)); 560 mutex_exit(&ds->ds_lock); 561 *dsp = ds; 562 } 563 } 564 565 dsl_dir_rele(dd, FTAG); 566 return (err); 567} 568 569int 570dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 571 void *tag, dsl_dataset_t **dsp) 572{ 573 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 574 if (err != 0) 575 return (err); 576 if (!dsl_dataset_tryown(*dsp, tag)) { 577 dsl_dataset_rele(*dsp, tag); 578 *dsp = NULL; 579 return (SET_ERROR(EBUSY)); 580 } 581 return (0); 582} 583 584int 585dsl_dataset_own(dsl_pool_t *dp, const char *name, 586 void *tag, dsl_dataset_t **dsp) 587{ 588 int err = dsl_dataset_hold(dp, name, tag, dsp); 589 if (err != 0) 590 return (err); 591 if (!dsl_dataset_tryown(*dsp, tag)) { 592 dsl_dataset_rele(*dsp, tag); 593 return (SET_ERROR(EBUSY)); 594 } 595 return (0); 596} 597 598/* 599 * See the comment above dsl_pool_hold() for details. In summary, a long 600 * hold is used to prevent destruction of a dataset while the pool hold 601 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 602 * 603 * The dataset and pool must be held when this function is called. After it 604 * is called, the pool hold may be released while the dataset is still held 605 * and accessed. 606 */ 607void 608dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 609{ 610 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 611 (void) refcount_add(&ds->ds_longholds, tag); 612} 613 614void 615dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 616{ 617 (void) refcount_remove(&ds->ds_longholds, tag); 618} 619 620/* Return B_TRUE if there are any long holds on this dataset. */ 621boolean_t 622dsl_dataset_long_held(dsl_dataset_t *ds) 623{ 624 return (!refcount_is_zero(&ds->ds_longholds)); 625} 626 627void 628dsl_dataset_name(dsl_dataset_t *ds, char *name) 629{ 630 if (ds == NULL) { 631 (void) strcpy(name, "mos"); 632 } else { 633 dsl_dir_name(ds->ds_dir, name); 634 VERIFY0(dsl_dataset_get_snapname(ds)); 635 if (ds->ds_snapname[0]) { 636 (void) strcat(name, "@"); 637 /* 638 * We use a "recursive" mutex so that we 639 * can call dprintf_ds() with ds_lock held. 640 */ 641 if (!MUTEX_HELD(&ds->ds_lock)) { 642 mutex_enter(&ds->ds_lock); 643 (void) strcat(name, ds->ds_snapname); 644 mutex_exit(&ds->ds_lock); 645 } else { 646 (void) strcat(name, ds->ds_snapname); 647 } 648 } 649 } 650} 651 652void 653dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 654{ 655 dmu_buf_rele(ds->ds_dbuf, tag); 656} 657 658void 659dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 660{ 661 ASSERT3P(ds->ds_owner, ==, tag); 662 ASSERT(ds->ds_dbuf != NULL); 663 664 mutex_enter(&ds->ds_lock); 665 ds->ds_owner = NULL; 666 mutex_exit(&ds->ds_lock); 667 dsl_dataset_long_rele(ds, tag); 668 dsl_dataset_rele(ds, tag); 669} 670 671boolean_t 672dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 673{ 674 boolean_t gotit = FALSE; 675 676 mutex_enter(&ds->ds_lock); 677 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 678 ds->ds_owner = tag; 679 dsl_dataset_long_hold(ds, tag); 680 gotit = TRUE; 681 } 682 mutex_exit(&ds->ds_lock); 683 return (gotit); 684} 685 686uint64_t 687dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 688 uint64_t flags, dmu_tx_t *tx) 689{ 690 dsl_pool_t *dp = dd->dd_pool; 691 dmu_buf_t *dbuf; 692 dsl_dataset_phys_t *dsphys; 693 uint64_t dsobj; 694 objset_t *mos = dp->dp_meta_objset; 695 696 if (origin == NULL) 697 origin = dp->dp_origin_snap; 698 699 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 700 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); 701 ASSERT(dmu_tx_is_syncing(tx)); 702 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 703 704 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 705 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 706 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 707 dmu_buf_will_dirty(dbuf, tx); 708 dsphys = dbuf->db_data; 709 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 710 dsphys->ds_dir_obj = dd->dd_object; 711 dsphys->ds_flags = flags; 712 dsphys->ds_fsid_guid = unique_create(); 713 do { 714 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 715 sizeof (dsphys->ds_guid)); 716 } while (dsphys->ds_guid == 0); 717 dsphys->ds_snapnames_zapobj = 718 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 719 DMU_OT_NONE, 0, tx); 720 dsphys->ds_creation_time = gethrestime_sec(); 721 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 722 723 if (origin == NULL) { 724 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 725 } else { 726 dsl_dataset_t *ohds; /* head of the origin snapshot */ 727 728 dsphys->ds_prev_snap_obj = origin->ds_object; 729 dsphys->ds_prev_snap_txg = 730 dsl_dataset_phys(origin)->ds_creation_txg; 731 dsphys->ds_referenced_bytes = 732 dsl_dataset_phys(origin)->ds_referenced_bytes; 733 dsphys->ds_compressed_bytes = 734 dsl_dataset_phys(origin)->ds_compressed_bytes; 735 dsphys->ds_uncompressed_bytes = 736 dsl_dataset_phys(origin)->ds_uncompressed_bytes; 737 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; 738 739 /* 740 * Inherit flags that describe the dataset's contents 741 * (INCONSISTENT) or properties (Case Insensitive). 742 */ 743 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & 744 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 745 746 if (origin->ds_large_blocks) 747 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); 748 749 dmu_buf_will_dirty(origin->ds_dbuf, tx); 750 dsl_dataset_phys(origin)->ds_num_children++; 751 752 VERIFY0(dsl_dataset_hold_obj(dp, 753 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, 754 FTAG, &ohds)); 755 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 756 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 757 dsl_dataset_rele(ohds, FTAG); 758 759 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 760 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { 761 dsl_dataset_phys(origin)->ds_next_clones_obj = 762 zap_create(mos, 763 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 764 } 765 VERIFY0(zap_add_int(mos, 766 dsl_dataset_phys(origin)->ds_next_clones_obj, 767 dsobj, tx)); 768 } 769 770 dmu_buf_will_dirty(dd->dd_dbuf, tx); 771 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; 772 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 773 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 774 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 775 dsl_dir_phys(origin->ds_dir)->dd_clones = 776 zap_create(mos, 777 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 778 } 779 VERIFY0(zap_add_int(mos, 780 dsl_dir_phys(origin->ds_dir)->dd_clones, 781 dsobj, tx)); 782 } 783 } 784 785 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 786 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 787 788 dmu_buf_rele(dbuf, FTAG); 789 790 dmu_buf_will_dirty(dd->dd_dbuf, tx); 791 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; 792 793 return (dsobj); 794} 795 796static void 797dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 798{ 799 objset_t *os; 800 801 VERIFY0(dmu_objset_from_ds(ds, &os)); 802 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 803 dsl_dataset_dirty(ds, tx); 804} 805 806uint64_t 807dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 808 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 809{ 810 dsl_pool_t *dp = pdd->dd_pool; 811 uint64_t dsobj, ddobj; 812 dsl_dir_t *dd; 813 814 ASSERT(dmu_tx_is_syncing(tx)); 815 ASSERT(lastname[0] != '@'); 816 817 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 818 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 819 820 dsobj = dsl_dataset_create_sync_dd(dd, origin, 821 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 822 823 dsl_deleg_set_create_perms(dd, tx, cr); 824 825 /* 826 * Since we're creating a new node we know it's a leaf, so we can 827 * initialize the counts if the limit feature is active. 828 */ 829 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 830 uint64_t cnt = 0; 831 objset_t *os = dd->dd_pool->dp_meta_objset; 832 833 dsl_dir_zapify(dd, tx); 834 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 835 sizeof (cnt), 1, &cnt, tx)); 836 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 837 sizeof (cnt), 1, &cnt, tx)); 838 } 839 840 dsl_dir_rele(dd, FTAG); 841 842 /* 843 * If we are creating a clone, make sure we zero out any stale 844 * data from the origin snapshots zil header. 845 */ 846 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 847 dsl_dataset_t *ds; 848 849 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 850 dsl_dataset_zero_zil(ds, tx); 851 dsl_dataset_rele(ds, FTAG); 852 } 853 854 return (dsobj); 855} 856 857#ifdef __FreeBSD__ 858/* FreeBSD ioctl compat begin */ 859struct destroyarg { 860 nvlist_t *nvl; 861 const char *snapname; 862}; 863 864static int 865dsl_check_snap_cb(const char *name, void *arg) 866{ 867 struct destroyarg *da = arg; 868 dsl_dataset_t *ds; 869 char *dsname; 870 871 dsname = kmem_asprintf("%s@%s", name, da->snapname); 872 fnvlist_add_boolean(da->nvl, dsname); 873 kmem_free(dsname, strlen(dsname) + 1); 874 875 return (0); 876} 877 878int 879dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname, 880 nvlist_t *snaps) 881{ 882 struct destroyarg *da; 883 int err; 884 885 da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP); 886 da->nvl = snaps; 887 da->snapname = snapname; 888 err = dmu_objset_find(fsname, dsl_check_snap_cb, da, 889 DS_FIND_CHILDREN); 890 kmem_free(da, sizeof (struct destroyarg)); 891 892 return (err); 893} 894/* FreeBSD ioctl compat end */ 895#endif /* __FreeBSD__ */ 896 897/* 898 * The unique space in the head dataset can be calculated by subtracting 899 * the space used in the most recent snapshot, that is still being used 900 * in this file system, from the space currently in use. To figure out 901 * the space in the most recent snapshot still in use, we need to take 902 * the total space used in the snapshot and subtract out the space that 903 * has been freed up since the snapshot was taken. 904 */ 905void 906dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 907{ 908 uint64_t mrs_used; 909 uint64_t dlused, dlcomp, dluncomp; 910 911 ASSERT(!dsl_dataset_is_snapshot(ds)); 912 913 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) 914 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; 915 else 916 mrs_used = 0; 917 918 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 919 920 ASSERT3U(dlused, <=, mrs_used); 921 dsl_dataset_phys(ds)->ds_unique_bytes = 922 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); 923 924 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 925 SPA_VERSION_UNIQUE_ACCURATE) 926 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 927} 928 929void 930dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 931 dmu_tx_t *tx) 932{ 933 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 934 uint64_t count; 935 int err; 936 937 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); 938 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 939 obj, tx); 940 /* 941 * The err should not be ENOENT, but a bug in a previous version 942 * of the code could cause upgrade_clones_cb() to not set 943 * ds_next_snap_obj when it should, leading to a missing entry. 