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 42 static char *dsl_reaper = "the grim reaper"; 43 44 static dsl_checkfunc_t dsl_dataset_destroy_begin_check; 45 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync; 46 static dsl_syncfunc_t dsl_dataset_set_reservation_sync; 47 48 #define DS_REF_MAX (1ULL << 62) 49 50 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE 51 52 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper) 53 54 55 /* 56 * Figure out how much of this delta should be propogated to the dsl_dir 57 * layer. If there's a refreservation, that space has already been 58 * partially accounted for in our ancestors. 59 */ 60 static int64_t 61 parent_delta(dsl_dataset_t *ds, int64_t delta) 62 { 63 uint64_t old_bytes, new_bytes; 64 65 if (ds->ds_reserved == 0) 66 return (delta); 67 68 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 69 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 70 71 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 72 return (new_bytes - old_bytes); 73 } 74 75 void 76 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 77 { 78 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 79 int compressed = BP_GET_PSIZE(bp); 80 int uncompressed = BP_GET_UCSIZE(bp); 81 int64_t delta; 82 83 dprintf_bp(bp, "born, ds=%p\n", ds); 84 85 ASSERT(dmu_tx_is_syncing(tx)); 86 /* It could have been compressed away to nothing */ 87 if (BP_IS_HOLE(bp)) 88 return; 89 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 90 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES); 91 if (ds == NULL) { 92 /* 93 * Account for the meta-objset space in its placeholder 94 * dsl_dir. 95 */ 96 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */ 97 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, 98 used, compressed, uncompressed, tx); 99 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); 100 return; 101 } 102 dmu_buf_will_dirty(ds->ds_dbuf, tx); 103 mutex_enter(&ds->ds_dir->dd_lock); 104 mutex_enter(&ds->ds_lock); 105 delta = parent_delta(ds, used); 106 ds->ds_phys->ds_used_bytes += used; 107 ds->ds_phys->ds_compressed_bytes += compressed; 108 ds->ds_phys->ds_uncompressed_bytes += uncompressed; 109 ds->ds_phys->ds_unique_bytes += used; 110 mutex_exit(&ds->ds_lock); 111 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 112 compressed, uncompressed, tx); 113 dsl_dir_transfer_space(ds->ds_dir, used - delta, 114 DD_USED_REFRSRV, DD_USED_HEAD, tx); 115 mutex_exit(&ds->ds_dir->dd_lock); 116 } 117 118 int 119 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 120 boolean_t async) 121 { 122 if (BP_IS_HOLE(bp)) 123 return (0); 124 125 ASSERT(dmu_tx_is_syncing(tx)); 126 ASSERT(bp->blk_birth <= tx->tx_txg); 127 128 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 129 int compressed = BP_GET_PSIZE(bp); 130 int uncompressed = BP_GET_UCSIZE(bp); 131 132 ASSERT(used > 0); 133 if (ds == NULL) { 134 /* 135 * Account for the meta-objset space in its placeholder 136 * dataset. 137 */ 138 dsl_free(tx->tx_pool, tx->tx_txg, bp); 139 140 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, 141 -used, -compressed, -uncompressed, tx); 142 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); 143 return (used); 144 } 145 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 146 147 ASSERT(!dsl_dataset_is_snapshot(ds)); 148 dmu_buf_will_dirty(ds->ds_dbuf, tx); 149 150 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { 151 int64_t delta; 152 153 dprintf_bp(bp, "freeing: %s", ""); 154 dsl_free(tx->tx_pool, tx->tx_txg, bp); 155 156 mutex_enter(&ds->ds_dir->dd_lock); 157 mutex_enter(&ds->ds_lock); 158 ASSERT(ds->ds_phys->ds_unique_bytes >= used || 159 !DS_UNIQUE_IS_ACCURATE(ds)); 160 delta = parent_delta(ds, -used); 161 ds->ds_phys->ds_unique_bytes -= used; 162 mutex_exit(&ds->ds_lock); 163 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 164 delta, -compressed, -uncompressed, tx); 165 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 166 DD_USED_REFRSRV, DD_USED_HEAD, tx); 167 mutex_exit(&ds->ds_dir->dd_lock); 168 } else { 169 dprintf_bp(bp, "putting on dead list: %s", ""); 170 if (async) { 171 /* 172 * We are here as part of zio's write done callback, 173 * which means we're a zio interrupt thread. We can't 174 * call bplist_enqueue() now because it may block 175 * waiting for I/O. Instead, put bp on the deferred 176 * queue and let dsl_pool_sync() finish the job. 177 */ 178 bplist_enqueue_deferred(&ds->ds_deadlist, bp); 179 } else { 180 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx)); 181 } 182 ASSERT3U(ds->ds_prev->ds_object, ==, 183 ds->ds_phys->ds_prev_snap_obj); 184 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); 185 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 186 if (ds->ds_prev->ds_phys->ds_next_snap_obj == 187 ds->ds_object && bp->blk_birth > 188 ds->ds_prev->ds_phys->ds_prev_snap_txg) { 189 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 190 mutex_enter(&ds->ds_prev->ds_lock); 191 ds->ds_prev->ds_phys->ds_unique_bytes += used; 192 mutex_exit(&ds->ds_prev->ds_lock); 193 } 194 if (bp->blk_birth > ds->ds_origin_txg) { 195 dsl_dir_transfer_space(ds->ds_dir, used, 196 DD_USED_HEAD, DD_USED_SNAP, tx); 197 } 198 } 199 mutex_enter(&ds->ds_lock); 200 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used); 201 ds->ds_phys->ds_used_bytes -= used; 202 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); 203 ds->ds_phys->ds_compressed_bytes -= compressed; 204 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); 205 ds->ds_phys->ds_uncompressed_bytes -= uncompressed; 206 mutex_exit(&ds->ds_lock); 207 208 return (used); 209 } 210 211 uint64_t 212 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 213 { 214 uint64_t trysnap = 0; 215 216 if (ds == NULL) 217 return (0); 218 /* 219 * The snapshot creation could fail, but that would cause an 220 * incorrect FALSE return, which would only result in an 221 * overestimation of the amount of space that an operation would 222 * consume, which is OK. 223 * 224 * There's also a small window where we could miss a pending 225 * snapshot, because we could set the sync task in the quiescing 226 * phase. So this should only be used as a guess. 227 */ 228 if (ds->ds_trysnap_txg > 229 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 230 trysnap = ds->ds_trysnap_txg; 231 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap)); 232 } 233 234 boolean_t 235 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth) 236 { 237 return (blk_birth > dsl_dataset_prev_snap_txg(ds)); 238 } 239 240 /* ARGSUSED */ 241 static void 242 dsl_dataset_evict(dmu_buf_t *db, void *dsv) 243 { 244 dsl_dataset_t *ds = dsv; 245 246 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds)); 247 248 unique_remove(ds->ds_fsid_guid); 249 250 if (ds->ds_objset != NULL) 251 dmu_objset_evict(ds->ds_objset); 252 253 if (ds->ds_prev) { 254 dsl_dataset_drop_ref(ds->ds_prev, ds); 255 ds->ds_prev = NULL; 256 } 257 258 bplist_close(&ds->ds_deadlist); 259 if (ds->ds_dir) 260 dsl_dir_close(ds->ds_dir, ds); 261 262 ASSERT(!list_link_active(&ds->ds_synced_link)); 263 264 mutex_destroy(&ds->ds_lock); 265 mutex_destroy(&ds->ds_recvlock); 266 mutex_destroy(&ds->ds_opening_lock); 267 rw_destroy(&ds->ds_rwlock); 268 cv_destroy(&ds->ds_exclusive_cv); 269 bplist_fini(&ds->ds_deadlist); 270 271 kmem_free(ds, sizeof (dsl_dataset_t)); 272 } 273 274 static int 275 dsl_dataset_get_snapname(dsl_dataset_t *ds) 276 { 277 dsl_dataset_phys_t *headphys; 278 int err; 279 dmu_buf_t *headdbuf; 280 dsl_pool_t *dp = ds->ds_dir->dd_pool; 281 objset_t *mos = dp->dp_meta_objset; 282 283 if (ds->ds_snapname[0]) 284 return (0); 285 if (ds->ds_phys->ds_next_snap_obj == 0) 286 return (0); 287 288 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, 289 FTAG, &headdbuf); 290 if (err) 291 return (err); 292 headphys = headdbuf->db_data; 293 err = zap_value_search(dp->dp_meta_objset, 294 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 295 dmu_buf_rele(headdbuf, FTAG); 296 return (err); 297 } 298 299 static int 300 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 301 { 302 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 303 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 304 matchtype_t mt; 305 int err; 306 307 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 308 mt = MT_FIRST; 309 else 310 mt = MT_EXACT; 311 312 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 313 value, mt, NULL, 0, NULL); 314 if (err == ENOTSUP && mt == MT_FIRST) 315 err = zap_lookup(mos, snapobj, name, 8, 1, value); 316 return (err); 317 } 318 319 static int 320 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx) 321 { 322 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 323 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 324 matchtype_t mt; 325 int err; 326 327 dsl_dir_snap_cmtime_update(ds->ds_dir); 328 329 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 330 mt = MT_FIRST; 331 else 332 mt = MT_EXACT; 333 334 err = zap_remove_norm(mos, snapobj, name, mt, tx); 335 if (err == ENOTSUP && mt == MT_FIRST) 336 err = zap_remove(mos, snapobj, name, tx); 337 return (err); 338 } 339 340 static int 341 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, 342 dsl_dataset_t **dsp) 343 { 344 objset_t *mos = dp->dp_meta_objset; 345 dmu_buf_t *dbuf; 346 dsl_dataset_t *ds; 347 int err; 348 349 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 350 dsl_pool_sync_context(dp)); 351 352 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 353 if (err) 354 return (err); 355 ds = dmu_buf_get_user(dbuf); 356 if (ds == NULL) { 357 dsl_dataset_t *winner; 358 359 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 360 ds->ds_dbuf = dbuf; 361 ds->ds_object = dsobj; 362 ds->ds_phys = dbuf->db_data; 363 364 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 365 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL); 366 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 367 rw_init(&ds->ds_rwlock, 0, 0, 0); 368 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL); 369 bplist_init(&ds->ds_deadlist); 370 371 err = bplist_open(&ds->ds_deadlist, 372 mos, ds->ds_phys->ds_deadlist_obj); 373 if (err == 0) { 374 err = dsl_dir_open_obj(dp, 375 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); 376 } 377 if (err) { 378 /* 379 * we don't really need to close the blist if we 380 * just opened it. 381 */ 382 mutex_destroy(&ds->ds_lock); 383 mutex_destroy(&ds->ds_recvlock); 384 mutex_destroy(&ds->ds_opening_lock); 385 rw_destroy(&ds->ds_rwlock); 386 cv_destroy(&ds->ds_exclusive_cv); 387 bplist_fini(&ds->ds_deadlist); 388 kmem_free(ds, sizeof (dsl_dataset_t)); 389 dmu_buf_rele(dbuf, tag); 390 return (err); 391 } 392 393 if (!dsl_dataset_is_snapshot(ds)) { 394 ds->ds_snapname[0] = '\0'; 395 if (ds->ds_phys->ds_prev_snap_obj) { 396 err = dsl_dataset_get_ref(dp, 397 ds->ds_phys->ds_prev_snap_obj, 398 ds, &ds->ds_prev); 399 } 400 401 if (err == 0 && dsl_dir_is_clone(ds->ds_dir)) { 402 dsl_dataset_t *origin; 403 404 err = dsl_dataset_hold_obj(dp, 405 ds->ds_dir->dd_phys->dd_origin_obj, 406 FTAG, &origin); 407 if (err == 0) { 408 ds->ds_origin_txg = 409 origin->ds_phys->ds_creation_txg; 410 dsl_dataset_rele(origin, FTAG); 411 } 412 } 413 } else { 414 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 415 err = dsl_dataset_get_snapname(ds); 416 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) { 417 err = zap_count( 418 ds->ds_dir->dd_pool->dp_meta_objset, 419 ds->ds_phys->ds_userrefs_obj, 420 &ds->ds_userrefs); 421 } 422 } 423 424 if (err == 0 && !dsl_dataset_is_snapshot(ds)) { 425 /* 426 * In sync context, we're called with either no lock 427 * or with the write lock. If we're not syncing, 428 * we're always called with the read lock held. 429 */ 430 boolean_t need_lock = 431 !RW_WRITE_HELD(&dp->dp_config_rwlock) && 432 dsl_pool_sync_context(dp); 433 434 if (need_lock) 435 rw_enter(&dp->dp_config_rwlock, RW_READER); 436 437 err = dsl_prop_get_ds(ds, 438 "refreservation", sizeof (uint64_t), 1, 439 &ds->ds_reserved, NULL); 440 if (err == 0) { 441 err = dsl_prop_get_ds(ds, 442 "refquota", sizeof (uint64_t), 1, 443 &ds->ds_quota, NULL); 444 } 445 446 if (need_lock) 447 rw_exit(&dp->dp_config_rwlock); 448 } else { 449 ds->ds_reserved = ds->ds_quota = 0; 450 } 451 452 if (err == 0) { 453 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys, 454 dsl_dataset_evict); 455 } 456 if (err || winner) { 457 bplist_close(&ds->ds_deadlist); 458 if (ds->ds_prev) 459 dsl_dataset_drop_ref(ds->ds_prev, ds); 460 dsl_dir_close(ds->ds_dir, ds); 461 mutex_destroy(&ds->ds_lock); 462 mutex_destroy(&ds->ds_recvlock); 463 mutex_destroy(&ds->ds_opening_lock); 464 rw_destroy(&ds->ds_rwlock); 465 cv_destroy(&ds->ds_exclusive_cv); 466 bplist_fini(&ds->ds_deadlist); 467 kmem_free(ds, sizeof (dsl_dataset_t)); 468 if (err) { 469 dmu_buf_rele(dbuf, tag); 470 return (err); 471 } 472 ds = winner; 473 } else { 474 ds->ds_fsid_guid = 475 unique_insert(ds->ds_phys->ds_fsid_guid); 476 } 477 } 478 ASSERT3P(ds->ds_dbuf, ==, dbuf); 479 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); 480 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 || 481 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 482 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 483 mutex_enter(&ds->ds_lock); 484 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) { 485 mutex_exit(&ds->ds_lock); 486 dmu_buf_rele(ds->ds_dbuf, tag); 487 return (ENOENT); 488 } 489 mutex_exit(&ds->ds_lock); 490 *dsp = ds; 491 return (0); 492 } 493 494 static int 495 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag) 496 { 497 dsl_pool_t *dp = ds->ds_dir->dd_pool; 498 499 /* 500 * In syncing context we don't want the rwlock lock: there 501 * may be an existing writer waiting for sync phase to 502 * finish. We don't need to worry about such writers, since 503 * sync phase is single-threaded, so the writer can't be 504 * doing anything while we are active. 