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 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #include <sys/dmu_objset.h> 27 #include <sys/dsl_dataset.h> 28 #include <sys/dsl_dir.h> 29 #include <sys/dsl_prop.h> 30 #include <sys/dsl_synctask.h> 31 #include <sys/dmu_traverse.h> 32 #include <sys/dmu_tx.h> 33 #include <sys/arc.h> 34 #include <sys/zio.h> 35 #include <sys/zap.h> 36 #include <sys/unique.h> 37 #include <sys/zfs_context.h> 38 #include <sys/zfs_ioctl.h> 39 #include <sys/spa.h> 40 #include <sys/zfs_znode.h> 41 #include <sys/sunddi.h> 42 #include <sys/zvol.h> 43 44 static char *dsl_reaper = "the grim reaper"; 45 46 static dsl_checkfunc_t dsl_dataset_destroy_begin_check; 47 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync; 48 static dsl_syncfunc_t dsl_dataset_set_reservation_sync; 49 50 #define DS_REF_MAX (1ULL << 62) 51 52 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE 53 54 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper) 55 56 57 /* 58 * Figure out how much of this delta should be propogated to the dsl_dir 59 * layer. If there's a refreservation, that space has already been 60 * partially accounted for in our ancestors. 61 */ 62 static int64_t 63 parent_delta(dsl_dataset_t *ds, int64_t delta) 64 { 65 uint64_t old_bytes, new_bytes; 66 67 if (ds->ds_reserved == 0) 68 return (delta); 69 70 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 71 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 72 73 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 74 return (new_bytes - old_bytes); 75 } 76 77 void 78 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 79 { 80 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 81 int compressed = BP_GET_PSIZE(bp); 82 int uncompressed = BP_GET_UCSIZE(bp); 83 int64_t delta; 84 85 dprintf_bp(bp, "born, ds=%p\n", ds); 86 87 ASSERT(dmu_tx_is_syncing(tx)); 88 /* It could have been compressed away to nothing */ 89 if (BP_IS_HOLE(bp)) 90 return; 91 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 92 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES); 93 if (ds == NULL) { 94 /* 95 * Account for the meta-objset space in its placeholder 96 * dsl_dir. 97 */ 98 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */ 99 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, 100 used, compressed, uncompressed, tx); 101 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); 102 return; 103 } 104 dmu_buf_will_dirty(ds->ds_dbuf, tx); 105 mutex_enter(&ds->ds_dir->dd_lock); 106 mutex_enter(&ds->ds_lock); 107 delta = parent_delta(ds, used); 108 ds->ds_phys->ds_used_bytes += used; 109 ds->ds_phys->ds_compressed_bytes += compressed; 110 ds->ds_phys->ds_uncompressed_bytes += uncompressed; 111 ds->ds_phys->ds_unique_bytes += used; 112 mutex_exit(&ds->ds_lock); 113 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 114 compressed, uncompressed, tx); 115 dsl_dir_transfer_space(ds->ds_dir, used - delta, 116 DD_USED_REFRSRV, DD_USED_HEAD, tx); 117 mutex_exit(&ds->ds_dir->dd_lock); 118 } 119 120 int 121 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 122 boolean_t async) 123 { 124 if (BP_IS_HOLE(bp)) 125 return (0); 126 127 ASSERT(dmu_tx_is_syncing(tx)); 128 ASSERT(bp->blk_birth <= tx->tx_txg); 129 130 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 131 int compressed = BP_GET_PSIZE(bp); 132 int uncompressed = BP_GET_UCSIZE(bp); 133 134 ASSERT(used > 0); 135 if (ds == NULL) { 136 /* 137 * Account for the meta-objset space in its placeholder 138 * dataset. 139 */ 140 dsl_free(tx->tx_pool, tx->tx_txg, bp); 141 142 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, 143 -used, -compressed, -uncompressed, tx); 144 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); 145 return (used); 146 } 147 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 148 149 ASSERT(!dsl_dataset_is_snapshot(ds)); 150 dmu_buf_will_dirty(ds->ds_dbuf, tx); 151 152 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { 153 int64_t delta; 154 155 dprintf_bp(bp, "freeing: %s", ""); 156 dsl_free(tx->tx_pool, tx->tx_txg, bp); 157 158 mutex_enter(&ds->ds_dir->dd_lock); 159 mutex_enter(&ds->ds_lock); 160 ASSERT(ds->ds_phys->ds_unique_bytes >= used || 161 !DS_UNIQUE_IS_ACCURATE(ds)); 162 delta = parent_delta(ds, -used); 163 ds->ds_phys->ds_unique_bytes -= used; 164 mutex_exit(&ds->ds_lock); 165 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 166 delta, -compressed, -uncompressed, tx); 167 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 168 DD_USED_REFRSRV, DD_USED_HEAD, tx); 169 mutex_exit(&ds->ds_dir->dd_lock); 170 } else { 171 dprintf_bp(bp, "putting on dead list: %s", ""); 172 if (async) { 173 /* 174 * We are here as part of zio's write done callback, 175 * which means we're a zio interrupt thread. We can't 176 * call bplist_enqueue() now because it may block 177 * waiting for I/O. Instead, put bp on the deferred 178 * queue and let dsl_pool_sync() finish the job. 179 */ 180 bplist_enqueue_deferred(&ds->ds_deadlist, bp); 181 } else { 182 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx)); 183 } 184 ASSERT3U(ds->ds_prev->ds_object, ==, 185 ds->ds_phys->ds_prev_snap_obj); 186 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); 187 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 188 if (ds->ds_prev->ds_phys->ds_next_snap_obj == 189 ds->ds_object && bp->blk_birth > 190 ds->ds_prev->ds_phys->ds_prev_snap_txg) { 191 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 192 mutex_enter(&ds->ds_prev->ds_lock); 193 ds->ds_prev->ds_phys->ds_unique_bytes += used; 194 mutex_exit(&ds->ds_prev->ds_lock); 195 } 196 if (bp->blk_birth > ds->ds_origin_txg) { 197 dsl_dir_transfer_space(ds->ds_dir, used, 198 DD_USED_HEAD, DD_USED_SNAP, tx); 199 } 200 } 201 mutex_enter(&ds->ds_lock); 202 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used); 203 ds->ds_phys->ds_used_bytes -= used; 204 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); 205 ds->ds_phys->ds_compressed_bytes -= compressed; 206 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); 207 ds->ds_phys->ds_uncompressed_bytes -= uncompressed; 208 mutex_exit(&ds->ds_lock); 209 210 return (used); 211 } 212 213 uint64_t 214 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 215 { 216 uint64_t trysnap = 0; 217 218 if (ds == NULL) 219 return (0); 220 /* 221 * The snapshot creation could fail, but that would cause an 222 * incorrect FALSE return, which would only result in an 223 * overestimation of the amount of space that an operation would 224 * consume, which is OK. 225 * 226 * There's also a small window where we could miss a pending 227 * snapshot, because we could set the sync task in the quiescing 228 * phase. So this should only be used as a guess. 229 */ 230 if (ds->ds_trysnap_txg > 231 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 232 trysnap = ds->ds_trysnap_txg; 233 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap)); 234 } 235 236 boolean_t 237 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth) 238 { 239 return (blk_birth > dsl_dataset_prev_snap_txg(ds)); 240 } 241 242 /* ARGSUSED */ 243 static void 244 dsl_dataset_evict(dmu_buf_t *db, void *dsv) 245 { 246 dsl_dataset_t *ds = dsv; 247 248 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds)); 249 250 unique_remove(ds->ds_fsid_guid); 251 252 if (ds->ds_objset != NULL) 253 dmu_objset_evict(ds->ds_objset); 254 255 if (ds->ds_prev) { 256 dsl_dataset_drop_ref(ds->ds_prev, ds); 257 ds->ds_prev = NULL; 258 } 259 260 bplist_close(&ds->ds_deadlist); 261 if (ds->ds_dir) 262 dsl_dir_close(ds->ds_dir, ds); 263 264 ASSERT(!list_link_active(&ds->ds_synced_link)); 265 266 mutex_destroy(&ds->ds_lock); 267 mutex_destroy(&ds->ds_recvlock); 268 mutex_destroy(&ds->ds_opening_lock); 269 rw_destroy(&ds->ds_rwlock); 270 cv_destroy(&ds->ds_exclusive_cv); 271 bplist_fini(&ds->ds_deadlist); 272 273 kmem_free(ds, sizeof (dsl_dataset_t)); 274 } 275 276 static int 277 dsl_dataset_get_snapname(dsl_dataset_t *ds) 278 { 279 dsl_dataset_phys_t *headphys; 280 int err; 281 dmu_buf_t *headdbuf; 282 dsl_pool_t *dp = ds->ds_dir->dd_pool; 283 objset_t *mos = dp->dp_meta_objset; 284 285 if (ds->ds_snapname[0]) 286 return (0); 287 if (ds->ds_phys->ds_next_snap_obj == 0) 288 return (0); 289 290 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, 291 FTAG, &headdbuf); 292 if (err) 293 return (err); 294 headphys = headdbuf->db_data; 295 err = zap_value_search(dp->dp_meta_objset, 296 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 297 dmu_buf_rele(headdbuf, FTAG); 298 return (err); 299 } 300 301 static int 302 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 303 { 304 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 305 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 306 matchtype_t mt; 307 int err; 308 309 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 310 mt = MT_FIRST; 311 else 312 mt = MT_EXACT; 313 314 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 315 value, mt, NULL, 0, NULL); 316 if (err == ENOTSUP && mt == MT_FIRST) 317 err = zap_lookup(mos, snapobj, name, 8, 1, value); 318 return (err); 319 } 320 321 static int 322 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx) 323 { 324 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 325 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 326 matchtype_t mt; 327 int err; 328 329 dsl_dir_snap_cmtime_update(ds->ds_dir); 330 331 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 332 mt = MT_FIRST; 333 else 334 mt = MT_EXACT; 335 336 err = zap_remove_norm(mos, snapobj, name, mt, tx); 337 if (err == ENOTSUP && mt == MT_FIRST) 338 err = zap_remove(mos, snapobj, name, tx); 339 return (err); 340 } 341 342 static int 343 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, 344 dsl_dataset_t **dsp) 345 { 346 objset_t *mos = dp->dp_meta_objset; 347 dmu_buf_t *dbuf; 348 dsl_dataset_t *ds; 349 int err; 350 351 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 352 dsl_pool_sync_context(dp)); 353 354 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 355 if (err) 356 return (err); 357 ds = dmu_buf_get_user(dbuf); 358 if (ds == NULL) { 359 dsl_dataset_t *winner; 360 361 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 362 ds->ds_dbuf = dbuf; 363 ds->ds_object = dsobj; 364 ds->ds_phys = dbuf->db_data; 365 366 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 367 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL); 368 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 369 rw_init(&ds->ds_rwlock, 0, 0, 0); 370 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL); 371 bplist_init(&ds->ds_deadlist); 372 373 err = bplist_open(&ds->ds_deadlist, 374 mos, ds->ds_phys->ds_deadlist_obj); 