944 * If we knew that the pool was created after 945 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 946 * ENOENT. However, at least we can check that we don't have 947 * too many entries in the next_clones_obj even after failing to 948 * remove this one. 949 */ 950 if (err != ENOENT) 951 VERIFY0(err); 952 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 953 &count)); 954 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); 955} 956 957 958blkptr_t * 959dsl_dataset_get_blkptr(dsl_dataset_t *ds) 960{ 961 return (&dsl_dataset_phys(ds)->ds_bp); 962} 963 964void 965dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 966{ 967 ASSERT(dmu_tx_is_syncing(tx)); 968 /* If it's the meta-objset, set dp_meta_rootbp */ 969 if (ds == NULL) { 970 tx->tx_pool->dp_meta_rootbp = *bp; 971 } else { 972 dmu_buf_will_dirty(ds->ds_dbuf, tx); 973 dsl_dataset_phys(ds)->ds_bp = *bp; 974 } 975} 976 977spa_t * 978dsl_dataset_get_spa(dsl_dataset_t *ds) 979{ 980 return (ds->ds_dir->dd_pool->dp_spa); 981} 982 983void 984dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 985{ 986 dsl_pool_t *dp; 987 988 if (ds == NULL) /* this is the meta-objset */ 989 return; 990 991 ASSERT(ds->ds_objset != NULL); 992 993 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) 994 panic("dirtying snapshot!"); 995 996 dp = ds->ds_dir->dd_pool; 997 998 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 999 /* up the hold count until we can be written out */ 1000 dmu_buf_add_ref(ds->ds_dbuf, ds); 1001 } 1002} 1003 1004boolean_t 1005dsl_dataset_is_dirty(dsl_dataset_t *ds) 1006{ 1007 for (int t = 0; t < TXG_SIZE; t++) { 1008 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1009 ds, t)) 1010 return (B_TRUE); 1011 } 1012 return (B_FALSE); 1013} 1014 1015static int 1016dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1017{ 1018 uint64_t asize; 1019 1020 if (!dmu_tx_is_syncing(tx)) 1021 return (0); 1022 1023 /* 1024 * If there's an fs-only reservation, any blocks that might become 1025 * owned by the snapshot dataset must be accommodated by space 1026 * outside of the reservation. 1027 */ 1028 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1029 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); 1030 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1031 return (SET_ERROR(ENOSPC)); 1032 1033 /* 1034 * Propagate any reserved space for this snapshot to other 1035 * snapshot checks in this sync group. 1036 */ 1037 if (asize > 0) 1038 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1039 1040 return (0); 1041} 1042 1043typedef struct dsl_dataset_snapshot_arg { 1044 nvlist_t *ddsa_snaps; 1045 nvlist_t *ddsa_props; 1046 nvlist_t *ddsa_errors; 1047 cred_t *ddsa_cr; 1048} dsl_dataset_snapshot_arg_t; 1049 1050int 1051dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1052 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 1053{ 1054 int error; 1055 uint64_t value; 1056 1057 ds->ds_trysnap_txg = tx->tx_txg; 1058 1059 if (!dmu_tx_is_syncing(tx)) 1060 return (0); 1061 1062 /* 1063 * We don't allow multiple snapshots of the same txg. If there 1064 * is already one, try again. 1065 */ 1066 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) 1067 return (SET_ERROR(EAGAIN)); 1068 1069 /* 1070 * Check for conflicting snapshot name. 1071 */ 1072 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1073 if (error == 0) 1074 return (SET_ERROR(EEXIST)); 1075 if (error != ENOENT) 1076 return (error); 1077 1078 /* 1079 * We don't allow taking snapshots of inconsistent datasets, such as 1080 * those into which we are currently receiving. However, if we are 1081 * creating this snapshot as part of a receive, this check will be 1082 * executed atomically with respect to the completion of the receive 1083 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1084 * case we ignore this, knowing it will be fixed up for us shortly in 1085 * dmu_recv_end_sync(). 1086 */ 1087 if (!recv && DS_IS_INCONSISTENT(ds)) 1088 return (SET_ERROR(EBUSY)); 1089 1090 /* 1091 * Skip the check for temporary snapshots or if we have already checked 1092 * the counts in dsl_dataset_snapshot_check. This means we really only 1093 * check the count here when we're receiving a stream. 1094 */ 1095 if (cnt != 0 && cr != NULL) { 1096 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1097 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 1098 if (error != 0) 1099 return (error); 1100 } 1101 1102 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1103 if (error != 0) 1104 return (error); 1105 1106 return (0); 1107} 1108 1109static int 1110dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1111{ 1112 dsl_dataset_snapshot_arg_t *ddsa = arg; 1113 dsl_pool_t *dp = dmu_tx_pool(tx); 1114 nvpair_t *pair; 1115 int rv = 0; 1116 1117 /* 1118 * Pre-compute how many total new snapshots will be created for each 1119 * level in the tree and below. This is needed for validating the 1120 * snapshot limit when either taking a recursive snapshot or when 1121 * taking multiple snapshots. 1122 * 1123 * The problem is that the counts are not actually adjusted when 1124 * we are checking, only when we finally sync. For a single snapshot, 1125 * this is easy, the count will increase by 1 at each node up the tree, 1126 * but its more complicated for the recursive/multiple snapshot case. 1127 * 1128 * The dsl_fs_ss_limit_check function does recursively check the count 1129 * at each level up the tree but since it is validating each snapshot 1130 * independently we need to be sure that we are validating the complete 1131 * count for the entire set of snapshots. We do this by rolling up the 1132 * counts for each component of the name into an nvlist and then 1133 * checking each of those cases with the aggregated count. 1134 * 1135 * This approach properly handles not only the recursive snapshot 1136 * case (where we get all of those on the ddsa_snaps list) but also 1137 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1138 * validate the limit on 'a' using a count of 2). 1139 * 1140 * We validate the snapshot names in the third loop and only report 1141 * name errors once. 1142 */ 1143 if (dmu_tx_is_syncing(tx)) { 1144 nvlist_t *cnt_track = NULL; 1145 cnt_track = fnvlist_alloc(); 1146 1147 /* Rollup aggregated counts into the cnt_track list */ 1148 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1149 pair != NULL; 1150 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1151 char *pdelim; 1152 uint64_t val; 1153 char nm[MAXPATHLEN]; 1154 1155 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1156 pdelim = strchr(nm, '@'); 1157 if (pdelim == NULL) 1158 continue; 1159 *pdelim = '\0'; 1160 1161 do { 1162 if (nvlist_lookup_uint64(cnt_track, nm, 1163 &val) == 0) { 1164 /* update existing entry */ 1165 fnvlist_add_uint64(cnt_track, nm, 1166 val + 1); 1167 } else { 1168 /* add to list */ 1169 fnvlist_add_uint64(cnt_track, nm, 1); 1170 } 1171 1172 pdelim = strrchr(nm, '/'); 1173 if (pdelim != NULL) 1174 *pdelim = '\0'; 1175 } while (pdelim != NULL); 1176 } 1177 1178 /* Check aggregated counts at each level */ 1179 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1180 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1181 int error = 0; 1182 char *name; 1183 uint64_t cnt = 0; 1184 dsl_dataset_t *ds; 1185 1186 name = nvpair_name(pair); 1187 cnt = fnvpair_value_uint64(pair); 1188 ASSERT(cnt > 0); 1189 1190 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1191 if (error == 0) { 1192 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1193 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1194 ddsa->ddsa_cr); 1195 dsl_dataset_rele(ds, FTAG); 1196 } 1197 1198 if (error != 0) { 1199 if (ddsa->ddsa_errors != NULL) 1200 fnvlist_add_int32(ddsa->ddsa_errors, 1201 name, error); 1202 rv = error; 1203 /* only report one error for this check */ 1204 break; 1205 } 1206 } 1207 nvlist_free(cnt_track); 1208 } 1209 1210 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1211 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1212 int error = 0; 1213 dsl_dataset_t *ds; 1214 char *name, *atp; 1215 char dsname[MAXNAMELEN]; 1216 1217 name = nvpair_name(pair); 1218 if (strlen(name) >= MAXNAMELEN) 1219 error = SET_ERROR(ENAMETOOLONG); 1220 if (error == 0) { 1221 atp = strchr(name, '@'); 1222 if (atp == NULL) 1223 error = SET_ERROR(EINVAL); 1224 if (error == 0) 1225 (void) strlcpy(dsname, name, atp - name + 1); 1226 } 1227 if (error == 0) 1228 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1229 if (error == 0) { 1230 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1231 error = dsl_dataset_snapshot_check_impl(ds, 1232 atp + 1, tx, B_FALSE, 0, NULL); 1233 dsl_dataset_rele(ds, FTAG); 1234 } 1235 1236 if (error != 0) { 1237 if (ddsa->ddsa_errors != NULL) { 1238 fnvlist_add_int32(ddsa->ddsa_errors, 1239 name, error); 1240 } 1241 rv = error; 1242 } 1243 } 1244 1245 return (rv); 1246} 1247 1248void 1249dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1250 dmu_tx_t *tx) 1251{ 1252 static zil_header_t zero_zil; 1253 1254 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1255 dmu_buf_t *dbuf; 1256 dsl_dataset_phys_t *dsphys; 1257 uint64_t dsobj, crtxg; 1258 objset_t *mos = dp->dp_meta_objset; 1259 objset_t *os; 1260 1261 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1262 1263 /* 1264 * If we are on an old pool, the zil must not be active, in which 1265 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1266 */ 1267 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1268 dmu_objset_from_ds(ds, &os) != 0 || 1269 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1270 sizeof (zero_zil)) == 0); 1271 1272 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1273 1274 /* 1275 * The origin's ds_creation_txg has to be < TXG_INITIAL 1276 */ 1277 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1278 crtxg = 1; 1279 else 1280 crtxg = tx->tx_txg; 1281 1282 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1283 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1284 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1285 dmu_buf_will_dirty(dbuf, tx); 1286 dsphys = dbuf->db_data; 1287 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1288 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1289 dsphys->ds_fsid_guid = unique_create(); 1290 do { 1291 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1292 sizeof (dsphys->ds_guid)); 1293 } while (dsphys->ds_guid == 0); 1294 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1295 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 1296 dsphys->ds_next_snap_obj = ds->ds_object; 1297 dsphys->ds_num_children = 1; 1298 dsphys->ds_creation_time = gethrestime_sec(); 1299 dsphys->ds_creation_txg = crtxg; 1300 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; 1301 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; 1302 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; 1303 dsphys->ds_uncompressed_bytes = 1304 dsl_dataset_phys(ds)->ds_uncompressed_bytes; 1305 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; 1306 