505 */ 506 if (dsl_pool_sync_context(dp)) { 507 ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); 508 return (0); 509 } 510 511 /* 512 * Normal users will hold the ds_rwlock as a READER until they 513 * are finished (i.e., call dsl_dataset_rele()). "Owners" will 514 * drop their READER lock after they set the ds_owner field. 515 * 516 * If the dataset is being destroyed, the destroy thread will 517 * obtain a WRITER lock for exclusive access after it's done its 518 * open-context work and then change the ds_owner to 519 * dsl_reaper once destruction is assured. So threads 520 * may block here temporarily, until the "destructability" of 521 * the dataset is determined. 522 */ 523 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock)); 524 mutex_enter(&ds->ds_lock); 525 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) { 526 rw_exit(&dp->dp_config_rwlock); 527 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock); 528 if (DSL_DATASET_IS_DESTROYED(ds)) { 529 mutex_exit(&ds->ds_lock); 530 dsl_dataset_drop_ref(ds, tag); 531 rw_enter(&dp->dp_config_rwlock, RW_READER); 532 return (ENOENT); 533 } 534 /* 535 * The dp_config_rwlock lives above the ds_lock. And 536 * we need to check DSL_DATASET_IS_DESTROYED() while 537 * holding the ds_lock, so we have to drop and reacquire 538 * the ds_lock here. 539 */ 540 mutex_exit(&ds->ds_lock); 541 rw_enter(&dp->dp_config_rwlock, RW_READER); 542 mutex_enter(&ds->ds_lock); 543 } 544 mutex_exit(&ds->ds_lock); 545 return (0); 546 } 547 548 int 549 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 550 dsl_dataset_t **dsp) 551 { 552 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp); 553 554 if (err) 555 return (err); 556 return (dsl_dataset_hold_ref(*dsp, tag)); 557 } 558 559 int 560 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok, 561 void *tag, dsl_dataset_t **dsp) 562 { 563 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 564 if (err) 565 return (err); 566 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { 567 dsl_dataset_rele(*dsp, tag); 568 *dsp = NULL; 569 return (EBUSY); 570 } 571 return (0); 572 } 573 574 int 575 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp) 576 { 577 dsl_dir_t *dd; 578 dsl_pool_t *dp; 579 const char *snapname; 580 uint64_t obj; 581 int err = 0; 582 583 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname); 584 if (err) 585 return (err); 586 587 dp = dd->dd_pool; 588 obj = dd->dd_phys->dd_head_dataset_obj; 589 rw_enter(&dp->dp_config_rwlock, RW_READER); 590 if (obj) 591 err = dsl_dataset_get_ref(dp, obj, tag, dsp); 592 else 593 err = ENOENT; 594 if (err) 595 goto out; 596 597 err = dsl_dataset_hold_ref(*dsp, tag); 598 599 /* we may be looking for a snapshot */ 600 if (err == 0 && snapname != NULL) { 601 dsl_dataset_t *ds = NULL; 602 603 if (*snapname++ != '@') { 604 dsl_dataset_rele(*dsp, tag); 605 err = ENOENT; 606 goto out; 607 } 608 609 dprintf("looking for snapshot '%s'\n", snapname); 610 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); 611 if (err == 0) 612 err = dsl_dataset_get_ref(dp, obj, tag, &ds); 613 dsl_dataset_rele(*dsp, tag); 614 615 ASSERT3U((err == 0), ==, (ds != NULL)); 616 617 if (ds) { 618 mutex_enter(&ds->ds_lock); 619 if (ds->ds_snapname[0] == 0) 620 (void) strlcpy(ds->ds_snapname, snapname, 621 sizeof (ds->ds_snapname)); 622 mutex_exit(&ds->ds_lock); 623 err = dsl_dataset_hold_ref(ds, tag); 624 *dsp = err ? NULL : ds; 625 } 626 } 627 out: 628 rw_exit(&dp->dp_config_rwlock); 629 dsl_dir_close(dd, FTAG); 630 return (err); 631 } 632 633 int 634 dsl_dataset_own(const char *name, boolean_t inconsistentok, 635 void *tag, dsl_dataset_t **dsp) 636 { 637 int err = dsl_dataset_hold(name, tag, dsp); 638 if (err) 639 return (err); 640 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { 641 dsl_dataset_rele(*dsp, tag); 642 return (EBUSY); 643 } 644 return (0); 645 } 646 647 void 648 dsl_dataset_name(dsl_dataset_t *ds, char *name) 649 { 650 if (ds == NULL) { 651 (void) strcpy(name, "mos"); 652 } else { 653 dsl_dir_name(ds->ds_dir, name); 654 VERIFY(0 == dsl_dataset_get_snapname(ds)); 655 if (ds->ds_snapname[0]) { 656 (void) strcat(name, "@"); 657 /* 658 * We use a "recursive" mutex so that we 659 * can call dprintf_ds() with ds_lock held. 660 */ 661 if (!MUTEX_HELD(&ds->ds_lock)) { 662 mutex_enter(&ds->ds_lock); 663 (void) strcat(name, ds->ds_snapname); 664 mutex_exit(&ds->ds_lock); 665 } else { 666 (void) strcat(name, ds->ds_snapname); 667 } 668 } 669 } 670 } 671 672 static int 673 dsl_dataset_namelen(dsl_dataset_t *ds) 674 { 675 int result; 676 677 if (ds == NULL) { 678 result = 3; /* "mos" */ 679 } else { 680 result = dsl_dir_namelen(ds->ds_dir); 681 VERIFY(0 == dsl_dataset_get_snapname(ds)); 682 if (ds->ds_snapname[0]) { 683 ++result; /* adding one for the @-sign */ 684 if (!MUTEX_HELD(&ds->ds_lock)) { 685 mutex_enter(&ds->ds_lock); 686 result += strlen(ds->ds_snapname); 687 mutex_exit(&ds->ds_lock); 688 } else { 689 result += strlen(ds->ds_snapname); 690 } 691 } 692 } 693 694 return (result); 695 } 696 697 void 698 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag) 699 { 700 dmu_buf_rele(ds->ds_dbuf, tag); 701 } 702 703 void 704 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 705 { 706 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) { 707 rw_exit(&ds->ds_rwlock); 708 } 709 dsl_dataset_drop_ref(ds, tag); 710 } 711 712 void 713 dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 714 { 715 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) || 716 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL)); 717 718 mutex_enter(&ds->ds_lock); 719 ds->ds_owner = NULL; 720 if (RW_WRITE_HELD(&ds->ds_rwlock)) { 721 rw_exit(&ds->ds_rwlock); 722 cv_broadcast(&ds->ds_exclusive_cv); 723 } 724 mutex_exit(&ds->ds_lock); 725 if (ds->ds_dbuf) 726 dsl_dataset_drop_ref(ds, tag); 727 else 728 dsl_dataset_evict(ds->ds_dbuf, ds); 729 } 730 731 boolean_t 732 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag) 733 { 734 boolean_t gotit = FALSE; 735 736 mutex_enter(&ds->ds_lock); 737 if (ds->ds_owner == NULL && 738 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) { 739 ds->ds_owner = tag; 740 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) 741 rw_exit(&ds->ds_rwlock); 742 gotit = TRUE; 743 } 744 mutex_exit(&ds->ds_lock); 745 return (gotit); 746 } 747 748 void 749 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner) 750 { 751 ASSERT3P(owner, ==, ds->ds_owner); 752 if (!RW_WRITE_HELD(&ds->ds_rwlock)) 753 rw_enter(&ds->ds_rwlock, RW_WRITER); 754 } 755 756 uint64_t 757 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 758 uint64_t flags, dmu_tx_t *tx) 759 { 760 dsl_pool_t *dp = dd->dd_pool; 761 dmu_buf_t *dbuf; 762 dsl_dataset_phys_t *dsphys; 763 uint64_t dsobj; 764 objset_t *mos = dp->dp_meta_objset; 765 766 if (origin == NULL) 767 origin = dp->dp_origin_snap; 768 769 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 770 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0); 771 ASSERT(dmu_tx_is_syncing(tx)); 772 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 773 774 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 775 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 776 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 777 dmu_buf_will_dirty(dbuf, tx); 778 dsphys = dbuf->db_data; 779 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 780 dsphys->ds_dir_obj = dd->dd_object; 781 dsphys->ds_flags = flags; 782 dsphys->ds_fsid_guid = unique_create(); 783 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 784 sizeof (dsphys->ds_guid)); 785 dsphys->ds_snapnames_zapobj = 786 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 787 DMU_OT_NONE, 0, tx); 788 dsphys->ds_creation_time = gethrestime_sec(); 789 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 790 dsphys->ds_deadlist_obj = 791 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); 792 793 if (origin) { 794 dsphys->ds_prev_snap_obj = origin->ds_object; 795 dsphys->ds_prev_snap_txg = 796 origin->ds_phys->ds_creation_txg; 797 dsphys->ds_used_bytes = 798 origin->ds_phys->ds_used_bytes; 799 dsphys->ds_compressed_bytes = 800 origin->ds_phys->ds_compressed_bytes; 801 dsphys->ds_uncompressed_bytes = 802 origin->ds_phys->ds_uncompressed_bytes; 803 dsphys->ds_bp = origin->ds_phys->ds_bp; 804 dsphys->ds_flags |= origin->ds_phys->ds_flags; 805 806 dmu_buf_will_dirty(origin->ds_dbuf, tx); 807 origin->ds_phys->ds_num_children++; 808 809 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 810 if (origin->ds_phys->ds_next_clones_obj == 0) { 811 origin->ds_phys->ds_next_clones_obj = 812 zap_create(mos, 813 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 814 } 815 VERIFY(0 == zap_add_int(mos, 816 origin->ds_phys->ds_next_clones_obj, 817 dsobj, tx)); 818 } 819 820 dmu_buf_will_dirty(dd->dd_dbuf, tx); 821 dd->dd_phys->dd_origin_obj = origin->ds_object; 822 } 823 824 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 825 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 826 827 dmu_buf_rele(dbuf, FTAG); 828 829 dmu_buf_will_dirty(dd->dd_dbuf, tx); 830 dd->dd_phys->dd_head_dataset_obj = dsobj; 831 832 return (dsobj); 833 } 834 835 uint64_t 836 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 837 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 838 { 839 dsl_pool_t *dp = pdd->dd_pool; 840 uint64_t dsobj, ddobj; 841 dsl_dir_t *dd; 842 843 ASSERT(lastname[0] != '@'); 844 845 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 846 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd)); 847 848 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx); 849 850 dsl_deleg_set_create_perms(dd, tx, cr); 851 852 dsl_dir_close(dd, FTAG); 853 854 return (dsobj); 855 } 856 857 struct destroyarg { 858 dsl_sync_task_group_t *dstg; 859 char *snapname; 860 char *failed; 861 boolean_t defer; 862 }; 863 864 static int 865 dsl_snapshot_destroy_one(const char *name, void *arg) 866 { 867 struct destroyarg *da = arg; 868 dsl_dataset_t *ds; 869 int err; 870 char *dsname; 871 872 dsname = kmem_asprintf("%s@%s", name, da->snapname); 873 err = dsl_dataset_own(dsname, B_TRUE, da->dstg, &ds); 874 strfree(dsname); 875 if (err == 0) { 876 struct dsl_ds_destroyarg *dsda; 877 878 dsl_dataset_make_exclusive(ds, da->dstg); 879 if (ds->ds_objset != NULL) { 880 dmu_objset_evict(ds->ds_objset); 881 ds->ds_objset = NULL; 882 } 883 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP); 884 dsda->ds = ds; 885 dsda->defer = da->defer; 886 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check, 887 dsl_dataset_destroy_sync, dsda, da->dstg, 0); 888 } else if (err == ENOENT) { 889 err = 0; 890 } else { 891 (void) strcpy(da->failed, name); 892 } 893 return (err); 894 } 895 896 /* 897 * Destroy 'snapname' in all descendants of 'fsname'. 898 */ 899 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy 900 int 901 dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer) 902 { 903 int err; 904 struct destroyarg da; 905 dsl_sync_task_t *dst; 906 spa_t *spa; 907 908 err = spa_open(fsname, &spa, FTAG); 909 if (err) 910 return (err); 911 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 912 da.snapname = snapname; 913 da.failed = fsname; 914 da.defer = defer; 915 916 err = dmu_objset_find(fsname, 917 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN); 918 919 if (err == 0) 920 err = dsl_sync_task_group_wait(da.dstg); 921 922 for (dst = list_head(&da.dstg->dstg_tasks); dst; 923 dst = list_next(&da.dstg->dstg_tasks, dst)) { 924 struct dsl_ds_destroyarg *dsda = dst->dst_arg1; 925 dsl_dataset_t *ds = dsda->ds; 926 927 /* 928 * Return the file system name that triggered the error 929 */ 930 if (dst->dst_err) { 931 dsl_dataset_name(ds, fsname); 932 *strchr(fsname, '@') = '\0'; 933 } 934 ASSERT3P(dsda->rm_origin, ==, NULL); 935 dsl_dataset_disown(ds, da.dstg); 936 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg)); 937 } 938 939 dsl_sync_task_group_destroy(da.dstg); 940 spa_close(spa, FTAG); 941 return (err); 942 } 943 944 static boolean_t 945 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds) 946 { 947 boolean_t might_destroy = B_FALSE; 948 949 mutex_enter(&ds->ds_lock); 950 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 && 951 DS_IS_DEFER_DESTROY(ds)) 952 might_destroy = B_TRUE; 953 mutex_exit(&ds->ds_lock); 954 955 return (might_destroy); 956 } 957 958 /* 959 * If we're removing a clone, and these three conditions are true: 960 * 1) the clone's origin has no other children 961 * 2) the clone's origin has no user references 962 * 3) the clone's origin has been marked for deferred destruction 963 * Then, prepare to remove the origin as part of this sync task group. 964 */ 965 static int 966 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag) 967 { 968 dsl_dataset_t *ds = dsda->ds; 969 dsl_dataset_t *origin = ds->ds_prev; 970 971 if (dsl_dataset_might_destroy_origin(origin)) { 972 char *name; 973 int namelen; 974 int error; 975 976 namelen = dsl_dataset_namelen(origin) + 1; 977 name = kmem_alloc(namelen, KM_SLEEP); 978 dsl_dataset_name(origin, name); 979 #ifdef _KERNEL 980 error = zfs_unmount_snap(name, NULL); 981 if (error) { 982 kmem_free(name, namelen); 983 return (error); 984 } 985 #endif 986 error = dsl_dataset_own(name, B_TRUE, tag, &origin); 987 kmem_free(name, namelen); 988 if (error) 989 return (error); 990 dsda->rm_origin = origin; 991 dsl_dataset_make_exclusive(origin, tag); 992 993 if (origin->ds_objset != NULL) { 994 dmu_objset_evict(origin->ds_objset); 995 origin->ds_objset = NULL; 996 } 997 } 998 999 return (0); 1000 } 1001 1002 /* 1003 * ds must be opened as OWNER. On return (whether successful or not), 1004 * ds will be closed and caller can no longer dereference it. 1005 */ 1006 int 1007 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer) 1008 { 1009 int err; 1010 dsl_sync_task_group_t *dstg; 1011 objset_t *os; 1012 dsl_dir_t *dd; 1013 uint64_t obj; 1014 struct dsl_ds_destroyarg dsda = { 0 }; 1015 dsl_dataset_t dummy_ds = { 0 }; 1016 1017 dsda.ds = ds; 1018 1019 if (dsl_dataset_is_snapshot(ds)) { 1020 /* Destroying a snapshot is simpler */ 1021 dsl_dataset_make_exclusive(ds, tag); 1022 1023 if (ds->ds_objset != NULL) { 1024 dmu_objset_evict(ds->ds_objset); 1025 ds->ds_objset = NULL; 1026 } 1027 dsda.defer = defer; 1028 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 1029 dsl_dataset_destroy_check, dsl_dataset_destroy_sync, 1030 &dsda, tag, 0); 1031 ASSERT3P(dsda.rm_origin, ==, NULL); 1032 goto out; 1033 } else if (defer) { 1034 err = EINVAL; 1035 goto out; 1036 } 1037 1038 dd = ds->ds_dir; 1039 dummy_ds.ds_dir = dd; 1040 dummy_ds.ds_object = ds->ds_object; 1041 1042 /* 1043 * Check for errors and mark this ds as inconsistent, in 1044 * case we crash while freeing the objects. 