375 if (err == 0) { 376 err = dsl_dir_open_obj(dp, 377 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); 378 } 379 if (err) { 380 /* 381 * we don't really need to close the blist if we 382 * just opened it. 383 */ 384 mutex_destroy(&ds->ds_lock); 385 mutex_destroy(&ds->ds_recvlock); 386 mutex_destroy(&ds->ds_opening_lock); 387 rw_destroy(&ds->ds_rwlock); 388 cv_destroy(&ds->ds_exclusive_cv); 389 bplist_fini(&ds->ds_deadlist); 390 kmem_free(ds, sizeof (dsl_dataset_t)); 391 dmu_buf_rele(dbuf, tag); 392 return (err); 393 } 394 395 if (!dsl_dataset_is_snapshot(ds)) { 396 ds->ds_snapname[0] = '\0'; 397 if (ds->ds_phys->ds_prev_snap_obj) { 398 err = dsl_dataset_get_ref(dp, 399 ds->ds_phys->ds_prev_snap_obj, 400 ds, &ds->ds_prev); 401 } 402 403 if (err == 0 && dsl_dir_is_clone(ds->ds_dir)) { 404 dsl_dataset_t *origin; 405 406 err = dsl_dataset_hold_obj(dp, 407 ds->ds_dir->dd_phys->dd_origin_obj, 408 FTAG, &origin); 409 if (err == 0) { 410 ds->ds_origin_txg = 411 origin->ds_phys->ds_creation_txg; 412 dsl_dataset_rele(origin, FTAG); 413 } 414 } 415 } else { 416 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 417 err = dsl_dataset_get_snapname(ds); 418 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) { 419 err = zap_count( 420 ds->ds_dir->dd_pool->dp_meta_objset, 421 ds->ds_phys->ds_userrefs_obj, 422 &ds->ds_userrefs); 423 } 424 } 425 426 if (err == 0 && !dsl_dataset_is_snapshot(ds)) { 427 /* 428 * In sync context, we're called with either no lock 429 * or with the write lock. If we're not syncing, 430 * we're always called with the read lock held. 431 */ 432 boolean_t need_lock = 433 !RW_WRITE_HELD(&dp->dp_config_rwlock) && 434 dsl_pool_sync_context(dp); 435 436 if (need_lock) 437 rw_enter(&dp->dp_config_rwlock, RW_READER); 438 439 err = dsl_prop_get_ds(ds, 440 "refreservation", sizeof (uint64_t), 1, 441 &ds->ds_reserved, NULL); 442 if (err == 0) { 443 err = dsl_prop_get_ds(ds, 444 "refquota", sizeof (uint64_t), 1, 445 &ds->ds_quota, NULL); 446 } 447 448 if (need_lock) 449 rw_exit(&dp->dp_config_rwlock); 450 } else { 451 ds->ds_reserved = ds->ds_quota = 0; 452 } 453 454 if (err == 0) { 455 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys, 456 dsl_dataset_evict); 457 } 458 if (err || winner) { 459 bplist_close(&ds->ds_deadlist); 460 if (ds->ds_prev) 461 dsl_dataset_drop_ref(ds->ds_prev, ds); 462 dsl_dir_close(ds->ds_dir, ds); 463 mutex_destroy(&ds->ds_lock); 464 mutex_destroy(&ds->ds_recvlock); 465 mutex_destroy(&ds->ds_opening_lock); 466 rw_destroy(&ds->ds_rwlock); 467 cv_destroy(&ds->ds_exclusive_cv); 468 bplist_fini(&ds->ds_deadlist); 469 kmem_free(ds, sizeof (dsl_dataset_t)); 470 if (err) { 471 dmu_buf_rele(dbuf, tag); 472 return (err); 473 } 474 ds = winner; 475 } else { 476 ds->ds_fsid_guid = 477 unique_insert(ds->ds_phys->ds_fsid_guid); 478 } 479 } 480 ASSERT3P(ds->ds_dbuf, ==, dbuf); 481 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); 482 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 || 483 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 484 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 485 mutex_enter(&ds->ds_lock); 486 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) { 487 mutex_exit(&ds->ds_lock); 488 dmu_buf_rele(ds->ds_dbuf, tag); 489 return (ENOENT); 490 } 491 mutex_exit(&ds->ds_lock); 492 *dsp = ds; 493 return (0); 494 } 495 496 static int 497 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag) 498 { 499 dsl_pool_t *dp = ds->ds_dir->dd_pool; 500 501 /* 502 * In syncing context we don't want the rwlock lock: there 503 * may be an existing writer waiting for sync phase to 504 * finish. We don't need to worry about such writers, since 505 * sync phase is single-threaded, so the writer can't be 506 * doing anything while we are active. 507 */ 508 if (dsl_pool_sync_context(dp)) { 509 ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); 510 return (0); 511 } 512 513 /* 514 * Normal users will hold the ds_rwlock as a READER until they 515 * are finished (i.e., call dsl_dataset_rele()). "Owners" will 516 * drop their READER lock after they set the ds_owner field. 517 * 518 * If the dataset is being destroyed, the destroy thread will 519 * obtain a WRITER lock for exclusive access after it's done its 520 * open-context work and then change the ds_owner to 521 * dsl_reaper once destruction is assured. So threads 522 * may block here temporarily, until the "destructability" of 523 * the dataset is determined. 524 */ 525 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock)); 526 mutex_enter(&ds->ds_lock); 527 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) { 528 rw_exit(&dp->dp_config_rwlock); 529 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock); 530 if (DSL_DATASET_IS_DESTROYED(ds)) { 531 mutex_exit(&ds->ds_lock); 532 dsl_dataset_drop_ref(ds, tag); 533 rw_enter(&dp->dp_config_rwlock, RW_READER); 534 return (ENOENT); 535 } 536 rw_enter(&dp->dp_config_rwlock, RW_READER); 537 } 538 mutex_exit(&ds->ds_lock); 539 return (0); 540 } 541 542 int 543 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 544 dsl_dataset_t **dsp) 545 { 546 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp); 547 548 if (err) 549 return (err); 550 return (dsl_dataset_hold_ref(*dsp, tag)); 551 } 552 553 int 554 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok, 555 void *tag, dsl_dataset_t **dsp) 556 { 557 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 558 if (err) 559 return (err); 560 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { 561 dsl_dataset_rele(*dsp, tag); 562 *dsp = NULL; 563 return (EBUSY); 564 } 565 return (0); 566 } 567 568 int 569 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp) 570 { 571 dsl_dir_t *dd; 572 dsl_pool_t *dp; 573 const char *snapname; 574 uint64_t obj; 575 int err = 0; 576 577 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname); 578 if (err) 579 return (err); 580 581 dp = dd->dd_pool; 582 obj = dd->dd_phys->dd_head_dataset_obj; 583 rw_enter(&dp->dp_config_rwlock, RW_READER); 584 if (obj) 585 err = dsl_dataset_get_ref(dp, obj, tag, dsp); 586 else 587 err = ENOENT; 588 if (err) 589 goto out; 590 591 err = dsl_dataset_hold_ref(*dsp, tag); 592 593 /* we may be looking for a snapshot */ 594 if (err == 0 && snapname != NULL) { 595 dsl_dataset_t *ds = NULL; 596 597 if (*snapname++ != '@') { 598 dsl_dataset_rele(*dsp, tag); 599 err = ENOENT; 600 goto out; 601 } 602 603 dprintf("looking for snapshot '%s'\n", snapname); 604 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); 605 if (err == 0) 606 err = dsl_dataset_get_ref(dp, obj, tag, &ds); 607 dsl_dataset_rele(*dsp, tag); 608 609 ASSERT3U((err == 0), ==, (ds != NULL)); 610 611 if (ds) { 612 mutex_enter(&ds->ds_lock); 613 if (ds->ds_snapname[0] == 0) 614 (void) strlcpy(ds->ds_snapname, snapname, 615 sizeof (ds->ds_snapname)); 616 mutex_exit(&ds->ds_lock); 617 err = dsl_dataset_hold_ref(ds, tag); 618 *dsp = err ? NULL : ds; 619 } 620 } 621 out: 622 rw_exit(&dp->dp_config_rwlock); 623 dsl_dir_close(dd, FTAG); 624 return (err); 625 } 626 627 int 628 dsl_dataset_own(const char *name, boolean_t inconsistentok, 629 void *tag, dsl_dataset_t **dsp) 630 { 631 int err = dsl_dataset_hold(name, tag, dsp); 632 if (err) 633 return (err); 634 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { 635 dsl_dataset_rele(*dsp, tag); 636 return (EBUSY); 637 } 638 return (0); 639 } 640 641 void 642 dsl_dataset_name(dsl_dataset_t *ds, char *name) 643 { 644 if (ds == NULL) { 645 (void) strcpy(name, "mos"); 646 } else { 647 dsl_dir_name(ds->ds_dir, name); 648 VERIFY(0 == dsl_dataset_get_snapname(ds)); 649 if (ds->ds_snapname[0]) { 650 (void) strcat(name, "@"); 651 /* 652 * We use a "recursive" mutex so that we 653 * can call dprintf_ds() with ds_lock held. 654 */ 655 if (!MUTEX_HELD(&ds->ds_lock)) { 656 mutex_enter(&ds->ds_lock); 657 (void) strcat(name, ds->ds_snapname); 658 mutex_exit(&ds->ds_lock); 659 } else { 660 (void) strcat(name, ds->ds_snapname); 661 } 662 } 663 } 664 } 665 666 static int 667 dsl_dataset_namelen(dsl_dataset_t *ds) 668 { 669 int result; 670 671 if (ds == NULL) { 672 result = 3; /* "mos" */ 673 } else { 674 result = dsl_dir_namelen(ds->ds_dir); 675 VERIFY(0 == dsl_dataset_get_snapname(ds)); 676 if (ds->ds_snapname[0]) { 677 ++result; /* adding one for the @-sign */ 678 if (!MUTEX_HELD(&ds->ds_lock)) { 679 mutex_enter(&ds->ds_lock); 680 result += strlen(ds->ds_snapname); 681 mutex_exit(&ds->ds_lock); 682 } else { 683 result += strlen(ds->ds_snapname); 684 } 685 } 686 } 687 688 return (result); 689 } 690 691 void 692 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag) 693 { 694 dmu_buf_rele(ds->ds_dbuf, tag); 695 } 696 697 void 698 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 699 { 700 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) { 701 rw_exit(&ds->ds_rwlock); 702 } 703 dsl_dataset_drop_ref(ds, tag); 704 } 705 706 void 707 dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 708 { 709 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) || 710 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL)); 711 712 mutex_enter(&ds->ds_lock); 713 ds->ds_owner = NULL; 714 if (RW_WRITE_HELD(&ds->ds_rwlock)) { 715 rw_exit(&ds->ds_rwlock); 716 cv_broadcast(&ds->ds_exclusive_cv); 717 } 718 mutex_exit(&ds->ds_lock); 719 if (ds->ds_dbuf) 720 dsl_dataset_drop_ref(ds, tag); 721 else 722 dsl_dataset_evict(ds->ds_dbuf, ds); 723 } 724 725 boolean_t 726 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag) 727 { 728 boolean_t gotit = FALSE; 729 730 mutex_enter(&ds->ds_lock); 731 if (ds->ds_owner == NULL && 732 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) { 733 ds->ds_owner = tag; 734 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) 735 rw_exit(&ds->ds_rwlock); 736 gotit = TRUE; 737 } 738 mutex_exit(&ds->ds_lock); 739 return (gotit); 740 } 741 742 void 743 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner) 744 { 745 ASSERT3P(owner, ==, ds->ds_owner); 746 if (!RW_WRITE_HELD(&ds->ds_rwlock)) 747 rw_enter(&ds->ds_rwlock, RW_WRITER); 748 } 749 750 uint64_t 751 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 752 uint64_t flags, dmu_tx_t *tx) 753 { 754 dsl_pool_t *dp = dd->dd_pool; 755 dmu_buf_t *dbuf; 756 dsl_dataset_phys_t *dsphys; 757 uint64_t dsobj; 758 objset_t *mos = dp->dp_meta_objset; 759 760 if (origin == NULL) 761 origin = dp->dp_origin_snap; 762 763 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 764 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0); 765 ASSERT(dmu_tx_is_syncing(tx)); 766 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 767 768 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 769 