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; 1307 dmu_buf_rele(dbuf, FTAG); 1308 1309 if (ds->ds_large_blocks) 1310 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); 1311 1312 ASSERT3U(ds->ds_prev != 0, ==, 1313 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); 1314 if (ds->ds_prev) { 1315 uint64_t next_clones_obj = 1316 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; 1317 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1318 ds->ds_object || 1319 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); 1320 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1321 ds->ds_object) { 1322 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1323 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, 1324 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); 1325 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; 1326 } else if (next_clones_obj != 0) { 1327 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1328 dsphys->ds_next_snap_obj, tx); 1329 VERIFY0(zap_add_int(mos, 1330 next_clones_obj, dsobj, tx)); 1331 } 1332 } 1333 1334 /* 1335 * If we have a reference-reservation on this dataset, we will 1336 * need to increase the amount of refreservation being charged 1337 * since our unique space is going to zero. 1338 */ 1339 if (ds->ds_reserved) { 1340 int64_t delta; 1341 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1342 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, 1343 ds->ds_reserved); 1344 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1345 delta, 0, 0, tx); 1346 } 1347 1348 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1349 dsl_dataset_phys(ds)->ds_deadlist_obj = 1350 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, 1351 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 1352 dsl_deadlist_close(&ds->ds_deadlist); 1353 dsl_deadlist_open(&ds->ds_deadlist, mos, 1354 dsl_dataset_phys(ds)->ds_deadlist_obj); 1355 dsl_deadlist_add_key(&ds->ds_deadlist, 1356 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); 1357 1358 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); 1359 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; 1360 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; 1361 dsl_dataset_phys(ds)->ds_unique_bytes = 0; 1362 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1363 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1364 1365 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1366 snapname, 8, 1, &dsobj, tx)); 1367 1368 if (ds->ds_prev) 1369 dsl_dataset_rele(ds->ds_prev, ds); 1370 VERIFY0(dsl_dataset_hold_obj(dp, 1371 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); 1372 1373 dsl_scan_ds_snapshotted(ds, tx); 1374 1375 dsl_dir_snap_cmtime_update(ds->ds_dir); 1376 1377 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1378} 1379 1380static void 1381dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1382{ 1383 dsl_dataset_snapshot_arg_t *ddsa = arg; 1384 dsl_pool_t *dp = dmu_tx_pool(tx); 1385 nvpair_t *pair; 1386 1387 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1388 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1389 dsl_dataset_t *ds; 1390 char *name, *atp; 1391 char dsname[MAXNAMELEN]; 1392 1393 name = nvpair_name(pair); 1394 atp = strchr(name, '@'); 1395 (void) strlcpy(dsname, name, atp - name + 1); 1396 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1397 1398 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1399 if (ddsa->ddsa_props != NULL) { 1400 dsl_props_set_sync_impl(ds->ds_prev, 1401 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1402 } 1403 dsl_dataset_rele(ds, FTAG); 1404 } 1405} 1406 1407/* 1408 * The snapshots must all be in the same pool. 1409 * All-or-nothing: if there are any failures, nothing will be modified. 1410 */ 1411int 1412dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1413{ 1414 dsl_dataset_snapshot_arg_t ddsa; 1415 nvpair_t *pair; 1416 boolean_t needsuspend; 1417 int error; 1418 spa_t *spa; 1419 char *firstname; 1420 nvlist_t *suspended = NULL; 1421 1422 pair = nvlist_next_nvpair(snaps, NULL); 1423 if (pair == NULL) 1424 return (0); 1425 firstname = nvpair_name(pair); 1426 1427 error = spa_open(firstname, &spa, FTAG); 1428 if (error != 0) 1429 return (error); 1430 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1431 spa_close(spa, FTAG); 1432 1433 if (needsuspend) { 1434 suspended = fnvlist_alloc(); 1435 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1436 pair = nvlist_next_nvpair(snaps, pair)) { 1437 char fsname[MAXNAMELEN]; 1438 char *snapname = nvpair_name(pair); 1439 char *atp; 1440 void *cookie; 1441 1442 atp = strchr(snapname, '@'); 1443 if (atp == NULL) { 1444 error = SET_ERROR(EINVAL); 1445 break; 1446 } 1447 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1448 1449 error = zil_suspend(fsname, &cookie); 1450 if (error != 0) 1451 break; 1452 fnvlist_add_uint64(suspended, fsname, 1453 (uintptr_t)cookie); 1454 } 1455 } 1456 1457 ddsa.ddsa_snaps = snaps; 1458 ddsa.ddsa_props = props; 1459 ddsa.ddsa_errors = errors; 1460 ddsa.ddsa_cr = CRED(); 1461 1462 if (error == 0) { 1463 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1464 dsl_dataset_snapshot_sync, &ddsa, 1465 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1466 } 1467 1468 if (suspended != NULL) { 1469 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1470 pair = nvlist_next_nvpair(suspended, pair)) { 1471 zil_resume((void *)(uintptr_t) 1472 fnvpair_value_uint64(pair)); 1473 } 1474 fnvlist_free(suspended); 1475 } 1476 1477#ifdef __FreeBSD__ 1478#ifdef _KERNEL 1479 if (error == 0) { 1480 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1481 pair = nvlist_next_nvpair(snaps, pair)) { 1482 char *snapname = nvpair_name(pair); 1483 zvol_create_minors(snapname); 1484 } 1485 } 1486#endif 1487#endif 1488 return (error); 1489} 1490 1491typedef struct dsl_dataset_snapshot_tmp_arg { 1492 const char *ddsta_fsname; 1493 const char *ddsta_snapname; 1494 minor_t ddsta_cleanup_minor; 1495 const char *ddsta_htag; 1496} dsl_dataset_snapshot_tmp_arg_t; 1497 1498static int 1499dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1500{ 1501 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1502 dsl_pool_t *dp = dmu_tx_pool(tx); 1503 dsl_dataset_t *ds; 1504 int error; 1505 1506 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1507 if (error != 0) 1508 return (error); 1509 1510 /* NULL cred means no limit check for tmp snapshot */ 1511 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1512 tx, B_FALSE, 0, NULL); 1513 if (error != 0) { 1514 dsl_dataset_rele(ds, FTAG); 1515 return (error); 1516 } 1517 1518 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1519 dsl_dataset_rele(ds, FTAG); 1520 return (SET_ERROR(ENOTSUP)); 1521 } 1522 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1523 B_TRUE, tx); 1524 if (error != 0) { 1525 dsl_dataset_rele(ds, FTAG); 1526 return (error); 1527 } 1528 1529 dsl_dataset_rele(ds, FTAG); 1530 return (0); 1531} 1532 1533static void 1534dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1535{ 1536 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1537 dsl_pool_t *dp = dmu_tx_pool(tx); 1538 dsl_dataset_t *ds; 1539 1540 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1541 1542 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1543 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1544 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1545 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1546 1547 dsl_dataset_rele(ds, FTAG); 1548} 1549 1550int 1551dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1552 minor_t cleanup_minor, const char *htag) 1553{ 1554 dsl_dataset_snapshot_tmp_arg_t ddsta; 1555 int error; 1556 spa_t *spa; 1557 boolean_t needsuspend; 1558 void *cookie; 1559 1560 ddsta.ddsta_fsname = fsname; 1561 ddsta.ddsta_snapname = snapname; 1562 ddsta.ddsta_cleanup_minor = cleanup_minor; 1563 ddsta.ddsta_htag = htag; 1564 1565 error = spa_open(fsname, &spa, FTAG); 1566 if (error != 0) 1567 return (error); 1568 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1569 spa_close(spa, FTAG); 1570 1571 if (needsuspend) { 1572 error = zil_suspend(fsname, &cookie); 1573 if (error != 0) 1574 return (error); 1575 } 1576 1577 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1578 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 1579 1580 if (needsuspend) 1581 zil_resume(cookie); 1582 return (error); 1583} 1584 1585 1586void 1587dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1588{ 1589 ASSERT(dmu_tx_is_syncing(tx)); 1590 ASSERT(ds->ds_objset != NULL); 1591 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); 1592 1593 /* 1594 * in case we had to change ds_fsid_guid when we opened it, 1595 * sync it out now. 1596 */ 1597 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1598 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; 1599 1600 dmu_objset_sync(ds->ds_objset, zio, tx); 1601 1602 if (ds->ds_need_large_blocks && !ds->ds_large_blocks) { 1603 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx); 1604 ds->ds_large_blocks = B_TRUE; 1605 } 1606} 1607 1608static void 1609get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1610{ 1611 uint64_t count = 0; 1612 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1613 zap_cursor_t zc; 1614 zap_attribute_t za; 1615 nvlist_t *propval = fnvlist_alloc(); 1616 nvlist_t *val = fnvlist_alloc(); 1617 1618 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1619 1620 /* 1621 * There may be missing entries in ds_next_clones_obj 1622 * due to a bug in a previous version of the code. 1623 * Only trust it if it has the right number of entries. 