1045 */ 1046 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check, 1047 dsl_dataset_destroy_begin_sync, ds, NULL, 0); 1048 if (err) 1049 goto out; 1050 1051 err = dmu_objset_from_ds(ds, &os); 1052 if (err) 1053 goto out; 1054 1055 /* 1056 * remove the objects in open context, so that we won't 1057 * have too much to do in syncing context. 1058 */ 1059 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 1060 ds->ds_phys->ds_prev_snap_txg)) { 1061 /* 1062 * Ignore errors, if there is not enough disk space 1063 * we will deal with it in dsl_dataset_destroy_sync(). 1064 */ 1065 (void) dmu_free_object(os, obj); 1066 } 1067 1068 /* 1069 * We need to sync out all in-flight IO before we try to evict 1070 * (the dataset evict func is trying to clear the cached entries 1071 * for this dataset in the ARC). 1072 */ 1073 txg_wait_synced(dd->dd_pool, 0); 1074 1075 /* 1076 * If we managed to free all the objects in open 1077 * context, the user space accounting should be zero. 1078 */ 1079 if (ds->ds_phys->ds_bp.blk_fill == 0 && 1080 dmu_objset_userused_enabled(os)) { 1081 uint64_t count; 1082 1083 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 || 1084 count == 0); 1085 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 || 1086 count == 0); 1087 } 1088 1089 if (err != ESRCH) 1090 goto out; 1091 1092 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); 1093 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd); 1094 rw_exit(&dd->dd_pool->dp_config_rwlock); 1095 1096 if (err) 1097 goto out; 1098 1099 if (ds->ds_objset) { 1100 /* 1101 * We need to sync out all in-flight IO before we try 1102 * to evict (the dataset evict func is trying to clear 1103 * the cached entries for this dataset in the ARC). 1104 */ 1105 txg_wait_synced(dd->dd_pool, 0); 1106 } 1107 1108 /* 1109 * Blow away the dsl_dir + head dataset. 1110 */ 1111 dsl_dataset_make_exclusive(ds, tag); 1112 if (ds->ds_objset) { 1113 dmu_objset_evict(ds->ds_objset); 1114 ds->ds_objset = NULL; 1115 } 1116 1117 /* 1118 * If we're removing a clone, we might also need to remove its 1119 * origin. 1120 */ 1121 do { 1122 dsda.need_prep = B_FALSE; 1123 if (dsl_dir_is_clone(dd)) { 1124 err = dsl_dataset_origin_rm_prep(&dsda, tag); 1125 if (err) { 1126 dsl_dir_close(dd, FTAG); 1127 goto out; 1128 } 1129 } 1130 1131 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool); 1132 dsl_sync_task_create(dstg, dsl_dataset_destroy_check, 1133 dsl_dataset_destroy_sync, &dsda, tag, 0); 1134 dsl_sync_task_create(dstg, dsl_dir_destroy_check, 1135 dsl_dir_destroy_sync, &dummy_ds, FTAG, 0); 1136 err = dsl_sync_task_group_wait(dstg); 1137 dsl_sync_task_group_destroy(dstg); 1138 1139 /* 1140 * We could be racing against 'zfs release' or 'zfs destroy -d' 1141 * on the origin snap, in which case we can get EBUSY if we 1142 * needed to destroy the origin snap but were not ready to 1143 * do so. 1144 */ 1145 if (dsda.need_prep) { 1146 ASSERT(err == EBUSY); 1147 ASSERT(dsl_dir_is_clone(dd)); 1148 ASSERT(dsda.rm_origin == NULL); 1149 } 1150 } while (dsda.need_prep); 1151 1152 if (dsda.rm_origin != NULL) 1153 dsl_dataset_disown(dsda.rm_origin, tag); 1154 1155 /* if it is successful, dsl_dir_destroy_sync will close the dd */ 1156 if (err) 1157 dsl_dir_close(dd, FTAG); 1158 out: 1159 dsl_dataset_disown(ds, tag); 1160 return (err); 1161 } 1162 1163 blkptr_t * 1164 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1165 { 1166 return (&ds->ds_phys->ds_bp); 1167 } 1168 1169 void 1170 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 1171 { 1172 ASSERT(dmu_tx_is_syncing(tx)); 1173 /* If it's the meta-objset, set dp_meta_rootbp */ 1174 if (ds == NULL) { 1175 tx->tx_pool->dp_meta_rootbp = *bp; 1176 } else { 1177 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1178 ds->ds_phys->ds_bp = *bp; 1179 } 1180 } 1181 1182 spa_t * 1183 dsl_dataset_get_spa(dsl_dataset_t *ds) 1184 { 1185 return (ds->ds_dir->dd_pool->dp_spa); 1186 } 1187 1188 void 1189 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1190 { 1191 dsl_pool_t *dp; 1192 1193 if (ds == NULL) /* this is the meta-objset */ 1194 return; 1195 1196 ASSERT(ds->ds_objset != NULL); 1197 1198 if (ds->ds_phys->ds_next_snap_obj != 0) 1199 panic("dirtying snapshot!"); 1200 1201 dp = ds->ds_dir->dd_pool; 1202 1203 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) { 1204 /* up the hold count until we can be written out */ 1205 dmu_buf_add_ref(ds->ds_dbuf, ds); 1206 } 1207 } 1208 1209 /* 1210 * The unique space in the head dataset can be calculated by subtracting 1211 * the space used in the most recent snapshot, that is still being used 1212 * in this file system, from the space currently in use. To figure out 1213 * the space in the most recent snapshot still in use, we need to take 1214 * the total space used in the snapshot and subtract out the space that 1215 * has been freed up since the snapshot was taken. 1216 */ 1217 static void 1218 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 1219 { 1220 uint64_t mrs_used; 1221 uint64_t dlused, dlcomp, dluncomp; 1222 1223 ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj); 1224 1225 if (ds->ds_phys->ds_prev_snap_obj != 0) 1226 mrs_used = ds->ds_prev->ds_phys->ds_used_bytes; 1227 else 1228 mrs_used = 0; 1229 1230 VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp, 1231 &dluncomp)); 1232 1233 ASSERT3U(dlused, <=, mrs_used); 1234 ds->ds_phys->ds_unique_bytes = 1235 ds->ds_phys->ds_used_bytes - (mrs_used - dlused); 1236 1237 if (!DS_UNIQUE_IS_ACCURATE(ds) && 1238 spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1239 SPA_VERSION_UNIQUE_ACCURATE) 1240 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1241 } 1242 1243 static uint64_t 1244 dsl_dataset_unique(dsl_dataset_t *ds) 1245 { 1246 if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds)) 1247 dsl_dataset_recalc_head_uniq(ds); 1248 1249 return (ds->ds_phys->ds_unique_bytes); 1250 } 1251 1252 struct killarg { 1253 dsl_dataset_t *ds; 1254 dmu_tx_t *tx; 1255 }; 1256 1257 /* ARGSUSED */ 1258 static int 1259 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1260 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 1261 { 1262 struct killarg *ka = arg; 1263 dmu_tx_t *tx = ka->tx; 1264 1265 if (bp == NULL) 1266 return (0); 1267 1268 if (zb->zb_level == ZB_ZIL_LEVEL) { 1269 ASSERT(zilog != NULL); 1270 /* 1271 * It's a block in the intent log. It has no 1272 * accounting, so just free it. 1273 */ 1274 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp); 1275 } else { 1276 ASSERT(zilog == NULL); 1277 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg); 1278 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE); 1279 } 1280 1281 return (0); 1282 } 1283 1284 /* ARGSUSED */ 1285 static int 1286 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx) 1287 { 1288 dsl_dataset_t *ds = arg1; 1289 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1290 uint64_t count; 1291 int err; 1292 1293 /* 1294 * Can't delete a head dataset if there are snapshots of it. 1295 * (Except if the only snapshots are from the branch we cloned 1296 * from.) 1297 */ 1298 if (ds->ds_prev != NULL && 1299 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) 1300 return (EBUSY); 1301 1302 /* 1303 * This is really a dsl_dir thing, but check it here so that 1304 * we'll be less likely to leave this dataset inconsistent & 1305 * nearly destroyed. 1306 */ 1307 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count); 1308 if (err) 1309 return (err); 1310 if (count != 0) 1311 return (EEXIST); 1312 1313 return (0); 1314 } 1315 1316 /* ARGSUSED */ 1317 static void 1318 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 1319 { 1320 dsl_dataset_t *ds = arg1; 1321 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1322 1323 /* Mark it as inconsistent on-disk, in case we crash */ 1324 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1325 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; 1326 1327 spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx, 1328 cr, "dataset = %llu", ds->ds_object); 1329 } 1330 1331 static int 1332 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag, 1333 dmu_tx_t *tx) 1334 { 1335 dsl_dataset_t *ds = dsda->ds; 1336 dsl_dataset_t *ds_prev = ds->ds_prev; 1337 1338 if (dsl_dataset_might_destroy_origin(ds_prev)) { 1339 struct dsl_ds_destroyarg ndsda = {0}; 1340 1341 /* 1342 * If we're not prepared to remove the origin, don't remove 1343 * the clone either. 1344 */ 1345 if (dsda->rm_origin == NULL) { 1346 dsda->need_prep = B_TRUE; 1347 return (EBUSY); 1348 } 1349 1350 ndsda.ds = ds_prev; 1351 ndsda.is_origin_rm = B_TRUE; 1352 return (dsl_dataset_destroy_check(&ndsda, tag, tx)); 1353 } 1354 1355 /* 1356 * If we're not going to remove the origin after all, 1357 * undo the open context setup. 1358 */ 1359 if (dsda->rm_origin != NULL) { 1360 dsl_dataset_disown(dsda->rm_origin, tag); 1361 dsda->rm_origin = NULL; 1362 } 1363 1364 return (0); 1365 } 1366 1367 /* ARGSUSED */ 1368 int 1369 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) 1370 { 1371 struct dsl_ds_destroyarg *dsda = arg1; 1372 dsl_dataset_t *ds = dsda->ds; 1373 1374 /* we have an owner hold, so noone else can destroy us */ 1375 ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); 1376 1377 /* 1378 * Only allow deferred destroy on pools that support it. 1379 * NOTE: deferred destroy is only supported on snapshots. 1380 */ 1381 if (dsda->defer) { 1382 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < 1383 SPA_VERSION_USERREFS) 1384 return (ENOTSUP); 1385 ASSERT(dsl_dataset_is_snapshot(ds)); 1386 return (0); 1387 } 1388 1389 /* 1390 * Can't delete a head dataset if there are snapshots of it. 1391 * (Except if the only snapshots are from the branch we cloned 1392 * from.) 1393 */ 1394 if (ds->ds_prev != NULL && 1395 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) 1396 return (EBUSY); 1397 1398 /* 1399 * If we made changes this txg, traverse_dsl_dataset won't find 1400 * them. Try again. 1401 */ 1402 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) 1403 return (EAGAIN); 1404 1405 if (dsl_dataset_is_snapshot(ds)) { 1406 /* 1407 * If this snapshot has an elevated user reference count, 1408 * we can't destroy it yet. 1409 */ 1410 if (ds->ds_userrefs > 0 && !dsda->releasing) 1411 return (EBUSY); 1412 1413 mutex_enter(&ds->ds_lock); 1414 /* 1415 * Can't delete a branch point. However, if we're destroying 1416 * a clone and removing its origin due to it having a user 1417 * hold count of 0 and having been marked for deferred destroy, 1418 * it's OK for the origin to have a single clone. 1419 */ 1420 if (ds->ds_phys->ds_num_children > 1421 (dsda->is_origin_rm ? 2 : 1)) { 1422 mutex_exit(&ds->ds_lock); 1423 return (EEXIST); 1424 } 1425 mutex_exit(&ds->ds_lock); 1426 } else if (dsl_dir_is_clone(ds->ds_dir)) { 1427 return (dsl_dataset_origin_check(dsda, arg2, tx)); 1428 } 1429 1430 /* XXX we should do some i/o error checking... */ 1431 return (0); 1432 } 1433 1434 struct refsarg { 1435 kmutex_t lock; 1436 boolean_t gone; 1437 kcondvar_t cv; 1438 }; 1439 1440 /* ARGSUSED */ 1441 static void 1442 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv) 1443 { 1444 struct refsarg *arg = argv; 1445 1446 mutex_enter(&arg->lock); 1447 arg->gone = TRUE; 1448 cv_signal(&arg->cv); 1449 mutex_exit(&arg->lock); 1450 } 1451 1452 static void 1453 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag) 1454 { 1455 struct refsarg arg; 1456 1457 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL); 1458 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL); 1459 arg.gone = FALSE; 1460 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys, 1461 dsl_dataset_refs_gone); 1462 dmu_buf_rele(ds->ds_dbuf, tag); 1463 mutex_enter(&arg.lock); 1464 while (!arg.gone) 1465 cv_wait(&arg.cv, &arg.lock); 1466 ASSERT(arg.gone); 1467 mutex_exit(&arg.lock); 1468 ds->ds_dbuf = NULL; 1469 ds->ds_phys = NULL; 1470 mutex_destroy(&arg.lock); 1471 cv_destroy(&arg.cv); 1472 } 1473 1474 static void 1475 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx) 1476 { 1477 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1478 uint64_t count; 1479 int err; 1480 1481 ASSERT(ds->ds_phys->ds_num_children >= 2); 1482 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx); 1483 /* 1484 * The err should not be ENOENT, but a bug in a previous version 1485 * of the code could cause upgrade_clones_cb() to not set 1486 * ds_next_snap_obj when it should, leading to a missing entry. 