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