1624 */ 1625 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { 1626 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1627 &count)); 1628 } 1629 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) 1630 goto fail; 1631 for (zap_cursor_init(&zc, mos, 1632 dsl_dataset_phys(ds)->ds_next_clones_obj); 1633 zap_cursor_retrieve(&zc, &za) == 0; 1634 zap_cursor_advance(&zc)) { 1635 dsl_dataset_t *clone; 1636 char buf[ZFS_MAXNAMELEN]; 1637 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1638 za.za_first_integer, FTAG, &clone)); 1639 dsl_dir_name(clone->ds_dir, buf); 1640 fnvlist_add_boolean(val, buf); 1641 dsl_dataset_rele(clone, FTAG); 1642 } 1643 zap_cursor_fini(&zc); 1644 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1645 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); 1646fail: 1647 nvlist_free(val); 1648 nvlist_free(propval); 1649} 1650 1651void 1652dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1653{ 1654 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1655 uint64_t refd, avail, uobjs, aobjs, ratio; 1656 1657 ASSERT(dsl_pool_config_held(dp)); 1658 1659 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : 1660 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / 1661 dsl_dataset_phys(ds)->ds_compressed_bytes); 1662 1663 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); 1664 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 1665 dsl_dataset_phys(ds)->ds_uncompressed_bytes); 1666 1667 if (dsl_dataset_is_snapshot(ds)) { 1668 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); 1669 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 1670 dsl_dataset_phys(ds)->ds_unique_bytes); 1671 get_clones_stat(ds, nv); 1672 } else { 1673 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 1674 char buf[MAXNAMELEN]; 1675 dsl_dataset_name(ds->ds_prev, buf); 1676 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf); 1677 } 1678 1679 dsl_dir_stats(ds->ds_dir, nv); 1680 } 1681 1682 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 1683 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 1684 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 1685 1686 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 1687 dsl_dataset_phys(ds)->ds_creation_time); 1688 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 1689 dsl_dataset_phys(ds)->ds_creation_txg); 1690 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 1691 ds->ds_quota); 1692 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 1693 ds->ds_reserved); 1694 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 1695 dsl_dataset_phys(ds)->ds_guid); 1696 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 1697 dsl_dataset_phys(ds)->ds_unique_bytes); 1698 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 1699 ds->ds_object); 1700 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 1701 ds->ds_userrefs); 1702 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 1703 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 1704 1705 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 1706 uint64_t written, comp, uncomp; 1707 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1708 dsl_dataset_t *prev; 1709 1710 int err = dsl_dataset_hold_obj(dp, 1711 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 1712 if (err == 0) { 1713 err = dsl_dataset_space_written(prev, ds, &written, 1714 &comp, &uncomp); 1715 dsl_dataset_rele(prev, FTAG); 1716 if (err == 0) { 1717 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 1718 written); 1719 } 1720 } 1721 } 1722} 1723 1724void 1725dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 1726{ 1727 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1728 ASSERT(dsl_pool_config_held(dp)); 1729 1730 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg; 1731 stat->dds_inconsistent = 1732 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT; 1733 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid; 1734 stat->dds_origin[0] = '\0'; 1735 if (dsl_dataset_is_snapshot(ds)) { 1736 stat->dds_is_snapshot = B_TRUE; 1737 stat->dds_num_clones = 1738 dsl_dataset_phys(ds)->ds_num_children - 1; 1739 } else { 1740 stat->dds_is_snapshot = B_FALSE; 1741 stat->dds_num_clones = 0; 1742 1743 if (dsl_dir_is_clone(ds->ds_dir)) { 1744 dsl_dataset_t *ods; 1745 1746 VERIFY0(dsl_dataset_hold_obj(dp, 1747 dsl_dir_phys(ds->ds_dir)->dd_origin_obj, 1748 FTAG, &ods)); 1749 dsl_dataset_name(ods, stat->dds_origin); 1750 dsl_dataset_rele(ods, FTAG); 1751 } 1752 } 1753} 1754 1755uint64_t 1756dsl_dataset_fsid_guid(dsl_dataset_t *ds) 1757{ 1758 return (ds->ds_fsid_guid); 1759} 1760 1761void 1762dsl_dataset_space(dsl_dataset_t *ds, 1763 uint64_t *refdbytesp, uint64_t *availbytesp, 1764 uint64_t *usedobjsp, uint64_t *availobjsp) 1765{ 1766 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; 1767 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 1768 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) 1769 *availbytesp += 1770 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 1771 if (ds->ds_quota != 0) { 1772 /* 1773 * Adjust available bytes according to refquota 1774 */ 1775 if (*refdbytesp < ds->ds_quota) 1776 *availbytesp = MIN(*availbytesp, 1777 ds->ds_quota - *refdbytesp); 1778 else 1779 *availbytesp = 0; 1780 } 1781 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); 1782 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 1783} 1784 1785boolean_t 1786dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 1787{ 1788 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1789 1790 ASSERT(dsl_pool_config_held(dp)); 1791 if (snap == NULL) 1792 return (B_FALSE); 1793 if (dsl_dataset_phys(ds)->ds_bp.blk_birth > 1794 dsl_dataset_phys(snap)->ds_creation_txg) { 1795 objset_t *os, *os_snap; 1796 /* 1797 * It may be that only the ZIL differs, because it was 1798 * reset in the head. Don't count that as being 1799 * modified. 1800 */ 1801 if (dmu_objset_from_ds(ds, &os) != 0) 1802 return (B_TRUE); 1803 if (dmu_objset_from_ds(snap, &os_snap) != 0) 1804 return (B_TRUE); 1805 return (bcmp(&os->os_phys->os_meta_dnode, 1806 &os_snap->os_phys->os_meta_dnode, 1807 sizeof (os->os_phys->os_meta_dnode)) != 0); 1808 } 1809 return (B_FALSE); 1810} 1811 1812typedef struct dsl_dataset_rename_snapshot_arg { 1813 const char *ddrsa_fsname; 1814 const char *ddrsa_oldsnapname; 1815 const char *ddrsa_newsnapname; 1816 boolean_t ddrsa_recursive; 1817 dmu_tx_t *ddrsa_tx; 1818} dsl_dataset_rename_snapshot_arg_t; 1819 1820/* ARGSUSED */ 1821static int 1822dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 1823 dsl_dataset_t *hds, void *arg) 1824{ 1825 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1826 int error; 1827 uint64_t val; 1828 1829 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1830 if (error != 0) { 1831 /* ignore nonexistent snapshots */ 1832 return (error == ENOENT ? 0 : error); 1833 } 1834 1835 /* new name should not exist */ 1836 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 1837 if (error == 0) 1838 error = SET_ERROR(EEXIST); 1839 else if (error == ENOENT) 1840 error = 0; 1841 1842 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 1843 if (dsl_dir_namelen(hds->ds_dir) + 1 + 1844 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN) 1845 error = SET_ERROR(ENAMETOOLONG); 1846 1847 return (error); 1848} 1849 1850static int 1851dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 1852{ 1853 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1854 dsl_pool_t *dp = dmu_tx_pool(tx); 1855 dsl_dataset_t *hds; 1856 int error; 1857 1858 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 1859 if (error != 0) 1860 return (error); 1861 1862 if (ddrsa->ddrsa_recursive) { 1863 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1864 dsl_dataset_rename_snapshot_check_impl, ddrsa, 1865 DS_FIND_CHILDREN); 1866 } else { 1867 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 1868 } 1869 dsl_dataset_rele(hds, FTAG); 1870 return (error); 1871} 1872 1873static int 1874dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 1875 dsl_dataset_t *hds, void *arg) 1876{ 1877#ifdef __FreeBSD__ 1878#ifdef _KERNEL 1879 char *oldname, *newname; 1880#endif 1881#endif 1882 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1883 dsl_dataset_t *ds; 1884 uint64_t val; 1885 dmu_tx_t *tx = ddrsa->ddrsa_tx; 1886 int error; 1887 1888 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1889 ASSERT(error == 0 || error == ENOENT); 1890 if (error == ENOENT) { 1891 /* ignore nonexistent snapshots */ 1892 return (0); 1893 } 1894 1895 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 1896 1897 /* log before we change the name */ 1898 spa_history_log_internal_ds(ds, "rename", tx, 1899 "-> @%s", ddrsa->ddrsa_newsnapname); 1900 1901 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 1902 B_FALSE)); 1903 mutex_enter(&ds->ds_lock); 1904 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 1905 mutex_exit(&ds->ds_lock); 1906 VERIFY0(zap_add(dp->dp_meta_objset, 1907 dsl_dataset_phys(hds)->ds_snapnames_zapobj, 1908 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 1909 1910#ifdef __FreeBSD__ 1911#ifdef _KERNEL 1912 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1913 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1914 snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname, 1915 ddrsa->ddrsa_oldsnapname); 1916 snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname, 1917 ddrsa->ddrsa_newsnapname); 1918 zfsvfs_update_fromname(oldname, newname); 1919 zvol_rename_minors(oldname, newname); 1920 kmem_free(newname, MAXPATHLEN); 1921 kmem_free(oldname, MAXPATHLEN); 1922#endif 1923#endif 1924 dsl_dataset_rele(ds, FTAG); 1925 1926 return (0); 1927} 1928 1929static void 1930dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 1931{ 1932 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1933 dsl_pool_t *dp = dmu_tx_pool(tx); 1934 dsl_dataset_t *hds; 1935 1936 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 1937 ddrsa->ddrsa_tx = tx; 1938 if (ddrsa->ddrsa_recursive) { 1939 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1940 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 1941 DS_FIND_CHILDREN)); 1942 } else { 1943 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 1944 } 1945 dsl_dataset_rele(hds, FTAG); 1946} 1947 1948int 1949dsl_dataset_rename_snapshot(const char *fsname, 1950 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 1951{ 1952 dsl_dataset_rename_snapshot_arg_t ddrsa; 1953 1954 ddrsa.ddrsa_fsname = fsname; 1955 ddrsa.ddrsa_oldsnapname = oldsnapname; 1956 ddrsa.ddrsa_newsnapname = newsnapname; 1957 ddrsa.ddrsa_recursive = recursive; 1958 1959 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 1960 dsl_dataset_rename_snapshot_sync, &ddrsa, 1961 1, ZFS_SPACE_CHECK_RESERVED)); 1962} 1963 1964/* 1965 * If we're doing an ownership handoff, we need to make sure that there is 1966 * only one long hold on the dataset. We're not allowed to change anything here 1967 * so we don't permanently release the long hold or regular hold here. We want 1968 * to do this only when syncing to avoid the dataset unexpectedly going away 1969 * when we release the long hold. 1970 */ 1971static int 1972dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 1973{ 1974 boolean_t held; 1975 1976 if (!dmu_tx_is_syncing(tx)) 1977 return (0); 1978 1979 if (owner != NULL) { 1980 VERIFY3P(ds->ds_owner, ==, owner); 1981 dsl_dataset_long_rele(ds, owner); 1982 } 1983 1984 held = dsl_dataset_long_held(ds); 1985 1986 if (owner != NULL) 1987 dsl_dataset_long_hold(ds, owner); 1988 1989 if (held) 1990 return (SET_ERROR(EBUSY)); 1991 1992 return (0); 1993} 1994 1995typedef struct dsl_dataset_rollback_arg { 1996 const char *ddra_fsname; 1997 void *ddra_owner; 1998 nvlist_t *ddra_result; 1999} dsl_dataset_rollback_arg_t; 2000 2001static int 2002dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 2003{ 2004 dsl_dataset_rollback_arg_t *ddra = arg; 2005 dsl_pool_t *dp = dmu_tx_pool(tx); 2006 dsl_dataset_t *ds; 2007 int64_t unused_refres_delta; 2008 int error; 2009 2010 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 2011 if (error != 0) 2012 return (error); 2013 2014 /* must not be a snapshot */ 2015 if (dsl_dataset_is_snapshot(ds)) { 2016 dsl_dataset_rele(ds, FTAG); 2017 return (SET_ERROR(EINVAL)); 2018 } 2019 2020 /* must have a most recent snapshot */ 2021 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { 2022 dsl_dataset_rele(ds, FTAG); 2023 return (SET_ERROR(EINVAL)); 2024 } 2025 2026 /* must not have any bookmarks after the most recent snapshot */ 2027 nvlist_t *proprequest = fnvlist_alloc(); 2028 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2029 nvlist_t *bookmarks = fnvlist_alloc(); 2030 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 2031 fnvlist_free(proprequest); 2032 if (error != 0) 2033 return (error); 2034 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 2035 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 2036 nvlist_t *valuenv = 2037 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 2038 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2039 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 2040 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 2041 fnvlist_free(bookmarks); 2042 dsl_dataset_rele(ds, FTAG); 2043 return (SET_ERROR(EEXIST)); 2044 } 2045 } 2046 fnvlist_free(bookmarks); 2047 2048 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 2049 if (error != 0) { 2050 dsl_dataset_rele(ds, FTAG); 2051 return (error); 2052 } 2053 2054 /* 2055 * Check if the snap we are rolling back to uses more than 2056 * the refquota. 