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 770 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 771 dmu_buf_will_dirty(dbuf, tx); 772 dsphys = dbuf->db_data; 773 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 774 dsphys->ds_dir_obj = dd->dd_object; 775 dsphys->ds_flags = flags; 776 dsphys->ds_fsid_guid = unique_create(); 777 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 778 sizeof (dsphys->ds_guid)); 779 dsphys->ds_snapnames_zapobj = 780 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 781 DMU_OT_NONE, 0, tx); 782 dsphys->ds_creation_time = gethrestime_sec(); 783 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 784 dsphys->ds_deadlist_obj = 785 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); 786 787 if (origin) { 788 dsphys->ds_prev_snap_obj = origin->ds_object; 789 dsphys->ds_prev_snap_txg = 790 origin->ds_phys->ds_creation_txg; 791 dsphys->ds_used_bytes = 792 origin->ds_phys->ds_used_bytes; 793 dsphys->ds_compressed_bytes = 794 origin->ds_phys->ds_compressed_bytes; 795 dsphys->ds_uncompressed_bytes = 796 origin->ds_phys->ds_uncompressed_bytes; 797 dsphys->ds_bp = origin->ds_phys->ds_bp; 798 dsphys->ds_flags |= origin->ds_phys->ds_flags; 799 800 dmu_buf_will_dirty(origin->ds_dbuf, tx); 801 origin->ds_phys->ds_num_children++; 802 803 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 804 if (origin->ds_phys->ds_next_clones_obj == 0) { 805 origin->ds_phys->ds_next_clones_obj = 806 zap_create(mos, 807 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 808 } 809 VERIFY(0 == zap_add_int(mos, 810 origin->ds_phys->ds_next_clones_obj, 811 dsobj, tx)); 812 } 813 814 dmu_buf_will_dirty(dd->dd_dbuf, tx); 815 dd->dd_phys->dd_origin_obj = origin->ds_object; 816 } 817 818 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 819 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 820 821 dmu_buf_rele(dbuf, FTAG); 822 823 dmu_buf_will_dirty(dd->dd_dbuf, tx); 824 dd->dd_phys->dd_head_dataset_obj = dsobj; 825 826 return (dsobj); 827 } 828 829 uint64_t 830 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 831 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 832 { 833 dsl_pool_t *dp = pdd->dd_pool; 834 uint64_t dsobj, ddobj; 835 dsl_dir_t *dd; 836 837 ASSERT(lastname[0] != '@'); 838 839 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 840 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd)); 841 842 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx); 843 844 dsl_deleg_set_create_perms(dd, tx, cr); 845 846 dsl_dir_close(dd, FTAG); 847 848 return (dsobj); 849 } 850 851 struct destroyarg { 852 dsl_sync_task_group_t *dstg; 853 char *snapname; 854 char *failed; 855 boolean_t defer; 856 }; 857 858 static int 859 dsl_snapshot_destroy_one(char *name, void *arg) 860 { 861 struct destroyarg *da = arg; 862 dsl_dataset_t *ds; 863 int err; 864 char *dsname; 865 866 dsname = kmem_asprintf("%s@%s", name, da->snapname); 867 err = dsl_dataset_own(dsname, B_TRUE, da->dstg, &ds); 868 strfree(dsname); 869 if (err == 0) { 870 struct dsl_ds_destroyarg *dsda; 871 872 dsl_dataset_make_exclusive(ds, da->dstg); 873 if (ds->ds_objset != NULL) { 874 dmu_objset_evict(ds->ds_objset); 875 ds->ds_objset = NULL; 876 } 877 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP); 878 dsda->ds = ds; 879 dsda->defer = da->defer; 880 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check, 881 dsl_dataset_destroy_sync, dsda, da->dstg, 0); 882 } else if (err == ENOENT) { 883 err = 0; 884 } else { 885 (void) strcpy(da->failed, name); 886 } 887 return (err); 888 } 889 890 /* 891 * Destroy 'snapname' in all descendants of 'fsname'. 892 */ 893 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy 894 int 895 dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer) 896 { 897 int err; 898 struct destroyarg da; 899 dsl_sync_task_t *dst; 900 spa_t *spa; 901 902 err = spa_open(fsname, &spa, FTAG); 903 if (err) 904 return (err); 905 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 906 da.snapname = snapname; 907 da.failed = fsname; 908 da.defer = defer; 909 910 err = dmu_objset_find(fsname, 911 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN); 912 913 if (err == 0) 914 err = dsl_sync_task_group_wait(da.dstg); 915 916 for (dst = list_head(&da.dstg->dstg_tasks); dst; 917 dst = list_next(&da.dstg->dstg_tasks, dst)) { 918 struct dsl_ds_destroyarg *dsda = dst->dst_arg1; 919 dsl_dataset_t *ds = dsda->ds; 920 921 /* 922 * Return the file system name that triggered the error 923 */ 924 if (dst->dst_err) { 925 dsl_dataset_name(ds, fsname); 926 *strchr(fsname, '@') = '\0'; 927 } 928 ASSERT3P(dsda->rm_origin, ==, NULL); 929 dsl_dataset_disown(ds, da.dstg); 930 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg)); 931 } 932 933 dsl_sync_task_group_destroy(da.dstg); 934 spa_close(spa, FTAG); 935 return (err); 936 } 937 938 static boolean_t 939 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds) 940 { 941 boolean_t might_destroy = B_FALSE; 942 943 mutex_enter(&ds->ds_lock); 944 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 && 945 DS_IS_DEFER_DESTROY(ds)) 946 might_destroy = B_TRUE; 947 mutex_exit(&ds->ds_lock); 948 949 return (might_destroy); 950 } 951 952 /* 953 * If we're removing a clone, and these three conditions are true: 954 * 1) the clone's origin has no other children 955 * 2) the clone's origin has no user references 956 * 3) the clone's origin has been marked for deferred destruction 957 * Then, prepare to remove the origin as part of this sync task group. 958 */ 959 static int 960 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag) 961 { 962 dsl_dataset_t *ds = dsda->ds; 963 dsl_dataset_t *origin = ds->ds_prev; 964 965 if (dsl_dataset_might_destroy_origin(origin)) { 966 char *name; 967 int namelen; 968 int error; 969 970 namelen = dsl_dataset_namelen(origin) + 1; 971 name = kmem_alloc(namelen, KM_SLEEP); 972 dsl_dataset_name(origin, name); 973 #ifdef _KERNEL 974 error = zfs_unmount_snap(name, NULL); 975 if (error) { 976 kmem_free(name, namelen); 977 return (error); 978 } 979 #endif 980 error = dsl_dataset_own(name, B_TRUE, tag, &origin); 981 kmem_free(name, namelen); 982 if (error) 983 return (error); 984 dsda->rm_origin = origin; 985 dsl_dataset_make_exclusive(origin, tag); 986 987 if (origin->ds_objset != NULL) { 988 dmu_objset_evict(origin->ds_objset); 989 origin->ds_objset = NULL; 990 } 991 } 992 993 return (0); 994 } 995 996 /* 997 * ds must be opened as OWNER. On return (whether successful or not), 998 * ds will be closed and caller can no longer dereference it. 999 */ 1000 int 1001 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer) 1002 { 1003 int err; 1004 dsl_sync_task_group_t *dstg; 1005 objset_t *os; 1006 dsl_dir_t *dd; 1007 uint64_t obj; 1008 struct dsl_ds_destroyarg dsda = {0}; 1009 1010 dsda.ds = ds; 1011 1012 if (dsl_dataset_is_snapshot(ds)) { 1013 /* Destroying a snapshot is simpler */ 1014 dsl_dataset_make_exclusive(ds, tag); 1015 1016 if (ds->ds_objset != NULL) { 1017 dmu_objset_evict(ds->ds_objset); 1018 ds->ds_objset = NULL; 1019 } 1020 dsda.defer = defer; 1021 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 1022 dsl_dataset_destroy_check, dsl_dataset_destroy_sync, 1023 &dsda, tag, 0); 1024 ASSERT3P(dsda.rm_origin, ==, NULL); 1025 goto out; 1026 } else if (defer) { 1027 err = EINVAL; 1028 goto out; 1029 } 1030 1031 dd = ds->ds_dir; 1032 1033 /* 1034 * Check for errors and mark this ds as inconsistent, in 1035 * case we crash while freeing the objects. 1036 */ 1037 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check, 1038 dsl_dataset_destroy_begin_sync, ds, NULL, 0); 1039 if (err) 1040 goto out; 1041 1042 err = dmu_objset_from_ds(ds, &os); 1043 if (err) 1044 goto out; 1045 1046 /* 1047 * remove the objects in open context, so that we won't 1048 * have too much to do in syncing context. 1049 */ 1050 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 1051 ds->ds_phys->ds_prev_snap_txg)) { 1052 /* 1053 * Ignore errors, if there is not enough disk space 1054 * we will deal with it in dsl_dataset_destroy_sync(). 1055 */ 1056 (void) dmu_free_object(os, obj); 1057 } 1058 1059 /* 1060 * We need to sync out all in-flight IO before we try to evict 1061 * (the dataset evict func is trying to clear the cached entries 1062 * for this dataset in the ARC). 1063 */ 1064 txg_wait_synced(dd->dd_pool, 0); 1065 1066 /* 1067 * If we managed to free all the objects in open 1068 * context, the user space accounting should be zero. 1069 */ 1070 if (ds->ds_phys->ds_bp.blk_fill == 0 && 1071 dmu_objset_userused_enabled(os)) { 1072 uint64_t count; 1073 1074 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 || 1075 count == 0); 1076 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 || 1077 count == 0); 1078 } 1079 1080 if (err != ESRCH) 1081 goto out; 1082 1083 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); 1084 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd); 1085 rw_exit(&dd->dd_pool->dp_config_rwlock); 1086 1087 if (err) 1088 goto out; 1089 1090 if (ds->ds_objset) { 1091 /* 1092 * We need to sync out all in-flight IO before we try 1093 * to evict (the dataset evict func is trying to clear 1094 * the cached entries for this dataset in the ARC). 1095 */ 1096 txg_wait_synced(dd->dd_pool, 0); 1097 } 1098 1099 /* 1100 * Blow away the dsl_dir + head dataset. 1101 */ 1102 dsl_dataset_make_exclusive(ds, tag); 1103 if (ds->ds_objset) { 1104 dmu_objset_evict(ds->ds_objset); 1105 ds->ds_objset = NULL; 1106 } 1107 1108 /* 1109 * If we're removing a clone, we might also need to remove its 1110 * origin. 