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 */ 24 25 #include <sys/dmu_objset.h> 26 #include <sys/dsl_dataset.h> 27 #include <sys/dsl_dir.h> 28 #include <sys/dsl_prop.h> 29 #include <sys/dsl_synctask.h> 30 #include <sys/dmu_traverse.h> 31 #include <sys/dmu_tx.h> 32 #include <sys/arc.h> 33 #include <sys/zio.h> 34 #include <sys/zap.h> 35 #include <sys/unique.h> 36 #include <sys/zfs_context.h> 37 #include <sys/zfs_ioctl.h> 38 #include <sys/spa.h> 39 #include <sys/zfs_znode.h> 40 #include <sys/zvol.h> 41 #include <sys/dsl_scan.h> 42 43 /* 44 * Enable/disable prefetching of dedup-ed blocks which are going to be freed. 45 */ 46 int zfs_dedup_prefetch = 1; 47 48 static char *dsl_reaper = "the grim reaper"; 49 50 static dsl_checkfunc_t dsl_dataset_destroy_begin_check; 51 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync; 52 static dsl_syncfunc_t dsl_dataset_set_reservation_sync; 53 54 #define DS_REF_MAX (1ULL << 62) 55 56 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE 57 58 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper) 59 60 61 /* 62 * Figure out how much of this delta should be propogated to the dsl_dir 63 * layer. If there's a refreservation, that space has already been 64 * partially accounted for in our ancestors. 65 */ 66 static int64_t 67 parent_delta(dsl_dataset_t *ds, int64_t delta) 68 { 69 uint64_t old_bytes, new_bytes; 70 71 if (ds->ds_reserved == 0) 72 return (delta); 73 74 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 75 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 76 77 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 78 return (new_bytes - old_bytes); 79 } 80 81 void 82 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 83 { 84 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 85 int compressed = BP_GET_PSIZE(bp); 86 int uncompressed = BP_GET_UCSIZE(bp); 87 int64_t delta; 88 89 dprintf_bp(bp, "ds=%p", ds); 90 91 ASSERT(dmu_tx_is_syncing(tx)); 92 /* It could have been compressed away to nothing */ 93 if (BP_IS_HOLE(bp)) 94 return; 95 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 96 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES); 97 if (ds == NULL) { 98 /* 99 * Account for the meta-objset space in its placeholder 100 * dsl_dir. 101 */ 102 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */ 103 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, 104 used, compressed, uncompressed, tx); 105 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); 106 return; 107 } 108 dmu_buf_will_dirty(ds->ds_dbuf, tx); 109 110 mutex_enter(&ds->ds_dir->dd_lock); 111 mutex_enter(&ds->ds_lock); 112 delta = parent_delta(ds, used); 113 ds->ds_phys->ds_used_bytes += used; 114 ds->ds_phys->ds_compressed_bytes += compressed; 115 ds->ds_phys->ds_uncompressed_bytes += uncompressed; 116 ds->ds_phys->ds_unique_bytes += used; 117 mutex_exit(&ds->ds_lock); 118 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 119 compressed, uncompressed, tx); 120 dsl_dir_transfer_space(ds->ds_dir, used - delta, 121 DD_USED_REFRSRV, DD_USED_HEAD, tx); 122 mutex_exit(&ds->ds_dir->dd_lock); 123 } 124 125 int 126 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 127 boolean_t async) 128 { 129 if (BP_IS_HOLE(bp)) 130 return (0); 131 132 ASSERT(dmu_tx_is_syncing(tx)); 133 ASSERT(bp->blk_birth <= tx->tx_txg); 134 135 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 136 int compressed = BP_GET_PSIZE(bp); 137 int uncompressed = BP_GET_UCSIZE(bp); 138 139 ASSERT(used > 0); 140 if (ds == NULL) { 141 /* 142 * Account for the meta-objset space in its placeholder 143 * dataset. 144 */ 145 dsl_free(tx->tx_pool, tx->tx_txg, bp); 146 147 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, 148 -used, -compressed, -uncompressed, tx); 149 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); 150 return (used); 151 } 152 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 153 154 ASSERT(!dsl_dataset_is_snapshot(ds)); 155 dmu_buf_will_dirty(ds->ds_dbuf, tx); 156 157 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { 158 int64_t delta; 159 160 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 161 dsl_free(tx->tx_pool, tx->tx_txg, bp); 162 163 mutex_enter(&ds->ds_dir->dd_lock); 164 mutex_enter(&ds->ds_lock); 165 ASSERT(ds->ds_phys->ds_unique_bytes >= used || 166 !DS_UNIQUE_IS_ACCURATE(ds)); 167 delta = parent_delta(ds, -used); 168 ds->ds_phys->ds_unique_bytes -= used; 169 mutex_exit(&ds->ds_lock); 170 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 171 delta, -compressed, -uncompressed, tx); 172 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 173 DD_USED_REFRSRV, DD_USED_HEAD, tx); 174 mutex_exit(&ds->ds_dir->dd_lock); 175 } else { 176 dprintf_bp(bp, "putting on dead list: %s", ""); 177 if (async) { 178 /* 179 * We are here as part of zio's write done callback, 180 * which means we're a zio interrupt thread. We can't 181 * call bplist_enqueue() now because it may block 182 * waiting for I/O. Instead, put bp on the deferred 183 * queue and let dsl_pool_sync() finish the job. 184 */ 185 bplist_enqueue_deferred(&ds->ds_deadlist, bp); 186 } else { 187 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx)); 188 } 189 ASSERT3U(ds->ds_prev->ds_object, ==, 190 ds->ds_phys->ds_prev_snap_obj); 191 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); 192 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 193 if (ds->ds_prev->ds_phys->ds_next_snap_obj == 194 ds->ds_object && bp->blk_birth > 195 ds->ds_prev->ds_phys->ds_prev_snap_txg) { 196 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 197 mutex_enter(&ds->ds_prev->ds_lock); 198 ds->ds_prev->ds_phys->ds_unique_bytes += used; 199 mutex_exit(&ds->ds_prev->ds_lock); 200 } 201 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 202 dsl_dir_transfer_space(ds->ds_dir, used, 203 DD_USED_HEAD, DD_USED_SNAP, tx); 204 } 205 } 206 mutex_enter(&ds->ds_lock); 207 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used); 208 ds->ds_phys->ds_used_bytes -= used; 209 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); 210 ds->ds_phys->ds_compressed_bytes -= compressed; 211 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); 212 ds->ds_phys->ds_uncompressed_bytes -= uncompressed; 213 mutex_exit(&ds->ds_lock); 214 215 return (used); 216 } 217 218 uint64_t 219 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 220 { 221 uint64_t trysnap = 0; 222 223 if (ds == NULL) 224 return (0); 225 /* 226 * The snapshot creation could fail, but that would cause an 227 * incorrect FALSE return, which would only result in an 228 * overestimation of the amount of space that an operation would 229 * consume, which is OK. 230 * 231 * There's also a small window where we could miss a pending 232 * snapshot, because we could set the sync task in the quiescing 233 * phase. So this should only be used as a guess. 234 */ 235 if (ds->ds_trysnap_txg > 236 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 237 trysnap = ds->ds_trysnap_txg; 238 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap)); 239 } 240 241 boolean_t 242 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp, 243 uint64_t blk_birth) 244 { 245 if (blk_birth <= dsl_dataset_prev_snap_txg(ds)) 246 return (B_FALSE); 247 248 if (zfs_dedup_prefetch && bp && BP_GET_DEDUP(bp)) 249 ddt_prefetch(dsl_dataset_get_spa(ds), bp); 250 251 return (B_TRUE); 252 } 253 254 /* ARGSUSED */ 255 static void 256 dsl_dataset_evict(dmu_buf_t *db, void *dsv) 257 { 258 dsl_dataset_t *ds = dsv; 259 260 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds)); 261 262 unique_remove(ds->ds_fsid_guid); 263 264 if (ds->ds_objset != NULL) 265 dmu_objset_evict(ds->ds_objset); 266 267 if (ds->ds_prev) { 268 dsl_dataset_drop_ref(ds->ds_prev, ds); 269 ds->ds_prev = NULL; 270 } 271 272 bplist_close(&ds->ds_deadlist); 273 if (ds->ds_dir) 274 dsl_dir_close(ds->ds_dir, ds); 275 276 ASSERT(!list_link_active(&ds->ds_synced_link)); 277 278 mutex_destroy(&ds->ds_lock); 279 mutex_destroy(&ds->ds_recvlock); 280 mutex_destroy(&ds->ds_opening_lock); 281 rw_destroy(&ds->ds_rwlock); 282 cv_destroy(&ds->ds_exclusive_cv); 283 bplist_fini(&ds->ds_deadlist); 284 285 kmem_free(ds, sizeof (dsl_dataset_t)); 286 } 287 288 static int 289 dsl_dataset_get_snapname(dsl_dataset_t *ds) 290 { 291 dsl_dataset_phys_t *headphys; 292 int err; 293 dmu_buf_t *headdbuf; 294 dsl_pool_t *dp = ds->ds_dir->dd_pool; 295 objset_t *mos = dp->dp_meta_objset; 296 297 if (ds->ds_snapname[0]) 298 return (0); 299 if (ds->ds_phys->ds_next_snap_obj == 0) 300 return (0); 301 302 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, 303 FTAG, &headdbuf); 304 if (err) 305 return (err); 306 headphys = headdbuf->db_data; 307 err = zap_value_search(dp->dp_meta_objset, 308 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 309 dmu_buf_rele(headdbuf, FTAG); 310 return (err); 311 } 312 313 static int 314 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 315 { 316 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 317 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 318 matchtype_t mt; 319 int err; 320 321 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 322 mt = MT_FIRST; 323 else 324 mt = MT_EXACT; 325 326 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 327 value, mt, NULL, 0, NULL); 328 if (err == ENOTSUP && mt == MT_FIRST) 329 err = zap_lookup(mos, snapobj, name, 8, 1, value); 330 return (err); 331 } 332 333 static int 334 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx) 335 { 336 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 337 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 338 matchtype_t mt; 339 int err; 340 341 dsl_dir_snap_cmtime_update(ds->ds_dir); 342 343 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 344 mt = MT_FIRST; 345 else 346 mt = MT_EXACT; 347 348 err = zap_remove_norm(mos, snapobj, name, mt, tx); 349 if (err == ENOTSUP && mt == MT_FIRST) 350 err = zap_remove(mos, snapobj, name, tx); 351 return (err); 352 } 353 354 static int 355 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, 356 dsl_dataset_t **dsp) 357 { 358 objset_t *mos = dp->dp_meta_objset; 359 dmu_buf_t *dbuf; 360 dsl_dataset_t *ds; 361 int err; 362 363 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 364 dsl_pool_sync_context(dp)); 365 366 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 367 if (err) 368 return (err); 369 ds = dmu_buf_get_user(dbuf); 370 if (ds == NULL) { 371 dsl_dataset_t *winner; 372 373 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 374 ds->ds_dbuf = dbuf; 375 ds->ds_object = dsobj; 376 ds->ds_phys = dbuf->db_data; 377 378 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 379 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL); 380 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 381 rw_init(&ds->ds_rwlock, 0, 0, 0); 382 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL); 383 bplist_init(&ds->ds_deadlist); 384 385 err = bplist_open(&ds->ds_deadlist, 386 mos, ds->ds_phys->ds_deadlist_obj); 387 if (err == 0) { 388 err = dsl_dir_open_obj(dp, 389 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); 390 } 391 if (err) { 392 /* 393 * we don't really need to close the blist if we 394 * just opened it. 395 */ 396 mutex_destroy(&ds->ds_lock); 397 mutex_destroy(&ds->ds_recvlock); 398 mutex_destroy(&ds->ds_opening_lock); 399 rw_destroy(&ds->ds_rwlock); 400 cv_destroy(&ds->ds_exclusive_cv); 401 bplist_fini(&ds->ds_deadlist); 402 kmem_free(ds, sizeof (dsl_dataset_t)); 403 dmu_buf_rele(dbuf, tag); 404 return (err); 405 } 406 407 if (!dsl_dataset_is_snapshot(ds)) { 408 ds->ds_snapname[0] = '\0'; 409 if (ds->ds_phys->ds_prev_snap_obj) { 410 err = dsl_dataset_get_ref(dp, 411 ds->ds_phys->ds_prev_snap_obj, 412 ds, &ds->ds_prev); 413 } 414 } else { 415 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 416 err = dsl_dataset_get_snapname(ds); 417 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) { 418 err = zap_count( 419 ds->ds_dir->dd_pool->dp_meta_objset, 420 ds->ds_phys->ds_userrefs_obj, 421 &ds->ds_userrefs); 422 } 423 } 424 425 if (err == 0 && !dsl_dataset_is_snapshot(ds)) { 426 /* 427 * In sync context, we're called with either no lock 428 * or with the write lock. If we're not syncing, 429 * we're always called with the read lock held. 430 */ 431 boolean_t need_lock = 432 !RW_WRITE_HELD(&dp->dp_config_rwlock) && 433 dsl_pool_sync_context(dp); 434 435 if (need_lock) 436 rw_enter(&dp->dp_config_rwlock, RW_READER); 437 438 err = dsl_prop_get_ds(ds, 439 "refreservation", sizeof (uint64_t), 1, 440 &ds->ds_reserved, NULL); 441 if (err == 0) { 442 err = dsl_prop_get_ds(ds, 443 "refquota", sizeof (uint64_t), 1, 444 &ds->ds_quota, NULL); 445 } 446 447 if (need_lock) 448 rw_exit(&dp->dp_config_rwlock); 449 } else { 450 ds->ds_reserved = ds->ds_quota = 0; 451 } 452 453 if (err == 0) { 454 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys, 455 dsl_dataset_evict); 456 } 457 if (err || winner) { 458 bplist_close(&ds->ds_deadlist); 459 if (ds->ds_prev) 460 dsl_dataset_drop_ref(ds->ds_prev, ds); 461 dsl_dir_close(ds->ds_dir, ds); 462 mutex_destroy(&ds->ds_lock); 463 mutex_destroy(&ds->ds_recvlock); 464 mutex_destroy(&ds->ds_opening_lock); 465 rw_destroy(&ds->ds_rwlock); 466 cv_destroy(&ds->ds_exclusive_cv); 467 bplist_fini(&ds->ds_deadlist); 468 kmem_free(ds, sizeof (dsl_dataset_t)); 469 if (err) { 470 dmu_buf_rele(dbuf, tag); 471 return (err); 472 } 473 ds = winner; 474 } else { 475 ds->ds_fsid_guid = 476 unique_insert(ds->ds_phys->ds_fsid_guid); 477 } 478 } 479 ASSERT3P(ds->ds_dbuf, ==, dbuf); 480 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); 481 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 || 482 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 483 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 484 mutex_enter(&ds->ds_lock); 485 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) { 486 mutex_exit(&ds->ds_lock); 487 dmu_buf_rele(ds->ds_dbuf, tag); 488 return (ENOENT); 489 } 490 mutex_exit(&ds->ds_lock); 491 *dsp = ds; 492 return (0); 493 } 494 495 static int 496 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag) 497 { 498 dsl_pool_t *dp = ds->ds_dir->dd_pool; 499 500 /* 501 * In syncing context we don't want the rwlock lock: there 502 * may be an existing writer waiting for sync phase to 503 * finish. We don't need to worry about such writers, since 504 * sync phase is single-threaded, so the writer can't be 505 * doing anything while we are active. 