1487 * If we knew that the pool was created after 1488 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 1489 * ENOENT. However, at least we can check that we don't have 1490 * too many entries in the next_clones_obj even after failing to 1491 * remove this one. 1492 */ 1493 if (err != ENOENT) { 1494 VERIFY3U(err, ==, 0); 1495 } 1496 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj, 1497 &count)); 1498 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2); 1499 } 1500 1501 void 1502 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx) 1503 { 1504 struct dsl_ds_destroyarg *dsda = arg1; 1505 dsl_dataset_t *ds = dsda->ds; 1506 int err; 1507 int after_branch_point = FALSE; 1508 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1509 objset_t *mos = dp->dp_meta_objset; 1510 dsl_dataset_t *ds_prev = NULL; 1511 uint64_t obj; 1512 1513 ASSERT(ds->ds_owner); 1514 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1); 1515 ASSERT(ds->ds_prev == NULL || 1516 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object); 1517 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); 1518 1519 if (dsda->defer) { 1520 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 1521 if (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1) { 1522 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1523 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY; 1524 return; 1525 } 1526 } 1527 1528 /* signal any waiters that this dataset is going away */ 1529 mutex_enter(&ds->ds_lock); 1530 ds->ds_owner = dsl_reaper; 1531 cv_broadcast(&ds->ds_exclusive_cv); 1532 mutex_exit(&ds->ds_lock); 1533 1534 /* Remove our reservation */ 1535 if (ds->ds_reserved != 0) { 1536 dsl_prop_setarg_t psa; 1537 uint64_t value = 0; 1538 1539 dsl_prop_setarg_init_uint64(&psa, "refreservation", 1540 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED), 1541 &value); 1542 psa.psa_effective_value = 0; /* predict default value */ 1543 1544 dsl_dataset_set_reservation_sync(ds, &psa, cr, tx); 1545 ASSERT3U(ds->ds_reserved, ==, 0); 1546 } 1547 1548 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); 1549 1550 dsl_pool_ds_destroyed(ds, tx); 1551 1552 obj = ds->ds_object; 1553 1554 if (ds->ds_phys->ds_prev_snap_obj != 0) { 1555 if (ds->ds_prev) { 1556 ds_prev = ds->ds_prev; 1557 } else { 1558 VERIFY(0 == dsl_dataset_hold_obj(dp, 1559 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev)); 1560 } 1561 after_branch_point = 1562 (ds_prev->ds_phys->ds_next_snap_obj != obj); 1563 1564 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); 1565 if (after_branch_point && 1566 ds_prev->ds_phys->ds_next_clones_obj != 0) { 1567 remove_from_next_clones(ds_prev, obj, tx); 1568 if (ds->ds_phys->ds_next_snap_obj != 0) { 1569 VERIFY(0 == zap_add_int(mos, 1570 ds_prev->ds_phys->ds_next_clones_obj, 1571 ds->ds_phys->ds_next_snap_obj, tx)); 1572 } 1573 } 1574 if (after_branch_point && 1575 ds->ds_phys->ds_next_snap_obj == 0) { 1576 /* This clone is toast. */ 1577 ASSERT(ds_prev->ds_phys->ds_num_children > 1); 1578 ds_prev->ds_phys->ds_num_children--; 1579 1580 /* 1581 * If the clone's origin has no other clones, no 1582 * user holds, and has been marked for deferred 1583 * deletion, then we should have done the necessary 1584 * destroy setup for it. 1585 */ 1586 if (ds_prev->ds_phys->ds_num_children == 1 && 1587 ds_prev->ds_userrefs == 0 && 1588 DS_IS_DEFER_DESTROY(ds_prev)) { 1589 ASSERT3P(dsda->rm_origin, !=, NULL); 1590 } else { 1591 ASSERT3P(dsda->rm_origin, ==, NULL); 1592 } 1593 } else if (!after_branch_point) { 1594 ds_prev->ds_phys->ds_next_snap_obj = 1595 ds->ds_phys->ds_next_snap_obj; 1596 } 1597 } 1598 1599 if (ds->ds_phys->ds_next_snap_obj != 0) { 1600 blkptr_t bp; 1601 zio_t *pio; 1602 dsl_dataset_t *ds_next; 1603 uint64_t itor = 0; 1604 uint64_t old_unique; 1605 int64_t used = 0, compressed = 0, uncompressed = 0; 1606 1607 VERIFY(0 == dsl_dataset_hold_obj(dp, 1608 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next)); 1609 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); 1610 1611 old_unique = dsl_dataset_unique(ds_next); 1612 1613 dmu_buf_will_dirty(ds_next->ds_dbuf, tx); 1614 ds_next->ds_phys->ds_prev_snap_obj = 1615 ds->ds_phys->ds_prev_snap_obj; 1616 ds_next->ds_phys->ds_prev_snap_txg = 1617 ds->ds_phys->ds_prev_snap_txg; 1618 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1619 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); 1620 1621 pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1622 1623 /* 1624 * Transfer to our deadlist (which will become next's 1625 * new deadlist) any entries from next's current 1626 * deadlist which were born before prev, and free the 1627 * other entries. 1628 * 1629 * XXX we're doing this long task with the config lock held 1630 */ 1631 while (bplist_iterate(&ds_next->ds_deadlist, &itor, &bp) == 0) { 1632 if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) { 1633 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, 1634 &bp, tx)); 1635 if (ds_prev && !after_branch_point && 1636 bp.blk_birth > 1637 ds_prev->ds_phys->ds_prev_snap_txg) { 1638 ds_prev->ds_phys->ds_unique_bytes += 1639 bp_get_dsize_sync(dp->dp_spa, &bp); 1640 } 1641 } else { 1642 used += bp_get_dsize_sync(dp->dp_spa, &bp); 1643 compressed += BP_GET_PSIZE(&bp); 1644 uncompressed += BP_GET_UCSIZE(&bp); 1645 dsl_free_sync(pio, dp, tx->tx_txg, &bp); 1646 } 1647 } 1648 VERIFY3U(zio_wait(pio), ==, 0); 1649 ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes); 1650 1651 /* change snapused */ 1652 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, 1653 -used, -compressed, -uncompressed, tx); 1654 1655 /* free next's deadlist */ 1656 bplist_close(&ds_next->ds_deadlist); 1657 bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx); 1658 1659 /* set next's deadlist to our deadlist */ 1660 bplist_close(&ds->ds_deadlist); 1661 ds_next->ds_phys->ds_deadlist_obj = 1662 ds->ds_phys->ds_deadlist_obj; 1663 VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos, 1664 ds_next->ds_phys->ds_deadlist_obj)); 1665 ds->ds_phys->ds_deadlist_obj = 0; 1666 1667 if (ds_next->ds_phys->ds_next_snap_obj != 0) { 1668 /* 1669 * Update next's unique to include blocks which 1670 * were previously shared by only this snapshot 1671 * and it. Those blocks will be born after the 1672 * prev snap and before this snap, and will have 1673 * died after the next snap and before the one 1674 * after that (ie. be on the snap after next's 1675 * deadlist). 1676 * 1677 * XXX we're doing this long task with the 1678 * config lock held 1679 */ 1680 dsl_dataset_t *ds_after_next; 1681 uint64_t space; 1682 1683 VERIFY(0 == dsl_dataset_hold_obj(dp, 1684 ds_next->ds_phys->ds_next_snap_obj, 1685 FTAG, &ds_after_next)); 1686 1687 VERIFY(0 == 1688 bplist_space_birthrange(&ds_after_next->ds_deadlist, 1689 ds->ds_phys->ds_prev_snap_txg, 1690 ds->ds_phys->ds_creation_txg, &space)); 1691 ds_next->ds_phys->ds_unique_bytes += space; 1692 1693 dsl_dataset_rele(ds_after_next, FTAG); 1694 ASSERT3P(ds_next->ds_prev, ==, NULL); 1695 } else { 1696 ASSERT3P(ds_next->ds_prev, ==, ds); 1697 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next); 1698 ds_next->ds_prev = NULL; 1699 if (ds_prev) { 1700 VERIFY(0 == dsl_dataset_get_ref(dp, 1701 ds->ds_phys->ds_prev_snap_obj, 1702 ds_next, &ds_next->ds_prev)); 1703 } 1704 1705 dsl_dataset_recalc_head_uniq(ds_next); 1706 1707 /* 1708 * Reduce the amount of our unconsmed refreservation 1709 * being charged to our parent by the amount of 1710 * new unique data we have gained. 1711 */ 1712 if (old_unique < ds_next->ds_reserved) { 1713 int64_t mrsdelta; 1714 uint64_t new_unique = 1715 ds_next->ds_phys->ds_unique_bytes; 1716 1717 ASSERT(old_unique <= new_unique); 1718 mrsdelta = MIN(new_unique - old_unique, 1719 ds_next->ds_reserved - old_unique); 1720 dsl_dir_diduse_space(ds->ds_dir, 1721 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx); 1722 } 1723 } 1724 dsl_dataset_rele(ds_next, FTAG); 1725 } else { 1726 /* 1727 * There's no next snapshot, so this is a head dataset. 1728 * Destroy the deadlist. Unless it's a clone, the 1729 * deadlist should be empty. (If it's a clone, it's 1730 * safe to ignore the deadlist contents.) 1731 */ 1732 struct killarg ka; 1733 1734 ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist)); 1735 bplist_close(&ds->ds_deadlist); 1736 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); 1737 ds->ds_phys->ds_deadlist_obj = 0; 1738 1739 /* 1740 * Free everything that we point to (that's born after 1741 * the previous snapshot, if we are a clone) 1742 * 1743 * NB: this should be very quick, because we already 1744 * freed all the objects in open context. 1745 */ 1746 ka.ds = ds; 1747 ka.tx = tx; 1748 err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg, 1749 TRAVERSE_POST, kill_blkptr, &ka); 1750 ASSERT3U(err, ==, 0); 1751 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || 1752 ds->ds_phys->ds_unique_bytes == 0); 1753 1754 if (ds->ds_prev != NULL) { 1755 dsl_dataset_rele(ds->ds_prev, ds); 1756 ds->ds_prev = ds_prev = NULL; 1757 } 1758 } 1759 1760 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) { 1761 /* Erase the link in the dir */ 1762 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 1763 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0; 1764 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0); 1765 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx); 1766 ASSERT(err == 0); 1767 } else { 1768 /* remove from snapshot namespace */ 1769 dsl_dataset_t *ds_head; 1770 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0); 1771 VERIFY(0 == dsl_dataset_hold_obj(dp, 1772 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head)); 1773 VERIFY(0 == dsl_dataset_get_snapname(ds)); 1774 #ifdef ZFS_DEBUG 1775 { 1776 uint64_t val; 1777 1778 err = dsl_dataset_snap_lookup(ds_head, 1779 ds->ds_snapname, &val); 1780 ASSERT3U(err, ==, 0); 1781 ASSERT3U(val, ==, obj); 1782 } 1783 #endif 1784 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx); 1785 ASSERT(err == 0); 1786 dsl_dataset_rele(ds_head, FTAG); 1787 } 1788 1789 if (ds_prev && ds->ds_prev != ds_prev) 1790 dsl_dataset_rele(ds_prev, FTAG); 1791 1792 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx); 1793 spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx, 1794 cr, "dataset = %llu", ds->ds_object); 1795 1796 if (ds->ds_phys->ds_next_clones_obj != 0) { 1797 uint64_t count; 1798 ASSERT(0 == zap_count(mos, 1799 ds->ds_phys->ds_next_clones_obj, &count) && count == 0); 1800 VERIFY(0 == dmu_object_free(mos, 1801 ds->ds_phys->ds_next_clones_obj, tx)); 1802 } 1803 if (ds->ds_phys->ds_props_obj != 0) 1804 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx)); 1805 if (ds->ds_phys->ds_userrefs_obj != 0) 1806 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx)); 1807 dsl_dir_close(ds->ds_dir, ds); 1808 ds->ds_dir = NULL; 1809 dsl_dataset_drain_refs(ds, tag); 1810 VERIFY(0 == dmu_object_free(mos, obj, tx)); 1811 1812 if (dsda->rm_origin) { 1813 /* 1814 * Remove the origin of the clone we just destroyed. 1815 */ 1816 struct dsl_ds_destroyarg ndsda = {0}; 1817 1818 ndsda.ds = dsda->rm_origin; 1819 dsl_dataset_destroy_sync(&ndsda, tag, cr, tx); 1820 } 1821 } 1822 1823 static int 1824 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1825 { 1826 uint64_t asize; 1827 1828 if (!dmu_tx_is_syncing(tx)) 1829 return (0); 1830 1831 /* 1832 * If there's an fs-only reservation, any blocks that might become 1833 * owned by the snapshot dataset must be accommodated by space 1834 * outside of the reservation. 1835 */ 1836 asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved); 1837 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE)) 1838 return (ENOSPC); 1839 1840 /* 1841 * Propogate any reserved space for this snapshot to other 1842 * snapshot checks in this sync group. 1843 */ 1844 if (asize > 0) 1845 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1846 1847 return (0); 1848 } 1849 1850 /* ARGSUSED */ 1851 int 1852 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx) 1853 { 1854 dsl_dataset_t *ds = arg1; 1855 const char *snapname = arg2; 1856 int err; 1857 uint64_t value; 1858 1859 /* 1860 * We don't allow multiple snapshots of the same txg. If there 1861 * is already one, try again. 1862 */ 1863 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) 1864 return (EAGAIN); 1865 1866 /* 1867 * Check for conflicting name snapshot name. 1868 */ 1869 err = dsl_dataset_snap_lookup(ds, snapname, &value); 1870 if (err == 0) 1871 return (EEXIST); 1872 if (err != ENOENT) 1873 return (err); 1874 1875 /* 1876 * Check that the dataset's name is not too long. Name consists 1877 * of the dataset's length + 1 for the @-sign + snapshot name's length 1878 */ 1879 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN) 1880 return (ENAMETOOLONG); 1881 1882 err = dsl_dataset_snapshot_reserve_space(ds, tx); 1883 if (err) 1884 return (err); 1885 1886 ds->ds_trysnap_txg = tx->tx_txg; 1887 return (0); 1888 } 1889 1890 void 1891 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 1892 { 1893 dsl_dataset_t *ds = arg1; 1894 const char *snapname = arg2; 1895 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1896 dmu_buf_t *dbuf; 1897 dsl_dataset_phys_t *dsphys; 1898 uint64_t dsobj, crtxg; 1899 objset_t *mos = dp->dp_meta_objset; 1900 int err; 1901 1902 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); 1903 1904 /* 1905 * The origin's ds_creation_txg has to be < TXG_INITIAL 1906 */ 1907 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1908 crtxg = 1; 1909 else 1910 crtxg = tx->tx_txg; 1911 1912 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1913 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1914 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1915 dmu_buf_will_dirty(dbuf, tx); 1916 dsphys = dbuf->db_data; 1917 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1918 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1919 dsphys->ds_fsid_guid = unique_create(); 1920 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1921 sizeof (dsphys->ds_guid)); 1922 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; 1923 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; 1924 dsphys->ds_next_snap_obj = ds->ds_object; 1925 dsphys->ds_num_children = 1; 1926 dsphys->ds_creation_time = gethrestime_sec(); 1927 dsphys->ds_creation_txg = crtxg; 1928 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; 