1111 */ 1112 do { 1113 dsda.need_prep = B_FALSE; 1114 if (dsl_dir_is_clone(dd)) { 1115 err = dsl_dataset_origin_rm_prep(&dsda, tag); 1116 if (err) { 1117 dsl_dir_close(dd, FTAG); 1118 goto out; 1119 } 1120 } 1121 1122 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool); 1123 dsl_sync_task_create(dstg, dsl_dataset_destroy_check, 1124 dsl_dataset_destroy_sync, &dsda, tag, 0); 1125 dsl_sync_task_create(dstg, dsl_dir_destroy_check, 1126 dsl_dir_destroy_sync, dd, FTAG, 0); 1127 err = dsl_sync_task_group_wait(dstg); 1128 dsl_sync_task_group_destroy(dstg); 1129 1130 /* 1131 * We could be racing against 'zfs release' or 'zfs destroy -d' 1132 * on the origin snap, in which case we can get EBUSY if we 1133 * needed to destroy the origin snap but were not ready to 1134 * do so. 1135 */ 1136 if (dsda.need_prep) { 1137 ASSERT(err == EBUSY); 1138 ASSERT(dsl_dir_is_clone(dd)); 1139 ASSERT(dsda.rm_origin == NULL); 1140 } 1141 } while (dsda.need_prep); 1142 1143 if (dsda.rm_origin != NULL) 1144 dsl_dataset_disown(dsda.rm_origin, tag); 1145 1146 /* if it is successful, dsl_dir_destroy_sync will close the dd */ 1147 if (err) 1148 dsl_dir_close(dd, FTAG); 1149 out: 1150 dsl_dataset_disown(ds, tag); 1151 return (err); 1152 } 1153 1154 blkptr_t * 1155 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1156 { 1157 return (&ds->ds_phys->ds_bp); 1158 } 1159 1160 void 1161 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 1162 { 1163 ASSERT(dmu_tx_is_syncing(tx)); 1164 /* If it's the meta-objset, set dp_meta_rootbp */ 1165 if (ds == NULL) { 1166 tx->tx_pool->dp_meta_rootbp = *bp; 1167 } else { 1168 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1169 ds->ds_phys->ds_bp = *bp; 1170 } 1171 } 1172 1173 spa_t * 1174 dsl_dataset_get_spa(dsl_dataset_t *ds) 1175 { 1176 return (ds->ds_dir->dd_pool->dp_spa); 1177 } 1178 1179 void 1180 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1181 { 1182 dsl_pool_t *dp; 1183 1184 if (ds == NULL) /* this is the meta-objset */ 1185 return; 1186 1187 ASSERT(ds->ds_objset != NULL); 1188 1189 if (ds->ds_phys->ds_next_snap_obj != 0) 1190 panic("dirtying snapshot!"); 1191 1192 dp = ds->ds_dir->dd_pool; 1193 1194 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) { 1195 /* up the hold count until we can be written out */ 1196 dmu_buf_add_ref(ds->ds_dbuf, ds); 1197 } 1198 } 1199 1200 /* 1201 * The unique space in the head dataset can be calculated by subtracting 1202 * the space used in the most recent snapshot, that is still being used 1203 * in this file system, from the space currently in use. To figure out 1204 * the space in the most recent snapshot still in use, we need to take 1205 * the total space used in the snapshot and subtract out the space that 1206 * has been freed up since the snapshot was taken. 1207 */ 1208 static void 1209 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 1210 { 1211 uint64_t mrs_used; 1212 uint64_t dlused, dlcomp, dluncomp; 1213 1214 ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj); 1215 1216 if (ds->ds_phys->ds_prev_snap_obj != 0) 1217 mrs_used = ds->ds_prev->ds_phys->ds_used_bytes; 1218 else 1219 mrs_used = 0; 1220 1221 VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp, 1222 &dluncomp)); 1223 1224 ASSERT3U(dlused, <=, mrs_used); 1225 ds->ds_phys->ds_unique_bytes = 1226 ds->ds_phys->ds_used_bytes - (mrs_used - dlused); 1227 1228 if (!DS_UNIQUE_IS_ACCURATE(ds) && 1229 spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1230 SPA_VERSION_UNIQUE_ACCURATE) 1231 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1232 } 1233 1234 static uint64_t 1235 dsl_dataset_unique(dsl_dataset_t *ds) 1236 { 1237 if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds)) 1238 dsl_dataset_recalc_head_uniq(ds); 1239 1240 return (ds->ds_phys->ds_unique_bytes); 1241 } 1242 1243 struct killarg { 1244 dsl_dataset_t *ds; 1245 dmu_tx_t *tx; 1246 }; 1247 1248 /* ARGSUSED */ 1249 static int 1250 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1251 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 1252 { 1253 struct killarg *ka = arg; 1254 dmu_tx_t *tx = ka->tx; 1255 1256 if (bp == NULL) 1257 return (0); 1258 1259 if (zb->zb_level == ZB_ZIL_LEVEL) { 1260 ASSERT(zilog != NULL); 1261 /* 1262 * It's a block in the intent log. It has no 1263 * accounting, so just free it. 1264 */ 1265 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp); 1266 } else { 1267 ASSERT(zilog == NULL); 1268 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg); 1269 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE); 1270 } 1271 1272 return (0); 1273 } 1274 1275 /* ARGSUSED */ 1276 static int 1277 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx) 1278 { 1279 dsl_dataset_t *ds = arg1; 1280 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1281 uint64_t count; 1282 int err; 1283 1284 /* 1285 * Can't delete a head dataset if there are snapshots of it. 1286 * (Except if the only snapshots are from the branch we cloned 1287 * from.) 1288 */ 1289 if (ds->ds_prev != NULL && 1290 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) 1291 return (EBUSY); 1292 1293 /* 1294 * This is really a dsl_dir thing, but check it here so that 1295 * we'll be less likely to leave this dataset inconsistent & 1296 * nearly destroyed. 1297 */ 1298 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count); 1299 if (err) 1300 return (err); 1301 if (count != 0) 1302 return (EEXIST); 1303 1304 return (0); 1305 } 1306 1307 /* ARGSUSED */ 1308 static void 1309 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 1310 { 1311 dsl_dataset_t *ds = arg1; 1312 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1313 1314 /* Mark it as inconsistent on-disk, in case we crash */ 1315 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1316 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; 1317 1318 spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx, 1319 cr, "dataset = %llu", ds->ds_object); 1320 } 1321 1322 static int 1323 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag, 1324 dmu_tx_t *tx) 1325 { 1326 dsl_dataset_t *ds = dsda->ds; 1327 dsl_dataset_t *ds_prev = ds->ds_prev; 1328 1329 if (dsl_dataset_might_destroy_origin(ds_prev)) { 1330 struct dsl_ds_destroyarg ndsda = {0}; 1331 1332 /* 1333 * If we're not prepared to remove the origin, don't remove 1334 * the clone either. 1335 */ 1336 if (dsda->rm_origin == NULL) { 1337 dsda->need_prep = B_TRUE; 1338 return (EBUSY); 1339 } 1340 1341 ndsda.ds = ds_prev; 1342 ndsda.is_origin_rm = B_TRUE; 1343 return (dsl_dataset_destroy_check(&ndsda, tag, tx)); 1344 } 1345 1346 /* 1347 * If we're not going to remove the origin after all, 1348 * undo the open context setup. 1349 */ 1350 if (dsda->rm_origin != NULL) { 1351 dsl_dataset_disown(dsda->rm_origin, tag); 1352 dsda->rm_origin = NULL; 1353 } 1354 1355 return (0); 1356 } 1357 1358 /* ARGSUSED */ 1359 int 1360 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) 1361 { 1362 struct dsl_ds_destroyarg *dsda = arg1; 1363 dsl_dataset_t *ds = dsda->ds; 1364 1365 /* we have an owner hold, so noone else can destroy us */ 1366 ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); 1367 1368 /* 1369 * Only allow deferred destroy on pools that support it. 1370 * NOTE: deferred destroy is only supported on snapshots. 1371 */ 1372 if (dsda->defer) { 1373 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < 1374 SPA_VERSION_USERREFS) 1375 return (ENOTSUP); 1376 ASSERT(dsl_dataset_is_snapshot(ds)); 1377 return (0); 1378 } 1379 1380 /* 1381 * Can't delete a head dataset if there are snapshots of it. 1382 * (Except if the only snapshots are from the branch we cloned 1383 * from.) 1384 */ 1385 if (ds->ds_prev != NULL && 1386 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) 1387 return (EBUSY); 1388 1389 /* 1390 * If we made changes this txg, traverse_dsl_dataset won't find 1391 * them. Try again. 1392 */ 1393 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) 1394 return (EAGAIN); 1395 1396 if (dsl_dataset_is_snapshot(ds)) { 1397 /* 1398 * If this snapshot has an elevated user reference count, 1399 * we can't destroy it yet. 1400 */ 1401 if (ds->ds_userrefs > 0 && !dsda->releasing) 1402 return (EBUSY); 1403 1404 mutex_enter(&ds->ds_lock); 1405 /* 1406 * Can't delete a branch point. However, if we're destroying 1407 * a clone and removing its origin due to it having a user 1408 * hold count of 0 and having been marked for deferred destroy, 1409 * it's OK for the origin to have a single clone. 1410 */ 1411 if (ds->ds_phys->ds_num_children > 1412 (dsda->is_origin_rm ? 2 : 1)) { 1413 mutex_exit(&ds->ds_lock); 1414 return (EEXIST); 1415 } 1416 mutex_exit(&ds->ds_lock); 1417 } else if (dsl_dir_is_clone(ds->ds_dir)) { 1418 return (dsl_dataset_origin_check(dsda, arg2, tx)); 1419 } 1420 1421 /* XXX we should do some i/o error checking... */ 1422 return (0); 1423 } 1424 1425 struct refsarg { 1426 kmutex_t lock; 1427 boolean_t gone; 1428 kcondvar_t cv; 1429 }; 1430 1431 /* ARGSUSED */ 1432 static void 1433 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv) 1434 { 1435 struct refsarg *arg = argv; 1436 1437 mutex_enter(&arg->lock); 1438 arg->gone = TRUE; 1439 cv_signal(&arg->cv); 1440 mutex_exit(&arg->lock); 1441 } 1442 1443 static void 1444 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag) 1445 { 1446 struct refsarg arg; 1447 1448 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL); 1449 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL); 1450 arg.gone = FALSE; 1451 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys, 1452 dsl_dataset_refs_gone); 1453 dmu_buf_rele(ds->ds_dbuf, tag); 1454 mutex_enter(&arg.lock); 1455 while (!arg.gone) 1456 cv_wait(&arg.cv, &arg.lock); 1457 ASSERT(arg.gone); 1458 mutex_exit(&arg.lock); 1459 ds->ds_dbuf = NULL; 1460 ds->ds_phys = NULL; 1461 mutex_destroy(&arg.lock); 1462 cv_destroy(&arg.cv); 1463 } 1464 1465 static void 1466 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx) 1467 { 1468 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1469 uint64_t count; 1470 int err; 1471 1472 ASSERT(ds->ds_phys->ds_num_children >= 2); 1473 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx); 1474 /* 1475 * The err should not be ENOENT, but a bug in a previous version 1476 * of the code could cause upgrade_clones_cb() to not set 1477 * ds_next_snap_obj when it should, leading to a missing entry. 