2057 */ 2058 if (ds->ds_quota != 0 && 2059 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { 2060 dsl_dataset_rele(ds, FTAG); 2061 return (SET_ERROR(EDQUOT)); 2062 } 2063 2064 /* 2065 * When we do the clone swap, we will temporarily use more space 2066 * due to the refreservation (the head will no longer have any 2067 * unique space, so the entire amount of the refreservation will need 2068 * to be free). We will immediately destroy the clone, freeing 2069 * this space, but the freeing happens over many txg's. 2070 */ 2071 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 2072 dsl_dataset_phys(ds)->ds_unique_bytes); 2073 2074 if (unused_refres_delta > 0 && 2075 unused_refres_delta > 2076 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 2077 dsl_dataset_rele(ds, FTAG); 2078 return (SET_ERROR(ENOSPC)); 2079 } 2080 2081 dsl_dataset_rele(ds, FTAG); 2082 return (0); 2083} 2084 2085static void 2086dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 2087{ 2088 dsl_dataset_rollback_arg_t *ddra = arg; 2089 dsl_pool_t *dp = dmu_tx_pool(tx); 2090 dsl_dataset_t *ds, *clone; 2091 uint64_t cloneobj; 2092 char namebuf[ZFS_MAXNAMELEN]; 2093 2094 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 2095 2096 dsl_dataset_name(ds->ds_prev, namebuf); 2097 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 2098 2099 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 2100 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 2101 2102 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 2103 2104 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 2105 dsl_dataset_zero_zil(ds, tx); 2106 2107 dsl_destroy_head_sync_impl(clone, tx); 2108 2109 dsl_dataset_rele(clone, FTAG); 2110 dsl_dataset_rele(ds, FTAG); 2111} 2112 2113/* 2114 * Rolls back the given filesystem or volume to the most recent snapshot. 2115 * The name of the most recent snapshot will be returned under key "target" 2116 * in the result nvlist. 2117 * 2118 * If owner != NULL: 2119 * - The existing dataset MUST be owned by the specified owner at entry 2120 * - Upon return, dataset will still be held by the same owner, whether we 2121 * succeed or not. 2122 * 2123 * This mode is required any time the existing filesystem is mounted. See 2124 * notes above zfs_suspend_fs() for further details. 2125 */ 2126int 2127dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result) 2128{ 2129 dsl_dataset_rollback_arg_t ddra; 2130 2131 ddra.ddra_fsname = fsname; 2132 ddra.ddra_owner = owner; 2133 ddra.ddra_result = result; 2134 2135 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 2136 dsl_dataset_rollback_sync, &ddra, 2137 1, ZFS_SPACE_CHECK_RESERVED)); 2138} 2139 2140struct promotenode { 2141 list_node_t link; 2142 dsl_dataset_t *ds; 2143}; 2144 2145typedef struct dsl_dataset_promote_arg { 2146 const char *ddpa_clonename; 2147 dsl_dataset_t *ddpa_clone; 2148 list_t shared_snaps, origin_snaps, clone_snaps; 2149 dsl_dataset_t *origin_origin; /* origin of the origin */ 2150 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2151 char *err_ds; 2152 cred_t *cr; 2153} dsl_dataset_promote_arg_t; 2154 2155static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2156static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2157 void *tag); 2158static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2159 2160static int 2161dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2162{ 2163 dsl_dataset_promote_arg_t *ddpa = arg; 2164 dsl_pool_t *dp = dmu_tx_pool(tx); 2165 dsl_dataset_t *hds; 2166 struct promotenode *snap; 2167 dsl_dataset_t *origin_ds; 2168 int err; 2169 uint64_t unused; 2170 uint64_t ss_mv_cnt; 2171 size_t max_snap_len; 2172 2173 err = promote_hold(ddpa, dp, FTAG); 2174 if (err != 0) 2175 return (err); 2176 2177 hds = ddpa->ddpa_clone; 2178 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; 2179 2180 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { 2181 promote_rele(ddpa, FTAG); 2182 return (SET_ERROR(EXDEV)); 2183 } 2184 2185 /* 2186 * Compute and check the amount of space to transfer. Since this is 2187 * so expensive, don't do the preliminary check. 2188 */ 2189 if (!dmu_tx_is_syncing(tx)) { 2190 promote_rele(ddpa, FTAG); 2191 return (0); 2192 } 2193 2194 snap = list_head(&ddpa->shared_snaps); 2195 origin_ds = snap->ds; 2196 2197 /* compute origin's new unique space */ 2198 snap = list_tail(&ddpa->clone_snaps); 2199 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2200 origin_ds->ds_object); 2201 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2202 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, 2203 &ddpa->unique, &unused, &unused); 2204 2205 /* 2206 * Walk the snapshots that we are moving 2207 * 2208 * Compute space to transfer. Consider the incremental changes 2209 * to used by each snapshot: 2210 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2211 * So each snapshot gave birth to: 2212 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2213 * So a sequence would look like: 2214 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2215 * Which simplifies to: 2216 * uN + kN + kN-1 + ... + k1 + k0 2217 * Note however, if we stop before we reach the ORIGIN we get: 2218 * uN + kN + kN-1 + ... + kM - uM-1 2219 */ 2220 ss_mv_cnt = 0; 2221 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; 2222 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; 2223 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; 2224 for (snap = list_head(&ddpa->shared_snaps); snap; 2225 snap = list_next(&ddpa->shared_snaps, snap)) { 2226 uint64_t val, dlused, dlcomp, dluncomp; 2227 dsl_dataset_t *ds = snap->ds; 2228 2229 ss_mv_cnt++; 2230 2231 /* 2232 * If there are long holds, we won't be able to evict 2233 * the objset. 2234 */ 2235 if (dsl_dataset_long_held(ds)) { 2236 err = SET_ERROR(EBUSY); 2237 goto out; 2238 } 2239 2240 /* Check that the snapshot name does not conflict */ 2241 VERIFY0(dsl_dataset_get_snapname(ds)); 2242 if (strlen(ds->ds_snapname) >= max_snap_len) { 2243 err = SET_ERROR(ENAMETOOLONG); 2244 goto out; 2245 } 2246 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2247 if (err == 0) { 2248 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname); 2249 err = SET_ERROR(EEXIST); 2250 goto out; 2251 } 2252 if (err != ENOENT) 2253 goto out; 2254 2255 /* The very first snapshot does not have a deadlist */ 2256 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) 2257 continue; 2258 2259 dsl_deadlist_space(&ds->ds_deadlist, 2260 &dlused, &dlcomp, &dluncomp); 2261 ddpa->used += dlused; 2262 ddpa->comp += dlcomp; 2263 ddpa->uncomp += dluncomp; 2264 } 2265 2266 /* 2267 * If we are a clone of a clone then we never reached ORIGIN, 2268 * so we need to subtract out the clone origin's used space. 2269 */ 2270 if (ddpa->origin_origin) { 2271 ddpa->used -= 2272 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; 2273 ddpa->comp -= 2274 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; 2275 ddpa->uncomp -= 2276 dsl_dataset_phys(ddpa->origin_origin)-> 2277 ds_uncompressed_bytes; 2278 } 2279 2280 /* Check that there is enough space and limit headroom here */ 2281 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2282 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2283 if (err != 0) 2284 goto out; 2285 2286 /* 2287 * Compute the amounts of space that will be used by snapshots 2288 * after the promotion (for both origin and clone). For each, 2289 * it is the amount of space that will be on all of their 2290 * deadlists (that was not born before their new origin). 2291 */ 2292 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2293 uint64_t space; 2294 2295 /* 2296 * Note, typically this will not be a clone of a clone, 2297 * so dd_origin_txg will be < TXG_INITIAL, so 2298 * these snaplist_space() -> dsl_deadlist_space_range() 2299 * calls will be fast because they do not have to 2300 * iterate over all bps. 2301 */ 2302 snap = list_head(&ddpa->origin_snaps); 2303 err = snaplist_space(&ddpa->shared_snaps, 2304 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2305 if (err != 0) 2306 goto out; 2307 2308 err = snaplist_space(&ddpa->clone_snaps, 2309 snap->ds->ds_dir->dd_origin_txg, &space); 2310 if (err != 0) 2311 goto out; 2312 ddpa->cloneusedsnap += space; 2313 } 2314 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & 2315 DD_FLAG_USED_BREAKDOWN) { 2316 err = snaplist_space(&ddpa->origin_snaps, 2317 dsl_dataset_phys(origin_ds)->ds_creation_txg, 2318 &ddpa->originusedsnap); 2319 if (err != 0) 2320 goto out; 2321 } 2322 2323out: 2324 promote_rele(ddpa, FTAG); 2325 return (err); 2326} 2327 2328static void 2329dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2330{ 2331 dsl_dataset_promote_arg_t *ddpa = arg; 2332 dsl_pool_t *dp = dmu_tx_pool(tx); 2333 dsl_dataset_t *hds; 2334 struct promotenode *snap; 2335 dsl_dataset_t *origin_ds; 2336 dsl_dataset_t *origin_head; 2337 dsl_dir_t *dd; 2338 dsl_dir_t *odd = NULL; 2339 uint64_t oldnext_obj; 2340 int64_t delta; 2341#if defined(__FreeBSD__) && defined(_KERNEL) 2342 char *oldname, *newname; 2343#endif 2344 2345 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2346 hds = ddpa->ddpa_clone; 2347 2348 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); 2349 2350 snap = list_head(&ddpa->shared_snaps); 2351 origin_ds = snap->ds; 2352 dd = hds->ds_dir; 2353 2354 snap = list_head(&ddpa->origin_snaps); 2355 origin_head = snap->ds; 2356 2357 /* 2358 * We need to explicitly open odd, since origin_ds's dd will be 2359 * changing. 