506 */ 507 if (dsl_pool_sync_context(dp)) { 508 ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); 509 return (0); 510 } 511 512 /* 513 * Normal users will hold the ds_rwlock as a READER until they 514 * are finished (i.e., call dsl_dataset_rele()). "Owners" will 515 * drop their READER lock after they set the ds_owner field. 516 * 517 * If the dataset is being destroyed, the destroy thread will 518 * obtain a WRITER lock for exclusive access after it's done its 519 * open-context work and then change the ds_owner to 520 * dsl_reaper once destruction is assured. So threads 521 * may block here temporarily, until the "destructability" of 522 * the dataset is determined. 523 */ 524 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock)); 525 mutex_enter(&ds->ds_lock); 526 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) { 527 rw_exit(&dp->dp_config_rwlock); 528 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock); 529 if (DSL_DATASET_IS_DESTROYED(ds)) { 530 mutex_exit(&ds->ds_lock); 531 dsl_dataset_drop_ref(ds, tag); 532 rw_enter(&dp->dp_config_rwlock, RW_READER); 533 return (ENOENT); 534 } 535 /* 536 * The dp_config_rwlock lives above the ds_lock. And 537 * we need to check DSL_DATASET_IS_DESTROYED() while 538 * holding the ds_lock, so we have to drop and reacquire 539 * the ds_lock here. 540 */ 541 mutex_exit(&ds->ds_lock); 542 rw_enter(&dp->dp_config_rwlock, RW_READER); 543 mutex_enter(&ds->ds_lock); 544 } 545 mutex_exit(&ds->ds_lock); 546 return (0); 547 } 548 549 int 550 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 551 dsl_dataset_t **dsp) 552 { 553 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp); 554 555 if (err) 556 return (err); 557 return (dsl_dataset_hold_ref(*dsp, tag)); 558 } 559 560 int 561 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok, 562 void *tag, dsl_dataset_t **dsp) 563 { 564 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 565 if (err) 566 return (err); 567 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { 568 dsl_dataset_rele(*dsp, tag); 569 *dsp = NULL; 570 return (EBUSY); 571 } 572 return (0); 573 } 574 575 int 576 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp) 577 { 578 dsl_dir_t *dd; 579 dsl_pool_t *dp; 580 const char *snapname; 581 uint64_t obj; 582 int err = 0; 583 584 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname); 585 if (err) 586 return (err); 587 588 dp = dd->dd_pool; 589 obj = dd->dd_phys->dd_head_dataset_obj; 590 rw_enter(&dp->dp_config_rwlock, RW_READER); 591 if (obj) 592 err = dsl_dataset_get_ref(dp, obj, tag, dsp); 593 else 594 err = ENOENT; 595 if (err) 596 goto out; 597 598 err = dsl_dataset_hold_ref(*dsp, tag); 599 600 /* we may be looking for a snapshot */ 601 if (err == 0 && snapname != NULL) { 602 dsl_dataset_t *ds = NULL; 603 604 if (*snapname++ != '@') { 605 dsl_dataset_rele(*dsp, tag); 606 err = ENOENT; 607 goto out; 608 } 609 610 dprintf("looking for snapshot '%s'\n", snapname); 611 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); 612 if (err == 0) 613 err = dsl_dataset_get_ref(dp, obj, tag, &ds); 614 dsl_dataset_rele(*dsp, tag); 615 616 ASSERT3U((err == 0), ==, (ds != NULL)); 617 618 if (ds) { 619 mutex_enter(&ds->ds_lock); 620 if (ds->ds_snapname[0] == 0) 621 (void) strlcpy(ds->ds_snapname, snapname, 622 sizeof (ds->ds_snapname)); 623 mutex_exit(&ds->ds_lock); 624 err = dsl_dataset_hold_ref(ds, tag); 625 *dsp = err ? NULL : ds; 626 } 627 } 628 out: 629 rw_exit(&dp->dp_config_rwlock); 630 dsl_dir_close(dd, FTAG); 631 return (err); 632 } 633 634 int 635 dsl_dataset_own(const char *name, boolean_t inconsistentok, 636 void *tag, dsl_dataset_t **dsp) 637 { 638 int err = dsl_dataset_hold(name, tag, dsp); 639 if (err) 640 return (err); 641 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { 642 dsl_dataset_rele(*dsp, tag); 643 return (EBUSY); 644 } 645 return (0); 646 } 647 648 void 649 dsl_dataset_name(dsl_dataset_t *ds, char *name) 650 { 651 if (ds == NULL) { 652 (void) strcpy(name, "mos"); 653 } else { 654 dsl_dir_name(ds->ds_dir, name); 655 VERIFY(0 == dsl_dataset_get_snapname(ds)); 656 if (ds->ds_snapname[0]) { 657 (void) strcat(name, "@"); 658 /* 659 * We use a "recursive" mutex so that we 660 * can call dprintf_ds() with ds_lock held. 661 */ 662 if (!MUTEX_HELD(&ds->ds_lock)) { 663 mutex_enter(&ds->ds_lock); 664 (void) strcat(name, ds->ds_snapname); 665 mutex_exit(&ds->ds_lock); 666 } else { 667 (void) strcat(name, ds->ds_snapname); 668 } 669 } 670 } 671 } 672 673 static int 674 dsl_dataset_namelen(dsl_dataset_t *ds) 675 { 676 int result; 677 678 if (ds == NULL) { 679 result = 3; /* "mos" */ 680 } else { 681 result = dsl_dir_namelen(ds->ds_dir); 682 VERIFY(0 == dsl_dataset_get_snapname(ds)); 683 if (ds->ds_snapname[0]) { 684 ++result; /* adding one for the @-sign */ 685 if (!MUTEX_HELD(&ds->ds_lock)) { 686 mutex_enter(&ds->ds_lock); 687 result += strlen(ds->ds_snapname); 688 mutex_exit(&ds->ds_lock); 689 } else { 690 result += strlen(ds->ds_snapname); 691 } 692 } 693 } 694 695 return (result); 696 } 697 698 void 699 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag) 700 { 701 dmu_buf_rele(ds->ds_dbuf, tag); 702 } 703 704 void 705 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 706 { 707 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) { 708 rw_exit(&ds->ds_rwlock); 709 } 710 dsl_dataset_drop_ref(ds, tag); 711 } 712 713 void 714 dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 715 { 716 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) || 717 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL)); 718 719 mutex_enter(&ds->ds_lock); 720 ds->ds_owner = NULL; 721 if (RW_WRITE_HELD(&ds->ds_rwlock)) { 722 rw_exit(&ds->ds_rwlock); 723 cv_broadcast(&ds->ds_exclusive_cv); 724 } 725 mutex_exit(&ds->ds_lock); 726 if (ds->ds_dbuf) 727 dsl_dataset_drop_ref(ds, tag); 728 else 729 dsl_dataset_evict(ds->ds_dbuf, ds); 730 } 731 732 boolean_t 733 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag) 734 { 735 boolean_t gotit = FALSE; 736 737 mutex_enter(&ds->ds_lock); 738 if (ds->ds_owner == NULL && 739 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) { 740 ds->ds_owner = tag; 741 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) 742 rw_exit(&ds->ds_rwlock); 743 gotit = TRUE; 744 } 745 mutex_exit(&ds->ds_lock); 746 return (gotit); 747 } 748 749 void 750 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner) 751 { 752 ASSERT3P(owner, ==, ds->ds_owner); 753 if (!RW_WRITE_HELD(&ds->ds_rwlock)) 754 rw_enter(&ds->ds_rwlock, RW_WRITER); 755 } 756 757 uint64_t 758 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 759 uint64_t flags, dmu_tx_t *tx) 760 { 761 dsl_pool_t *dp = dd->dd_pool; 762 dmu_buf_t *dbuf; 763 dsl_dataset_phys_t *dsphys; 764 uint64_t dsobj; 765 objset_t *mos = dp->dp_meta_objset; 766 767 if (origin == NULL) 768 origin = dp->dp_origin_snap; 769 770 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 771 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0); 772 ASSERT(dmu_tx_is_syncing(tx)); 773 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 774 775 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 776 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 777 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 778 dmu_buf_will_dirty(dbuf, tx); 779 dsphys = dbuf->db_data; 780 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 781 dsphys->ds_dir_obj = dd->dd_object; 782 dsphys->ds_flags = flags; 783 dsphys->ds_fsid_guid = unique_create(); 784 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 785 sizeof (dsphys->ds_guid)); 786 dsphys->ds_snapnames_zapobj = 787 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 788 DMU_OT_NONE, 0, tx); 789 dsphys->ds_creation_time = gethrestime_sec(); 790 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 791 dsphys->ds_deadlist_obj = 792 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); 793 794 if (origin) { 795 dsphys->ds_prev_snap_obj = origin->ds_object; 796 dsphys->ds_prev_snap_txg = 797 origin->ds_phys->ds_creation_txg; 798 dsphys->ds_used_bytes = 799 origin->ds_phys->ds_used_bytes; 800 dsphys->ds_compressed_bytes = 801 origin->ds_phys->ds_compressed_bytes; 802 dsphys->ds_uncompressed_bytes = 803 origin->ds_phys->ds_uncompressed_bytes; 804 dsphys->ds_bp = origin->ds_phys->ds_bp; 805 dsphys->ds_flags |= origin->ds_phys->ds_flags; 806 807 dmu_buf_will_dirty(origin->ds_dbuf, tx); 808 origin->ds_phys->ds_num_children++; 809 810 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 811 if (origin->ds_phys->ds_next_clones_obj == 0) { 812 origin->ds_phys->ds_next_clones_obj = 813 zap_create(mos, 814 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 815 } 816 VERIFY(0 == zap_add_int(mos, 817 origin->ds_phys->ds_next_clones_obj, 818 dsobj, tx)); 819 } 820 821 dmu_buf_will_dirty(dd->dd_dbuf, tx); 822 dd->dd_phys->dd_origin_obj = origin->ds_object; 823 } 824 825 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 826 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 827 828 dmu_buf_rele(dbuf, FTAG); 829 830 dmu_buf_will_dirty(dd->dd_dbuf, tx); 831 dd->dd_phys->dd_head_dataset_obj = dsobj; 832 833 return (dsobj); 834 } 835 836 uint64_t 837 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 838 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 839 { 840 dsl_pool_t *dp = pdd->dd_pool; 841 uint64_t dsobj, ddobj; 842 dsl_dir_t *dd; 843 844 ASSERT(lastname[0] != '@'); 845 846 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 847 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd)); 848 849 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx); 850 851 dsl_deleg_set_create_perms(dd, tx, cr); 852 853 dsl_dir_close(dd, FTAG); 854 855 return (dsobj); 856 } 857 858 struct destroyarg { 859 dsl_sync_task_group_t *dstg; 860 char *snapname; 861 char *failed; 862 boolean_t defer; 863 }; 864 865 static int 866 dsl_snapshot_destroy_one(const char *name, void *arg) 867 { 868 struct destroyarg *da = arg; 869 dsl_dataset_t *ds; 870 int err; 871 char *dsname; 872 873 dsname = kmem_asprintf("%s@%s", name, da->snapname); 874 err = dsl_dataset_own(dsname, B_TRUE, da->dstg, &ds); 875 strfree(dsname); 876 if (err == 0) { 877 struct dsl_ds_destroyarg *dsda; 878 879 dsl_dataset_make_exclusive(ds, da->dstg); 880 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP); 881 dsda->ds = ds; 882 dsda->defer = da->defer; 883 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check, 884 dsl_dataset_destroy_sync, dsda, da->dstg, 0); 885 } else if (err == ENOENT) { 886 err = 0; 887 } else { 888 (void) strcpy(da->failed, name); 889 } 890 return (err); 891 } 892 893 /* 894 * Destroy 'snapname' in all descendants of 'fsname'. 895 */ 896 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy 897 int 898 dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer) 899 { 900 int err; 901 struct destroyarg da; 902 dsl_sync_task_t *dst; 903 spa_t *spa; 904 905 err = spa_open(fsname, &spa, FTAG); 906 if (err) 907 return (err); 908 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 909 da.snapname = snapname; 910 da.failed = fsname; 911 da.defer = defer; 912 913 err = dmu_objset_find(fsname, 914 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN); 915 916 if (err == 0) 917 err = dsl_sync_task_group_wait(da.dstg); 918 919 for (dst = list_head(&da.dstg->dstg_tasks); dst; 920 dst = list_next(&da.dstg->dstg_tasks, dst)) { 921 struct dsl_ds_destroyarg *dsda = dst->dst_arg1; 922 dsl_dataset_t *ds = dsda->ds; 923 924 /* 925 * Return the file system name that triggered the error 926 */ 927 if (dst->dst_err) { 928 dsl_dataset_name(ds, fsname); 929 *strchr(fsname, '@') = '\0'; 930 } 931 ASSERT3P(dsda->rm_origin, ==, NULL); 932 dsl_dataset_disown(ds, da.dstg); 933 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg)); 934 } 935 936 dsl_sync_task_group_destroy(da.dstg); 937 spa_close(spa, FTAG); 938 return (err); 939 } 940 941 static boolean_t 942 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds) 943 { 944 boolean_t might_destroy = B_FALSE; 945 946 mutex_enter(&ds->ds_lock); 947 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 && 948 DS_IS_DEFER_DESTROY(ds)) 949 might_destroy = B_TRUE; 950 mutex_exit(&ds->ds_lock); 951 952 return (might_destroy); 953 } 954 955 /* 956 * If we're removing a clone, and these three conditions are true: 957 * 1) the clone's origin has no other children 958 * 2) the clone's origin has no user references 959 * 3) the clone's origin has been marked for deferred destruction 960 * Then, prepare to remove the origin as part of this sync task group. 