1929 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes; 1930 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; 1931 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; 1932 dsphys->ds_flags = ds->ds_phys->ds_flags; 1933 dsphys->ds_bp = ds->ds_phys->ds_bp; 1934 dmu_buf_rele(dbuf, FTAG); 1935 1936 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); 1937 if (ds->ds_prev) { 1938 uint64_t next_clones_obj = 1939 ds->ds_prev->ds_phys->ds_next_clones_obj; 1940 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == 1941 ds->ds_object || 1942 ds->ds_prev->ds_phys->ds_num_children > 1); 1943 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { 1944 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1945 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1946 ds->ds_prev->ds_phys->ds_creation_txg); 1947 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; 1948 } else if (next_clones_obj != 0) { 1949 remove_from_next_clones(ds->ds_prev, 1950 dsphys->ds_next_snap_obj, tx); 1951 VERIFY3U(0, ==, zap_add_int(mos, 1952 next_clones_obj, dsobj, tx)); 1953 } 1954 } 1955 1956 /* 1957 * If we have a reference-reservation on this dataset, we will 1958 * need to increase the amount of refreservation being charged 1959 * since our unique space is going to zero. 1960 */ 1961 if (ds->ds_reserved) { 1962 int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved); 1963 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1964 add, 0, 0, tx); 1965 } 1966 1967 bplist_close(&ds->ds_deadlist); 1968 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1969 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg); 1970 ds->ds_phys->ds_prev_snap_obj = dsobj; 1971 ds->ds_phys->ds_prev_snap_txg = crtxg; 1972 ds->ds_phys->ds_unique_bytes = 0; 1973 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1974 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1975 ds->ds_phys->ds_deadlist_obj = 1976 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); 1977 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos, 1978 ds->ds_phys->ds_deadlist_obj)); 1979 1980 dprintf("snap '%s' -> obj %llu\n", snapname, dsobj); 1981 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, 1982 snapname, 8, 1, &dsobj, tx); 1983 ASSERT(err == 0); 1984 1985 if (ds->ds_prev) 1986 dsl_dataset_drop_ref(ds->ds_prev, ds); 1987 VERIFY(0 == dsl_dataset_get_ref(dp, 1988 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 1989 1990 dsl_pool_ds_snapshotted(ds, tx); 1991 1992 dsl_dir_snap_cmtime_update(ds->ds_dir); 1993 1994 spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr, 1995 "dataset = %llu", dsobj); 1996 } 1997 1998 void 1999 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 2000 { 2001 ASSERT(dmu_tx_is_syncing(tx)); 2002 ASSERT(ds->ds_objset != NULL); 2003 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); 2004 2005 /* 2006 * in case we had to change ds_fsid_guid when we opened it, 2007 * sync it out now. 2008 */ 2009 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2010 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; 2011 2012 dsl_dir_dirty(ds->ds_dir, tx); 2013 dmu_objset_sync(ds->ds_objset, zio, tx); 2014 } 2015 2016 void 2017 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 2018 { 2019 uint64_t refd, avail, uobjs, aobjs; 2020 2021 dsl_dir_stats(ds->ds_dir, nv); 2022 2023 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 2024 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 2025 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 2026 2027 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 2028 ds->ds_phys->ds_creation_time); 2029 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 2030 ds->ds_phys->ds_creation_txg); 2031 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 2032 ds->ds_quota); 2033 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 2034 ds->ds_reserved); 2035 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 2036 ds->ds_phys->ds_guid); 2037 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 2038 dsl_dataset_unique(ds)); 2039 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 2040 ds->ds_object); 2041 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 2042 ds->ds_userrefs); 2043 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 2044 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 2045 2046 if (ds->ds_phys->ds_next_snap_obj) { 2047 /* 2048 * This is a snapshot; override the dd's space used with 2049 * our unique space and compression ratio. 2050 */ 2051 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 2052 ds->ds_phys->ds_unique_bytes); 2053 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 2054 ds->ds_phys->ds_compressed_bytes == 0 ? 100 : 2055 (ds->ds_phys->ds_uncompressed_bytes * 100 / 2056 ds->ds_phys->ds_compressed_bytes)); 2057 } 2058 } 2059 2060 void 2061 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 2062 { 2063 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; 2064 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; 2065 stat->dds_guid = ds->ds_phys->ds_guid; 2066 if (ds->ds_phys->ds_next_snap_obj) { 2067 stat->dds_is_snapshot = B_TRUE; 2068 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; 2069 } else { 2070 stat->dds_is_snapshot = B_FALSE; 2071 stat->dds_num_clones = 0; 2072 } 2073 2074 /* clone origin is really a dsl_dir thing... */ 2075 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); 2076 if (dsl_dir_is_clone(ds->ds_dir)) { 2077 dsl_dataset_t *ods; 2078 2079 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool, 2080 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods)); 2081 dsl_dataset_name(ods, stat->dds_origin); 2082 dsl_dataset_drop_ref(ods, FTAG); 2083 } else { 2084 stat->dds_origin[0] = '\0'; 2085 } 2086 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); 2087 } 2088 2089 uint64_t 2090 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 2091 { 2092 return (ds->ds_fsid_guid); 2093 } 2094 2095 void 2096 dsl_dataset_space(dsl_dataset_t *ds, 2097 uint64_t *refdbytesp, uint64_t *availbytesp, 2098 uint64_t *usedobjsp, uint64_t *availobjsp) 2099 { 2100 *refdbytesp = ds->ds_phys->ds_used_bytes; 2101 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 2102 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) 2103 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes; 2104 if (ds->ds_quota != 0) { 2105 /* 2106 * Adjust available bytes according to refquota 2107 */ 2108 if (*refdbytesp < ds->ds_quota) 2109 *availbytesp = MIN(*availbytesp, 2110 ds->ds_quota - *refdbytesp); 2111 else 2112 *availbytesp = 0; 2113 } 2114 *usedobjsp = ds->ds_phys->ds_bp.blk_fill; 2115 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 2116 } 2117 2118 boolean_t 2119 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds) 2120 { 2121 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2122 2123 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 2124 dsl_pool_sync_context(dp)); 2125 if (ds->ds_prev == NULL) 2126 return (B_FALSE); 2127 if (ds->ds_phys->ds_bp.blk_birth > 2128 ds->ds_prev->ds_phys->ds_creation_txg) 2129 return (B_TRUE); 2130 return (B_FALSE); 2131 } 2132 2133 /* ARGSUSED */ 2134 static int 2135 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) 2136 { 2137 dsl_dataset_t *ds = arg1; 2138 char *newsnapname = arg2; 2139 dsl_dir_t *dd = ds->ds_dir; 2140 dsl_dataset_t *hds; 2141 uint64_t val; 2142 int err; 2143 2144 err = dsl_dataset_hold_obj(dd->dd_pool, 2145 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds); 2146 if (err) 2147 return (err); 2148 2149 /* new name better not be in use */ 2150 err = dsl_dataset_snap_lookup(hds, newsnapname, &val); 2151 dsl_dataset_rele(hds, FTAG); 2152 2153 if (err == 0) 2154 err = EEXIST; 2155 else if (err == ENOENT) 2156 err = 0; 2157 2158 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2159 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN) 2160 err = ENAMETOOLONG; 2161 2162 return (err); 2163 } 2164 2165 static void 2166 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, 2167 cred_t *cr, dmu_tx_t *tx) 2168 { 2169 dsl_dataset_t *ds = arg1; 2170 const char *newsnapname = arg2; 2171 dsl_dir_t *dd = ds->ds_dir; 2172 objset_t *mos = dd->dd_pool->dp_meta_objset; 2173 dsl_dataset_t *hds; 2174 int err; 2175 2176 ASSERT(ds->ds_phys->ds_next_snap_obj != 0); 2177 2178 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, 2179 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds)); 2180 2181 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2182 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx); 2183 ASSERT3U(err, ==, 0); 2184 mutex_enter(&ds->ds_lock); 2185 (void) strcpy(ds->ds_snapname, newsnapname); 2186 mutex_exit(&ds->ds_lock); 2187 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj, 2188 ds->ds_snapname, 8, 1, &ds->ds_object, tx); 2189 ASSERT3U(err, ==, 0); 2190 2191 spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx, 2192 cr, "dataset = %llu", ds->ds_object); 2193 dsl_dataset_rele(hds, FTAG); 2194 } 2195 2196 struct renamesnaparg { 2197 dsl_sync_task_group_t *dstg; 2198 char failed[MAXPATHLEN]; 2199 char *oldsnap; 2200 char *newsnap; 2201 }; 2202 2203 static int 2204 dsl_snapshot_rename_one(const char *name, void *arg) 2205 { 2206 struct renamesnaparg *ra = arg; 2207 dsl_dataset_t *ds = NULL; 2208 char *snapname; 2209 int err; 2210 2211 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap); 2212 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed)); 2213 2214 /* 2215 * For recursive snapshot renames the parent won't be changing 2216 * so we just pass name for both the to/from argument. 2217 */ 2218 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED()); 2219 if (err != 0) { 2220 strfree(snapname); 2221 return (err == ENOENT ? 0 : err); 2222 } 2223 2224 #ifdef _KERNEL 2225 /* 2226 * For all filesystems undergoing rename, we'll need to unmount it. 2227 */ 2228 (void) zfs_unmount_snap(snapname, NULL); 2229 #endif 2230 err = dsl_dataset_hold(snapname, ra->dstg, &ds); 2231 strfree(snapname); 2232 if (err != 0) 2233 return (err == ENOENT ? 0 : err); 2234 2235 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check, 2236 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0); 2237 2238 return (0); 2239 } 2240 2241 static int 2242 dsl_recursive_rename(char *oldname, const char *newname) 2243 { 2244 int err; 2245 struct renamesnaparg *ra; 2246 dsl_sync_task_t *dst; 2247 spa_t *spa; 2248 char *cp, *fsname = spa_strdup(oldname); 2249 int len = strlen(oldname) + 1; 2250 2251 /* truncate the snapshot name to get the fsname */ 2252 cp = strchr(fsname, '@'); 2253 *cp = '\0'; 2254 2255 err = spa_open(fsname, &spa, FTAG); 2256 if (err) { 2257 kmem_free(fsname, len); 2258 return (err); 2259 } 2260 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP); 2261 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 2262 2263 ra->oldsnap = strchr(oldname, '@') + 1; 2264 ra->newsnap = strchr(newname, '@') + 1; 2265 *ra->failed = '\0'; 2266 2267 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra, 2268 DS_FIND_CHILDREN); 2269 kmem_free(fsname, len); 2270 2271 if (err == 0) { 2272 err = dsl_sync_task_group_wait(ra->dstg); 2273 } 2274 2275 for (dst = list_head(&ra->dstg->dstg_tasks); dst; 2276 dst = list_next(&ra->dstg->dstg_tasks, dst)) { 2277 dsl_dataset_t *ds = dst->dst_arg1; 2278 if (dst->dst_err) { 2279 dsl_dir_name(ds->ds_dir, ra->failed); 2280 (void) strlcat(ra->failed, "@", sizeof (ra->failed)); 2281 (void) strlcat(ra->failed, ra->newsnap, 2282 sizeof (ra->failed)); 2283 } 2284 dsl_dataset_rele(ds, ra->dstg); 2285 } 2286 2287 if (err) 2288 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed)); 2289 2290 dsl_sync_task_group_destroy(ra->dstg); 2291 kmem_free(ra, sizeof (struct renamesnaparg)); 2292 spa_close(spa, FTAG); 2293 return (err); 2294 } 2295 2296 static int 2297 dsl_valid_rename(const char *oldname, void *arg) 2298 { 2299 int delta = *(int *)arg; 2300 2301 if (strlen(oldname) + delta >= MAXNAMELEN) 2302 return (ENAMETOOLONG); 2303 2304 return (0); 2305 } 2306 2307 #pragma weak dmu_objset_rename = dsl_dataset_rename 2308 int 2309 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive) 2310 { 2311 dsl_dir_t *dd; 2312 dsl_dataset_t *ds; 2313 const char *tail; 2314 int err; 2315 2316 err = dsl_dir_open(oldname, FTAG, &dd, &tail); 2317 if (err) 2318 return (err); 2319 2320 if (tail == NULL) { 2321 int delta = strlen(newname) - strlen(oldname); 2322 2323 /* if we're growing, validate child name lengths */ 2324 if (delta > 0) 2325 err = dmu_objset_find(oldname, dsl_valid_rename, 2326 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 2327 2328 if (err == 0) 2329 err = dsl_dir_rename(dd, newname); 2330 dsl_dir_close(dd, FTAG); 2331 return (err); 2332 } 2333 2334 if (tail[0] != '@') { 2335 /* the name ended in a nonexistent component */ 2336 dsl_dir_close(dd, FTAG); 2337 return (ENOENT); 2338 } 2339 2340 dsl_dir_close(dd, FTAG); 2341 2342 /* new name must be snapshot in same filesystem */ 2343 tail = strchr(newname, '@'); 2344 if (tail == NULL) 2345 return (EINVAL); 2346 tail++; 2347 if (strncmp(oldname, newname, tail - newname) != 0) 2348 return (EXDEV); 2349 2350 if (recursive) { 2351 err = dsl_recursive_rename(oldname, newname); 2352 } else { 2353 err = dsl_dataset_hold(oldname, FTAG, &ds); 2354 if (err) 2355 return (err); 2356 2357 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 2358 dsl_dataset_snapshot_rename_check, 2359 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1); 2360 2361 dsl_dataset_rele(ds, FTAG); 2362 } 2363 2364 return (err); 2365 } 2366 2367 struct promotenode { 2368 list_node_t link; 2369 dsl_dataset_t *ds; 2370 }; 2371 2372 struct promotearg { 2373 list_t shared_snaps, origin_snaps, clone_snaps; 2374 dsl_dataset_t *origin_origin, *origin_head; 2375 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2376 char *err_ds; 2377 }; 2378 2379 