1478 * If we knew that the pool was created after 1479 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 1480 * ENOENT. However, at least we can check that we don't have 1481 * too many entries in the next_clones_obj even after failing to 1482 * remove this one. 1483 */ 1484 if (err != ENOENT) { 1485 VERIFY3U(err, ==, 0); 1486 } 1487 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj, 1488 &count)); 1489 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2); 1490 } 1491 1492 void 1493 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx) 1494 { 1495 struct dsl_ds_destroyarg *dsda = arg1; 1496 dsl_dataset_t *ds = dsda->ds; 1497 int err; 1498 int after_branch_point = FALSE; 1499 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1500 objset_t *mos = dp->dp_meta_objset; 1501 dsl_dataset_t *ds_prev = NULL; 1502 uint64_t obj; 1503 1504 ASSERT(ds->ds_owner); 1505 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1); 1506 ASSERT(ds->ds_prev == NULL || 1507 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object); 1508 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); 1509 1510 if (dsda->defer) { 1511 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 1512 if (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1) { 1513 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1514 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY; 1515 return; 1516 } 1517 } 1518 1519 /* signal any waiters that this dataset is going away */ 1520 mutex_enter(&ds->ds_lock); 1521 ds->ds_owner = dsl_reaper; 1522 cv_broadcast(&ds->ds_exclusive_cv); 1523 mutex_exit(&ds->ds_lock); 1524 1525 /* Remove our reservation */ 1526 if (ds->ds_reserved != 0) { 1527 uint64_t val = 0; 1528 dsl_dataset_set_reservation_sync(ds, &val, cr, tx); 1529 ASSERT3U(ds->ds_reserved, ==, 0); 1530 } 1531 1532 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); 1533 1534 dsl_pool_ds_destroyed(ds, tx); 1535 1536 obj = ds->ds_object; 1537 1538 if (ds->ds_phys->ds_prev_snap_obj != 0) { 1539 if (ds->ds_prev) { 1540 ds_prev = ds->ds_prev; 1541 } else { 1542 VERIFY(0 == dsl_dataset_hold_obj(dp, 1543 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev)); 1544 } 1545 after_branch_point = 1546 (ds_prev->ds_phys->ds_next_snap_obj != obj); 1547 1548 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); 1549 if (after_branch_point && 1550 ds_prev->ds_phys->ds_next_clones_obj != 0) { 1551 remove_from_next_clones(ds_prev, obj, tx); 1552 if (ds->ds_phys->ds_next_snap_obj != 0) { 1553 VERIFY(0 == zap_add_int(mos, 1554 ds_prev->ds_phys->ds_next_clones_obj, 1555 ds->ds_phys->ds_next_snap_obj, tx)); 1556 } 1557 } 1558 if (after_branch_point && 1559 ds->ds_phys->ds_next_snap_obj == 0) { 1560 /* This clone is toast. */ 1561 ASSERT(ds_prev->ds_phys->ds_num_children > 1); 1562 ds_prev->ds_phys->ds_num_children--; 1563 1564 /* 1565 * If the clone's origin has no other clones, no 1566 * user holds, and has been marked for deferred 1567 * deletion, then we should have done the necessary 1568 * destroy setup for it. 1569 */ 1570 if (ds_prev->ds_phys->ds_num_children == 1 && 1571 ds_prev->ds_userrefs == 0 && 1572 DS_IS_DEFER_DESTROY(ds_prev)) { 1573 ASSERT3P(dsda->rm_origin, !=, NULL); 1574 } else { 1575 ASSERT3P(dsda->rm_origin, ==, NULL); 1576 } 1577 } else if (!after_branch_point) { 1578 ds_prev->ds_phys->ds_next_snap_obj = 1579 ds->ds_phys->ds_next_snap_obj; 1580 } 1581 } 1582 1583 if (ds->ds_phys->ds_next_snap_obj != 0) { 1584 blkptr_t bp; 1585 dsl_dataset_t *ds_next; 1586 uint64_t itor = 0; 1587 uint64_t old_unique; 1588 int64_t used = 0, compressed = 0, uncompressed = 0; 1589 1590 VERIFY(0 == dsl_dataset_hold_obj(dp, 1591 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next)); 1592 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); 1593 1594 old_unique = dsl_dataset_unique(ds_next); 1595 1596 dmu_buf_will_dirty(ds_next->ds_dbuf, tx); 1597 ds_next->ds_phys->ds_prev_snap_obj = 1598 ds->ds_phys->ds_prev_snap_obj; 1599 ds_next->ds_phys->ds_prev_snap_txg = 1600 ds->ds_phys->ds_prev_snap_txg; 1601 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1602 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); 1603 1604 /* 1605 * Transfer to our deadlist (which will become next's 1606 * new deadlist) any entries from next's current 1607 * deadlist which were born before prev, and free the 1608 * other entries. 1609 * 1610 * XXX we're doing this long task with the config lock held 1611 */ 1612 while (bplist_iterate(&ds_next->ds_deadlist, &itor, &bp) == 0) { 1613 if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) { 1614 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, 1615 &bp, tx)); 1616 if (ds_prev && !after_branch_point && 1617 bp.blk_birth > 1618 ds_prev->ds_phys->ds_prev_snap_txg) { 1619 ds_prev->ds_phys->ds_unique_bytes += 1620 bp_get_dsize_sync(dp->dp_spa, &bp); 1621 } 1622 } else { 1623 used += bp_get_dsize_sync(dp->dp_spa, &bp); 1624 compressed += BP_GET_PSIZE(&bp); 1625 uncompressed += BP_GET_UCSIZE(&bp); 1626 dsl_free(dp, tx->tx_txg, &bp); 1627 } 1628 } 1629 1630 ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes); 1631 1632 /* change snapused */ 1633 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, 1634 -used, -compressed, -uncompressed, tx); 1635 1636 /* free next's deadlist */ 1637 bplist_close(&ds_next->ds_deadlist); 1638 bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx); 1639 1640 /* set next's deadlist to our deadlist */ 1641 bplist_close(&ds->ds_deadlist); 1642 ds_next->ds_phys->ds_deadlist_obj = 1643 ds->ds_phys->ds_deadlist_obj; 1644 VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos, 1645 ds_next->ds_phys->ds_deadlist_obj)); 1646 ds->ds_phys->ds_deadlist_obj = 0; 1647 1648 if (ds_next->ds_phys->ds_next_snap_obj != 0) { 1649 /* 1650 * Update next's unique to include blocks which 1651 * were previously shared by only this snapshot 1652 * and it. Those blocks will be born after the 1653 * prev snap and before this snap, and will have 1654 * died after the next snap and before the one 1655 * after that (ie. be on the snap after next's 1656 * deadlist). 1657 * 1658 * XXX we're doing this long task with the 1659 * config lock held 1660 */ 1661 dsl_dataset_t *ds_after_next; 1662 uint64_t space; 1663 1664 VERIFY(0 == dsl_dataset_hold_obj(dp, 1665 ds_next->ds_phys->ds_next_snap_obj, 1666 FTAG, &ds_after_next)); 1667 1668 VERIFY(0 == 1669 bplist_space_birthrange(&ds_after_next->ds_deadlist, 1670 ds->ds_phys->ds_prev_snap_txg, 1671 ds->ds_phys->ds_creation_txg, &space)); 1672 ds_next->ds_phys->ds_unique_bytes += space; 1673 1674 dsl_dataset_rele(ds_after_next, FTAG); 1675 ASSERT3P(ds_next->ds_prev, ==, NULL); 1676 } else { 1677 ASSERT3P(ds_next->ds_prev, ==, ds); 1678 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next); 1679 ds_next->ds_prev = NULL; 1680 if (ds_prev) { 1681 VERIFY(0 == dsl_dataset_get_ref(dp, 1682 ds->ds_phys->ds_prev_snap_obj, 1683 ds_next, &ds_next->ds_prev)); 1684 } 1685 1686 dsl_dataset_recalc_head_uniq(ds_next); 1687 1688 /* 1689 * Reduce the amount of our unconsmed refreservation 1690 * being charged to our parent by the amount of 1691 * new unique data we have gained. 1692 */ 1693 if (old_unique < ds_next->ds_reserved) { 1694 int64_t mrsdelta; 1695 uint64_t new_unique = 1696 ds_next->ds_phys->ds_unique_bytes; 1697 1698 ASSERT(old_unique <= new_unique); 1699 mrsdelta = MIN(new_unique - old_unique, 1700 ds_next->ds_reserved - old_unique); 1701 dsl_dir_diduse_space(ds->ds_dir, 1702 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx); 1703 } 1704 } 1705 dsl_dataset_rele(ds_next, FTAG); 1706 } else { 1707 /* 1708 * There's no next snapshot, so this is a head dataset. 1709 * Destroy the deadlist. Unless it's a clone, the 1710 * deadlist should be empty. (If it's a clone, it's 1711 * safe to ignore the deadlist contents.) 1712 */ 1713 struct killarg ka; 1714 1715 ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist)); 1716 bplist_close(&ds->ds_deadlist); 1717 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); 1718 ds->ds_phys->ds_deadlist_obj = 0; 1719 1720 /* 1721 * Free everything that we point to (that's born after 1722 * the previous snapshot, if we are a clone) 1723 * 1724 * NB: this should be very quick, because we already 1725 * freed all the objects in open context. 1726 */ 1727 ka.ds = ds; 1728 ka.tx = tx; 1729 err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg, 1730 TRAVERSE_POST, kill_blkptr, &ka); 1731 ASSERT3U(err, ==, 0); 1732 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || 1733 ds->ds_phys->ds_unique_bytes == 0); 1734 1735 if (ds->ds_prev != NULL) { 1736 dsl_dataset_rele(ds->ds_prev, ds); 1737 ds->ds_prev = ds_prev = NULL; 1738 } 1739 } 1740 1741 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) { 1742 /* Erase the link in the dir */ 1743 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 1744 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0; 1745 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0); 1746 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx); 1747 ASSERT(err == 0); 1748 } else { 1749 /* remove from snapshot namespace */ 1750 dsl_dataset_t *ds_head; 1751 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0); 1752 VERIFY(0 == dsl_dataset_hold_obj(dp, 1753 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head)); 1754 VERIFY(0 == dsl_dataset_get_snapname(ds)); 1755 #ifdef ZFS_DEBUG 1756 { 1757 uint64_t val; 1758 1759 err = dsl_dataset_snap_lookup(ds_head, 1760 ds->ds_snapname, &val); 1761 ASSERT3U(err, ==, 0); 1762 ASSERT3U(val, ==, obj); 1763 } 1764 #endif 1765 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx); 1766 ASSERT(err == 0); 1767 dsl_dataset_rele(ds_head, FTAG); 1768 } 1769 1770 if (ds_prev && ds->ds_prev != ds_prev) 1771 dsl_dataset_rele(ds_prev, FTAG); 1772 1773 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx); 1774 spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx, 1775 cr, "dataset = %llu", ds->ds_object); 1776 1777 if (ds->ds_phys->ds_next_clones_obj != 0) { 1778 uint64_t count; 1779 ASSERT(0 == zap_count(mos, 1780 ds->ds_phys->ds_next_clones_obj, &count) && count == 0); 1781 VERIFY(0 == dmu_object_free(mos, 1782 ds->ds_phys->ds_next_clones_obj, tx)); 1783 } 1784 if (ds->ds_phys->ds_props_obj != 0) 1785 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx)); 1786 if (ds->ds_phys->ds_userrefs_obj != 0) 1787 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx)); 1788 dsl_dir_close(ds->ds_dir, ds); 1789 ds->ds_dir = NULL; 1790 dsl_dataset_drain_refs(ds, tag); 1791 VERIFY(0 == dmu_object_free(mos, obj, tx)); 1792 1793 if (dsda->rm_origin) { 1794 /* 1795 * Remove the origin of the clone we just destroyed. 1796 */ 1797 struct dsl_ds_destroyarg ndsda = {0}; 1798 1799 ndsda.ds = dsda->rm_origin; 1800 dsl_dataset_destroy_sync(&ndsda, tag, cr, tx); 1801 } 1802 } 1803 1804 static int 1805 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1806 { 1807 uint64_t asize; 1808 1809 if (!dmu_tx_is_syncing(tx)) 1810 return (0); 1811 1812 /* 1813 * If there's an fs-only reservation, any blocks that might become 1814 * owned by the snapshot dataset must be accommodated by space 1815 * outside of the reservation. 1816 */ 1817 asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved); 1818 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE)) 1819 return (ENOSPC); 1820 1821 /* 1822 * Propogate any reserved space for this snapshot to other 1823 * snapshot checks in this sync group. 1824 */ 1825 if (asize > 0) 1826 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1827 1828 return (0); 1829 } 1830 1831 /* ARGSUSED */ 1832 int 1833 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx) 1834 { 1835 dsl_dataset_t *ds = arg1; 1836 const char *snapname = arg2; 1837 int err; 1838 uint64_t value; 1839 1840 /* 1841 * We don't allow multiple snapshots of the same txg. If there 1842 * is already one, try again. 