961 */ 962 static int 963 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag) 964 { 965 dsl_dataset_t *ds = dsda->ds; 966 dsl_dataset_t *origin = ds->ds_prev; 967 968 if (dsl_dataset_might_destroy_origin(origin)) { 969 char *name; 970 int namelen; 971 int error; 972 973 namelen = dsl_dataset_namelen(origin) + 1; 974 name = kmem_alloc(namelen, KM_SLEEP); 975 dsl_dataset_name(origin, name); 976 #ifdef _KERNEL 977 error = zfs_unmount_snap(name, NULL); 978 if (error) { 979 kmem_free(name, namelen); 980 return (error); 981 } 982 #endif 983 error = dsl_dataset_own(name, B_TRUE, tag, &origin); 984 kmem_free(name, namelen); 985 if (error) 986 return (error); 987 dsda->rm_origin = origin; 988 dsl_dataset_make_exclusive(origin, tag); 989 } 990 991 return (0); 992 } 993 994 /* 995 * ds must be opened as OWNER. On return (whether successful or not), 996 * ds will be closed and caller can no longer dereference it. 997 */ 998 int 999 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer) 1000 { 1001 int err; 1002 dsl_sync_task_group_t *dstg; 1003 objset_t *os; 1004 dsl_dir_t *dd; 1005 uint64_t obj; 1006 struct dsl_ds_destroyarg dsda = { 0 }; 1007 dsl_dataset_t dummy_ds = { 0 }; 1008 1009 dsda.ds = ds; 1010 1011 if (dsl_dataset_is_snapshot(ds)) { 1012 /* Destroying a snapshot is simpler */ 1013 dsl_dataset_make_exclusive(ds, tag); 1014 1015 dsda.defer = defer; 1016 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 1017 dsl_dataset_destroy_check, dsl_dataset_destroy_sync, 1018 &dsda, tag, 0); 1019 ASSERT3P(dsda.rm_origin, ==, NULL); 1020 goto out; 1021 } else if (defer) { 1022 err = EINVAL; 1023 goto out; 1024 } 1025 1026 dd = ds->ds_dir; 1027 dummy_ds.ds_dir = dd; 1028 dummy_ds.ds_object = ds->ds_object; 1029 1030 /* 1031 * Check for errors and mark this ds as inconsistent, in 1032 * case we crash while freeing the objects. 1033 */ 1034 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check, 1035 dsl_dataset_destroy_begin_sync, ds, NULL, 0); 1036 if (err) 1037 goto out; 1038 1039 err = dmu_objset_from_ds(ds, &os); 1040 if (err) 1041 goto out; 1042 1043 /* 1044 * remove the objects in open context, so that we won't 1045 * have too much to do in syncing context. 1046 */ 1047 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 1048 ds->ds_phys->ds_prev_snap_txg)) { 1049 /* 1050 * Ignore errors, if there is not enough disk space 1051 * we will deal with it in dsl_dataset_destroy_sync(). 1052 */ 1053 (void) dmu_free_object(os, obj); 1054 } 1055 1056 /* 1057 * We need to sync out all in-flight IO before we try to evict 1058 * (the dataset evict func is trying to clear the cached entries 1059 * for this dataset in the ARC). 1060 */ 1061 txg_wait_synced(dd->dd_pool, 0); 1062 1063 /* 1064 * If we managed to free all the objects in open 1065 * context, the user space accounting should be zero. 1066 */ 1067 if (ds->ds_phys->ds_bp.blk_fill == 0 && 1068 dmu_objset_userused_enabled(os)) { 1069 uint64_t count; 1070 1071 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 || 1072 count == 0); 1073 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 || 1074 count == 0); 1075 } 1076 1077 if (err != ESRCH) 1078 goto out; 1079 1080 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); 1081 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd); 1082 rw_exit(&dd->dd_pool->dp_config_rwlock); 1083 1084 if (err) 1085 goto out; 1086 1087 /* 1088 * Blow away the dsl_dir + head dataset. 1089 */ 1090 dsl_dataset_make_exclusive(ds, tag); 1091 /* 1092 * If we're removing a clone, we might also need to remove its 1093 * origin. 1094 */ 1095 do { 1096 dsda.need_prep = B_FALSE; 1097 if (dsl_dir_is_clone(dd)) { 1098 err = dsl_dataset_origin_rm_prep(&dsda, tag); 1099 if (err) { 1100 dsl_dir_close(dd, FTAG); 1101 goto out; 1102 } 1103 } 1104 1105 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool); 1106 dsl_sync_task_create(dstg, dsl_dataset_destroy_check, 1107 dsl_dataset_destroy_sync, &dsda, tag, 0); 1108 dsl_sync_task_create(dstg, dsl_dir_destroy_check, 1109 dsl_dir_destroy_sync, &dummy_ds, FTAG, 0); 1110 err = dsl_sync_task_group_wait(dstg); 1111 dsl_sync_task_group_destroy(dstg); 1112 1113 /* 1114 * We could be racing against 'zfs release' or 'zfs destroy -d' 1115 * on the origin snap, in which case we can get EBUSY if we 1116 * needed to destroy the origin snap but were not ready to 1117 * do so. 1118 */ 1119 if (dsda.need_prep) { 1120 ASSERT(err == EBUSY); 1121 ASSERT(dsl_dir_is_clone(dd)); 1122 ASSERT(dsda.rm_origin == NULL); 1123 } 1124 } while (dsda.need_prep); 1125 1126 if (dsda.rm_origin != NULL) 1127 dsl_dataset_disown(dsda.rm_origin, tag); 1128 1129 /* if it is successful, dsl_dir_destroy_sync will close the dd */ 1130 if (err) 1131 dsl_dir_close(dd, FTAG); 1132 out: 1133 dsl_dataset_disown(ds, tag); 1134 return (err); 1135 } 1136 1137 blkptr_t * 1138 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1139 { 1140 return (&ds->ds_phys->ds_bp); 1141 } 1142 1143 void 1144 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 1145 { 1146 ASSERT(dmu_tx_is_syncing(tx)); 1147 /* If it's the meta-objset, set dp_meta_rootbp */ 1148 if (ds == NULL) { 1149 tx->tx_pool->dp_meta_rootbp = *bp; 1150 } else { 1151 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1152 ds->ds_phys->ds_bp = *bp; 1153 } 1154 } 1155 1156 spa_t * 1157 dsl_dataset_get_spa(dsl_dataset_t *ds) 1158 { 1159 return (ds->ds_dir->dd_pool->dp_spa); 1160 } 1161 1162 void 1163 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1164 { 1165 dsl_pool_t *dp; 1166 1167 if (ds == NULL) /* this is the meta-objset */ 1168 return; 1169 1170 ASSERT(ds->ds_objset != NULL); 1171 1172 if (ds->ds_phys->ds_next_snap_obj != 0) 1173 panic("dirtying snapshot!"); 1174 1175 dp = ds->ds_dir->dd_pool; 1176 1177 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) { 1178 /* up the hold count until we can be written out */ 1179 dmu_buf_add_ref(ds->ds_dbuf, ds); 1180 } 1181 } 1182 1183 /* 1184 * The unique space in the head dataset can be calculated by subtracting 1185 * the space used in the most recent snapshot, that is still being used 1186 * in this file system, from the space currently in use. To figure out 1187 * the space in the most recent snapshot still in use, we need to take 1188 * the total space used in the snapshot and subtract out the space that 1189 * has been freed up since the snapshot was taken. 1190 */ 1191 static void 1192 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 1193 { 1194 uint64_t mrs_used; 1195 uint64_t dlused, dlcomp, dluncomp; 1196 1197 ASSERT(!dsl_dataset_is_snapshot(ds)); 1198 1199 if (ds->ds_phys->ds_prev_snap_obj != 0) 1200 mrs_used = ds->ds_prev->ds_phys->ds_used_bytes; 1201 else 1202 mrs_used = 0; 1203 1204 VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp, 1205 &dluncomp)); 1206 1207 ASSERT3U(dlused, <=, mrs_used); 1208 ds->ds_phys->ds_unique_bytes = 1209 ds->ds_phys->ds_used_bytes - (mrs_used - dlused); 1210 1211 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1212 SPA_VERSION_UNIQUE_ACCURATE) 1213 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1214 } 1215 1216 struct killarg { 1217 dsl_dataset_t *ds; 1218 dmu_tx_t *tx; 1219 }; 1220 1221 /* ARGSUSED */ 1222 static int 1223 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf, 1224 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 1225 { 1226 struct killarg *ka = arg; 1227 dmu_tx_t *tx = ka->tx; 1228 1229 if (bp == NULL) 1230 return (0); 1231 1232 if (zb->zb_level == ZB_ZIL_LEVEL) { 1233 ASSERT(zilog != NULL); 1234 /* 1235 * It's a block in the intent log. It has no 1236 * accounting, so just free it. 1237 */ 1238 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp); 1239 } else { 1240 ASSERT(zilog == NULL); 1241 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg); 1242 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE); 1243 } 1244 1245 return (0); 1246 } 1247 1248 /* ARGSUSED */ 1249 static int 1250 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx) 1251 { 1252 dsl_dataset_t *ds = arg1; 1253 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1254 uint64_t count; 1255 int err; 1256 1257 /* 1258 * Can't delete a head dataset if there are snapshots of it. 1259 * (Except if the only snapshots are from the branch we cloned 1260 * from.) 1261 */ 1262 if (ds->ds_prev != NULL && 1263 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) 1264 return (EBUSY); 1265 1266 /* 1267 * This is really a dsl_dir thing, but check it here so that 1268 * we'll be less likely to leave this dataset inconsistent & 1269 * nearly destroyed. 1270 */ 1271 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count); 1272 if (err) 1273 return (err); 1274 if (count != 0) 1275 return (EEXIST); 1276 1277 return (0); 1278 } 1279 1280 /* ARGSUSED */ 1281 static void 1282 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx) 1283 { 1284 dsl_dataset_t *ds = arg1; 1285 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1286 1287 /* Mark it as inconsistent on-disk, in case we crash */ 1288 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1289 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; 1290 1291 spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx, 1292 "dataset = %llu", ds->ds_object); 1293 } 1294 1295 static int 1296 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag, 1297 dmu_tx_t *tx) 1298 { 1299 dsl_dataset_t *ds = dsda->ds; 1300 dsl_dataset_t *ds_prev = ds->ds_prev; 1301 1302 if (dsl_dataset_might_destroy_origin(ds_prev)) { 1303 struct dsl_ds_destroyarg ndsda = {0}; 1304 1305 /* 1306 * If we're not prepared to remove the origin, don't remove 1307 * the clone either. 1308 */ 1309 if (dsda->rm_origin == NULL) { 1310 dsda->need_prep = B_TRUE; 1311 return (EBUSY); 1312 } 1313 1314 ndsda.ds = ds_prev; 1315 ndsda.is_origin_rm = B_TRUE; 1316 return (dsl_dataset_destroy_check(&ndsda, tag, tx)); 1317 } 1318 1319 /* 1320 * If we're not going to remove the origin after all, 1321 * undo the open context setup. 1322 */ 1323 if (dsda->rm_origin != NULL) { 1324 dsl_dataset_disown(dsda->rm_origin, tag); 1325 dsda->rm_origin = NULL; 1326 } 1327 1328 return (0); 1329 } 1330 1331 /* ARGSUSED */ 1332 int 1333 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) 1334 { 1335 struct dsl_ds_destroyarg *dsda = arg1; 1336 dsl_dataset_t *ds = dsda->ds; 1337 1338 /* we have an owner hold, so noone else can destroy us */ 1339 ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); 1340 1341 /* 1342 * Only allow deferred destroy on pools that support it. 1343 * NOTE: deferred destroy is only supported on snapshots. 1344 */ 1345 if (dsda->defer) { 1346 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < 1347 SPA_VERSION_USERREFS) 1348 return (ENOTSUP); 1349 ASSERT(dsl_dataset_is_snapshot(ds)); 1350 return (0); 1351 } 1352 1353 /* 1354 * Can't delete a head dataset if there are snapshots of it. 1355 * (Except if the only snapshots are from the branch we cloned 1356 * from.) 1357 */ 1358 if (ds->ds_prev != NULL && 1359 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) 1360 return (EBUSY); 1361 1362 /* 1363 * If we made changes this txg, traverse_dsl_dataset won't find 1364 * them. Try again. 1365 */ 1366 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) 1367 return (EAGAIN); 1368 1369 if (dsl_dataset_is_snapshot(ds)) { 1370 /* 1371 * If this snapshot has an elevated user reference count, 1372 * we can't destroy it yet. 1373 */ 1374 if (ds->ds_userrefs > 0 && !dsda->releasing) 1375 return (EBUSY); 1376 1377 mutex_enter(&ds->ds_lock); 1378 /* 1379 * Can't delete a branch point. However, if we're destroying 1380 * a clone and removing its origin due to it having a user 1381 * hold count of 0 and having been marked for deferred destroy, 1382 * it's OK for the origin to have a single clone. 1383 */ 1384 if (ds->ds_phys->ds_num_children > 1385 (dsda->is_origin_rm ? 2 : 1)) { 1386 mutex_exit(&ds->ds_lock); 1387 return (EEXIST); 1388 } 1389 mutex_exit(&ds->ds_lock); 1390 } else if (dsl_dir_is_clone(ds->ds_dir)) { 1391 return (dsl_dataset_origin_check(dsda, arg2, tx)); 1392 } 1393 1394 /* XXX we should do some i/o error checking... */ 1395 return (0); 1396 } 1397 1398 struct refsarg { 1399 kmutex_t lock; 1400 boolean_t gone; 1401 kcondvar_t cv; 1402 }; 1403 1404 /* ARGSUSED */ 1405 static void 1406 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv) 1407 { 1408 struct refsarg *arg = argv; 1409 1410 mutex_enter(&arg->lock); 1411 arg->gone = TRUE; 1412 cv_signal(&arg->cv); 1413 mutex_exit(&arg->lock); 1414 } 1415 1416 static void 1417 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag) 1418 { 1419 struct refsarg arg; 1420 1421 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL); 1422 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL); 1423 arg.gone = FALSE; 1424 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys, 1425 dsl_dataset_refs_gone); 1426 dmu_buf_rele(ds->ds_dbuf, tag); 1427 mutex_enter(&arg.lock); 1428 while (!arg.gone) 1429 cv_wait(&arg.cv, &arg.lock); 1430 ASSERT(arg.gone); 1431 mutex_exit(&arg.lock); 1432 ds->ds_dbuf = NULL; 1433 ds->ds_phys = NULL; 1434 mutex_destroy(&arg.lock); 1435 cv_destroy(&arg.cv); 1436 } 1437 1438 static void 1439 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx) 1440 { 1441 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1442 uint64_t count; 1443 int err; 1444 1445 ASSERT(ds->ds_phys->ds_num_children >= 2); 1446 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx); 1447 /* 1448 * The err should not be ENOENT, but a bug in a previous version 1449 * of the code could cause upgrade_clones_cb() to not set 1450 * ds_next_snap_obj when it should, leading to a missing entry. 