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2380 2381 /* ARGSUSED */ 2382 static int 2383 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) 2384 { 2385 dsl_dataset_t *hds = arg1; 2386 struct promotearg *pa = arg2; 2387 struct promotenode *snap = list_head(&pa->shared_snaps); 2388 dsl_dataset_t *origin_ds = snap->ds; 2389 int err; 2390 2391 /* Check that it is a real clone */ 2392 if (!dsl_dir_is_clone(hds->ds_dir)) 2393 return (EINVAL); 2394 2395 /* Since this is so expensive, don't do the preliminary check */ 2396 if (!dmu_tx_is_syncing(tx)) 2397 return (0); 2398 2399 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) 2400 return (EXDEV); 2401 2402 /* compute origin's new unique space */ 2403 snap = list_tail(&pa->clone_snaps); 2404 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2405 err = bplist_space_birthrange(&snap->ds->ds_deadlist, 2406 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, &pa->unique); 2407 if (err) 2408 return (err); 2409 2410 /* 2411 * Walk the snapshots that we are moving 2412 * 2413 * Compute space to transfer. Consider the incremental changes 2414 * to used for each snapshot: 2415 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2416 * So each snapshot gave birth to: 2417 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2418 * So a sequence would look like: 2419 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2420 * Which simplifies to: 2421 * uN + kN + kN-1 + ... + k1 + k0 2422 * Note however, if we stop before we reach the ORIGIN we get: 2423 * uN + kN + kN-1 + ... + kM - uM-1 2424 */ 2425 pa->used = origin_ds->ds_phys->ds_used_bytes; 2426 pa->comp = origin_ds->ds_phys->ds_compressed_bytes; 2427 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; 2428 for (snap = list_head(&pa->shared_snaps); snap; 2429 snap = list_next(&pa->shared_snaps, snap)) { 2430 uint64_t val, dlused, dlcomp, dluncomp; 2431 dsl_dataset_t *ds = snap->ds; 2432 2433 /* Check that the snapshot name does not conflict */ 2434 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2435 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2436 if (err == 0) { 2437 err = EEXIST; 2438 goto out; 2439 } 2440 if (err != ENOENT) 2441 goto out; 2442 2443 /* The very first snapshot does not have a deadlist */ 2444 if (ds->ds_phys->ds_prev_snap_obj == 0) 2445 continue; 2446 2447 if (err = bplist_space(&ds->ds_deadlist, 2448 &dlused, &dlcomp, &dluncomp)) 2449 goto out; 2450 pa->used += dlused; 2451 pa->comp += dlcomp; 2452 pa->uncomp += dluncomp; 2453 } 2454 2455 /* 2456 * If we are a clone of a clone then we never reached ORIGIN, 2457 * so we need to subtract out the clone origin's used space. 2458 */ 2459 if (pa->origin_origin) { 2460 pa->used -= pa->origin_origin->ds_phys->ds_used_bytes; 2461 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes; 2462 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes; 2463 } 2464 2465 /* Check that there is enough space here */ 2466 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2467 pa->used); 2468 if (err) 2469 return (err); 2470 2471 /* 2472 * Compute the amounts of space that will be used by snapshots 2473 * after the promotion (for both origin and clone). For each, 2474 * it is the amount of space that will be on all of their 2475 * deadlists (that was not born before their new origin). 2476 */ 2477 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2478 uint64_t space; 2479 2480 /* 2481 * Note, typically this will not be a clone of a clone, 2482 * so snap->ds->ds_origin_txg will be < TXG_INITIAL, so 2483 * these snaplist_space() -> bplist_space_birthrange() 2484 * calls will be fast because they do not have to 2485 * iterate over all bps. 2486 */ 2487 snap = list_head(&pa->origin_snaps); 2488 err = snaplist_space(&pa->shared_snaps, 2489 snap->ds->ds_origin_txg, &pa->cloneusedsnap); 2490 if (err) 2491 return (err); 2492 2493 err = snaplist_space(&pa->clone_snaps, 2494 snap->ds->ds_origin_txg, &space); 2495 if (err) 2496 return (err); 2497 pa->cloneusedsnap += space; 2498 } 2499 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2500 err = snaplist_space(&pa->origin_snaps, 2501 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap); 2502 if (err) 2503 return (err); 2504 } 2505 2506 return (0); 2507 out: 2508 pa->err_ds = snap->ds->ds_snapname; 2509 return (err); 2510 } 2511 2512 static void 2513 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 2514 { 2515 dsl_dataset_t *hds = arg1; 2516 struct promotearg *pa = arg2; 2517 struct promotenode *snap = list_head(&pa->shared_snaps); 2518 dsl_dataset_t *origin_ds = snap->ds; 2519 dsl_dataset_t *origin_head; 2520 dsl_dir_t *dd = hds->ds_dir; 2521 dsl_pool_t *dp = hds->ds_dir->dd_pool; 2522 dsl_dir_t *odd = NULL; 2523 uint64_t oldnext_obj; 2524 int64_t delta; 2525 2526 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)); 2527 2528 snap = list_head(&pa->origin_snaps); 2529 origin_head = snap->ds; 2530 2531 /* 2532 * We need to explicitly open odd, since origin_ds's dd will be 2533 * changing. 2534 */ 2535 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object, 2536 NULL, FTAG, &odd)); 2537 2538 /* change origin's next snap */ 2539 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2540 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj; 2541 snap = list_tail(&pa->clone_snaps); 2542 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2543 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object; 2544 2545 /* change the origin's next clone */ 2546 if (origin_ds->ds_phys->ds_next_clones_obj) { 2547 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx); 2548 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, 2549 origin_ds->ds_phys->ds_next_clones_obj, 2550 oldnext_obj, tx)); 2551 } 2552 2553 /* change origin */ 2554 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2555 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object); 2556 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj; 2557 hds->ds_origin_txg = origin_head->ds_origin_txg; 2558 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2559 odd->dd_phys->dd_origin_obj = origin_ds->ds_object; 2560 origin_head->ds_origin_txg = origin_ds->ds_phys->ds_creation_txg; 2561 2562 /* move snapshots to this dir */ 2563 for (snap = list_head(&pa->shared_snaps); snap; 2564 snap = list_next(&pa->shared_snaps, snap)) { 2565 dsl_dataset_t *ds = snap->ds; 2566 2567 /* unregister props as dsl_dir is changing */ 2568 if (ds->ds_objset) { 2569 dmu_objset_evict(ds->ds_objset); 2570 ds->ds_objset = NULL; 2571 } 2572 /* move snap name entry */ 2573 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2574 VERIFY(0 == dsl_dataset_snap_remove(origin_head, 2575 ds->ds_snapname, tx)); 2576 VERIFY(0 == zap_add(dp->dp_meta_objset, 2577 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, 2578 8, 1, &ds->ds_object, tx)); 2579 /* change containing dsl_dir */ 2580 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2581 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object); 2582 ds->ds_phys->ds_dir_obj = dd->dd_object; 2583 ASSERT3P(ds->ds_dir, ==, odd); 2584 dsl_dir_close(ds->ds_dir, ds); 2585 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object, 2586 NULL, ds, &ds->ds_dir)); 2587 2588 ASSERT3U(dsl_prop_numcb(ds), ==, 0); 2589 } 2590 2591 /* 2592 * Change space accounting. 2593 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2594 * both be valid, or both be 0 (resulting in delta == 0). This 2595 * is true for each of {clone,origin} independently. 2596 */ 2597 2598 delta = pa->cloneusedsnap - 2599 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2600 ASSERT3S(delta, >=, 0); 2601 ASSERT3U(pa->used, >=, delta); 2602 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2603 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2604 pa->used - delta, pa->comp, pa->uncomp, tx); 2605 2606 delta = pa->originusedsnap - 2607 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2608 ASSERT3S(delta, <=, 0); 2609 ASSERT3U(pa->used, >=, -delta); 2610 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2611 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2612 -pa->used - delta, -pa->comp, -pa->uncomp, tx); 2613 2614 origin_ds->ds_phys->ds_unique_bytes = pa->unique; 2615 2616 /* log history record */ 2617 spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx, 2618 cr, "dataset = %llu", hds->ds_object); 2619 2620 dsl_dir_close(odd, FTAG); 2621 } 2622 2623 static char *snaplist_tag = "snaplist"; 2624 /* 2625 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2626 * (exclusive) and last_obj (inclusive). The list will be in reverse 2627 * order (last_obj will be the list_head()). If first_obj == 0, do all 2628 * snapshots back to this dataset's origin. 2629 */ 2630 static int 2631 snaplist_make(dsl_pool_t *dp, boolean_t own, 2632 uint64_t first_obj, uint64_t last_obj, list_t *l) 2633 { 2634 uint64_t obj = last_obj; 2635 2636 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock)); 2637 2638 list_create(l, sizeof (struct promotenode), 2639 offsetof(struct promotenode, link)); 2640 2641 while (obj != first_obj) { 2642 dsl_dataset_t *ds; 2643 struct promotenode *snap; 2644 int err; 2645 2646 if (own) { 2647 err = dsl_dataset_own_obj(dp, obj, 2648 0, snaplist_tag, &ds); 2649 if (err == 0) 2650 dsl_dataset_make_exclusive(ds, snaplist_tag); 2651 } else { 2652 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds); 2653 } 2654 if (err == ENOENT) { 2655 /* lost race with snapshot destroy */ 2656 struct promotenode *last = list_tail(l); 2657 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj); 2658 obj = last->ds->ds_phys->ds_prev_snap_obj; 2659 continue; 2660 } else if (err) { 2661 return (err); 2662 } 2663 2664 if (first_obj == 0) 2665 first_obj = ds->ds_dir->dd_phys->dd_origin_obj; 2666 2667 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP); 2668 snap->ds = ds; 2669 list_insert_tail(l, snap); 2670 obj = ds->ds_phys->ds_prev_snap_obj; 2671 } 2672 2673 return (0); 2674 } 2675 2676 static int 2677 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2678 { 2679 struct promotenode *snap; 2680 2681 *spacep = 0; 2682 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2683 uint64_t used; 2684 int err = bplist_space_birthrange(&snap->ds->ds_deadlist, 2685 mintxg, UINT64_MAX, &used); 2686 if (err) 2687 return (err); 2688 *spacep += used; 2689 } 2690 return (0); 2691 } 2692 2693 static void 2694 snaplist_destroy(list_t *l, boolean_t own) 2695 { 2696 struct promotenode *snap; 2697 2698 if (!l || !list_link_active(&l->list_head)) 2699 return; 2700 2701 while ((snap = list_tail(l)) != NULL) { 2702 list_remove(l, snap); 2703 if (own) 2704 dsl_dataset_disown(snap->ds, snaplist_tag); 2705 else 2706 dsl_dataset_rele(snap->ds, snaplist_tag); 2707 kmem_free(snap, sizeof (struct promotenode)); 2708 } 2709 list_destroy(l); 2710 } 2711 2712 /* 2713 * Promote a clone. Nomenclature note: 2714 * "clone" or "cds": the original clone which is being promoted 2715 * "origin" or "ods": the snapshot which is originally clone's origin 2716 * "origin head" or "ohds": the dataset which is the head 2717 * (filesystem/volume) for the origin 2718 * "origin origin": the origin of the origin's filesystem (typically 2719 * NULL, indicating that the clone is not a clone of a clone). 2720 */ 2721 int 2722 dsl_dataset_promote(const char *name, char *conflsnap) 2723 { 2724 dsl_dataset_t *ds; 2725 dsl_dir_t *dd; 2726 dsl_pool_t *dp; 2727 dmu_object_info_t doi; 2728 struct promotearg pa = { 0 }; 2729 struct promotenode *snap; 2730 int err; 2731 2732 err = dsl_dataset_hold(name, FTAG, &ds); 2733 if (err) 2734 return (err); 2735 dd = ds->ds_dir; 2736 dp = dd->dd_pool; 2737 2738 err = dmu_object_info(dp->dp_meta_objset, 2739 ds->ds_phys->ds_snapnames_zapobj, &doi); 2740 if (err) { 2741 dsl_dataset_rele(ds, FTAG); 2742 return (err); 2743 } 2744 2745 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) { 2746 dsl_dataset_rele(ds, FTAG); 2747 return (EINVAL); 2748 } 2749 2750 /* 2751 * We are going to inherit all the snapshots taken before our 2752 * origin (i.e., our new origin will be our parent's origin). 2753 * Take ownership of them so that we can rename them into our 2754 * namespace. 2755 */ 2756 rw_enter(&dp->dp_config_rwlock, RW_READER); 2757 2758 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj, 2759 &pa.shared_snaps); 2760 if (err != 0) 2761 goto out; 2762 2763 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps); 2764 if (err != 0) 2765 goto out; 2766 2767 snap = list_head(&pa.shared_snaps); 2768 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); 2769 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj, 2770 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps); 2771 if (err != 0) 2772 goto out; 2773 2774 if (dsl_dir_is_clone(snap->ds->ds_dir)) { 2775 err = dsl_dataset_own_obj(dp, 2776 snap->ds->ds_dir->dd_phys->dd_origin_obj, 2777 0, FTAG, &pa.origin_origin); 2778 if (err != 0) 2779 goto out; 2780 } 2781 2782 out: 2783 rw_exit(&dp->dp_config_rwlock); 2784 2785 /* 2786 * Add in 128x the snapnames zapobj size, since we will be moving 2787 * a bunch of snapnames to the promoted ds, and dirtying their 2788 * bonus buffers. 