1843 */ 1844 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) 1845 return (EAGAIN); 1846 1847 /* 1848 * Check for conflicting name snapshot name. 1849 */ 1850 err = dsl_dataset_snap_lookup(ds, snapname, &value); 1851 if (err == 0) 1852 return (EEXIST); 1853 if (err != ENOENT) 1854 return (err); 1855 1856 /* 1857 * Check that the dataset's name is not too long. Name consists 1858 * of the dataset's length + 1 for the @-sign + snapshot name's length 1859 */ 1860 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN) 1861 return (ENAMETOOLONG); 1862 1863 err = dsl_dataset_snapshot_reserve_space(ds, tx); 1864 if (err) 1865 return (err); 1866 1867 ds->ds_trysnap_txg = tx->tx_txg; 1868 return (0); 1869 } 1870 1871 void 1872 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 1873 { 1874 dsl_dataset_t *ds = arg1; 1875 const char *snapname = arg2; 1876 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1877 dmu_buf_t *dbuf; 1878 dsl_dataset_phys_t *dsphys; 1879 uint64_t dsobj, crtxg; 1880 objset_t *mos = dp->dp_meta_objset; 1881 int err; 1882 1883 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); 1884 1885 /* 1886 * The origin's ds_creation_txg has to be < TXG_INITIAL 1887 */ 1888 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1889 crtxg = 1; 1890 else 1891 crtxg = tx->tx_txg; 1892 1893 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1894 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1895 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1896 dmu_buf_will_dirty(dbuf, tx); 1897 dsphys = dbuf->db_data; 1898 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1899 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1900 dsphys->ds_fsid_guid = unique_create(); 1901 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1902 sizeof (dsphys->ds_guid)); 1903 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; 1904 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; 1905 dsphys->ds_next_snap_obj = ds->ds_object; 1906 dsphys->ds_num_children = 1; 1907 dsphys->ds_creation_time = gethrestime_sec(); 1908 dsphys->ds_creation_txg = crtxg; 1909 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; 1910 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes; 1911 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; 1912 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; 1913 dsphys->ds_flags = ds->ds_phys->ds_flags; 1914 dsphys->ds_bp = ds->ds_phys->ds_bp; 1915 dmu_buf_rele(dbuf, FTAG); 1916 1917 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); 1918 if (ds->ds_prev) { 1919 uint64_t next_clones_obj = 1920 ds->ds_prev->ds_phys->ds_next_clones_obj; 1921 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == 1922 ds->ds_object || 1923 ds->ds_prev->ds_phys->ds_num_children > 1); 1924 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { 1925 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1926 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1927 ds->ds_prev->ds_phys->ds_creation_txg); 1928 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; 1929 } else if (next_clones_obj != 0) { 1930 remove_from_next_clones(ds->ds_prev, 1931 dsphys->ds_next_snap_obj, tx); 1932 VERIFY3U(0, ==, zap_add_int(mos, 1933 next_clones_obj, dsobj, tx)); 1934 } 1935 } 1936 1937 /* 1938 * If we have a reference-reservation on this dataset, we will 1939 * need to increase the amount of refreservation being charged 1940 * since our unique space is going to zero. 1941 */ 1942 if (ds->ds_reserved) { 1943 int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved); 1944 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1945 add, 0, 0, tx); 1946 } 1947 1948 bplist_close(&ds->ds_deadlist); 1949 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1950 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg); 1951 ds->ds_phys->ds_prev_snap_obj = dsobj; 1952 ds->ds_phys->ds_prev_snap_txg = crtxg; 1953 ds->ds_phys->ds_unique_bytes = 0; 1954 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1955 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1956 ds->ds_phys->ds_deadlist_obj = 1957 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); 1958 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos, 1959 ds->ds_phys->ds_deadlist_obj)); 1960 1961 dprintf("snap '%s' -> obj %llu\n", snapname, dsobj); 1962 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, 1963 snapname, 8, 1, &dsobj, tx); 1964 ASSERT(err == 0); 1965 1966 if (ds->ds_prev) 1967 dsl_dataset_drop_ref(ds->ds_prev, ds); 1968 VERIFY(0 == dsl_dataset_get_ref(dp, 1969 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 1970 1971 dsl_pool_ds_snapshotted(ds, tx); 1972 1973 dsl_dir_snap_cmtime_update(ds->ds_dir); 1974 1975 spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr, 1976 "dataset = %llu", dsobj); 1977 } 1978 1979 void 1980 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1981 { 1982 ASSERT(dmu_tx_is_syncing(tx)); 1983 ASSERT(ds->ds_objset != NULL); 1984 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); 1985 1986 /* 1987 * in case we had to change ds_fsid_guid when we opened it, 1988 * sync it out now. 1989 */ 1990 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1991 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; 1992 1993 dsl_dir_dirty(ds->ds_dir, tx); 1994 dmu_objset_sync(ds->ds_objset, zio, tx); 1995 } 1996 1997 void 1998 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1999 { 2000 uint64_t refd, avail, uobjs, aobjs; 2001 2002 dsl_dir_stats(ds->ds_dir, nv); 2003 2004 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 2005 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 2006 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 2007 2008 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 2009 ds->ds_phys->ds_creation_time); 2010 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 2011 ds->ds_phys->ds_creation_txg); 2012 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 2013 ds->ds_quota); 2014 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 2015 ds->ds_reserved); 2016 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 2017 ds->ds_phys->ds_guid); 2018 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 2019 dsl_dataset_unique(ds)); 2020 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 2021 ds->ds_object); 2022 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, ds->ds_userrefs); 2023 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 2024 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 2025 2026 if (ds->ds_phys->ds_next_snap_obj) { 2027 /* 2028 * This is a snapshot; override the dd's space used with 2029 * our unique space and compression ratio. 2030 */ 2031 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 2032 ds->ds_phys->ds_unique_bytes); 2033 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 2034 ds->ds_phys->ds_compressed_bytes == 0 ? 100 : 2035 (ds->ds_phys->ds_uncompressed_bytes * 100 / 2036 ds->ds_phys->ds_compressed_bytes)); 2037 } 2038 } 2039 2040 void 2041 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 2042 { 2043 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; 2044 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; 2045 stat->dds_guid = ds->ds_phys->ds_guid; 2046 if (ds->ds_phys->ds_next_snap_obj) { 2047 stat->dds_is_snapshot = B_TRUE; 2048 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; 2049 } else { 2050 stat->dds_is_snapshot = B_FALSE; 2051 stat->dds_num_clones = 0; 2052 } 2053 2054 /* clone origin is really a dsl_dir thing... */ 2055 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); 2056 if (dsl_dir_is_clone(ds->ds_dir)) { 2057 dsl_dataset_t *ods; 2058 2059 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool, 2060 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods)); 2061 dsl_dataset_name(ods, stat->dds_origin); 2062 dsl_dataset_drop_ref(ods, FTAG); 2063 } else { 2064 stat->dds_origin[0] = '\0'; 2065 } 2066 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); 2067 } 2068 2069 uint64_t 2070 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 2071 { 2072 return (ds->ds_fsid_guid); 2073 } 2074 2075 void 2076 dsl_dataset_space(dsl_dataset_t *ds, 2077 uint64_t *refdbytesp, uint64_t *availbytesp, 2078 uint64_t *usedobjsp, uint64_t *availobjsp) 2079 { 2080 *refdbytesp = ds->ds_phys->ds_used_bytes; 2081 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 2082 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) 2083 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes; 2084 if (ds->ds_quota != 0) { 2085 /* 2086 * Adjust available bytes according to refquota 2087 */ 2088 if (*refdbytesp < ds->ds_quota) 2089 *availbytesp = MIN(*availbytesp, 2090 ds->ds_quota - *refdbytesp); 2091 else 2092 *availbytesp = 0; 2093 } 2094 *usedobjsp = ds->ds_phys->ds_bp.blk_fill; 2095 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 2096 } 2097 2098 boolean_t 2099 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds) 2100 { 2101 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2102 2103 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 2104 dsl_pool_sync_context(dp)); 2105 if (ds->ds_prev == NULL) 2106 return (B_FALSE); 2107 if (ds->ds_phys->ds_bp.blk_birth > 2108 ds->ds_prev->ds_phys->ds_creation_txg) 2109 return (B_TRUE); 2110 return (B_FALSE); 2111 } 2112 2113 /* ARGSUSED */ 2114 static int 2115 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) 2116 { 2117 dsl_dataset_t *ds = arg1; 2118 char *newsnapname = arg2; 2119 dsl_dir_t *dd = ds->ds_dir; 2120 dsl_dataset_t *hds; 2121 uint64_t val; 2122 int err; 2123 2124 err = dsl_dataset_hold_obj(dd->dd_pool, 2125 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds); 2126 if (err) 2127 return (err); 2128 2129 /* new name better not be in use */ 2130 err = dsl_dataset_snap_lookup(hds, newsnapname, &val); 2131 dsl_dataset_rele(hds, FTAG); 2132 2133 if (err == 0) 2134 err = EEXIST; 2135 else if (err == ENOENT) 2136 err = 0; 2137 2138 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2139 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN) 2140 err = ENAMETOOLONG; 2141 2142 return (err); 2143 } 2144 2145 static void 2146 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, 2147 cred_t *cr, dmu_tx_t *tx) 2148 { 2149 dsl_dataset_t *ds = arg1; 2150 const char *newsnapname = arg2; 2151 dsl_dir_t *dd = ds->ds_dir; 2152 objset_t *mos = dd->dd_pool->dp_meta_objset; 2153 dsl_dataset_t *hds; 2154 int err; 2155 2156 ASSERT(ds->ds_phys->ds_next_snap_obj != 0); 2157 2158 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, 2159 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds)); 2160 2161 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2162 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx); 2163 ASSERT3U(err, ==, 0); 2164 mutex_enter(&ds->ds_lock); 2165 (void) strcpy(ds->ds_snapname, newsnapname); 2166 mutex_exit(&ds->ds_lock); 2167 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj, 2168 ds->ds_snapname, 8, 1, &ds->ds_object, tx); 2169 ASSERT3U(err, ==, 0); 2170 2171 spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx, 2172 cr, "dataset = %llu", ds->ds_object); 2173 dsl_dataset_rele(hds, FTAG); 2174 } 2175 2176 struct renamesnaparg { 2177 dsl_sync_task_group_t *dstg; 2178 char failed[MAXPATHLEN]; 2179 char *oldsnap; 2180 char *newsnap; 2181 }; 2182 2183 static int 2184 dsl_snapshot_rename_one(char *name, void *arg) 2185 { 2186 struct renamesnaparg *ra = arg; 2187 dsl_dataset_t *ds = NULL; 2188 char *cp; 2189 int err; 2190 2191 cp = name + strlen(name); 2192 *cp = '@'; 2193 (void) strcpy(cp + 1, ra->oldsnap); 2194 2195 /* 2196 * For recursive snapshot renames the parent won't be changing 2197 * so we just pass name for both the to/from argument. 