2360 */ 2361 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2362 NULL, FTAG, &odd)); 2363 2364 /* change origin's next snap */ 2365 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2366 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; 2367 snap = list_tail(&ddpa->clone_snaps); 2368 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2369 origin_ds->ds_object); 2370 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; 2371 2372 /* change the origin's next clone */ 2373 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { 2374 dsl_dataset_remove_from_next_clones(origin_ds, 2375 snap->ds->ds_object, tx); 2376 VERIFY0(zap_add_int(dp->dp_meta_objset, 2377 dsl_dataset_phys(origin_ds)->ds_next_clones_obj, 2378 oldnext_obj, tx)); 2379 } 2380 2381 /* change origin */ 2382 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2383 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); 2384 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; 2385 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2386 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2387 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; 2388 origin_head->ds_dir->dd_origin_txg = 2389 dsl_dataset_phys(origin_ds)->ds_creation_txg; 2390 2391 /* change dd_clone entries */ 2392 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2393 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2394 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); 2395 VERIFY0(zap_add_int(dp->dp_meta_objset, 2396 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2397 hds->ds_object, tx)); 2398 2399 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2400 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2401 origin_head->ds_object, tx)); 2402 if (dsl_dir_phys(dd)->dd_clones == 0) { 2403 dsl_dir_phys(dd)->dd_clones = 2404 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, 2405 DMU_OT_NONE, 0, tx); 2406 } 2407 VERIFY0(zap_add_int(dp->dp_meta_objset, 2408 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); 2409 } 2410 2411#if defined(__FreeBSD__) && defined(_KERNEL) 2412 /* Take the spa_namespace_lock early so zvol renames don't deadlock. */ 2413 mutex_enter(&spa_namespace_lock); 2414 2415 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2416 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2417#endif 2418 2419 /* move snapshots to this dir */ 2420 for (snap = list_head(&ddpa->shared_snaps); snap; 2421 snap = list_next(&ddpa->shared_snaps, snap)) { 2422 dsl_dataset_t *ds = snap->ds; 2423 2424 /* 2425 * Property callbacks are registered to a particular 2426 * dsl_dir. Since ours is changing, evict the objset 2427 * so that they will be unregistered from the old dsl_dir. 2428 */ 2429 if (ds->ds_objset) { 2430 dmu_objset_evict(ds->ds_objset); 2431 ds->ds_objset = NULL; 2432 } 2433 2434 /* move snap name entry */ 2435 VERIFY0(dsl_dataset_get_snapname(ds)); 2436 VERIFY0(dsl_dataset_snap_remove(origin_head, 2437 ds->ds_snapname, tx, B_TRUE)); 2438 VERIFY0(zap_add(dp->dp_meta_objset, 2439 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 2440 8, 1, &ds->ds_object, tx)); 2441 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 2442 DD_FIELD_SNAPSHOT_COUNT, tx); 2443 2444 /* change containing dsl_dir */ 2445 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2446 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); 2447 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; 2448 ASSERT3P(ds->ds_dir, ==, odd); 2449 dsl_dir_rele(ds->ds_dir, ds); 2450 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 2451 NULL, ds, &ds->ds_dir)); 2452 2453#if defined(__FreeBSD__) && defined(_KERNEL) 2454 dsl_dataset_name(ds, newname); 2455 zfsvfs_update_fromname(oldname, newname); 2456 zvol_rename_minors(oldname, newname); 2457#endif 2458 2459 /* move any clone references */ 2460 if (dsl_dataset_phys(ds)->ds_next_clones_obj && 2461 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2462 zap_cursor_t zc; 2463 zap_attribute_t za; 2464 2465 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2466 dsl_dataset_phys(ds)->ds_next_clones_obj); 2467 zap_cursor_retrieve(&zc, &za) == 0; 2468 zap_cursor_advance(&zc)) { 2469 dsl_dataset_t *cnds; 2470 uint64_t o; 2471 2472 if (za.za_first_integer == oldnext_obj) { 2473 /* 2474 * We've already moved the 2475 * origin's reference. 2476 */ 2477 continue; 2478 } 2479 2480 VERIFY0(dsl_dataset_hold_obj(dp, 2481 za.za_first_integer, FTAG, &cnds)); 2482 o = dsl_dir_phys(cnds->ds_dir)-> 2483 dd_head_dataset_obj; 2484 2485 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2486 dsl_dir_phys(odd)->dd_clones, o, tx)); 2487 VERIFY0(zap_add_int(dp->dp_meta_objset, 2488 dsl_dir_phys(dd)->dd_clones, o, tx)); 2489 dsl_dataset_rele(cnds, FTAG); 2490 } 2491 zap_cursor_fini(&zc); 2492 } 2493 2494 ASSERT(!dsl_prop_hascb(ds)); 2495 } 2496 2497#if defined(__FreeBSD__) && defined(_KERNEL) 2498 mutex_exit(&spa_namespace_lock); 2499 2500 kmem_free(newname, MAXPATHLEN); 2501 kmem_free(oldname, MAXPATHLEN); 2502#endif 2503 /* 2504 * Change space accounting. 2505 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2506 * both be valid, or both be 0 (resulting in delta == 0). This 2507 * is true for each of {clone,origin} independently. 2508 */ 2509 2510 delta = ddpa->cloneusedsnap - 2511 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; 2512 ASSERT3S(delta, >=, 0); 2513 ASSERT3U(ddpa->used, >=, delta); 2514 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2515 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2516 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 2517 2518 delta = ddpa->originusedsnap - 2519 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; 2520 ASSERT3S(delta, <=, 0); 2521 ASSERT3U(ddpa->used, >=, -delta); 2522 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2523 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2524 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 2525 2526 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; 2527 2528 /* log history record */ 2529 spa_history_log_internal_ds(hds, "promote", tx, ""); 2530 2531 dsl_dir_rele(odd, FTAG); 2532 promote_rele(ddpa, FTAG); 2533} 2534 2535/* 2536 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2537 * (exclusive) and last_obj (inclusive). The list will be in reverse 2538 * order (last_obj will be the list_head()). If first_obj == 0, do all 2539 * snapshots back to this dataset's origin. 2540 */ 2541static int 2542snaplist_make(dsl_pool_t *dp, 2543 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 2544{ 2545 uint64_t obj = last_obj; 2546 2547 list_create(l, sizeof (struct promotenode), 2548 offsetof(struct promotenode, link)); 2549 2550 while (obj != first_obj) { 2551 dsl_dataset_t *ds; 2552 struct promotenode *snap; 2553 int err; 2554 2555 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 2556 ASSERT(err != ENOENT); 2557 if (err != 0) 2558 return (err); 2559 2560 if (first_obj == 0) 2561 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; 2562 2563 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 2564 snap->ds = ds; 2565 list_insert_tail(l, snap); 2566 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 2567 } 2568 2569 return (0); 2570} 2571 2572static int 2573snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2574{ 2575 struct promotenode *snap; 2576 2577 *spacep = 0; 2578 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2579 uint64_t used, comp, uncomp; 2580 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2581 mintxg, UINT64_MAX, &used, &comp, &uncomp); 2582 *spacep += used; 2583 } 2584 return (0); 2585} 2586 2587static void 2588snaplist_destroy(list_t *l, void *tag) 2589{ 2590 struct promotenode *snap; 2591 2592 if (l == NULL || !list_link_active(&l->list_head)) 2593 return; 2594 2595 while ((snap = list_tail(l)) != NULL) { 2596 list_remove(l, snap); 2597 dsl_dataset_rele(snap->ds, tag); 2598 kmem_free(snap, sizeof (*snap)); 2599 } 2600 list_destroy(l); 2601} 2602 2603static int 2604promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 2605{ 2606 int error; 2607 dsl_dir_t *dd; 2608 struct promotenode *snap; 2609 2610 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 2611 &ddpa->ddpa_clone); 2612 if (error != 0) 2613 return (error); 2614 dd = ddpa->ddpa_clone->ds_dir; 2615 2616 if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) || 2617 !dsl_dir_is_clone(dd)) { 2618 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2619 return (SET_ERROR(EINVAL)); 2620 } 2621 2622 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, 2623 &ddpa->shared_snaps, tag); 2624 if (error != 0) 2625 goto out; 2626 2627 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 2628 &ddpa->clone_snaps, tag); 2629 if (error != 0) 2630 goto out; 2631 2632 snap = list_head(&ddpa->shared_snaps); 2633 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); 2634 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, 2635 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, 2636 &ddpa->origin_snaps, tag); 2637 if (error != 0) 2638 goto out; 2639 2640 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { 2641 error = dsl_dataset_hold_obj(dp, 2642 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, 2643 tag, &ddpa->origin_origin); 2644 if (error != 0) 2645 goto out; 2646 } 2647out: 2648 if (error != 0) 2649 promote_rele(ddpa, tag); 2650 return (error); 2651} 2652 2653static void 2654promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 2655{ 2656 snaplist_destroy(&ddpa->shared_snaps, tag); 2657 snaplist_destroy(&ddpa->clone_snaps, tag); 2658 snaplist_destroy(&ddpa->origin_snaps, tag); 2659 if (ddpa->origin_origin != NULL) 2660 dsl_dataset_rele(ddpa->origin_origin, tag); 2661 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2662} 2663 2664/* 2665 * Promote a clone. 2666 * 2667 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 2668 * in with the name. (It must be at least MAXNAMELEN bytes long.) 2669 */ 2670int 2671dsl_dataset_promote(const char *name, char *conflsnap) 2672{ 2673 dsl_dataset_promote_arg_t ddpa = { 0 }; 2674 uint64_t numsnaps; 2675 int error; 2676 objset_t *os; 2677 2678 /* 2679 * We will modify space proportional to the number of 2680 * snapshots. Compute numsnaps. 