2789 */ 2790 if (err == 0) { 2791 err = dsl_sync_task_do(dp, dsl_dataset_promote_check, 2792 dsl_dataset_promote_sync, ds, &pa, 2793 2 + 2 * doi.doi_physical_blocks_512); 2794 if (err && pa.err_ds && conflsnap) 2795 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN); 2796 } 2797 2798 snaplist_destroy(&pa.shared_snaps, B_TRUE); 2799 snaplist_destroy(&pa.clone_snaps, B_FALSE); 2800 snaplist_destroy(&pa.origin_snaps, B_FALSE); 2801 if (pa.origin_origin) 2802 dsl_dataset_disown(pa.origin_origin, FTAG); 2803 dsl_dataset_rele(ds, FTAG); 2804 return (err); 2805 } 2806 2807 struct cloneswaparg { 2808 dsl_dataset_t *cds; /* clone dataset */ 2809 dsl_dataset_t *ohds; /* origin's head dataset */ 2810 boolean_t force; 2811 int64_t unused_refres_delta; /* change in unconsumed refreservation */ 2812 }; 2813 2814 /* ARGSUSED */ 2815 static int 2816 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx) 2817 { 2818 struct cloneswaparg *csa = arg1; 2819 2820 /* they should both be heads */ 2821 if (dsl_dataset_is_snapshot(csa->cds) || 2822 dsl_dataset_is_snapshot(csa->ohds)) 2823 return (EINVAL); 2824 2825 /* the branch point should be just before them */ 2826 if (csa->cds->ds_prev != csa->ohds->ds_prev) 2827 return (EINVAL); 2828 2829 /* cds should be the clone (unless they are unrelated) */ 2830 if (csa->cds->ds_prev != NULL && 2831 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap && 2832 csa->ohds->ds_object != 2833 csa->cds->ds_prev->ds_phys->ds_next_snap_obj) 2834 return (EINVAL); 2835 2836 /* the clone should be a child of the origin */ 2837 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir) 2838 return (EINVAL); 2839 2840 /* ohds shouldn't be modified unless 'force' */ 2841 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds)) 2842 return (ETXTBSY); 2843 2844 /* adjust amount of any unconsumed refreservation */ 2845 csa->unused_refres_delta = 2846 (int64_t)MIN(csa->ohds->ds_reserved, 2847 csa->ohds->ds_phys->ds_unique_bytes) - 2848 (int64_t)MIN(csa->ohds->ds_reserved, 2849 csa->cds->ds_phys->ds_unique_bytes); 2850 2851 if (csa->unused_refres_delta > 0 && 2852 csa->unused_refres_delta > 2853 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE)) 2854 return (ENOSPC); 2855 2856 if (csa->ohds->ds_quota != 0 && 2857 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota) 2858 return (EDQUOT); 2859 2860 return (0); 2861 } 2862 2863 /* ARGSUSED */ 2864 static void 2865 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 2866 { 2867 struct cloneswaparg *csa = arg1; 2868 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool; 2869 2870 ASSERT(csa->cds->ds_reserved == 0); 2871 ASSERT(csa->ohds->ds_quota == 0 || 2872 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota); 2873 2874 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx); 2875 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx); 2876 2877 if (csa->cds->ds_objset != NULL) { 2878 dmu_objset_evict(csa->cds->ds_objset); 2879 csa->cds->ds_objset = NULL; 2880 } 2881 2882 if (csa->ohds->ds_objset != NULL) { 2883 dmu_objset_evict(csa->ohds->ds_objset); 2884 csa->ohds->ds_objset = NULL; 2885 } 2886 2887 /* 2888 * Reset origin's unique bytes, if it exists. 2889 */ 2890 if (csa->cds->ds_prev) { 2891 dsl_dataset_t *origin = csa->cds->ds_prev; 2892 dmu_buf_will_dirty(origin->ds_dbuf, tx); 2893 VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist, 2894 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2895 &origin->ds_phys->ds_unique_bytes)); 2896 } 2897 2898 /* swap blkptrs */ 2899 { 2900 blkptr_t tmp; 2901 tmp = csa->ohds->ds_phys->ds_bp; 2902 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp; 2903 csa->cds->ds_phys->ds_bp = tmp; 2904 } 2905 2906 /* set dd_*_bytes */ 2907 { 2908 int64_t dused, dcomp, duncomp; 2909 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2910 uint64_t odl_used, odl_comp, odl_uncomp; 2911 2912 ASSERT3U(csa->cds->ds_dir->dd_phys-> 2913 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2914 2915 VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used, 2916 &cdl_comp, &cdl_uncomp)); 2917 VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used, 2918 &odl_comp, &odl_uncomp)); 2919 2920 dused = csa->cds->ds_phys->ds_used_bytes + cdl_used - 2921 (csa->ohds->ds_phys->ds_used_bytes + odl_used); 2922 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp - 2923 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp); 2924 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes + 2925 cdl_uncomp - 2926 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp); 2927 2928 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD, 2929 dused, dcomp, duncomp, tx); 2930 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD, 2931 -dused, -dcomp, -duncomp, tx); 2932 2933 /* 2934 * The difference in the space used by snapshots is the 2935 * difference in snapshot space due to the head's 2936 * deadlist (since that's the only thing that's 2937 * changing that affects the snapused). 2938 */ 2939 VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist, 2940 csa->ohds->ds_origin_txg, UINT64_MAX, &cdl_used)); 2941 VERIFY(0 == bplist_space_birthrange(&csa->ohds->ds_deadlist, 2942 csa->ohds->ds_origin_txg, UINT64_MAX, &odl_used)); 2943 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used, 2944 DD_USED_HEAD, DD_USED_SNAP, tx); 2945 } 2946 2947 #define SWITCH64(x, y) \ 2948 { \ 2949 uint64_t __tmp = (x); \ 2950 (x) = (y); \ 2951 (y) = __tmp; \ 2952 } 2953 2954 /* swap ds_*_bytes */ 2955 SWITCH64(csa->ohds->ds_phys->ds_used_bytes, 2956 csa->cds->ds_phys->ds_used_bytes); 2957 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes, 2958 csa->cds->ds_phys->ds_compressed_bytes); 2959 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes, 2960 csa->cds->ds_phys->ds_uncompressed_bytes); 2961 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes, 2962 csa->cds->ds_phys->ds_unique_bytes); 2963 2964 /* apply any parent delta for change in unconsumed refreservation */ 2965 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV, 2966 csa->unused_refres_delta, 0, 0, tx); 2967 2968 /* swap deadlists */ 2969 bplist_close(&csa->cds->ds_deadlist); 2970 bplist_close(&csa->ohds->ds_deadlist); 2971 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj, 2972 csa->cds->ds_phys->ds_deadlist_obj); 2973 VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset, 2974 csa->cds->ds_phys->ds_deadlist_obj)); 2975 VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset, 2976 csa->ohds->ds_phys->ds_deadlist_obj)); 2977 2978 dsl_pool_ds_clone_swapped(csa->ohds, csa->cds, tx); 2979 } 2980 2981 /* 2982 * Swap 'clone' with its origin head datasets. Used at the end of "zfs 2983 * recv" into an existing fs to swizzle the file system to the new 2984 * version, and by "zfs rollback". Can also be used to swap two 2985 * independent head datasets if neither has any snapshots. 2986 */ 2987 int 2988 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head, 2989 boolean_t force) 2990 { 2991 struct cloneswaparg csa; 2992 int error; 2993 2994 ASSERT(clone->ds_owner); 2995 ASSERT(origin_head->ds_owner); 2996 retry: 2997 /* Need exclusive access for the swap */ 2998 rw_enter(&clone->ds_rwlock, RW_WRITER); 2999 if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) { 3000 rw_exit(&clone->ds_rwlock); 3001 rw_enter(&origin_head->ds_rwlock, RW_WRITER); 3002 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) { 3003 rw_exit(&origin_head->ds_rwlock); 3004 goto retry; 3005 } 3006 } 3007 csa.cds = clone; 3008 csa.ohds = origin_head; 3009 csa.force = force; 3010 error = dsl_sync_task_do(clone->ds_dir->dd_pool, 3011 dsl_dataset_clone_swap_check, 3012 dsl_dataset_clone_swap_sync, &csa, NULL, 9); 3013 return (error); 3014 } 3015 3016 /* 3017 * Given a pool name and a dataset object number in that pool, 3018 * return the name of that dataset. 3019 */ 3020 int 3021 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 3022 { 3023 spa_t *spa; 3024 dsl_pool_t *dp; 3025 dsl_dataset_t *ds; 3026 int error; 3027 3028 if ((error = spa_open(pname, &spa, FTAG)) != 0) 3029 return (error); 3030 dp = spa_get_dsl(spa); 3031 rw_enter(&dp->dp_config_rwlock, RW_READER); 3032 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) { 3033 dsl_dataset_name(ds, buf); 3034 dsl_dataset_rele(ds, FTAG); 3035 } 3036 rw_exit(&dp->dp_config_rwlock); 3037 spa_close(spa, FTAG); 3038 3039 return (error); 3040 } 3041 3042 int 3043 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 3044 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 3045 { 3046 int error = 0; 3047 3048 ASSERT3S(asize, >, 0); 3049 3050 /* 3051 * *ref_rsrv is the portion of asize that will come from any 3052 * unconsumed refreservation space. 3053 */ 3054 *ref_rsrv = 0; 3055 3056 mutex_enter(&ds->ds_lock); 3057 /* 3058 * Make a space adjustment for reserved bytes. 3059 */ 3060 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) { 3061 ASSERT3U(*used, >=, 3062 ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 3063 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 3064 *ref_rsrv = 3065 asize - MIN(asize, parent_delta(ds, asize + inflight)); 3066 } 3067 3068 if (!check_quota || ds->ds_quota == 0) { 3069 mutex_exit(&ds->ds_lock); 3070 return (0); 3071 } 3072 /* 3073 * If they are requesting more space, and our current estimate 3074 * is over quota, they get to try again unless the actual 3075 * on-disk is over quota and there are no pending changes (which 3076 * may free up space for us). 3077 */ 3078 if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) { 3079 if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota) 3080 error = ERESTART; 3081 else 3082 error = EDQUOT; 3083 } 3084 mutex_exit(&ds->ds_lock); 3085 3086 return (error); 3087 } 3088 3089 /* ARGSUSED */ 3090 static int 3091 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) 3092 { 3093 dsl_dataset_t *ds = arg1; 3094 dsl_prop_setarg_t *psa = arg2; 3095 int err; 3096 3097 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA) 3098 return (ENOTSUP); 3099 3100 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) 3101 return (err); 3102 3103 if (psa->psa_effective_value == 0) 3104 return (0); 3105 3106 if (psa->psa_effective_value < ds->ds_phys->ds_used_bytes || 3107 psa->psa_effective_value < ds->ds_reserved) 3108 return (ENOSPC); 3109 3110 return (0); 3111 } 3112 3113 extern void dsl_prop_set_sync(void *, void *, cred_t *, dmu_tx_t *); 3114 3115 void 3116 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 3117 { 3118 dsl_dataset_t *ds = arg1; 3119 dsl_prop_setarg_t *psa = arg2; 3120 uint64_t effective_value = psa->psa_effective_value; 3121 3122 dsl_prop_set_sync(ds, psa, cr, tx); 3123 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa); 3124 3125 if (ds->ds_quota != effective_value) { 3126 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3127 ds->ds_quota = effective_value; 3128 3129 spa_history_internal_log(LOG_DS_REFQUOTA, 3130 ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu ", 3131 (longlong_t)ds->ds_quota, ds->ds_object); 3132 } 3133 } 3134 3135 int 3136 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota) 3137 { 3138 dsl_dataset_t *ds; 3139 dsl_prop_setarg_t psa; 3140 int err; 3141 3142 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a); 3143 3144 err = dsl_dataset_hold(dsname, FTAG, &ds); 3145 if (err) 3146 return (err); 3147 3148 /* 3149 * If someone removes a file, then tries to set the quota, we 3150 * want to make sure the file freeing takes effect. 3151 */ 3152 txg_wait_open(ds->ds_dir->dd_pool, 0); 3153 3154 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 3155 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync, 3156 ds, &psa, 0); 3157 3158 dsl_dataset_rele(ds, FTAG); 3159 return (err); 3160 } 3161 3162 static int 3163 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) 3164 { 3165 dsl_dataset_t *ds = arg1; 3166 dsl_prop_setarg_t *psa = arg2; 3167 uint64_t effective_value; 3168 uint64_t unique; 3169 int err; 3170 3171 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < 3172 SPA_VERSION_REFRESERVATION) 3173 return (ENOTSUP); 3174 3175 if (dsl_dataset_is_snapshot(ds)) 3176 return (EINVAL); 3177 3178 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) 3179 return (err); 3180 3181 effective_value = psa->psa_effective_value; 3182 3183 /* 3184 * If we are doing the preliminary check in open context, the 3185 * space estimates may be inaccurate. 3186 */ 3187 if (!dmu_tx_is_syncing(tx)) 3188 return (0); 3189 3190 mutex_enter(&ds->ds_lock); 3191 unique = dsl_dataset_unique(ds); 3192 mutex_exit(&ds->ds_lock); 3193 3194 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) { 3195 uint64_t delta = MAX(unique, effective_value) - 3196 MAX(unique, ds->ds_reserved); 3197 3198 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 3199 return (ENOSPC); 3200 if (ds->ds_quota > 0 && 3201 effective_value > ds->ds_quota) 3202 return (ENOSPC); 3203 } 3204 3205 return (0); 3206 } 3207 3208 /* ARGSUSED */ 3209 static void 3210 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr, 3211 dmu_tx_t *tx) 3212 { 3213 dsl_dataset_t *ds = arg1; 3214 dsl_prop_setarg_t *psa = arg2; 3215 uint64_t effective_value = psa->psa_effective_value; 3216 uint64_t unique; 3217 int64_t delta; 3218 3219 dsl_prop_set_sync(ds, psa, cr, tx); 3220 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa); 3221 3222 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3223 3224 mutex_enter(&ds->ds_dir->dd_lock); 3225 mutex_enter(&ds->ds_lock); 3226 unique = dsl_dataset_unique(ds); 3227 delta = MAX(0, (int64_t)(effective_value - unique)) - 3228 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3229 ds->ds_reserved = effective_value; 3230 mutex_exit(&ds->ds_lock); 3231 3232 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3233 mutex_exit(&ds->ds_dir->dd_lock); 3234 3235 spa_history_internal_log(LOG_DS_REFRESERV, 3236 ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu", 3237 (longlong_t)effective_value, ds->ds_object); 3238 } 3239 3240 int 3241 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source, 3242 uint64_t reservation) 3243 { 3244 dsl_dataset_t *ds; 3245 dsl_prop_setarg_t psa; 3246 int err; 3247 3248 dsl_prop_setarg_init_uint64(&psa, "refreservation", source, 3249 &reservation); 3250 3251 err = dsl_dataset_hold(dsname, FTAG, &ds); 3252 if (err) 3253 return (err); 3254 3255 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 3256 dsl_dataset_set_reservation_check, 3257 dsl_dataset_set_reservation_sync, ds, &psa, 0); 3258 3259 dsl_dataset_rele(ds, FTAG); 3260 return (err); 3261 } 3262 3263 struct dsl_ds_holdarg { 3264 dsl_sync_task_group_t *dstg; 3265 char *htag; 3266 char *snapname; 3267 boolean_t recursive; 3268 boolean_t gotone; 3269 boolean_t temphold; 3270 char failed[MAXPATHLEN]; 3271 }; 3272 3273 /* 3274 * The max length of a temporary tag prefix is the number of hex digits 3275 * required to express UINT64_MAX plus one for the hyphen. 