2198 */ 2199 err = zfs_secpolicy_rename_perms(name, name, CRED()); 2200 if (err == ENOENT) { 2201 return (0); 2202 } else if (err) { 2203 (void) strcpy(ra->failed, name); 2204 return (err); 2205 } 2206 2207 #ifdef _KERNEL 2208 /* 2209 * For all filesystems undergoing rename, we'll need to unmount it. 2210 */ 2211 (void) zfs_unmount_snap(name, NULL); 2212 #endif 2213 err = dsl_dataset_hold(name, ra->dstg, &ds); 2214 *cp = '\0'; 2215 if (err == ENOENT) { 2216 return (0); 2217 } else if (err) { 2218 (void) strcpy(ra->failed, name); 2219 return (err); 2220 } 2221 2222 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check, 2223 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0); 2224 2225 return (0); 2226 } 2227 2228 static int 2229 dsl_recursive_rename(char *oldname, const char *newname) 2230 { 2231 int err; 2232 struct renamesnaparg *ra; 2233 dsl_sync_task_t *dst; 2234 spa_t *spa; 2235 char *cp, *fsname = spa_strdup(oldname); 2236 int len = strlen(oldname); 2237 2238 /* truncate the snapshot name to get the fsname */ 2239 cp = strchr(fsname, '@'); 2240 *cp = '\0'; 2241 2242 err = spa_open(fsname, &spa, FTAG); 2243 if (err) { 2244 kmem_free(fsname, len + 1); 2245 return (err); 2246 } 2247 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP); 2248 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 2249 2250 ra->oldsnap = strchr(oldname, '@') + 1; 2251 ra->newsnap = strchr(newname, '@') + 1; 2252 *ra->failed = '\0'; 2253 2254 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra, 2255 DS_FIND_CHILDREN); 2256 kmem_free(fsname, len + 1); 2257 2258 if (err == 0) { 2259 err = dsl_sync_task_group_wait(ra->dstg); 2260 } 2261 2262 for (dst = list_head(&ra->dstg->dstg_tasks); dst; 2263 dst = list_next(&ra->dstg->dstg_tasks, dst)) { 2264 dsl_dataset_t *ds = dst->dst_arg1; 2265 if (dst->dst_err) { 2266 dsl_dir_name(ds->ds_dir, ra->failed); 2267 (void) strcat(ra->failed, "@"); 2268 (void) strcat(ra->failed, ra->newsnap); 2269 } 2270 dsl_dataset_rele(ds, ra->dstg); 2271 } 2272 2273 if (err) 2274 (void) strcpy(oldname, ra->failed); 2275 2276 dsl_sync_task_group_destroy(ra->dstg); 2277 kmem_free(ra, sizeof (struct renamesnaparg)); 2278 spa_close(spa, FTAG); 2279 return (err); 2280 } 2281 2282 static int 2283 dsl_valid_rename(char *oldname, void *arg) 2284 { 2285 int delta = *(int *)arg; 2286 2287 if (strlen(oldname) + delta >= MAXNAMELEN) 2288 return (ENAMETOOLONG); 2289 2290 return (0); 2291 } 2292 2293 #pragma weak dmu_objset_rename = dsl_dataset_rename 2294 int 2295 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive) 2296 { 2297 dsl_dir_t *dd; 2298 dsl_dataset_t *ds; 2299 const char *tail; 2300 int err; 2301 2302 err = dsl_dir_open(oldname, FTAG, &dd, &tail); 2303 if (err) 2304 return (err); 2305 /* 2306 * If there are more than 2 references there may be holds 2307 * hanging around that haven't been cleared out yet. 2308 */ 2309 if (dmu_buf_refcount(dd->dd_dbuf) > 2) 2310 txg_wait_synced(dd->dd_pool, 0); 2311 if (tail == NULL) { 2312 int delta = strlen(newname) - strlen(oldname); 2313 2314 /* if we're growing, validate child name lengths */ 2315 if (delta > 0) 2316 err = dmu_objset_find(oldname, dsl_valid_rename, 2317 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 2318 2319 if (!err) 2320 err = dsl_dir_rename(dd, newname); 2321 dsl_dir_close(dd, FTAG); 2322 return (err); 2323 } 2324 if (tail[0] != '@') { 2325 /* the name ended in a nonexistent component */ 2326 dsl_dir_close(dd, FTAG); 2327 return (ENOENT); 2328 } 2329 2330 dsl_dir_close(dd, FTAG); 2331 2332 /* new name must be snapshot in same filesystem */ 2333 tail = strchr(newname, '@'); 2334 if (tail == NULL) 2335 return (EINVAL); 2336 tail++; 2337 if (strncmp(oldname, newname, tail - newname) != 0) 2338 return (EXDEV); 2339 2340 if (recursive) { 2341 err = dsl_recursive_rename(oldname, newname); 2342 } else { 2343 err = dsl_dataset_hold(oldname, FTAG, &ds); 2344 if (err) 2345 return (err); 2346 2347 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 2348 dsl_dataset_snapshot_rename_check, 2349 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1); 2350 2351 dsl_dataset_rele(ds, FTAG); 2352 } 2353 2354 return (err); 2355 } 2356 2357 struct promotenode { 2358 list_node_t link; 2359 dsl_dataset_t *ds; 2360 }; 2361 2362 struct promotearg { 2363 list_t shared_snaps, origin_snaps, clone_snaps; 2364 dsl_dataset_t *origin_origin, *origin_head; 2365 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2366 char *err_ds; 2367 }; 2368 2369 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2370 2371 /* ARGSUSED */ 2372 static int 2373 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) 2374 { 2375 dsl_dataset_t *hds = arg1; 2376 struct promotearg *pa = arg2; 2377 struct promotenode *snap = list_head(&pa->shared_snaps); 2378 dsl_dataset_t *origin_ds = snap->ds; 2379 int err; 2380 2381 /* Check that it is a real clone */ 2382 if (!dsl_dir_is_clone(hds->ds_dir)) 2383 return (EINVAL); 2384 2385 /* Since this is so expensive, don't do the preliminary check */ 2386 if (!dmu_tx_is_syncing(tx)) 2387 return (0); 2388 2389 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) 2390 return (EXDEV); 2391 2392 /* compute origin's new unique space */ 2393 snap = list_tail(&pa->clone_snaps); 2394 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2395 err = bplist_space_birthrange(&snap->ds->ds_deadlist, 2396 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, &pa->unique); 2397 if (err) 2398 return (err); 2399 2400 /* 2401 * Walk the snapshots that we are moving 2402 * 2403 * Compute space to transfer. Consider the incremental changes 2404 * to used for each snapshot: 2405 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2406 * So each snapshot gave birth to: 2407 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2408 * So a sequence would look like: 2409 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2410 * Which simplifies to: 2411 * uN + kN + kN-1 + ... + k1 + k0 2412 * Note however, if we stop before we reach the ORIGIN we get: 2413 * uN + kN + kN-1 + ... + kM - uM-1 2414 */ 2415 pa->used = origin_ds->ds_phys->ds_used_bytes; 2416 pa->comp = origin_ds->ds_phys->ds_compressed_bytes; 2417 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; 2418 for (snap = list_head(&pa->shared_snaps); snap; 2419 snap = list_next(&pa->shared_snaps, snap)) { 2420 uint64_t val, dlused, dlcomp, dluncomp; 2421 dsl_dataset_t *ds = snap->ds; 2422 2423 /* Check that the snapshot name does not conflict */ 2424 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2425 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2426 if (err == 0) { 2427 err = EEXIST; 2428 goto out; 2429 } 2430 if (err != ENOENT) 2431 goto out; 2432 2433 /* The very first snapshot does not have a deadlist */ 2434 if (ds->ds_phys->ds_prev_snap_obj == 0) 2435 continue; 2436 2437 if (err = bplist_space(&ds->ds_deadlist, 2438 &dlused, &dlcomp, &dluncomp)) 2439 goto out; 2440 pa->used += dlused; 2441 pa->comp += dlcomp; 2442 pa->uncomp += dluncomp; 2443 } 2444 2445 /* 2446 * If we are a clone of a clone then we never reached ORIGIN, 2447 * so we need to subtract out the clone origin's used space. 2448 */ 2449 if (pa->origin_origin) { 2450 pa->used -= pa->origin_origin->ds_phys->ds_used_bytes; 2451 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes; 2452 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes; 2453 } 2454 2455 /* Check that there is enough space here */ 2456 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2457 pa->used); 2458 if (err) 2459 return (err); 2460 2461 /* 2462 * Compute the amounts of space that will be used by snapshots 2463 * after the promotion (for both origin and clone). For each, 2464 * it is the amount of space that will be on all of their 2465 * deadlists (that was not born before their new origin). 2466 */ 2467 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2468 uint64_t space; 2469 2470 /* 2471 * Note, typically this will not be a clone of a clone, 2472 * so snap->ds->ds_origin_txg will be < TXG_INITIAL, so 2473 * these snaplist_space() -> bplist_space_birthrange() 2474 * calls will be fast because they do not have to 2475 * iterate over all bps. 2476 */ 2477 snap = list_head(&pa->origin_snaps); 2478 err = snaplist_space(&pa->shared_snaps, 2479 snap->ds->ds_origin_txg, &pa->cloneusedsnap); 2480 if (err) 2481 return (err); 2482 2483 err = snaplist_space(&pa->clone_snaps, 2484 snap->ds->ds_origin_txg, &space); 2485 if (err) 2486 return (err); 2487 pa->cloneusedsnap += space; 2488 } 2489 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2490 err = snaplist_space(&pa->origin_snaps, 2491 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap); 2492 if (err) 2493 return (err); 2494 } 2495 2496 return (0); 2497 out: 2498 pa->err_ds = snap->ds->ds_snapname; 2499 return (err); 2500 } 2501 2502 static void 2503 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 2504 { 2505 dsl_dataset_t *hds = arg1; 2506 struct promotearg *pa = arg2; 2507 struct promotenode *snap = list_head(&pa->shared_snaps); 2508 dsl_dataset_t *origin_ds = snap->ds; 2509 dsl_dataset_t *origin_head; 2510 dsl_dir_t *dd = hds->ds_dir; 2511 dsl_pool_t *dp = hds->ds_dir->dd_pool; 2512 dsl_dir_t *odd = NULL; 2513 uint64_t oldnext_obj; 2514 int64_t delta; 2515 2516 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)); 2517 2518 snap = list_head(&pa->origin_snaps); 2519 origin_head = snap->ds; 2520 2521 /* 2522 * We need to explicitly open odd, since origin_ds's dd will be 2523 * changing. 