1451 * If we knew that the pool was created after 1452 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 1453 * ENOENT. However, at least we can check that we don't have 1454 * too many entries in the next_clones_obj even after failing to 1455 * remove this one. 1456 */ 1457 if (err != ENOENT) { 1458 VERIFY3U(err, ==, 0); 1459 } 1460 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj, 1461 &count)); 1462 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2); 1463 } 1464 1465 void 1466 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) 1467 { 1468 struct dsl_ds_destroyarg *dsda = arg1; 1469 dsl_dataset_t *ds = dsda->ds; 1470 int err; 1471 int after_branch_point = FALSE; 1472 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1473 objset_t *mos = dp->dp_meta_objset; 1474 dsl_dataset_t *ds_prev = NULL; 1475 uint64_t obj; 1476 1477 ASSERT(ds->ds_owner); 1478 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1); 1479 ASSERT(ds->ds_prev == NULL || 1480 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object); 1481 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); 1482 1483 if (dsda->defer) { 1484 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 1485 if (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1) { 1486 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1487 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY; 1488 return; 1489 } 1490 } 1491 1492 /* signal any waiters that this dataset is going away */ 1493 mutex_enter(&ds->ds_lock); 1494 ds->ds_owner = dsl_reaper; 1495 cv_broadcast(&ds->ds_exclusive_cv); 1496 mutex_exit(&ds->ds_lock); 1497 1498 if (ds->ds_objset) { 1499 dmu_objset_evict(ds->ds_objset); 1500 ds->ds_objset = NULL; 1501 } 1502 1503 /* Remove our reservation */ 1504 if (ds->ds_reserved != 0) { 1505 dsl_prop_setarg_t psa; 1506 uint64_t value = 0; 1507 1508 dsl_prop_setarg_init_uint64(&psa, "refreservation", 1509 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED), 1510 &value); 1511 psa.psa_effective_value = 0; /* predict default value */ 1512 1513 dsl_dataset_set_reservation_sync(ds, &psa, tx); 1514 ASSERT3U(ds->ds_reserved, ==, 0); 1515 } 1516 1517 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); 1518 1519 dsl_scan_ds_destroyed(ds, tx); 1520 1521 obj = ds->ds_object; 1522 1523 if (ds->ds_phys->ds_prev_snap_obj != 0) { 1524 if (ds->ds_prev) { 1525 ds_prev = ds->ds_prev; 1526 } else { 1527 VERIFY(0 == dsl_dataset_hold_obj(dp, 1528 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev)); 1529 } 1530 after_branch_point = 1531 (ds_prev->ds_phys->ds_next_snap_obj != obj); 1532 1533 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); 1534 if (after_branch_point && 1535 ds_prev->ds_phys->ds_next_clones_obj != 0) { 1536 remove_from_next_clones(ds_prev, obj, tx); 1537 if (ds->ds_phys->ds_next_snap_obj != 0) { 1538 VERIFY(0 == zap_add_int(mos, 1539 ds_prev->ds_phys->ds_next_clones_obj, 1540 ds->ds_phys->ds_next_snap_obj, tx)); 1541 } 1542 } 1543 if (after_branch_point && 1544 ds->ds_phys->ds_next_snap_obj == 0) { 1545 /* This clone is toast. */ 1546 ASSERT(ds_prev->ds_phys->ds_num_children > 1); 1547 ds_prev->ds_phys->ds_num_children--; 1548 1549 /* 1550 * If the clone's origin has no other clones, no 1551 * user holds, and has been marked for deferred 1552 * deletion, then we should have done the necessary 1553 * destroy setup for it. 1554 */ 1555 if (ds_prev->ds_phys->ds_num_children == 1 && 1556 ds_prev->ds_userrefs == 0 && 1557 DS_IS_DEFER_DESTROY(ds_prev)) { 1558 ASSERT3P(dsda->rm_origin, !=, NULL); 1559 } else { 1560 ASSERT3P(dsda->rm_origin, ==, NULL); 1561 } 1562 } else if (!after_branch_point) { 1563 ds_prev->ds_phys->ds_next_snap_obj = 1564 ds->ds_phys->ds_next_snap_obj; 1565 } 1566 } 1567 1568 if (dsl_dataset_is_snapshot(ds)) { 1569 blkptr_t bp; 1570 zio_t *pio; 1571 dsl_dataset_t *ds_next; 1572 uint64_t itor = 0; 1573 uint64_t old_unique; 1574 int64_t used = 0, compressed = 0, uncompressed = 0; 1575 1576 VERIFY(0 == dsl_dataset_hold_obj(dp, 1577 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next)); 1578 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); 1579 1580 old_unique = ds_next->ds_phys->ds_unique_bytes; 1581 1582 dmu_buf_will_dirty(ds_next->ds_dbuf, tx); 1583 ds_next->ds_phys->ds_prev_snap_obj = 1584 ds->ds_phys->ds_prev_snap_obj; 1585 ds_next->ds_phys->ds_prev_snap_txg = 1586 ds->ds_phys->ds_prev_snap_txg; 1587 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1588 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); 1589 1590 pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1591 1592 /* 1593 * Transfer to our deadlist (which will become next's 1594 * new deadlist) any entries from next's current 1595 * deadlist which were born before prev, and free the 1596 * other entries. 1597 * 1598 * XXX we're doing this long task with the config lock held 1599 */ 1600 while (bplist_iterate(&ds_next->ds_deadlist, &itor, &bp) == 0) { 1601 if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) { 1602 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, 1603 &bp, tx)); 1604 if (ds_prev && !after_branch_point && 1605 bp.blk_birth > 1606 ds_prev->ds_phys->ds_prev_snap_txg) { 1607 ds_prev->ds_phys->ds_unique_bytes += 1608 bp_get_dsize_sync(dp->dp_spa, &bp); 1609 } 1610 } else { 1611 used += bp_get_dsize_sync(dp->dp_spa, &bp); 1612 compressed += BP_GET_PSIZE(&bp); 1613 uncompressed += BP_GET_UCSIZE(&bp); 1614 dsl_free_sync(pio, dp, tx->tx_txg, &bp); 1615 } 1616 } 1617 VERIFY3U(zio_wait(pio), ==, 0); 1618 ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes); 1619 1620 /* change snapused */ 1621 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, 1622 -used, -compressed, -uncompressed, tx); 1623 1624 /* free next's deadlist */ 1625 bplist_close(&ds_next->ds_deadlist); 1626 bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx); 1627 1628 /* set next's deadlist to our deadlist */ 1629 bplist_close(&ds->ds_deadlist); 1630 ds_next->ds_phys->ds_deadlist_obj = 1631 ds->ds_phys->ds_deadlist_obj; 1632 VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos, 1633 ds_next->ds_phys->ds_deadlist_obj)); 1634 ds->ds_phys->ds_deadlist_obj = 0; 1635 1636 if (dsl_dataset_is_snapshot(ds_next)) { 1637 /* 1638 * Update next's unique to include blocks which 1639 * were previously shared by only this snapshot 1640 * and it. Those blocks will be born after the 1641 * prev snap and before this snap, and will have 1642 * died after the next snap and before the one 1643 * after that (ie. be on the snap after next's 1644 * deadlist). 1645 * 1646 * XXX we're doing this long task with the 1647 * config lock held 1648 */ 1649 dsl_dataset_t *ds_after_next; 1650 uint64_t space; 1651 1652 VERIFY(0 == dsl_dataset_hold_obj(dp, 1653 ds_next->ds_phys->ds_next_snap_obj, 1654 FTAG, &ds_after_next)); 1655 1656 VERIFY(0 == 1657 bplist_space_birthrange(&ds_after_next->ds_deadlist, 1658 ds->ds_phys->ds_prev_snap_txg, 1659 ds->ds_phys->ds_creation_txg, &space)); 1660 ds_next->ds_phys->ds_unique_bytes += space; 1661 1662 dsl_dataset_rele(ds_after_next, FTAG); 1663 ASSERT3P(ds_next->ds_prev, ==, NULL); 1664 } else { 1665 ASSERT3P(ds_next->ds_prev, ==, ds); 1666 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next); 1667 ds_next->ds_prev = NULL; 1668 if (ds_prev) { 1669 VERIFY(0 == dsl_dataset_get_ref(dp, 1670 ds->ds_phys->ds_prev_snap_obj, 1671 ds_next, &ds_next->ds_prev)); 1672 } 1673 1674 dsl_dataset_recalc_head_uniq(ds_next); 1675 1676 /* 1677 * Reduce the amount of our unconsmed refreservation 1678 * being charged to our parent by the amount of 1679 * new unique data we have gained. 1680 */ 1681 if (old_unique < ds_next->ds_reserved) { 1682 int64_t mrsdelta; 1683 uint64_t new_unique = 1684 ds_next->ds_phys->ds_unique_bytes; 1685 1686 ASSERT(old_unique <= new_unique); 1687 mrsdelta = MIN(new_unique - old_unique, 1688 ds_next->ds_reserved - old_unique); 1689 dsl_dir_diduse_space(ds->ds_dir, 1690 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx); 1691 } 1692 } 1693 dsl_dataset_rele(ds_next, FTAG); 1694 } else { 1695 /* 1696 * There's no next snapshot, so this is a head dataset. 1697 * Destroy the deadlist. Unless it's a clone, the 1698 * deadlist should be empty. (If it's a clone, it's 1699 * safe to ignore the deadlist contents.) 1700 */ 1701 struct killarg ka; 1702 1703 ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist)); 1704 bplist_close(&ds->ds_deadlist); 1705 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); 1706 ds->ds_phys->ds_deadlist_obj = 0; 1707 1708 /* 1709 * Free everything that we point to (that's born after 1710 * the previous snapshot, if we are a clone) 1711 * 1712 * NB: this should be very quick, because we already 1713 * freed all the objects in open context. 1714 */ 1715 ka.ds = ds; 1716 ka.tx = tx; 1717 err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg, 1718 TRAVERSE_POST, kill_blkptr, &ka); 1719 ASSERT3U(err, ==, 0); 1720 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || 1721 ds->ds_phys->ds_unique_bytes == 0); 1722 1723 if (ds->ds_prev != NULL) { 1724 dsl_dataset_rele(ds->ds_prev, ds); 1725 ds->ds_prev = ds_prev = NULL; 1726 } 1727 } 1728 1729 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) { 1730 /* Erase the link in the dir */ 1731 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 1732 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0; 1733 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0); 1734 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx); 1735 ASSERT(err == 0); 1736 } else { 1737 /* remove from snapshot namespace */ 1738 dsl_dataset_t *ds_head; 1739 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0); 1740 VERIFY(0 == dsl_dataset_hold_obj(dp, 1741 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head)); 1742 VERIFY(0 == dsl_dataset_get_snapname(ds)); 1743 #ifdef ZFS_DEBUG 1744 { 1745 uint64_t val; 1746 1747 err = dsl_dataset_snap_lookup(ds_head, 1748 ds->ds_snapname, &val); 1749 ASSERT3U(err, ==, 0); 1750 ASSERT3U(val, ==, obj); 1751 } 1752 #endif 1753 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx); 1754 ASSERT(err == 0); 1755 dsl_dataset_rele(ds_head, FTAG); 1756 } 1757 1758 if (ds_prev && ds->ds_prev != ds_prev) 1759 dsl_dataset_rele(ds_prev, FTAG); 1760 1761 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx); 1762 spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx, 1763 "dataset = %llu", ds->ds_object); 1764 1765 if (ds->ds_phys->ds_next_clones_obj != 0) { 1766 uint64_t count; 1767 ASSERT(0 == zap_count(mos, 1768 ds->ds_phys->ds_next_clones_obj, &count) && count == 0); 1769 VERIFY(0 == dmu_object_free(mos, 1770 ds->ds_phys->ds_next_clones_obj, tx)); 1771 } 1772 if (ds->ds_phys->ds_props_obj != 0) 1773 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx)); 1774 if (ds->ds_phys->ds_userrefs_obj != 0) 1775 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx)); 1776 dsl_dir_close(ds->ds_dir, ds); 1777 ds->ds_dir = NULL; 1778 dsl_dataset_drain_refs(ds, tag); 1779 VERIFY(0 == dmu_object_free(mos, obj, tx)); 1780 1781 if (dsda->rm_origin) { 1782 /* 1783 * Remove the origin of the clone we just destroyed. 1784 */ 1785 struct dsl_ds_destroyarg ndsda = {0}; 1786 1787 ndsda.ds = dsda->rm_origin; 1788 dsl_dataset_destroy_sync(&ndsda, tag, tx); 1789 } 1790 } 1791 1792 static int 1793 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1794 { 1795 uint64_t asize; 1796 1797 if (!dmu_tx_is_syncing(tx)) 1798 return (0); 1799 1800 /* 1801 * If there's an fs-only reservation, any blocks that might become 1802 * owned by the snapshot dataset must be accommodated by space 1803 * outside of the reservation. 1804 */ 1805 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1806 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 1807 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE)) 1808 return (ENOSPC); 1809 1810 /* 1811 * Propogate any reserved space for this snapshot to other 1812 * snapshot checks in this sync group. 1813 */ 1814 if (asize > 0) 1815 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1816 1817 return (0); 1818 } 1819 1820 int 1821 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx) 1822 { 1823 dsl_dataset_t *ds = arg1; 1824 const char *snapname = arg2; 1825 int err; 1826 uint64_t value; 1827 1828 /* 1829 * We don't allow multiple snapshots of the same txg. If there 1830 * is already one, try again. 1831 */ 1832 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) 1833 return (EAGAIN); 1834 1835 /* 1836 * Check for conflicting name snapshot name. 1837 */ 1838 err = dsl_dataset_snap_lookup(ds, snapname, &value); 1839 if (err == 0) 1840 return (EEXIST); 1841 if (err != ENOENT) 1842 return (err); 1843 1844 /* 1845 * Check that the dataset's name is not too long. Name consists 1846 * of the dataset's length + 1 for the @-sign + snapshot name's length 1847 */ 1848 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN) 1849 return (ENAMETOOLONG); 1850 1851 err = dsl_dataset_snapshot_reserve_space(ds, tx); 1852 if (err) 1853 return (err); 1854 1855 ds->ds_trysnap_txg = tx->tx_txg; 1856 return (0); 1857 } 1858 1859 void 1860 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx) 1861 { 1862 dsl_dataset_t *ds = arg1; 1863 const char *snapname = arg2; 1864 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1865 dmu_buf_t *dbuf; 1866 dsl_dataset_phys_t *dsphys; 1867 uint64_t dsobj, crtxg; 1868 objset_t *mos = dp->dp_meta_objset; 1869 int err; 1870 1871 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); 1872 1873 /* 1874 * The origin's ds_creation_txg has to be < TXG_INITIAL 1875 */ 1876 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1877 crtxg = 1; 1878 else 1879 crtxg = tx->tx_txg; 1880 1881 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1882 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1883 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1884 dmu_buf_will_dirty(dbuf, tx); 1885 dsphys = dbuf->db_data; 1886 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1887 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1888 dsphys->ds_fsid_guid = unique_create(); 1889 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1890 sizeof (dsphys->ds_guid)); 1891 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; 1892 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; 1893 dsphys->ds_next_snap_obj = ds->ds_object; 1894 dsphys->ds_num_children = 1; 1895 dsphys->ds_creation_time = gethrestime_sec(); 1896 dsphys->ds_creation_txg = crtxg; 1897 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; 1898 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes; 1899 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; 1900 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; 1901 dsphys->ds_flags = ds->ds_phys->ds_flags; 1902 dsphys->ds_bp = ds->ds_phys->ds_bp; 