2681 */ 2682 error = dmu_objset_hold(name, FTAG, &os); 2683 if (error != 0) 2684 return (error); 2685 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 2686 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, 2687 &numsnaps); 2688 dmu_objset_rele(os, FTAG); 2689 if (error != 0) 2690 return (error); 2691 2692 ddpa.ddpa_clonename = name; 2693 ddpa.err_ds = conflsnap; 2694 ddpa.cr = CRED(); 2695 2696 return (dsl_sync_task(name, dsl_dataset_promote_check, 2697 dsl_dataset_promote_sync, &ddpa, 2698 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED)); 2699} 2700 2701int 2702dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 2703 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 2704{ 2705 int64_t unused_refres_delta; 2706 2707 /* they should both be heads */ 2708 if (dsl_dataset_is_snapshot(clone) || 2709 dsl_dataset_is_snapshot(origin_head)) 2710 return (SET_ERROR(EINVAL)); 2711 2712 /* if we are not forcing, the branch point should be just before them */ 2713 if (!force && clone->ds_prev != origin_head->ds_prev) 2714 return (SET_ERROR(EINVAL)); 2715 2716 /* clone should be the clone (unless they are unrelated) */ 2717 if (clone->ds_prev != NULL && 2718 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 2719 origin_head->ds_dir != clone->ds_prev->ds_dir) 2720 return (SET_ERROR(EINVAL)); 2721 2722 /* the clone should be a child of the origin */ 2723 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 2724 return (SET_ERROR(EINVAL)); 2725 2726 /* origin_head shouldn't be modified unless 'force' */ 2727 if (!force && 2728 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 2729 return (SET_ERROR(ETXTBSY)); 2730 2731 /* origin_head should have no long holds (e.g. is not mounted) */ 2732 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 2733 return (SET_ERROR(EBUSY)); 2734 2735 /* check amount of any unconsumed refreservation */ 2736 unused_refres_delta = 2737 (int64_t)MIN(origin_head->ds_reserved, 2738 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 2739 (int64_t)MIN(origin_head->ds_reserved, 2740 dsl_dataset_phys(clone)->ds_unique_bytes); 2741 2742 if (unused_refres_delta > 0 && 2743 unused_refres_delta > 2744 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 2745 return (SET_ERROR(ENOSPC)); 2746 2747 /* clone can't be over the head's refquota */ 2748 if (origin_head->ds_quota != 0 && 2749 dsl_dataset_phys(clone)->ds_referenced_bytes > 2750 origin_head->ds_quota) 2751 return (SET_ERROR(EDQUOT)); 2752 2753 return (0); 2754} 2755 2756void 2757dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 2758 dsl_dataset_t *origin_head, dmu_tx_t *tx) 2759{ 2760 dsl_pool_t *dp = dmu_tx_pool(tx); 2761 int64_t unused_refres_delta; 2762 2763 ASSERT(clone->ds_reserved == 0); 2764 ASSERT(origin_head->ds_quota == 0 || 2765 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota); 2766 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 2767 2768 dmu_buf_will_dirty(clone->ds_dbuf, tx); 2769 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 2770 2771 if (clone->ds_objset != NULL) { 2772 dmu_objset_evict(clone->ds_objset); 2773 clone->ds_objset = NULL; 2774 } 2775 2776 if (origin_head->ds_objset != NULL) { 2777 dmu_objset_evict(origin_head->ds_objset); 2778 origin_head->ds_objset = NULL; 2779 } 2780 2781 unused_refres_delta = 2782 (int64_t)MIN(origin_head->ds_reserved, 2783 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 2784 (int64_t)MIN(origin_head->ds_reserved, 2785 dsl_dataset_phys(clone)->ds_unique_bytes); 2786 2787 /* 2788 * Reset origin's unique bytes, if it exists. 2789 */ 2790 if (clone->ds_prev) { 2791 dsl_dataset_t *origin = clone->ds_prev; 2792 uint64_t comp, uncomp; 2793 2794 dmu_buf_will_dirty(origin->ds_dbuf, tx); 2795 dsl_deadlist_space_range(&clone->ds_deadlist, 2796 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, 2797 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); 2798 } 2799 2800 /* swap blkptrs */ 2801 { 2802 blkptr_t tmp; 2803 tmp = dsl_dataset_phys(origin_head)->ds_bp; 2804 dsl_dataset_phys(origin_head)->ds_bp = 2805 dsl_dataset_phys(clone)->ds_bp; 2806 dsl_dataset_phys(clone)->ds_bp = tmp; 2807 } 2808 2809 /* set dd_*_bytes */ 2810 { 2811 int64_t dused, dcomp, duncomp; 2812 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2813 uint64_t odl_used, odl_comp, odl_uncomp; 2814 2815 ASSERT3U(dsl_dir_phys(clone->ds_dir)-> 2816 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2817 2818 dsl_deadlist_space(&clone->ds_deadlist, 2819 &cdl_used, &cdl_comp, &cdl_uncomp); 2820 dsl_deadlist_space(&origin_head->ds_deadlist, 2821 &odl_used, &odl_comp, &odl_uncomp); 2822 2823 dused = dsl_dataset_phys(clone)->ds_referenced_bytes + 2824 cdl_used - 2825 (dsl_dataset_phys(origin_head)->ds_referenced_bytes + 2826 odl_used); 2827 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + 2828 cdl_comp - 2829 (dsl_dataset_phys(origin_head)->ds_compressed_bytes + 2830 odl_comp); 2831 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + 2832 cdl_uncomp - 2833 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + 2834 odl_uncomp); 2835 2836 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 2837 dused, dcomp, duncomp, tx); 2838 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 2839 -dused, -dcomp, -duncomp, tx); 2840 2841 /* 2842 * The difference in the space used by snapshots is the 2843 * difference in snapshot space due to the head's 2844 * deadlist (since that's the only thing that's 2845 * changing that affects the snapused). 2846 */ 2847 dsl_deadlist_space_range(&clone->ds_deadlist, 2848 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2849 &cdl_used, &cdl_comp, &cdl_uncomp); 2850 dsl_deadlist_space_range(&origin_head->ds_deadlist, 2851 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2852 &odl_used, &odl_comp, &odl_uncomp); 2853 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 2854 DD_USED_HEAD, DD_USED_SNAP, NULL); 2855 } 2856 2857 /* swap ds_*_bytes */ 2858 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, 2859 dsl_dataset_phys(clone)->ds_referenced_bytes); 2860 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, 2861 dsl_dataset_phys(clone)->ds_compressed_bytes); 2862 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, 2863 dsl_dataset_phys(clone)->ds_uncompressed_bytes); 2864 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, 2865 dsl_dataset_phys(clone)->ds_unique_bytes); 2866 2867 /* apply any parent delta for change in unconsumed refreservation */ 2868 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 2869 unused_refres_delta, 0, 0, tx); 2870 2871 /* 2872 * Swap deadlists. 2873 */ 2874 dsl_deadlist_close(&clone->ds_deadlist); 2875 dsl_deadlist_close(&origin_head->ds_deadlist); 2876 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, 2877 dsl_dataset_phys(clone)->ds_deadlist_obj); 2878 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 2879 dsl_dataset_phys(clone)->ds_deadlist_obj); 2880 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 2881 dsl_dataset_phys(origin_head)->ds_deadlist_obj); 2882 2883 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 2884 2885 spa_history_log_internal_ds(clone, "clone swap", tx, 2886 "parent=%s", origin_head->ds_dir->dd_myname); 2887} 2888 2889/* 2890 * Given a pool name and a dataset object number in that pool, 2891 * return the name of that dataset. 2892 */ 2893int 2894dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 2895{ 2896 dsl_pool_t *dp; 2897 dsl_dataset_t *ds; 2898 int error; 2899 2900 error = dsl_pool_hold(pname, FTAG, &dp); 2901 if (error != 0) 2902 return (error); 2903 2904 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 2905 if (error == 0) { 2906 dsl_dataset_name(ds, buf); 2907 dsl_dataset_rele(ds, FTAG); 2908 } 2909 dsl_pool_rele(dp, FTAG); 2910 2911 return (error); 2912} 2913 2914int 2915dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 2916 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 2917{ 2918 int error = 0; 2919 2920 ASSERT3S(asize, >, 0); 2921 2922 /* 2923 * *ref_rsrv is the portion of asize that will come from any 2924 * unconsumed refreservation space. 2925 */ 2926 *ref_rsrv = 0; 2927 2928 mutex_enter(&ds->ds_lock); 2929 /* 2930 * Make a space adjustment for reserved bytes. 2931 */ 2932 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 2933 ASSERT3U(*used, >=, 2934 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 2935 *used -= 2936 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 2937 *ref_rsrv = 2938 asize - MIN(asize, parent_delta(ds, asize + inflight)); 2939 } 2940 2941 if (!check_quota || ds->ds_quota == 0) { 2942 mutex_exit(&ds->ds_lock); 2943 return (0); 2944 } 2945 /* 2946 * If they are requesting more space, and our current estimate 2947 * is over quota, they get to try again unless the actual 2948 * on-disk is over quota and there are no pending changes (which 2949 * may free up space for us). 2950 */ 2951 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= 2952 ds->ds_quota) { 2953 if (inflight > 0 || 2954 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) 2955 error = SET_ERROR(ERESTART); 2956 else 2957 error = SET_ERROR(EDQUOT); 2958 } 2959 mutex_exit(&ds->ds_lock); 2960 2961 return (error); 2962} 2963 2964typedef struct dsl_dataset_set_qr_arg { 2965 const char *ddsqra_name; 2966 zprop_source_t ddsqra_source; 2967 uint64_t ddsqra_value; 2968} dsl_dataset_set_qr_arg_t; 2969 2970 2971/* ARGSUSED */ 2972static int 2973dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 2974{ 2975 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2976 dsl_pool_t *dp = dmu_tx_pool(tx); 2977 dsl_dataset_t *ds; 2978 int error; 2979 uint64_t newval; 2980 2981 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 2982 return (SET_ERROR(ENOTSUP)); 2983 2984 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 2985 if (error != 0) 2986 return (error); 2987 2988 if (dsl_dataset_is_snapshot(ds)) { 2989 dsl_dataset_rele(ds, FTAG); 2990 return (SET_ERROR(EINVAL)); 2991 } 2992 2993 error = dsl_prop_predict(ds->ds_dir, 2994 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 2995 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 2996 if (error != 0) { 2997 dsl_dataset_rele(ds, FTAG); 2998 return (error); 2999 } 3000 3001 if (newval == 0) { 3002 dsl_dataset_rele(ds, FTAG); 3003 return (0); 3004 } 3005 3006 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || 3007 newval < ds->ds_reserved) { 3008 dsl_dataset_rele(ds, FTAG); 3009 return (SET_ERROR(ENOSPC)); 3010 } 3011 3012 dsl_dataset_rele(ds, FTAG); 3013 return (0); 3014} 3015 3016static void 3017dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 3018{ 3019 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3020 dsl_pool_t *dp = dmu_tx_pool(tx); 3021 dsl_dataset_t *ds; 3022 uint64_t newval; 3023 3024 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3025 3026 dsl_prop_set_sync_impl(ds, 3027 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3028 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 3029 &ddsqra->ddsqra_value, tx); 3030 3031 VERIFY0(dsl_prop_get_int_ds(ds, 3032 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 3033 3034 if (ds->ds_quota != newval) { 3035 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3036 ds->ds_quota = newval; 3037 } 3038 dsl_dataset_rele(ds, FTAG); 3039} 3040 3041int 3042dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 3043 uint64_t refquota) 3044{ 3045 dsl_dataset_set_qr_arg_t ddsqra; 3046 3047 ddsqra.ddsqra_name = dsname; 3048 ddsqra.ddsqra_source = source; 3049 ddsqra.ddsqra_value = refquota; 3050 3051 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 3052 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 3053} 3054 3055static int 3056dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 3057{ 3058 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3059 dsl_pool_t *dp = dmu_tx_pool(tx); 3060 dsl_dataset_t *ds; 3061 int error; 3062 uint64_t newval, unique; 3063 3064 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 3065 return (SET_ERROR(ENOTSUP)); 3066 3067 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3068 if (error != 0) 3069 return (error); 3070 3071 if (dsl_dataset_is_snapshot(ds)) { 3072 dsl_dataset_rele(ds, FTAG); 3073 return (SET_ERROR(EINVAL)); 3074 } 3075 3076 error = dsl_prop_predict(ds->ds_dir, 3077 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3078 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3079 if (error != 0) { 3080 dsl_dataset_rele(ds, FTAG); 3081 return (error); 3082 } 3083 3084 /* 3085 * If we are doing the preliminary check in open context, the 3086 * space estimates may be inaccurate. 