3276 */ 3277 #define MAX_TAG_PREFIX_LEN 17 3278 3279 static int 3280 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx) 3281 { 3282 dsl_dataset_t *ds = arg1; 3283 struct dsl_ds_holdarg *ha = arg2; 3284 char *htag = ha->htag; 3285 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3286 int error = 0; 3287 3288 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) 3289 return (ENOTSUP); 3290 3291 if (!dsl_dataset_is_snapshot(ds)) 3292 return (EINVAL); 3293 3294 /* tags must be unique */ 3295 mutex_enter(&ds->ds_lock); 3296 if (ds->ds_phys->ds_userrefs_obj) { 3297 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag, 3298 8, 1, tx); 3299 if (error == 0) 3300 error = EEXIST; 3301 else if (error == ENOENT) 3302 error = 0; 3303 } 3304 mutex_exit(&ds->ds_lock); 3305 3306 if (error == 0 && ha->temphold && 3307 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN) 3308 error = E2BIG; 3309 3310 return (error); 3311 } 3312 3313 static void 3314 dsl_dataset_user_hold_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 3315 { 3316 dsl_dataset_t *ds = arg1; 3317 struct dsl_ds_holdarg *ha = arg2; 3318 char *htag = ha->htag; 3319 dsl_pool_t *dp = ds->ds_dir->dd_pool; 3320 objset_t *mos = dp->dp_meta_objset; 3321 uint64_t now = gethrestime_sec(); 3322 uint64_t zapobj; 3323 3324 mutex_enter(&ds->ds_lock); 3325 if (ds->ds_phys->ds_userrefs_obj == 0) { 3326 /* 3327 * This is the first user hold for this dataset. Create 3328 * the userrefs zap object. 3329 */ 3330 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3331 zapobj = ds->ds_phys->ds_userrefs_obj = 3332 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx); 3333 } else { 3334 zapobj = ds->ds_phys->ds_userrefs_obj; 3335 } 3336 ds->ds_userrefs++; 3337 mutex_exit(&ds->ds_lock); 3338 3339 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx)); 3340 3341 if (ha->temphold) { 3342 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object, 3343 htag, &now, tx)); 3344 } 3345 3346 spa_history_internal_log(LOG_DS_USER_HOLD, 3347 dp->dp_spa, tx, cr, "<%s> temp = %d dataset = %llu", htag, 3348 (int)ha->temphold, ds->ds_object); 3349 } 3350 3351 static int 3352 dsl_dataset_user_hold_one(const char *dsname, void *arg) 3353 { 3354 struct dsl_ds_holdarg *ha = arg; 3355 dsl_dataset_t *ds; 3356 int error; 3357 char *name; 3358 3359 /* alloc a buffer to hold dsname@snapname plus terminating NULL */ 3360 name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3361 error = dsl_dataset_hold(name, ha->dstg, &ds); 3362 strfree(name); 3363 if (error == 0) { 3364 ha->gotone = B_TRUE; 3365 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check, 3366 dsl_dataset_user_hold_sync, ds, ha, 0); 3367 } else if (error == ENOENT && ha->recursive) { 3368 error = 0; 3369 } else { 3370 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3371 } 3372 return (error); 3373 } 3374 3375 int 3376 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag, 3377 boolean_t recursive, boolean_t temphold) 3378 { 3379 struct dsl_ds_holdarg *ha; 3380 dsl_sync_task_t *dst; 3381 spa_t *spa; 3382 int error; 3383 3384 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); 3385 3386 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3387 3388 error = spa_open(dsname, &spa, FTAG); 3389 if (error) { 3390 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3391 return (error); 3392 } 3393 3394 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 3395 ha->htag = htag; 3396 ha->snapname = snapname; 3397 ha->recursive = recursive; 3398 ha->temphold = temphold; 3399 if (recursive) { 3400 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one, 3401 ha, DS_FIND_CHILDREN); 3402 } else { 3403 error = dsl_dataset_user_hold_one(dsname, ha); 3404 } 3405 if (error == 0) 3406 error = dsl_sync_task_group_wait(ha->dstg); 3407 3408 for (dst = list_head(&ha->dstg->dstg_tasks); dst; 3409 dst = list_next(&ha->dstg->dstg_tasks, dst)) { 3410 dsl_dataset_t *ds = dst->dst_arg1; 3411 3412 if (dst->dst_err) { 3413 dsl_dataset_name(ds, ha->failed); 3414 *strchr(ha->failed, '@') = '\0'; 3415 } 3416 dsl_dataset_rele(ds, ha->dstg); 3417 } 3418 3419 if (error == 0 && recursive && !ha->gotone) 3420 error = ENOENT; 3421 3422 if (error) 3423 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed)); 3424 3425 dsl_sync_task_group_destroy(ha->dstg); 3426 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3427 spa_close(spa, FTAG); 3428 return (error); 3429 } 3430 3431 struct dsl_ds_releasearg { 3432 dsl_dataset_t *ds; 3433 const char *htag; 3434 boolean_t own; /* do we own or just hold ds? */ 3435 }; 3436 3437 static int 3438 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag, 3439 boolean_t *might_destroy) 3440 { 3441 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3442 uint64_t zapobj; 3443 uint64_t tmp; 3444 int error; 3445 3446 *might_destroy = B_FALSE; 3447 3448 mutex_enter(&ds->ds_lock); 3449 zapobj = ds->ds_phys->ds_userrefs_obj; 3450 if (zapobj == 0) { 3451 /* The tag can't possibly exist */ 3452 mutex_exit(&ds->ds_lock); 3453 return (ESRCH); 3454 } 3455 3456 /* Make sure the tag exists */ 3457 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp); 3458 if (error) { 3459 mutex_exit(&ds->ds_lock); 3460 if (error == ENOENT) 3461 error = ESRCH; 3462 return (error); 3463 } 3464 3465 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 && 3466 DS_IS_DEFER_DESTROY(ds)) 3467 *might_destroy = B_TRUE; 3468 3469 mutex_exit(&ds->ds_lock); 3470 return (0); 3471 } 3472 3473 static int 3474 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx) 3475 { 3476 struct dsl_ds_releasearg *ra = arg1; 3477 dsl_dataset_t *ds = ra->ds; 3478 boolean_t might_destroy; 3479 int error; 3480 3481 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) 3482 return (ENOTSUP); 3483 3484 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy); 3485 if (error) 3486 return (error); 3487 3488 if (might_destroy) { 3489 struct dsl_ds_destroyarg dsda = {0}; 3490 3491 if (dmu_tx_is_syncing(tx)) { 3492 /* 3493 * If we're not prepared to remove the snapshot, 3494 * we can't allow the release to happen right now. 3495 */ 3496 if (!ra->own) 3497 return (EBUSY); 3498 if (ds->ds_objset) { 3499 dmu_objset_evict(ds->ds_objset); 3500 ds->ds_objset = NULL; 3501 } 3502 } 3503 dsda.ds = ds; 3504 dsda.releasing = B_TRUE; 3505 return (dsl_dataset_destroy_check(&dsda, tag, tx)); 3506 } 3507 3508 return (0); 3509 } 3510 3511 static void 3512 dsl_dataset_user_release_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx) 3513 { 3514 struct dsl_ds_releasearg *ra = arg1; 3515 dsl_dataset_t *ds = ra->ds; 3516 dsl_pool_t *dp = ds->ds_dir->dd_pool; 3517 objset_t *mos = dp->dp_meta_objset; 3518 uint64_t zapobj; 3519 uint64_t dsobj = ds->ds_object; 3520 uint64_t refs; 3521 int error; 3522 3523 mutex_enter(&ds->ds_lock); 3524 ds->ds_userrefs--; 3525 refs = ds->ds_userrefs; 3526 mutex_exit(&ds->ds_lock); 3527 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx); 3528 VERIFY(error == 0 || error == ENOENT); 3529 zapobj = ds->ds_phys->ds_userrefs_obj; 3530 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx)); 3531 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 && 3532 DS_IS_DEFER_DESTROY(ds)) { 3533 struct dsl_ds_destroyarg dsda = {0}; 3534 3535 ASSERT(ra->own); 3536 dsda.ds = ds; 3537 dsda.releasing = B_TRUE; 3538 /* We already did the destroy_check */ 3539 dsl_dataset_destroy_sync(&dsda, tag, cr, tx); 3540 } 3541 3542 spa_history_internal_log(LOG_DS_USER_RELEASE, 3543 dp->dp_spa, tx, cr, "<%s> %lld dataset = %llu", 3544 ra->htag, (longlong_t)refs, dsobj); 3545 } 3546 3547 static int 3548 dsl_dataset_user_release_one(const char *dsname, void *arg) 3549 { 3550 struct dsl_ds_holdarg *ha = arg; 3551 struct dsl_ds_releasearg *ra; 3552 dsl_dataset_t *ds; 3553 int error; 3554 void *dtag = ha->dstg; 3555 char *name; 3556 boolean_t own = B_FALSE; 3557 boolean_t might_destroy; 3558 3559 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */ 3560 name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3561 error = dsl_dataset_hold(name, dtag, &ds); 3562 strfree(name); 3563 if (error == ENOENT && ha->recursive) 3564 return (0); 3565 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3566 if (error) 3567 return (error); 3568 3569 ha->gotone = B_TRUE; 3570 3571 ASSERT(dsl_dataset_is_snapshot(ds)); 3572 3573 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy); 3574 if (error) { 3575 dsl_dataset_rele(ds, dtag); 3576 return (error); 3577 } 3578 3579 if (might_destroy) { 3580 #ifdef _KERNEL 3581 name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3582 error = zfs_unmount_snap(name, NULL); 3583 strfree(name); 3584 if (error) { 3585 dsl_dataset_rele(ds, dtag); 3586 return (error); 3587 } 3588 #endif 3589 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) { 3590 dsl_dataset_rele(ds, dtag); 3591 return (EBUSY); 3592 } else { 3593 own = B_TRUE; 3594 dsl_dataset_make_exclusive(ds, dtag); 3595 } 3596 } 3597 3598 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP); 3599 ra->ds = ds; 3600 ra->htag = ha->htag; 3601 ra->own = own; 3602 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check, 3603 dsl_dataset_user_release_sync, ra, dtag, 0); 3604 3605 return (0); 3606 } 3607 3608 int 3609 dsl_dataset_user_release(char *dsname, char *snapname, char *htag, 3610 boolean_t recursive) 3611 { 3612 struct dsl_ds_holdarg *ha; 3613 dsl_sync_task_t *dst; 3614 spa_t *spa; 3615 int error; 3616 3617 top: 3618 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); 3619 3620 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3621 3622 error = spa_open(dsname, &spa, FTAG); 3623 if (error) { 3624 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3625 return (error); 3626 } 3627 3628 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 3629 ha->htag = htag; 3630 ha->snapname = snapname; 3631 ha->recursive = recursive; 3632 if (recursive) { 3633 error = dmu_objset_find(dsname, dsl_dataset_user_release_one, 3634 ha, DS_FIND_CHILDREN); 3635 } else { 3636 error = dsl_dataset_user_release_one(dsname, ha); 3637 } 3638 if (error == 0) 3639 error = dsl_sync_task_group_wait(ha->dstg); 3640 3641 for (dst = list_head(&ha->dstg->dstg_tasks); dst; 3642 dst = list_next(&ha->dstg->dstg_tasks, dst)) { 3643 struct dsl_ds_releasearg *ra = dst->dst_arg1; 3644 dsl_dataset_t *ds = ra->ds; 3645 3646 if (dst->dst_err) 3647 dsl_dataset_name(ds, ha->failed); 3648 3649 if (ra->own) 3650 dsl_dataset_disown(ds, ha->dstg); 3651 else 3652 dsl_dataset_rele(ds, ha->dstg); 3653 3654 kmem_free(ra, sizeof (struct dsl_ds_releasearg)); 3655 } 3656 3657 if (error == 0 && recursive && !ha->gotone) 3658 error = ENOENT; 3659 3660 if (error && error != EBUSY) 3661 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed)); 3662 3663 dsl_sync_task_group_destroy(ha->dstg); 3664 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3665 spa_close(spa, FTAG); 3666 3667 /* 3668 * We can get EBUSY if we were racing with deferred destroy and 3669 * dsl_dataset_user_release_check() hadn't done the necessary 3670 * open context setup. We can also get EBUSY if we're racing 3671 * with destroy and that thread is the ds_owner. Either way 3672 * the busy condition should be transient, and we should retry 3673 * the release operation. 3674 */ 3675 if (error == EBUSY) 3676 goto top; 3677 3678 return (error); 3679 } 3680 3681 /* 3682 * Called at spa_load time to release a stale temporary user hold. 3683 */ 3684 int 3685 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag) 3686 { 3687 dsl_dataset_t *ds; 3688 char *snap; 3689 char *name; 3690 int namelen; 3691 int error; 3692 3693 rw_enter(&dp->dp_config_rwlock, RW_READER); 3694 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); 3695 rw_exit(&dp->dp_config_rwlock); 3696 if (error) 3697 return (error); 3698 namelen = dsl_dataset_namelen(ds)+1; 3699 name = kmem_alloc(namelen, KM_SLEEP); 3700 dsl_dataset_name(ds, name); 3701 dsl_dataset_rele(ds, FTAG); 3702 3703 snap = strchr(name, '@'); 3704 *snap = '\0'; 3705 ++snap; 3706 return (dsl_dataset_user_release(name, snap, htag, B_FALSE)); 3707 } 3708 3709 int 3710 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp) 3711 { 3712 dsl_dataset_t *ds; 3713 int err; 3714 3715 err = dsl_dataset_hold(dsname, FTAG, &ds); 3716 if (err) 3717 return (err); 3718 3719 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP)); 3720 if (ds->ds_phys->ds_userrefs_obj != 0) { 3721 zap_attribute_t *za; 3722 zap_cursor_t zc; 3723 3724 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 3725 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset, 3726 ds->ds_phys->ds_userrefs_obj); 3727 zap_cursor_retrieve(&zc, za) == 0; 3728 zap_cursor_advance(&zc)) { 3729 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name, 3730 za->za_first_integer)); 3731 } 3732 zap_cursor_fini(&zc); 3733 kmem_free(za, sizeof (zap_attribute_t)); 3734 } 3735 dsl_dataset_rele(ds, FTAG); 3736 return (0); 3737 } 3738 3739 /* 3740 * Note, this fuction is used as the callback for dmu_objset_find(). We 3741 * always return 0 so that we will continue to find and process 3742 * inconsistent datasets, even if we encounter an error trying to 3743 * process one of them. 3744 */ 3745 /* ARGSUSED */ 3746 int 3747 dsl_destroy_inconsistent(const char *dsname, void *arg) 3748 { 3749 dsl_dataset_t *ds; 3750 3751 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) { 3752 if (DS_IS_INCONSISTENT(ds)) 3753 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE); 3754 else 3755 dsl_dataset_disown(ds, FTAG); 3756 } 3757 return (0); 3758 } 3759