1903 dmu_buf_rele(dbuf, FTAG); 1904 1905 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); 1906 if (ds->ds_prev) { 1907 uint64_t next_clones_obj = 1908 ds->ds_prev->ds_phys->ds_next_clones_obj; 1909 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == 1910 ds->ds_object || 1911 ds->ds_prev->ds_phys->ds_num_children > 1); 1912 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { 1913 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1914 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1915 ds->ds_prev->ds_phys->ds_creation_txg); 1916 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; 1917 } else if (next_clones_obj != 0) { 1918 remove_from_next_clones(ds->ds_prev, 1919 dsphys->ds_next_snap_obj, tx); 1920 VERIFY3U(0, ==, zap_add_int(mos, 1921 next_clones_obj, dsobj, tx)); 1922 } 1923 } 1924 1925 /* 1926 * If we have a reference-reservation on this dataset, we will 1927 * need to increase the amount of refreservation being charged 1928 * since our unique space is going to zero. 1929 */ 1930 if (ds->ds_reserved) { 1931 int64_t delta; 1932 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1933 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 1934 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1935 delta, 0, 0, tx); 1936 } 1937 1938 bplist_close(&ds->ds_deadlist); 1939 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1940 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg); 1941 ds->ds_phys->ds_prev_snap_obj = dsobj; 1942 ds->ds_phys->ds_prev_snap_txg = crtxg; 1943 ds->ds_phys->ds_unique_bytes = 0; 1944 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1945 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1946 ds->ds_phys->ds_deadlist_obj = 1947 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); 1948 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos, 1949 ds->ds_phys->ds_deadlist_obj)); 1950 1951 dprintf("snap '%s' -> obj %llu\n", snapname, dsobj); 1952 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, 1953 snapname, 8, 1, &dsobj, tx); 1954 ASSERT(err == 0); 1955 1956 if (ds->ds_prev) 1957 dsl_dataset_drop_ref(ds->ds_prev, ds); 1958 VERIFY(0 == dsl_dataset_get_ref(dp, 1959 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 1960 1961 dsl_scan_ds_snapshotted(ds, tx); 1962 1963 dsl_dir_snap_cmtime_update(ds->ds_dir); 1964 1965 spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx, 1966 "dataset = %llu", dsobj); 1967 } 1968 1969 void 1970 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1971 { 1972 ASSERT(dmu_tx_is_syncing(tx)); 1973 ASSERT(ds->ds_objset != NULL); 1974 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); 1975 1976 /* 1977 * in case we had to change ds_fsid_guid when we opened it, 1978 * sync it out now. 1979 */ 1980 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1981 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; 1982 1983 dsl_dir_dirty(ds->ds_dir, tx); 1984 dmu_objset_sync(ds->ds_objset, zio, tx); 1985 } 1986 1987 void 1988 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1989 { 1990 uint64_t refd, avail, uobjs, aobjs; 1991 1992 dsl_dir_stats(ds->ds_dir, nv); 1993 1994 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 1995 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 1996 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 1997 1998 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 1999 ds->ds_phys->ds_creation_time); 2000 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 2001 ds->ds_phys->ds_creation_txg); 2002 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 2003 ds->ds_quota); 2004 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 2005 ds->ds_reserved); 2006 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 2007 ds->ds_phys->ds_guid); 2008 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 2009 ds->ds_phys->ds_unique_bytes); 2010 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 2011 ds->ds_object); 2012 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 2013 ds->ds_userrefs); 2014 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 2015 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 2016 2017 if (ds->ds_phys->ds_next_snap_obj) { 2018 /* 2019 * This is a snapshot; override the dd's space used with 2020 * our unique space and compression ratio. 2021 */ 2022 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 2023 ds->ds_phys->ds_unique_bytes); 2024 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 2025 ds->ds_phys->ds_compressed_bytes == 0 ? 100 : 2026 (ds->ds_phys->ds_uncompressed_bytes * 100 / 2027 ds->ds_phys->ds_compressed_bytes)); 2028 } 2029 } 2030 2031 void 2032 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 2033 { 2034 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; 2035 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; 2036 stat->dds_guid = ds->ds_phys->ds_guid; 2037 if (ds->ds_phys->ds_next_snap_obj) { 2038 stat->dds_is_snapshot = B_TRUE; 2039 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; 2040 } else { 2041 stat->dds_is_snapshot = B_FALSE; 2042 stat->dds_num_clones = 0; 2043 } 2044 2045 /* clone origin is really a dsl_dir thing... */ 2046 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); 2047 if (dsl_dir_is_clone(ds->ds_dir)) { 2048 dsl_dataset_t *ods; 2049 2050 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool, 2051 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods)); 2052 dsl_dataset_name(ods, stat->dds_origin); 2053 dsl_dataset_drop_ref(ods, FTAG); 2054 } else { 2055 stat->dds_origin[0] = '\0'; 2056 } 2057 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); 2058 } 2059 2060 uint64_t 2061 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 2062 { 2063 return (ds->ds_fsid_guid); 2064 } 2065 2066 void 2067 dsl_dataset_space(dsl_dataset_t *ds, 2068 uint64_t *refdbytesp, uint64_t *availbytesp, 2069 uint64_t *usedobjsp, uint64_t *availobjsp) 2070 { 2071 *refdbytesp = ds->ds_phys->ds_used_bytes; 2072 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 2073 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) 2074 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes; 2075 if (ds->ds_quota != 0) { 2076 /* 2077 * Adjust available bytes according to refquota 2078 */ 2079 if (*refdbytesp < ds->ds_quota) 2080 *availbytesp = MIN(*availbytesp, 2081 ds->ds_quota - *refdbytesp); 2082 else 2083 *availbytesp = 0; 2084 } 2085 *usedobjsp = ds->ds_phys->ds_bp.blk_fill; 2086 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 2087 } 2088 2089 boolean_t 2090 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds) 2091 { 2092 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2093 2094 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 2095 dsl_pool_sync_context(dp)); 2096 if (ds->ds_prev == NULL) 2097 return (B_FALSE); 2098 if (ds->ds_phys->ds_bp.blk_birth > 2099 ds->ds_prev->ds_phys->ds_creation_txg) 2100 return (B_TRUE); 2101 return (B_FALSE); 2102 } 2103 2104 /* ARGSUSED */ 2105 static int 2106 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) 2107 { 2108 dsl_dataset_t *ds = arg1; 2109 char *newsnapname = arg2; 2110 dsl_dir_t *dd = ds->ds_dir; 2111 dsl_dataset_t *hds; 2112 uint64_t val; 2113 int err; 2114 2115 err = dsl_dataset_hold_obj(dd->dd_pool, 2116 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds); 2117 if (err) 2118 return (err); 2119 2120 /* new name better not be in use */ 2121 err = dsl_dataset_snap_lookup(hds, newsnapname, &val); 2122 dsl_dataset_rele(hds, FTAG); 2123 2124 if (err == 0) 2125 err = EEXIST; 2126 else if (err == ENOENT) 2127 err = 0; 2128 2129 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2130 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN) 2131 err = ENAMETOOLONG; 2132 2133 return (err); 2134 } 2135 2136 static void 2137 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx) 2138 { 2139 dsl_dataset_t *ds = arg1; 2140 const char *newsnapname = arg2; 2141 dsl_dir_t *dd = ds->ds_dir; 2142 objset_t *mos = dd->dd_pool->dp_meta_objset; 2143 dsl_dataset_t *hds; 2144 int err; 2145 2146 ASSERT(ds->ds_phys->ds_next_snap_obj != 0); 2147 2148 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, 2149 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds)); 2150 2151 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2152 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx); 2153 ASSERT3U(err, ==, 0); 2154 mutex_enter(&ds->ds_lock); 2155 (void) strcpy(ds->ds_snapname, newsnapname); 2156 mutex_exit(&ds->ds_lock); 2157 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj, 2158 ds->ds_snapname, 8, 1, &ds->ds_object, tx); 2159 ASSERT3U(err, ==, 0); 2160 2161 spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx, 2162 "dataset = %llu", ds->ds_object); 2163 dsl_dataset_rele(hds, FTAG); 2164 } 2165 2166 struct renamesnaparg { 2167 dsl_sync_task_group_t *dstg; 2168 char failed[MAXPATHLEN]; 2169 char *oldsnap; 2170 char *newsnap; 2171 }; 2172 2173 static int 2174 dsl_snapshot_rename_one(const char *name, void *arg) 2175 { 2176 struct renamesnaparg *ra = arg; 2177 dsl_dataset_t *ds = NULL; 2178 char *snapname; 2179 int err; 2180 2181 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap); 2182 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed)); 2183 2184 /* 2185 * For recursive snapshot renames the parent won't be changing 2186 * so we just pass name for both the to/from argument. 2187 */ 2188 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED()); 2189 if (err != 0) { 2190 strfree(snapname); 2191 return (err == ENOENT ? 0 : err); 2192 } 2193 2194 #ifdef _KERNEL 2195 /* 2196 * For all filesystems undergoing rename, we'll need to unmount it. 2197 */ 2198 (void) zfs_unmount_snap(snapname, NULL); 2199 #endif 2200 err = dsl_dataset_hold(snapname, ra->dstg, &ds); 2201 strfree(snapname); 2202 if (err != 0) 2203 return (err == ENOENT ? 0 : err); 2204 2205 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check, 2206 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0); 2207 2208 return (0); 2209 } 2210 2211 static int 2212 dsl_recursive_rename(char *oldname, const char *newname) 2213 { 2214 int err; 2215 struct renamesnaparg *ra; 2216 dsl_sync_task_t *dst; 2217 spa_t *spa; 2218 char *cp, *fsname = spa_strdup(oldname); 2219 int len = strlen(oldname) + 1; 2220 2221 /* truncate the snapshot name to get the fsname */ 2222 cp = strchr(fsname, '@'); 2223 *cp = '\0'; 2224 2225 err = spa_open(fsname, &spa, FTAG); 2226 if (err) { 2227 kmem_free(fsname, len); 2228 return (err); 2229 } 2230 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP); 2231 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 2232 2233 ra->oldsnap = strchr(oldname, '@') + 1; 2234 ra->newsnap = strchr(newname, '@') + 1; 2235 *ra->failed = '\0'; 2236 2237 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra, 2238 DS_FIND_CHILDREN); 2239 kmem_free(fsname, len); 2240 2241 if (err == 0) { 2242 err = dsl_sync_task_group_wait(ra->dstg); 2243 } 2244 2245 for (dst = list_head(&ra->dstg->dstg_tasks); dst; 2246 dst = list_next(&ra->dstg->dstg_tasks, dst)) { 2247 dsl_dataset_t *ds = dst->dst_arg1; 2248 if (dst->dst_err) { 2249 dsl_dir_name(ds->ds_dir, ra->failed); 2250 (void) strlcat(ra->failed, "@", sizeof (ra->failed)); 2251 (void) strlcat(ra->failed, ra->newsnap, 2252 sizeof (ra->failed)); 2253 } 2254 dsl_dataset_rele(ds, ra->dstg); 2255 } 2256 2257 if (err) 2258 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed)); 2259 2260 dsl_sync_task_group_destroy(ra->dstg); 2261 kmem_free(ra, sizeof (struct renamesnaparg)); 2262 spa_close(spa, FTAG); 2263 return (err); 2264 } 2265 2266 static int 2267 dsl_valid_rename(const char *oldname, void *arg) 2268 { 2269 int delta = *(int *)arg; 2270 2271 if (strlen(oldname) + delta >= MAXNAMELEN) 2272 return (ENAMETOOLONG); 2273 2274 return (0); 2275 } 2276 2277 #pragma weak dmu_objset_rename = dsl_dataset_rename 2278 int 2279 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive) 2280 { 2281 dsl_dir_t *dd; 2282 dsl_dataset_t *ds; 2283 const char *tail; 2284 int err; 2285 2286 err = dsl_dir_open(oldname, FTAG, &dd, &tail); 2287 if (err) 2288 return (err); 2289 2290 if (tail == NULL) { 2291 int delta = strlen(newname) - strlen(oldname); 2292 2293 /* if we're growing, validate child name lengths */ 2294 if (delta > 0) 2295 err = dmu_objset_find(oldname, dsl_valid_rename, 2296 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 2297 2298 if (err == 0) 2299 err = dsl_dir_rename(dd, newname); 2300 dsl_dir_close(dd, FTAG); 2301 return (err); 2302 } 2303 2304 if (tail[0] != '@') { 2305 /* the name ended in a nonexistent component */ 2306 dsl_dir_close(dd, FTAG); 2307 return (ENOENT); 2308 } 2309 2310 dsl_dir_close(dd, FTAG); 2311 2312 /* new name must be snapshot in same filesystem */ 2313 tail = strchr(newname, '@'); 2314 if (tail == NULL) 2315 return (EINVAL); 2316 tail++; 2317 if (strncmp(oldname, newname, tail - newname) != 0) 2318 return (EXDEV); 2319 2320 if (recursive) { 2321 err = dsl_recursive_rename(oldname, newname); 2322 } else { 2323 err = dsl_dataset_hold(oldname, FTAG, &ds); 2324 if (err) 2325 return (err); 2326 2327 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 2328 dsl_dataset_snapshot_rename_check, 2329 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1); 2330 2331 dsl_dataset_rele(ds, FTAG); 2332 } 2333 2334 return (err); 2335 } 2336 2337 struct promotenode { 2338 list_node_t link; 2339 dsl_dataset_t *ds; 2340 }; 2341 2342 struct promotearg { 2343 list_t shared_snaps, origin_snaps, clone_snaps; 2344 dsl_dataset_t *origin_origin; 2345 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2346 char *err_ds; 2347 }; 2348 2349 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2350 static boolean_t snaplist_unstable(list_t *l); 2351 2352 static int 2353 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) 2354 { 2355 dsl_dataset_t *hds = arg1; 2356 struct