3087 */ 3088 if (!dmu_tx_is_syncing(tx)) { 3089 dsl_dataset_rele(ds, FTAG); 3090 return (0); 3091 } 3092 3093 mutex_enter(&ds->ds_lock); 3094 if (!DS_UNIQUE_IS_ACCURATE(ds)) 3095 dsl_dataset_recalc_head_uniq(ds); 3096 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3097 mutex_exit(&ds->ds_lock); 3098 3099 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 3100 uint64_t delta = MAX(unique, newval) - 3101 MAX(unique, ds->ds_reserved); 3102 3103 if (delta > 3104 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 3105 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 3106 dsl_dataset_rele(ds, FTAG); 3107 return (SET_ERROR(ENOSPC)); 3108 } 3109 } 3110 3111 dsl_dataset_rele(ds, FTAG); 3112 return (0); 3113} 3114 3115void 3116dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 3117 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 3118{ 3119 uint64_t newval; 3120 uint64_t unique; 3121 int64_t delta; 3122 3123 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3124 source, sizeof (value), 1, &value, tx); 3125 3126 VERIFY0(dsl_prop_get_int_ds(ds, 3127 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 3128 3129 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3130 mutex_enter(&ds->ds_dir->dd_lock); 3131 mutex_enter(&ds->ds_lock); 3132 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3133 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3134 delta = MAX(0, (int64_t)(newval - unique)) - 3135 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3136 ds->ds_reserved = newval; 3137 mutex_exit(&ds->ds_lock); 3138 3139 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3140 mutex_exit(&ds->ds_dir->dd_lock); 3141} 3142 3143static void 3144dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 3145{ 3146 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3147 dsl_pool_t *dp = dmu_tx_pool(tx); 3148 dsl_dataset_t *ds; 3149 3150 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3151 dsl_dataset_set_refreservation_sync_impl(ds, 3152 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 3153 dsl_dataset_rele(ds, FTAG); 3154} 3155 3156int 3157dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 3158 uint64_t refreservation) 3159{ 3160 dsl_dataset_set_qr_arg_t ddsqra; 3161 3162 ddsqra.ddsqra_name = dsname; 3163 ddsqra.ddsqra_source = source; 3164 ddsqra.ddsqra_value = refreservation; 3165 3166 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 3167 dsl_dataset_set_refreservation_sync, &ddsqra, 3168 0, ZFS_SPACE_CHECK_NONE)); 3169} 3170 3171/* 3172 * Return (in *usedp) the amount of space written in new that is not 3173 * present in oldsnap. New may be a snapshot or the head. Old must be 3174 * a snapshot before new, in new's filesystem (or its origin). If not then 3175 * fail and return EINVAL. 3176 * 3177 * The written space is calculated by considering two components: First, we 3178 * ignore any freed space, and calculate the written as new's used space 3179 * minus old's used space. Next, we add in the amount of space that was freed 3180 * between the two snapshots, thus reducing new's used space relative to old's. 3181 * Specifically, this is the space that was born before old->ds_creation_txg, 3182 * and freed before new (ie. on new's deadlist or a previous deadlist). 3183 * 3184 * space freed [---------------------] 3185 * snapshots ---O-------O--------O-------O------ 3186 * oldsnap new 3187 */ 3188int 3189dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 3190 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3191{ 3192 int err = 0; 3193 uint64_t snapobj; 3194 dsl_pool_t *dp = new->ds_dir->dd_pool; 3195 3196 ASSERT(dsl_pool_config_held(dp)); 3197 3198 *usedp = 0; 3199 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; 3200 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes; 3201 3202 *compp = 0; 3203 *compp += dsl_dataset_phys(new)->ds_compressed_bytes; 3204 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes; 3205 3206 *uncompp = 0; 3207 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; 3208 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes; 3209 3210 snapobj = new->ds_object; 3211 while (snapobj != oldsnap->ds_object) { 3212 dsl_dataset_t *snap; 3213 uint64_t used, comp, uncomp; 3214 3215 if (snapobj == new->ds_object) { 3216 snap = new; 3217 } else { 3218 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3219 if (err != 0) 3220 break; 3221 } 3222 3223 if (dsl_dataset_phys(snap)->ds_prev_snap_txg == 3224 dsl_dataset_phys(oldsnap)->ds_creation_txg) { 3225 /* 3226 * The blocks in the deadlist can not be born after 3227 * ds_prev_snap_txg, so get the whole deadlist space, 3228 * which is more efficient (especially for old-format 3229 * deadlists). Unfortunately the deadlist code 3230 * doesn't have enough information to make this 3231 * optimization itself. 3232 */ 3233 dsl_deadlist_space(&snap->ds_deadlist, 3234 &used, &comp, &uncomp); 3235 } else { 3236 dsl_deadlist_space_range(&snap->ds_deadlist, 3237 0, dsl_dataset_phys(oldsnap)->ds_creation_txg, 3238 &used, &comp, &uncomp); 3239 } 3240 *usedp += used; 3241 *compp += comp; 3242 *uncompp += uncomp; 3243 3244 /* 3245 * If we get to the beginning of the chain of snapshots 3246 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3247 * was not a snapshot of/before new. 3248 */ 3249 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 3250 if (snap != new) 3251 dsl_dataset_rele(snap, FTAG); 3252 if (snapobj == 0) { 3253 err = SET_ERROR(EINVAL); 3254 break; 3255 } 3256 3257 } 3258 return (err); 3259} 3260 3261/* 3262 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3263 * lastsnap, and all snapshots in between are deleted. 3264 * 3265 * blocks that would be freed [---------------------------] 3266 * snapshots ---O-------O--------O-------O--------O 3267 * firstsnap lastsnap 3268 * 3269 * This is the set of blocks that were born after the snap before firstsnap, 3270 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3271 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3272 * We calculate this by iterating over the relevant deadlists (from the snap 3273 * after lastsnap, backward to the snap after firstsnap), summing up the 3274 * space on the deadlist that was born after the snap before firstsnap. 3275 */ 3276int 3277dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3278 dsl_dataset_t *lastsnap, 3279 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3280{ 3281 int err = 0; 3282 uint64_t snapobj; 3283 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3284 3285 ASSERT(dsl_dataset_is_snapshot(firstsnap)); 3286 ASSERT(dsl_dataset_is_snapshot(lastsnap)); 3287 3288 /* 3289 * Check that the snapshots are in the same dsl_dir, and firstsnap 3290 * is before lastsnap. 3291 */ 3292 if (firstsnap->ds_dir != lastsnap->ds_dir || 3293 dsl_dataset_phys(firstsnap)->ds_creation_txg > 3294 dsl_dataset_phys(lastsnap)->ds_creation_txg) 3295 return (SET_ERROR(EINVAL)); 3296 3297 *usedp = *compp = *uncompp = 0; 3298 3299 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; 3300 while (snapobj != firstsnap->ds_object) { 3301 dsl_dataset_t *ds; 3302 uint64_t used, comp, uncomp; 3303 3304 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3305 if (err != 0) 3306 break; 3307 3308 dsl_deadlist_space_range(&ds->ds_deadlist, 3309 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, 3310 &used, &comp, &uncomp); 3311 *usedp += used; 3312 *compp += comp; 3313 *uncompp += uncomp; 3314 3315 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3316 ASSERT3U(snapobj, !=, 0); 3317 dsl_dataset_rele(ds, FTAG); 3318 } 3319 return (err); 3320} 3321 3322static int 3323dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx) 3324{ 3325 const char *dsname = arg; 3326 dsl_dataset_t *ds; 3327 dsl_pool_t *dp = dmu_tx_pool(tx); 3328 int error = 0; 3329 3330 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) 3331 return (SET_ERROR(ENOTSUP)); 3332 3333 ASSERT(spa_feature_is_enabled(dp->dp_spa, 3334 SPA_FEATURE_EXTENSIBLE_DATASET)); 3335 3336 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 3337 if (error != 0) 3338 return (error); 3339 3340 if (ds->ds_large_blocks) 3341 error = EALREADY; 3342 dsl_dataset_rele(ds, FTAG); 3343 3344 return (error); 3345} 3346 3347void 3348dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx) 3349{ 3350 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 3351 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 3352 uint64_t zero = 0; 3353 3354 spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx); 3355 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 3356 3357 VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS, 3358 sizeof (zero), 1, &zero, tx)); 3359} 3360 3361static void 3362dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx) 3363{ 3364 const char *dsname = arg; 3365 dsl_dataset_t *ds; 3366 3367 VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds)); 3368 3369 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx); 3370 ASSERT(!ds->ds_large_blocks); 3371 ds->ds_large_blocks = B_TRUE; 3372 dsl_dataset_rele(ds, FTAG); 3373} 3374 3375int 3376dsl_dataset_activate_large_blocks(const char *dsname) 3377{ 3378 int error; 3379 3380 error = dsl_sync_task(dsname, 3381 dsl_dataset_activate_large_blocks_check, 3382 dsl_dataset_activate_large_blocks_sync, (void *)dsname, 3383 1, ZFS_SPACE_CHECK_RESERVED); 3384 3385 /* 3386 * EALREADY indicates that this dataset already supports large blocks. 3387 */ 3388 if (error == EALREADY) 3389 error = 0; 3390 return (error); 3391} 3392 3393/* 3394 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 3395 * For example, they could both be snapshots of the same filesystem, and 3396 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 3397 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 3398 * filesystem. Or 'earlier' could be the origin's origin. 3399 * 3400 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 3401 */ 3402boolean_t 3403dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 3404 uint64_t earlier_txg) 3405{ 3406 dsl_pool_t *dp = later->ds_dir->dd_pool; 3407 int error; 3408 boolean_t ret; 3409 3410 ASSERT(dsl_pool_config_held(dp)); 3411 ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0); 3412 3413 if (earlier_txg == 0) 3414 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; 3415 3416 if (dsl_dataset_is_snapshot(later) && 3417 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) 3418 return (B_FALSE); 3419 3420 if (later->ds_dir == earlier->ds_dir) 3421 return (B_TRUE); 3422 if (!dsl_dir_is_clone(later->ds_dir)) 3423 return (B_FALSE); 3424 3425 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object) 3426 return (B_TRUE); 3427 dsl_dataset_t *origin; 3428 error = dsl_dataset_hold_obj(dp, 3429 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); 3430 if (error != 0) 3431 return (B_FALSE); 3432 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 3433 dsl_dataset_rele(origin, FTAG); 3434 return (ret); 3435} 3436 3437 3438void 3439dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 3440{ 3441 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3442 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 3443} 3444