promotearg *pa = arg2; 2357 struct promotenode *snap = list_head(&pa->shared_snaps); 2358 dsl_dataset_t *origin_ds = snap->ds; 2359 int err; 2360 2361 /* Check that it is a real clone */ 2362 if (!dsl_dir_is_clone(hds->ds_dir)) 2363 return (EINVAL); 2364 2365 /* Since this is so expensive, don't do the preliminary check */ 2366 if (!dmu_tx_is_syncing(tx)) 2367 return (0); 2368 2369 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) 2370 return (EXDEV); 2371 2372 /* compute origin's new unique space */ 2373 snap = list_tail(&pa->clone_snaps); 2374 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2375 err = bplist_space_birthrange(&snap->ds->ds_deadlist, 2376 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, &pa->unique); 2377 if (err) 2378 return (err); 2379 2380 /* 2381 * Walk the snapshots that we are moving 2382 * 2383 * Compute space to transfer. Consider the incremental changes 2384 * to used for each snapshot: 2385 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2386 * So each snapshot gave birth to: 2387 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2388 * So a sequence would look like: 2389 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2390 * Which simplifies to: 2391 * uN + kN + kN-1 + ... + k1 + k0 2392 * Note however, if we stop before we reach the ORIGIN we get: 2393 * uN + kN + kN-1 + ... + kM - uM-1 2394 */ 2395 pa->used = origin_ds->ds_phys->ds_used_bytes; 2396 pa->comp = origin_ds->ds_phys->ds_compressed_bytes; 2397 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; 2398 for (snap = list_head(&pa->shared_snaps); snap; 2399 snap = list_next(&pa->shared_snaps, snap)) { 2400 uint64_t val, dlused, dlcomp, dluncomp; 2401 dsl_dataset_t *ds = snap->ds; 2402 2403 /* Check that the snapshot name does not conflict */ 2404 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2405 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2406 if (err == 0) { 2407 err = EEXIST; 2408 goto out; 2409 } 2410 if (err != ENOENT) 2411 goto out; 2412 2413 /* The very first snapshot does not have a deadlist */ 2414 if (ds->ds_phys->ds_prev_snap_obj == 0) 2415 continue; 2416 2417 if (err = bplist_space(&ds->ds_deadlist, 2418 &dlused, &dlcomp, &dluncomp)) 2419 goto out; 2420 pa->used += dlused; 2421 pa->comp += dlcomp; 2422 pa->uncomp += dluncomp; 2423 } 2424 2425 /* 2426 * If we are a clone of a clone then we never reached ORIGIN, 2427 * so we need to subtract out the clone origin's used space. 2428 */ 2429 if (pa->origin_origin) { 2430 pa->used -= pa->origin_origin->ds_phys->ds_used_bytes; 2431 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes; 2432 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes; 2433 } 2434 2435 /* Check that there is enough space here */ 2436 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2437 pa->used); 2438 if (err) 2439 return (err); 2440 2441 /* 2442 * Compute the amounts of space that will be used by snapshots 2443 * after the promotion (for both origin and clone). For each, 2444 * it is the amount of space that will be on all of their 2445 * deadlists (that was not born before their new origin). 2446 */ 2447 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2448 uint64_t space; 2449 2450 /* 2451 * Note, typically this will not be a clone of a clone, 2452 * so dd_origin_txg will be < TXG_INITIAL, so 2453 * these snaplist_space() -> bplist_space_birthrange() 2454 * calls will be fast because they do not have to 2455 * iterate over all bps. 2456 */ 2457 snap = list_head(&pa->origin_snaps); 2458 err = snaplist_space(&pa->shared_snaps, 2459 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap); 2460 if (err) 2461 return (err); 2462 2463 err = snaplist_space(&pa->clone_snaps, 2464 snap->ds->ds_dir->dd_origin_txg, &space); 2465 if (err) 2466 return (err); 2467 pa->cloneusedsnap += space; 2468 } 2469 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2470 err = snaplist_space(&pa->origin_snaps, 2471 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap); 2472 if (err) 2473 return (err); 2474 } 2475 2476 return (0); 2477 out: 2478 pa->err_ds = snap->ds->ds_snapname; 2479 return (err); 2480 } 2481 2482 static void 2483 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx) 2484 { 2485 dsl_dataset_t *hds = arg1; 2486 struct promotearg *pa = arg2; 2487 struct promotenode *snap = list_head(&pa->shared_snaps); 2488 dsl_dataset_t *origin_ds = snap->ds; 2489 dsl_dataset_t *origin_head; 2490 dsl_dir_t *dd = hds->ds_dir; 2491 dsl_pool_t *dp = hds->ds_dir->dd_pool; 2492 dsl_dir_t *odd = NULL; 2493 uint64_t oldnext_obj; 2494 int64_t delta; 2495 2496 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)); 2497 2498 snap = list_head(&pa->origin_snaps); 2499 origin_head = snap->ds; 2500 2501 /* 2502 * We need to explicitly open odd, since origin_ds's dd will be 2503 * changing. 2504 */ 2505 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object, 2506 NULL, FTAG, &odd)); 2507 2508 /* change origin's next snap */ 2509 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2510 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj; 2511 snap = list_tail(&pa->clone_snaps); 2512 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2513 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object; 2514 2515 /* change the origin's next clone */ 2516 if (origin_ds->ds_phys->ds_next_clones_obj) { 2517 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx); 2518 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, 2519 origin_ds->ds_phys->ds_next_clones_obj, 2520 oldnext_obj, tx)); 2521 } 2522 2523 /* change origin */ 2524 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2525 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object); 2526 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj; 2527 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2528 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2529 odd->dd_phys->dd_origin_obj = origin_ds->ds_object; 2530 origin_head->ds_dir->dd_origin_txg = 2531 origin_ds->ds_phys->ds_creation_txg; 2532 2533 /* move snapshots to this dir */ 2534 for (snap = list_head(&pa->shared_snaps); snap; 2535 snap = list_next(&pa->shared_snaps, snap)) { 2536 dsl_dataset_t *ds = snap->ds; 2537 2538 /* unregister props as dsl_dir is changing */ 2539 if (ds->ds_objset) { 2540 dmu_objset_evict(ds->ds_objset); 2541 ds->ds_objset = NULL; 2542 } 2543 /* move snap name entry */ 2544 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2545 VERIFY(0 == dsl_dataset_snap_remove(origin_head, 2546 ds->ds_snapname, tx)); 2547 VERIFY(0 == zap_add(dp->dp_meta_objset, 2548 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, 2549 8, 1, &ds->ds_object, tx)); 2550 /* change containing dsl_dir */ 2551 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2552 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object); 2553 ds->ds_phys->ds_dir_obj = dd->dd_object; 2554 ASSERT3P(ds->ds_dir, ==, odd); 2555 dsl_dir_close(ds->ds_dir, ds); 2556 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object, 2557 NULL, ds, &ds->ds_dir)); 2558 2559 ASSERT3U(dsl_prop_numcb(ds), ==, 0); 2560 } 2561 2562 /* 2563 * Change space accounting. 2564 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2565 * both be valid, or both be 0 (resulting in delta == 0). This 2566 * is true for each of {clone,origin} independently. 2567 */ 2568 2569 delta = pa->cloneusedsnap - 2570 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2571 ASSERT3S(delta, >=, 0); 2572 ASSERT3U(pa->used, >=, delta); 2573 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2574 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2575 pa->used - delta, pa->comp, pa->uncomp, tx); 2576 2577 delta = pa->originusedsnap - 2578 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2579 ASSERT3S(delta, <=, 0); 2580 ASSERT3U(pa->used, >=, -delta); 2581 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2582 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2583 -pa->used - delta, -pa->comp, -pa->uncomp, tx); 2584 2585 origin_ds->ds_phys->ds_unique_bytes = pa->unique; 2586 2587 /* log history record */ 2588 spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx, 2589 "dataset = %llu", hds->ds_object); 2590 2591 dsl_dir_close(odd, FTAG); 2592 } 2593 2594 static char *snaplist_tag = "snaplist"; 2595 /* 2596 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2597 * (exclusive) and last_obj (inclusive). The list will be in reverse 2598 * order (last_obj will be the list_head()). If first_obj == 0, do all 2599 * snapshots back to this dataset's origin. 2600 */ 2601 static int 2602 snaplist_make(dsl_pool_t *dp, boolean_t own, 2603 uint64_t first_obj, uint64_t last_obj, list_t *l) 2604 { 2605 uint64_t obj = last_obj; 2606 2607 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock)); 2608 2609 list_create(l, sizeof (struct promotenode), 2610 offsetof(struct promotenode, link)); 2611 2612 while (obj != first_obj) { 2613 dsl_dataset_t *ds; 2614 struct promotenode *snap; 2615 int err; 2616 2617 if (own) { 2618 err = dsl_dataset_own_obj(dp, obj, 2619 0, snaplist_tag, &ds); 2620 if (err == 0) 2621 dsl_dataset_make_exclusive(ds, snaplist_tag); 2622 } else { 2623 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds); 2624 } 2625 if (err == ENOENT) { 2626 /* lost race with snapshot destroy */ 2627 struct promotenode *last = list_tail(l); 2628 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj); 2629 obj = last->ds->ds_phys->ds_prev_snap_obj; 2630 continue; 2631 } else if (err) { 2632 return (err); 2633 } 2634 2635 if (first_obj == 0) 2636 first_obj = ds->ds_dir->dd_phys->dd_origin_obj; 2637 2638 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP); 2639 snap->ds = ds; 2640 list_insert_tail(l, snap); 2641 obj = ds->ds_phys->ds_prev_snap_obj; 2642 } 2643 2644 return (0); 2645 } 2646 2647 static int 2648 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2649 { 2650 struct promotenode *snap; 2651 2652 *spacep = 0; 2653 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2654 uint64_t used; 2655 int err = bplist_space_birthrange(&snap->ds->ds_deadlist, 2656 mintxg, UINT64_MAX, &used); 2657 if (err) 2658 return (err); 2659 *spacep += used; 2660 } 2661 return (0); 2662 } 2663 2664 static void 2665 snaplist_destroy(list_t *l, boolean_t own) 2666 { 2667 struct promotenode *snap; 2668 2669 if (!l || !list_link_active(&l->list_head)) 2670 return; 2671 2672 while ((snap = list_tail(l)) != NULL) { 2673 list_remove(l, snap); 2674 if (own) 2675 dsl_dataset_disown(snap->ds, snaplist_tag); 2676 else 2677 dsl_dataset_rele(snap->ds, snaplist_tag); 2678 kmem_free(snap, sizeof (struct promotenode)); 2679 } 2680 list_destroy(l); 2681 } 2682 2683 /* 2684 * Promote a clone. Nomenclature note: 2685 * "clone" or "cds": the original clone which is being promoted 2686 * "origin" or "ods": the snapshot which is originally clone's origin 2687 * "origin head" or "ohds": the dataset which is the head 2688 * (filesystem/volume) for the origin 2689 * "origin origin": the origin of the origin's filesystem (typically 2690 * NULL, indicating that the clone is not a clone of a clone). 2691 */ 2692 int 2693 dsl_dataset_promote(const char *name, char *conflsnap) 2694 { 2695 dsl_dataset_t *ds; 2696 dsl_dir_t *dd; 2697 dsl_pool_t *dp; 2698 dmu_object_info_t doi; 2699 struct promotearg pa = { 0 }; 2700 struct promotenode *snap; 2701 int err; 2702 2703 err = dsl_dataset_hold(name, FTAG, &ds); 2704 if (err) 2705 return (err); 2706 dd = ds->ds_dir; 2707 dp = dd->dd_pool; 2708 2709 err = dmu_object_info(dp->dp_meta_objset, 2710 ds->ds_phys->ds_snapnames_zapobj, &doi); 2711 if (err) { 2712 dsl_dataset_rele(ds, FTAG); 2713 return (err); 2714 } 2715 2716 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) { 2717 dsl_dataset_rele(ds, FTAG); 2718 return (EINVAL); 2719 } 2720 2721 /* 2722 * We are going to inherit all the snapshots taken before our 2723 * origin (i.e., our new origin will be our parent's origin). 2724 * Take ownership of them so that we can rename them into our 2725 * namespace. 2726 */ 2727 rw_enter(&dp->dp_config_rwlock, RW_READER); 2728 2729 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj, 2730 &pa.shared_snaps); 2731 if (err != 0) 2732 goto out; 2733 2734 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps); 2735 if (err != 0) 2736 goto out; 2737 2738 snap = list_head(&pa.shared_snaps); 2739 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); 2740 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj, 2741 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps); 2742 if (err != 0) 2743 goto out; 2744 2745 if (dsl_dir_is_clone(snap->ds->ds_dir)) { 2746 err = dsl_dataset_own_obj(dp, 2747 snap->ds->ds_dir->dd_phys->dd_origin_obj, 2748 0, FTAG, &pa.origin_origin); 2749 if (err != 0) 2750 goto out; 2751 } 2752 2753 out: 2754 rw_exit(&dp->dp_config_rwlock); 2755 2756 /* 2757 * Add in 128x the snapnames zapobj size, since we will be moving 2758 * a bunch of snapnames to the promoted ds, and dirtying their 2759 * bonus buffers. 2760 */ 2761 if (err == 0) { 2762 err = dsl_sync_task_do(dp, dsl_dataset_promote_check, 2763 dsl_dataset_promote_sync, ds, &pa, 2764 2 + 2 * doi.doi_physical_blocks_512); 2765 if (err && pa.err_ds && conflsnap) 2766 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN); 2767 } 2768 2769 snaplist_destroy(&pa.shared_snaps, B_TRUE); 2770 snaplist_destroy(&pa.clone_snaps, B_FALSE); 2771 snaplist_destroy(&pa.origin_snaps, B_FALSE); 2772 if (pa.origin_origin) 2773 dsl_dataset_disown(pa.origin_origin, FTAG); 2774 dsl_dataset_rele(ds, FTAG); 2775 return (err); 2776 } 2777 2778 struct cloneswaparg { 2779 dsl_dataset_t *cds; /* clone dataset */ 2780 dsl_dataset_t *ohds; /* origin's head dataset */ 2781 boolean_t force; 2782 int64_t unused_refres_delta; /* change in unconsumed refreservation */ 2783 }; 2784 2785 /* ARGSUSED */ 2786 static int 2787 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx) 2788 { 2789 struct cloneswaparg *csa = arg1; 2790 2791 /* they should both be heads */ 2792 if (dsl_dataset_is_snapshot(csa->cds) || 2793 dsl_dataset_is_snapshot(csa->ohds)) 2794 return (EINVAL); 2795 2796 /* the branch point should be just before them */ 2797 if (csa->cds->ds_prev != csa->ohds->ds_prev) 2798 return (EINVAL); 2799 2800 /* cds should be the clone (unless they are unrelated) */ 2801 if (csa->cds->ds_prev != NULL && 2802 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap && 2803 csa->ohds->ds_object != 2804 csa->cds->ds_prev->ds_phys->ds_next_snap_obj) 2805 return (EINVAL); 2806 2807 /* the clone should be a child of the origin */ 2808 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir) 2809 return (EINVAL); 2810 2811 /* ohds shouldn't be modified unless 'force' */ 2812 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds)) 2813 return (ETXTBSY); 2814 2815 /* adjust amount of any unconsumed refreservation */ 2816 csa->unused_refres_delta = 2817 (int64_t)MIN(csa->ohds->ds_reserved, 2818 csa->ohds->ds_phys->ds_unique_bytes) - 2819 (int64_t)MIN(csa->ohds->ds_reserved, 2820 csa->cds->ds_phys->ds_unique_bytes); 2821 2822 if (csa->unused_refres_delta > 0 && 2823 csa->unused_refres_delta > 2824 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE)) 2825 return (ENOSPC); 2826 2827 if (csa->ohds->ds_quota != 0 && 2828 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota) 2829 return (EDQUOT); 2830 2831 return (0); 2832 } 2833 2834 /* ARGSUSED */ 2835 static void 2836 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx) 2837 { 2838 struct cloneswaparg *csa = arg1; 2839 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool; 2840 2841 ASSERT(csa->cds->ds_reserved == 0); 2842 ASSERT(csa->ohds->ds_quota == 0 || 2843 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota); 2844 2845 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx); 2846 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx); 2847 2848 if (csa->cds->ds_objset != NULL) { 2849 dmu_objset_evict(csa->cds->ds_objset); 2850 csa->cds->ds_objset = NULL; 2851 } 2852 2853 if (csa->ohds->ds_objset != NULL) { 2854 dmu_objset_evict(csa->ohds->ds_objset); 2855 csa->ohds->ds_objset = NULL; 2856 } 2857 2858 /* 2859 * Reset origin's unique bytes, if it exists. 2860 */ 2861 if (csa->cds->ds_prev) { 2862 dsl_dataset_t *origin = csa->cds->ds_prev; 2863 dmu_buf_will_dirty(origin->ds_dbuf, tx); 2864 VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist, 2865 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2866 &origin->ds_phys->ds_unique_bytes)); 2867 } 2868 2869 /* swap blkptrs */ 2870 { 2871 blkptr_t tmp; 2872 tmp = csa->ohds->ds_phys->ds_bp; 2873 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp; 2874 csa->cds->ds_phys->ds_bp = tmp; 2875 } 2876 2877 /* set dd_*_bytes */ 2878 { 2879 int64_t dused, dcomp, duncomp; 2880 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2881 uint64_t odl_used, odl_comp, odl_uncomp; 2882 2883 ASSERT3U(csa->cds->ds_dir->dd_phys-> 2884 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2885 2886 VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used, 2887 &cdl_comp, &cdl_uncomp)); 2888 VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used, 2889 &odl_comp, &odl_uncomp)); 2890 2891 dused = csa->cds->ds_phys->ds_used_bytes + cdl_used - 2892 (csa->ohds->ds_phys->ds_used_bytes + odl_used); 2893 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp - 2894 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp); 2895 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes + 2896 cdl_uncomp - 2897 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp); 2898 2899 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD, 2900 dused, dcomp, duncomp, tx); 2901 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD, 2902 -dused, -dcomp, -duncomp, tx); 2903 2904 /* 2905 * The difference in the space used by snapshots is the 2906 * difference in snapshot space due to the head's 2907 * deadlist (since that's the only thing that's 2908 * changing that affects the snapused). 2909 */ 2910 VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist, 2911 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX, &cdl_used)); 2912 VERIFY(0 == bplist_space_birthrange(&csa->ohds->ds_deadlist, 2913 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX, &odl_used)); 2914 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used, 2915 DD_USED_HEAD, DD_USED_SNAP, tx); 2916 } 2917 2918 #define SWITCH64(x, y) \ 2919 { \ 2920 uint64_t __tmp = (x); \ 2921 (x) = (y); \ 2922 (y) = __tmp; \ 2923 } 2924 2925 /* swap ds_*_bytes */ 2926 SWITCH64(csa->ohds->ds_phys->ds_used_bytes, 2927 csa->cds->ds_phys->ds_used_bytes); 2928 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes, 2929 csa->cds->ds_phys->ds_compressed_bytes); 2930 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes, 2931 csa->cds->ds_phys->ds_uncompressed_bytes); 2932 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes, 2933 csa->cds->ds_phys->ds_unique_bytes); 2934 2935 /* apply any parent delta for change in unconsumed refreservation */ 2936 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV, 2937 csa->unused_refres_delta, 0, 0, tx); 2938 2939 /* swap deadlists */ 2940 bplist_close(&csa->cds->ds_deadlist); 2941 bplist_close(&csa->ohds->ds_deadlist); 2942 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj, 2943 csa->cds->ds_phys->ds_deadlist_obj); 2944 VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset, 2945 csa->cds->ds_phys->ds_deadlist_obj)); 2946 VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset, 2947 csa->ohds->ds_phys->ds_deadlist_obj)); 2948 2949 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx); 2950 } 2951 2952 /* 2953 * Swap 'clone' with its origin head datasets. Used at the end of "zfs 2954 * recv" into an existing fs to swizzle the file system to the new 2955 * version, and by "zfs rollback". Can also be used to swap two 2956 * independent head datasets if neither has any snapshots. 2957 */ 2958 int 2959 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head, 2960 boolean_t force) 2961 { 2962 struct cloneswaparg csa; 2963 int error; 2964 2965 ASSERT(clone->ds_owner); 2966 ASSERT(origin_head->ds_owner); 2967 retry: 2968 /* Need exclusive access for the swap */ 2969 rw_enter(&clone->ds_rwlock, RW_WRITER); 2970 if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) { 2971 rw_exit(&clone->ds_rwlock); 2972 rw_enter(&origin_head->ds_rwlock, RW_WRITER); 2973 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) { 2974 rw_exit(&origin_head->ds_rwlock); 2975 goto retry; 2976 } 2977 } 2978 csa.cds = clone; 2979 csa.ohds = origin_head; 2980 csa.force = force; 2981 error = dsl_sync_task_do(clone->ds_dir->dd_pool, 2982 dsl_dataset_clone_swap_check, 2983 dsl_dataset_clone_swap_sync, &csa, NULL, 9); 2984 return (error); 2985 } 2986 2987 /* 2988 * Given a pool name and a dataset object number in that pool, 2989 * return the name of that dataset. 2990 */ 2991 int 2992 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 2993 { 2994 spa_t *spa; 2995 dsl_pool_t *dp; 2996 dsl_dataset_t *ds; 2997 int error; 2998 2999 if ((error = spa_open(pname, &spa, FTAG)) != 0) 3000 return (error); 3001 dp = spa_get_dsl(spa); 3002 rw_enter(&dp->dp_config_rwlock, RW_READER); 3003 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) { 3004 dsl_dataset_name(ds, buf); 3005 dsl_dataset_rele(ds, FTAG); 3006 } 3007 rw_exit(&dp->dp_config_rwlock); 3008 spa_close(spa, FTAG); 3009 3010 return (error); 3011 } 3012 3013 int 3014 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 3015 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 3016 { 3017 int error = 0; 3018 3019 ASSERT3S(asize, >, 0); 3020 3021 /* 3022 * *ref_rsrv is the portion of asize that will come from any 3023 * unconsumed refreservation space. 3024 */ 3025 *ref_rsrv = 0; 3026 3027 mutex_enter(&ds->ds_lock); 3028 /* 3029 * Make a space adjustment for reserved bytes. 3030 */ 3031 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) { 3032 ASSERT3U(*used, >=, 3033 ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 3034 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 3035 *ref_rsrv = 3036 asize - MIN(asize, parent_delta(ds, asize + inflight)); 3037 } 3038 3039 if (!check_quota || ds->ds_quota == 0) { 3040 mutex_exit(&ds->ds_lock); 3041 return (0); 3042 } 3043 /* 3044 * If they are requesting more space, and our current estimate 3045 * is over quota, they get to try again unless the actual 3046 * on-disk is over quota and there are no pending changes (which 3047 * may free up space for us). 3048 */ 3049 if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) { 3050 if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota) 3051 error = ERESTART; 3052 else 3053 error = EDQUOT; 3054 } 3055 mutex_exit(&ds->ds_lock); 3056 3057 return (error); 3058 } 3059 3060 /* ARGSUSED */ 3061 static int 3062 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) 3063 { 3064 dsl_dataset_t *ds = arg1; 3065 dsl_prop_setarg_t *psa = arg2; 3066 int err; 3067 3068 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA) 3069 return (ENOTSUP); 3070 3071 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) 3072 return (err); 3073 3074 if (psa->psa_effective_value == 0) 3075 return (0); 3076 3077 if (psa->psa_effective_value < ds->ds_phys->ds_used_bytes || 3078 psa->psa_effective_value < ds->ds_reserved) 3079 return (ENOSPC); 3080 3081 return (0); 3082 } 3083 3084 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *); 3085 3086 void 3087 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx) 3088 { 3089 dsl_dataset_t *ds = arg1; 3090 dsl_prop_setarg_t *psa = arg2; 3091 uint64_t effective_value = psa->psa_effective_value; 3092 3093 dsl_prop_set_sync(ds, psa, tx); 3094 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa); 3095 3096 if (ds->ds_quota != effective_value) { 3097 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3098 ds->ds_quota = effective_value; 3099 3100 spa_history_log_internal(LOG_DS_REFQUOTA, 3101 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ", 3102 (longlong_t)ds->ds_quota, ds->ds_object); 3103 } 3104 } 3105 3106 int 3107 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota) 3108 { 3109 dsl_dataset_t *ds; 3110 dsl_prop_setarg_t psa; 3111 int err; 3112 3113 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a); 3114 3115 err = dsl_dataset_hold(dsname, FTAG, &ds); 3116 if (err) 3117 return (err); 3118 3119 /* 3120 * If someone removes a file, then tries to set the quota, we 3121 * want to make sure the file freeing takes effect. 3122 */ 3123 txg_wait_open(ds->ds_dir->dd_pool, 0); 3124 3125 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 3126 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync, 3127 ds, &psa, 0); 3128 3129 dsl_dataset_rele(ds, FTAG); 3130 return (err); 3131 } 3132 3133 static int 3134 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) 3135 { 3136 dsl_dataset_t *ds = arg1; 3137 dsl_prop_setarg_t *psa = arg2; 3138 uint64_t effective_value; 3139 uint64_t unique; 3140 int err; 3141 3142 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < 3143 SPA_VERSION_REFRESERVATION) 3144 return (ENOTSUP); 3145 3146 if (dsl_dataset_is_snapshot(ds)) 3147 return (EINVAL); 3148 3149 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) 3150 return (err); 3151 3152 effective_value = psa->psa_effective_value; 3153 3154 /* 3155 * If we are doing the preliminary check in open context, the 3156 * space estimates may be inaccurate. 3157 */ 3158 if (!dmu_tx_is_syncing(tx)) 3159 return (0); 3160 3161 mutex_enter(&ds->ds_lock); 3162 if (!DS_UNIQUE_IS_ACCURATE(ds)) 3163 dsl_dataset_recalc_head_uniq(ds); 3164 unique = ds->ds_phys->ds_unique_bytes; 3165 mutex_exit(&ds->ds_lock); 3166 3167 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) { 3168 uint64_t delta = MAX(unique, effective_value) - 3169 MAX(unique, ds->ds_reserved); 3170 3171 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 3172 return (ENOSPC); 3173 if (ds->ds_quota > 0 && 3174 effective_value > ds->ds_quota) 3175 return (ENOSPC); 3176 } 3177 3178 return (0); 3179 } 3180 3181 static void 3182 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx) 3183 { 3184 dsl_dataset_t *ds = arg1; 3185 dsl_prop_setarg_t *psa = arg2; 3186 uint64_t effective_value = psa->psa_effective_value; 3187 uint64_t unique; 3188 int64_t delta; 3189 3190 dsl_prop_set_sync(ds, psa, tx); 3191 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa); 3192 3193 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3194 3195 mutex_enter(&ds->ds_dir->dd_lock); 3196 mutex_enter(&ds->ds_lock); 3197 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3198 unique = ds->ds_phys->ds_unique_bytes; 3199 delta = MAX(0, (int64_t)(effective_value - unique)) - 3200 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3201 ds->ds_reserved = effective_value; 3202 mutex_exit(&ds->ds_lock); 3203 3204 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3205 mutex_exit(&ds->ds_dir->dd_lock); 3206 3207 spa_history_log_internal(LOG_DS_REFRESERV, 3208 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu", 3209 (longlong_t)effective_value, ds->ds_object); 3210 } 3211 3212 int 3213 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source, 3214 uint64_t reservation) 3215 { 3216 dsl_dataset_t *ds; 3217 dsl_prop_setarg_t psa; 3218 int err; 3219 3220 dsl_prop_setarg_init_uint64(&psa, "refreservation", source, 3221 &reservation); 3222 3223 err = dsl_dataset_hold(dsname, FTAG, &ds); 3224 if (err) 3225 return (err); 3226 3227 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 3228 dsl_dataset_set_reservation_check, 3229 dsl_dataset_set_reservation_sync, ds, &psa, 0); 3230 3231 dsl_dataset_rele(ds, FTAG); 3232 return (err); 3233 } 3234 3235 struct dsl_ds_holdarg { 3236 dsl_sync_task_group_t *dstg; 3237 char *htag; 3238 char *snapname; 3239 boolean_t recursive; 3240 boolean_t gotone; 3241 boolean_t temphold; 3242 char failed[MAXPATHLEN]; 3243 }; 3244 3245 /* 3246 * The max length of a temporary tag prefix is the number of hex digits 3247 * required to express UINT64_MAX plus one for the hyphen. 3248 */ 3249 #define MAX_TAG_PREFIX_LEN 17 3250 3251 static int 3252 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx) 3253 { 3254 dsl_dataset_t *ds = arg1; 3255 struct dsl_ds_holdarg *ha = arg2; 3256 char *htag = ha->htag; 3257 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3258 int error = 0; 3259 3260 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) 3261 return (ENOTSUP); 3262 3263 if (!dsl_dataset_is_snapshot(ds)) 3264 return (EINVAL); 3265 3266 /* tags must be unique */ 3267 mutex_enter(&ds->ds_lock); 3268 if (ds->ds_phys->ds_userrefs_obj) { 3269 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag, 3270 8, 1, tx); 3271 if (error == 0) 3272 error = EEXIST; 3273 else if (error == ENOENT) 3274 error = 0; 3275 } 3276 mutex_exit(&ds->ds_lock); 3277 3278 if (error == 0 && ha->temphold && 3279 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN) 3280 error = E2BIG; 3281 3282 return (error); 3283 } 3284 3285 static void 3286 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx) 3287 { 3288 dsl_dataset_t *ds = arg1; 3289 struct dsl_ds_holdarg *ha = arg2; 3290 char *htag = ha->htag; 3291 dsl_pool_t *dp = ds->ds_dir->dd_pool; 3292 objset_t *mos = dp->dp_meta_objset; 3293 uint64_t now = gethrestime_sec(); 3294 uint64_t zapobj; 3295 3296 mutex_enter(&ds->ds_lock); 3297 if (ds->ds_phys->ds_userrefs_obj == 0) { 3298 /* 3299 * This is the first user hold for this dataset. Create 3300 * the userrefs zap object. 3301 */ 3302 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3303 zapobj = ds->ds_phys->ds_userrefs_obj = 3304 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx); 3305 } else { 3306 zapobj = ds->ds_phys->ds_userrefs_obj; 3307 } 3308 ds->ds_userrefs++; 3309 mutex_exit(&ds->ds_lock); 3310 3311 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx)); 3312 3313 if (ha->temphold) { 3314 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object, 3315 htag, &now, tx)); 3316 } 3317 3318 spa_history_log_internal(LOG_DS_USER_HOLD, 3319 dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag, 3320 (int)ha->temphold, ds->ds_object); 3321 } 3322 3323 static int 3324 dsl_dataset_user_hold_one(const char *dsname, void *arg) 3325 { 3326 struct dsl_ds_holdarg *ha = arg; 3327 dsl_dataset_t *ds; 3328 int error; 3329 char *name; 3330 3331 /* alloc a buffer to hold dsname@snapname plus terminating NULL */ 3332 name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3333 error = dsl_dataset_hold(name, ha->dstg, &ds); 3334 strfree(name); 3335 if (error == 0) { 3336 ha->gotone = B_TRUE; 3337 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check, 3338 dsl_dataset_user_hold_sync, ds, ha, 0); 3339 } else if (error == ENOENT && ha->recursive) { 3340 error = 0; 3341 } else { 3342 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3343 } 3344 return (error); 3345 } 3346 3347 int 3348 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag, 3349 boolean_t recursive, boolean_t temphold) 3350 { 3351 struct dsl_ds_holdarg *ha; 3352 dsl_sync_task_t *dst; 3353 spa_t *spa; 3354 int error; 3355 3356 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); 3357 3358 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3359 3360 error = spa_open(dsname, &spa, FTAG); 3361 if (error) { 3362 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3363 return (error); 3364 } 3365 3366 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 3367 ha->htag = htag; 3368 ha->snapname = snapname; 3369 ha->recursive = recursive; 3370 ha->temphold = temphold; 3371 if (recursive) { 3372 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one, 3373 ha, DS_FIND_CHILDREN); 3374 } else { 3375 error = dsl_dataset_user_hold_one(dsname, ha); 3376 } 3377 if (error == 0) 3378 error = dsl_sync_task_group_wait(ha->dstg); 3379 3380 for (dst = list_head(&ha->dstg->dstg_tasks); dst; 3381 dst = list_next(&ha->dstg->dstg_tasks, dst)) { 3382 dsl_dataset_t *ds = dst->dst_arg1; 3383 3384 if (dst->dst_err) { 3385 dsl_dataset_name(ds, ha->failed); 3386 *strchr(ha->failed, '@') = '\0'; 3387 } 3388 dsl_dataset_rele(ds, ha->dstg); 3389 } 3390 3391 if (error == 0 && recursive && !ha->gotone) 3392 error = ENOENT; 3393 3394 if (error) 3395 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed)); 3396 3397 dsl_sync_task_group_destroy(ha->dstg); 3398 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3399 spa_close(spa, FTAG); 3400 return (error); 3401 } 3402 3403 struct dsl_ds_releasearg { 3404 dsl_dataset_t *ds; 3405 const char *htag; 3406 boolean_t own; /* do we own or just hold ds? */ 3407 }; 3408 3409 static int 3410 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag, 3411 boolean_t *might_destroy) 3412 { 3413 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3414 uint64_t zapobj; 3415 uint64_t tmp; 3416 int error; 3417 3418 *might_destroy = B_FALSE; 3419 3420 mutex_enter(&ds->ds_lock); 3421 zapobj = ds->ds_phys->ds_userrefs_obj; 3422 if (zapobj == 0) { 3423 /* The tag can't possibly exist */ 3424 mutex_exit(&ds->ds_lock); 3425 return (ESRCH); 3426 } 3427 3428 /* Make sure the tag exists */ 3429 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp); 3430 if (error) { 3431 mutex_exit(&ds->ds_lock); 3432 if (error == ENOENT) 3433 error = ESRCH; 3434 return (error); 3435 } 3436 3437 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 && 3438 DS_IS_DEFER_DESTROY(ds)) 3439 *might_destroy = B_TRUE; 3440 3441 mutex_exit(&ds->ds_lock); 3442 return (0); 3443 } 3444 3445 static int 3446 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx) 3447 { 3448 struct dsl_ds_releasearg *ra = arg1; 3449 dsl_dataset_t *ds = ra->ds; 3450 boolean_t might_destroy; 3451 int error; 3452 3453 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) 3454 return (ENOTSUP); 3455 3456 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy); 3457 if (error) 3458 return (error); 3459 3460 if (might_destroy) { 3461 struct dsl_ds_destroyarg dsda = {0}; 3462 3463 if (dmu_tx_is_syncing(tx)) { 3464 /* 3465 * If we're not prepared to remove the snapshot, 3466 * we can't allow the release to happen right now. 3467 */ 3468 if (!ra->own) 3469 return (EBUSY); 3470 } 3471 dsda.ds = ds; 3472 dsda.releasing = B_TRUE; 3473 return (dsl_dataset_destroy_check(&dsda, tag, tx)); 3474 } 3475 3476 return (0); 3477 } 3478 3479 static void 3480 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx) 3481 { 3482 struct dsl_ds_releasearg *ra = arg1; 3483 dsl_dataset_t *ds = ra->ds; 3484 dsl_pool_t *dp = ds->ds_dir->dd_pool; 3485 objset_t *mos = dp->dp_meta_objset; 3486 uint64_t zapobj; 3487 uint64_t dsobj = ds->ds_object; 3488 uint64_t refs; 3489 int error; 3490 3491 if (ds->ds_objset) { 3492 dmu_objset_evict(ds->ds_objset); 3493 ds->ds_objset = NULL; 3494 } 3495 3496 mutex_enter(&ds->ds_lock); 3497 ds->ds_userrefs--; 3498 refs = ds->ds_userrefs; 3499 mutex_exit(&ds->ds_lock); 3500 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx); 3501 VERIFY(error == 0 || error == ENOENT); 3502 zapobj = ds->ds_phys->ds_userrefs_obj; 3503 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx)); 3504 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 && 3505 DS_IS_DEFER_DESTROY(ds)) { 3506 struct dsl_ds_destroyarg dsda = {0}; 3507 3508 ASSERT(ra->own); 3509 dsda.ds = ds; 3510 dsda.releasing = B_TRUE; 3511 /* We already did the destroy_check */ 3512 dsl_dataset_destroy_sync(&dsda, tag, tx); 3513 } 3514 3515 spa_history_log_internal(LOG_DS_USER_RELEASE, 3516 dp->dp_spa, tx, "<%s> %lld dataset = %llu", 3517 ra->htag, (longlong_t)refs, dsobj); 3518 } 3519 3520 static int 3521 dsl_dataset_user_release_one(const char *dsname, void *arg) 3522 { 3523 struct dsl_ds_holdarg *ha = arg; 3524 struct dsl_ds_releasearg *ra; 3525 dsl_dataset_t *ds; 3526 int error; 3527 void *dtag = ha->dstg; 3528 char *name; 3529 boolean_t own = B_FALSE; 3530 boolean_t might_destroy; 3531 3532 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */ 3533 name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3534 error = dsl_dataset_hold(name, dtag, &ds); 3535 strfree(name); 3536 if (error == ENOENT && ha->recursive) 3537 return (0); 3538 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3539 if (error) 3540 return (error); 3541 3542 ha->gotone = B_TRUE; 3543 3544 ASSERT(dsl_dataset_is_snapshot(ds)); 3545 3546 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy); 3547 if (error) { 3548 dsl_dataset_rele(ds, dtag); 3549 return (error); 3550 } 3551 3552 if (might_destroy) { 3553 #ifdef _KERNEL 3554 name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3555 error = zfs_unmount_snap(name, NULL); 3556 strfree(name); 3557 if (error) { 3558 dsl_dataset_rele(ds, dtag); 3559 return (error); 3560 } 3561 #endif 3562 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) { 3563 dsl_dataset_rele(ds, dtag); 3564 return (EBUSY); 3565 } else { 3566 own = B_TRUE; 3567 dsl_dataset_make_exclusive(ds, dtag); 3568 } 3569 } 3570 3571 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP); 3572 ra->ds = ds; 3573 ra->htag = ha->htag; 3574 ra->own = own; 3575 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check, 3576 dsl_dataset_user_release_sync, ra, dtag, 0); 3577 3578 return (0); 3579 } 3580 3581 int 3582 dsl_dataset_user_release(char *dsname, char *snapname, char *htag, 3583 boolean_t recursive) 3584 { 3585 struct dsl_ds_holdarg *ha; 3586 dsl_sync_task_t *dst; 3587 spa_t *spa; 3588 int error; 3589 3590 top: 3591 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); 3592 3593 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3594 3595 error = spa_open(dsname, &spa, FTAG); 3596 if (error) { 3597 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3598 return (error); 3599 } 3600 3601 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 3602 ha->htag = htag; 3603 ha->snapname = snapname; 3604 ha->recursive = recursive; 3605 if (recursive) { 3606 error = dmu_objset_find(dsname, dsl_dataset_user_release_one, 3607 ha, DS_FIND_CHILDREN); 3608 } else { 3609 error = dsl_dataset_user_release_one(dsname, ha); 3610 } 3611 if (error == 0) 3612 error = dsl_sync_task_group_wait(ha->dstg); 3613 3614 for (dst = list_head(&ha->dstg->dstg_tasks); dst; 3615 dst = list_next(&ha->dstg->dstg_tasks, dst)) { 3616 struct dsl_ds_releasearg *ra = dst->dst_arg1; 3617 dsl_dataset_t *ds = ra->ds; 3618 3619 if (dst->dst_err) 3620 dsl_dataset_name(ds, ha->failed); 3621 3622 if (ra->own) 3623 dsl_dataset_disown(ds, ha->dstg); 3624 else 3625 dsl_dataset_rele(ds, ha->dstg); 3626 3627 kmem_free(ra, sizeof (struct dsl_ds_releasearg)); 3628 } 3629 3630 if (error == 0 && recursive && !ha->gotone) 3631 error = ENOENT; 3632 3633 if (error && error != EBUSY) 3634 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed)); 3635 3636 dsl_sync_task_group_destroy(ha->dstg); 3637 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3638 spa_close(spa, FTAG); 3639 3640 /* 3641 * We can get EBUSY if we were racing with deferred destroy and 3642 * dsl_dataset_user_release_check() hadn't done the necessary 3643 * open context setup. We can also get EBUSY if we're racing 3644 * with destroy and that thread is the ds_owner. Either way 3645 * the busy condition should be transient, and we should retry 3646 * the release operation. 3647 */ 3648 if (error == EBUSY) 3649 goto top; 3650 3651 return (error); 3652 } 3653 3654 /* 3655 * Called at spa_load time to release a stale temporary user hold. 3656 */ 3657 int 3658 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag) 3659 { 3660 dsl_dataset_t *ds; 3661 char *snap; 3662 char *name; 3663 int namelen; 3664 int error; 3665 3666 rw_enter(&dp->dp_config_rwlock, RW_READER); 3667 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); 3668 rw_exit(&dp->dp_config_rwlock); 3669 if (error) 3670 return (error); 3671 namelen = dsl_dataset_namelen(ds)+1; 3672 name = kmem_alloc(namelen, KM_SLEEP); 3673 dsl_dataset_name(ds, name); 3674 dsl_dataset_rele(ds, FTAG); 3675 3676 snap = strchr(name, '@'); 3677 *snap = '\0'; 3678 ++snap; 3679 return (dsl_dataset_user_release(name, snap, htag, B_FALSE)); 3680 } 3681 3682 int 3683 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp) 3684 { 3685 dsl_dataset_t *ds; 3686 int err; 3687 3688 err = dsl_dataset_hold(dsname, FTAG, &ds); 3689 if (err) 3690 return (err); 3691 3692 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP)); 3693 if (ds->ds_phys->ds_userrefs_obj != 0) { 3694 zap_attribute_t *za; 3695 zap_cursor_t zc; 3696 3697 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 3698 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset, 3699 ds->ds_phys->ds_userrefs_obj); 3700 zap_cursor_retrieve(&zc, za) == 0; 3701 zap_cursor_advance(&zc)) { 3702 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name, 3703 za->za_first_integer)); 3704 } 3705 zap_cursor_fini(&zc); 3706 kmem_free(za, sizeof (zap_attribute_t)); 3707 } 3708 dsl_dataset_rele(ds, FTAG); 3709 return (0); 3710 } 3711 3712 /* 3713 * Note, this fuction is used as the callback for dmu_objset_find(). We 3714 * always return 0 so that we will continue to find and process 3715 * inconsistent datasets, even if we encounter an error trying to 3716 * process one of them. 3717 */ 3718 /* ARGSUSED */ 3719 int 3720 dsl_destroy_inconsistent(const char *dsname, void *arg) 3721 { 3722 dsl_dataset_t *ds; 3723 3724 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) { 3725 if (DS_IS_INCONSISTENT(ds)) 3726 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE); 3727 else 3728 dsl_dataset_disown(ds, FTAG); 3729 } 3730 return (0); 3731 } 3732