2524 */ 2525 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object, 2526 NULL, FTAG, &odd)); 2527 2528 /* change origin's next snap */ 2529 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2530 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj; 2531 snap = list_tail(&pa->clone_snaps); 2532 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2533 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object; 2534 2535 /* change the origin's next clone */ 2536 if (origin_ds->ds_phys->ds_next_clones_obj) { 2537 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx); 2538 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, 2539 origin_ds->ds_phys->ds_next_clones_obj, 2540 oldnext_obj, tx)); 2541 } 2542 2543 /* change origin */ 2544 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2545 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object); 2546 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj; 2547 hds->ds_origin_txg = origin_head->ds_origin_txg; 2548 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2549 odd->dd_phys->dd_origin_obj = origin_ds->ds_object; 2550 origin_head->ds_origin_txg = origin_ds->ds_phys->ds_creation_txg; 2551 2552 /* move snapshots to this dir */ 2553 for (snap = list_head(&pa->shared_snaps); snap; 2554 snap = list_next(&pa->shared_snaps, snap)) { 2555 dsl_dataset_t *ds = snap->ds; 2556 2557 /* unregister props as dsl_dir is changing */ 2558 if (ds->ds_objset) { 2559 dmu_objset_evict(ds->ds_objset); 2560 ds->ds_objset = NULL; 2561 } 2562 /* move snap name entry */ 2563 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2564 VERIFY(0 == dsl_dataset_snap_remove(origin_head, 2565 ds->ds_snapname, tx)); 2566 VERIFY(0 == zap_add(dp->dp_meta_objset, 2567 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, 2568 8, 1, &ds->ds_object, tx)); 2569 /* change containing dsl_dir */ 2570 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2571 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object); 2572 ds->ds_phys->ds_dir_obj = dd->dd_object; 2573 ASSERT3P(ds->ds_dir, ==, odd); 2574 dsl_dir_close(ds->ds_dir, ds); 2575 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object, 2576 NULL, ds, &ds->ds_dir)); 2577 2578 ASSERT3U(dsl_prop_numcb(ds), ==, 0); 2579 } 2580 2581 /* 2582 * Change space accounting. 2583 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2584 * both be valid, or both be 0 (resulting in delta == 0). This 2585 * is true for each of {clone,origin} independently. 2586 */ 2587 2588 delta = pa->cloneusedsnap - 2589 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2590 ASSERT3S(delta, >=, 0); 2591 ASSERT3U(pa->used, >=, delta); 2592 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2593 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2594 pa->used - delta, pa->comp, pa->uncomp, tx); 2595 2596 delta = pa->originusedsnap - 2597 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2598 ASSERT3S(delta, <=, 0); 2599 ASSERT3U(pa->used, >=, -delta); 2600 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2601 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2602 -pa->used - delta, -pa->comp, -pa->uncomp, tx); 2603 2604 origin_ds->ds_phys->ds_unique_bytes = pa->unique; 2605 2606 /* log history record */ 2607 spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx, 2608 cr, "dataset = %llu", hds->ds_object); 2609 2610 dsl_dir_close(odd, FTAG); 2611 } 2612 2613 static char *snaplist_tag = "snaplist"; 2614 /* 2615 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2616 * (exclusive) and last_obj (inclusive). The list will be in reverse 2617 * order (last_obj will be the list_head()). If first_obj == 0, do all 2618 * snapshots back to this dataset's origin. 2619 */ 2620 static int 2621 snaplist_make(dsl_pool_t *dp, boolean_t own, 2622 uint64_t first_obj, uint64_t last_obj, list_t *l) 2623 { 2624 uint64_t obj = last_obj; 2625 2626 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock)); 2627 2628 list_create(l, sizeof (struct promotenode), 2629 offsetof(struct promotenode, link)); 2630 2631 while (obj != first_obj) { 2632 dsl_dataset_t *ds; 2633 struct promotenode *snap; 2634 int err; 2635 2636 if (own) { 2637 err = dsl_dataset_own_obj(dp, obj, 2638 0, snaplist_tag, &ds); 2639 if (err == 0) 2640 dsl_dataset_make_exclusive(ds, snaplist_tag); 2641 } else { 2642 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds); 2643 } 2644 if (err == ENOENT) { 2645 /* lost race with snapshot destroy */ 2646 struct promotenode *last = list_tail(l); 2647 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj); 2648 obj = last->ds->ds_phys->ds_prev_snap_obj; 2649 continue; 2650 } else if (err) { 2651 return (err); 2652 } 2653 2654 if (first_obj == 0) 2655 first_obj = ds->ds_dir->dd_phys->dd_origin_obj; 2656 2657 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP); 2658 snap->ds = ds; 2659 list_insert_tail(l, snap); 2660 obj = ds->ds_phys->ds_prev_snap_obj; 2661 } 2662 2663 return (0); 2664 } 2665 2666 static int 2667 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2668 { 2669 struct promotenode *snap; 2670 2671 *spacep = 0; 2672 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2673 uint64_t used; 2674 int err = bplist_space_birthrange(&snap->ds->ds_deadlist, 2675 mintxg, UINT64_MAX, &used); 2676 if (err) 2677 return (err); 2678 *spacep += used; 2679 } 2680 return (0); 2681 } 2682 2683 static void 2684 snaplist_destroy(list_t *l, boolean_t own) 2685 { 2686 struct promotenode *snap; 2687 2688 if (!l || !list_link_active(&l->list_head)) 2689 return; 2690 2691 while ((snap = list_tail(l)) != NULL) { 2692 list_remove(l, snap); 2693 if (own) 2694 dsl_dataset_disown(snap->ds, snaplist_tag); 2695 else 2696 dsl_dataset_rele(snap->ds, snaplist_tag); 2697 kmem_free(snap, sizeof (struct promotenode)); 2698 } 2699 list_destroy(l); 2700 } 2701 2702 /* 2703 * Promote a clone. Nomenclature note: 2704 * "clone" or "cds": the original clone which is being promoted 2705 * "origin" or "ods": the snapshot which is originally clone's origin 2706 * "origin head" or "ohds": the dataset which is the head 2707 * (filesystem/volume) for the origin 2708 * "origin origin": the origin of the origin's filesystem (typically 2709 * NULL, indicating that the clone is not a clone of a clone). 2710 */ 2711 int 2712 dsl_dataset_promote(const char *name, char *conflsnap) 2713 { 2714 dsl_dataset_t *ds; 2715 dsl_dir_t *dd; 2716 dsl_pool_t *dp; 2717 dmu_object_info_t doi; 2718 struct promotearg pa = { 0 }; 2719 struct promotenode *snap; 2720 int err; 2721 2722 err = dsl_dataset_hold(name, FTAG, &ds); 2723 if (err) 2724 return (err); 2725 dd = ds->ds_dir; 2726 dp = dd->dd_pool; 2727 2728 err = dmu_object_info(dp->dp_meta_objset, 2729 ds->ds_phys->ds_snapnames_zapobj, &doi); 2730 if (err) { 2731 dsl_dataset_rele(ds, FTAG); 2732 return (err); 2733 } 2734 2735 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) { 2736 dsl_dataset_rele(ds, FTAG); 2737 return (EINVAL); 2738 } 2739 2740 /* 2741 * We are going to inherit all the snapshots taken before our 2742 * origin (i.e., our new origin will be our parent's origin). 2743 * Take ownership of them so that we can rename them into our 2744 * namespace. 2745 */ 2746 rw_enter(&dp->dp_config_rwlock, RW_READER); 2747 2748 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj, 2749 &pa.shared_snaps); 2750 if (err != 0) 2751 goto out; 2752 2753 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps); 2754 if (err != 0) 2755 goto out; 2756 2757 snap = list_head(&pa.shared_snaps); 2758 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); 2759 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj, 2760 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps); 2761 if (err != 0) 2762 goto out; 2763 2764 if (dsl_dir_is_clone(snap->ds->ds_dir)) { 2765 err = dsl_dataset_own_obj(dp, 2766 snap->ds->ds_dir->dd_phys->dd_origin_obj, 2767 0, FTAG, &pa.origin_origin); 2768 if (err != 0) 2769 goto out; 2770 } 2771 2772 out: 2773 rw_exit(&dp->dp_config_rwlock); 2774 2775 /* 2776 * Add in 128x the snapnames zapobj size, since we will be moving 2777 * a bunch of snapnames to the promoted ds, and dirtying their 2778 * bonus buffers. 2779 */ 2780 if (err == 0) { 2781 err = dsl_sync_task_do(dp, dsl_dataset_promote_check, 2782 dsl_dataset_promote_sync, ds, &pa, 2783 2 + 2 * doi.doi_physical_blocks_512); 2784 if (err && pa.err_ds && conflsnap) 2785 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN); 2786 } 2787 2788 snaplist_destroy(&pa.shared_snaps, B_TRUE); 2789 snaplist_destroy(&pa.clone_snaps, B_FALSE); 2790 snaplist_destroy(&pa.origin_snaps, B_FALSE); 2791 if (pa.origin_origin) 2792 dsl_dataset_disown(pa.origin_origin, FTAG); 2793 dsl_dataset_rele(ds, FTAG); 2794 return (err); 2795 } 2796 2797 struct cloneswaparg { 2798 dsl_dataset_t *cds; /* clone dataset */ 2799 dsl_dataset_t *ohds; /* origin's head dataset */ 2800 boolean_t force; 2801 int64_t unused_refres_delta; /* change in unconsumed refreservation */ 2802 }; 2803 2804 /* ARGSUSED */ 2805 static int 2806 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx) 2807 { 2808 struct cloneswaparg *csa = arg1; 2809 2810 /* they should both be heads */ 2811 if (dsl_dataset_is_snapshot(csa->cds) || 2812 dsl_dataset_is_snapshot(csa->ohds)) 2813 return (EINVAL); 2814 2815 /* the branch point should be just before them */ 2816 if (csa->cds->ds_prev != csa->ohds->ds_prev) 2817 return (EINVAL); 2818 2819 /* cds should be the clone (unless they are unrelated) */ 2820 if (csa->cds->ds_prev != NULL && 2821 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap && 2822 csa->ohds->ds_object != 2823 csa->cds->ds_prev->ds_phys->ds_next_snap_obj) 2824 return (EINVAL); 2825 2826 /* the clone should be a child of the origin */ 2827 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir) 2828 return (EINVAL); 2829 2830 /* ohds shouldn't be modified unless 'force' */ 2831 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds)) 2832 return (ETXTBSY); 2833 2834 /* adjust amount of any unconsumed refreservation */ 2835 csa->unused_refres_delta = 2836 (int64_t)MIN(csa->ohds->ds_reserved, 2837 csa->ohds->ds_phys->ds_unique_bytes) - 2838 (int64_t)MIN(csa->ohds->ds_reserved, 2839 csa->cds->ds_phys->ds_unique_bytes); 2840 2841 if (csa->unused_refres_delta > 0 && 2842 csa->unused_refres_delta > 2843 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE)) 2844 return (ENOSPC); 2845 2846 if (csa->ohds->ds_quota != 0 && 2847 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota) 2848 return (EDQUOT); 2849 2850 return (0); 2851 } 2852 2853 /* ARGSUSED */ 2854 static void 2855 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 2856 { 2857 struct cloneswaparg *csa = arg1; 2858 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool; 2859 2860 ASSERT(csa->cds->ds_reserved == 0); 2861 ASSERT(csa->ohds->ds_quota == 0 || 2862 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota); 2863 2864 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx); 2865 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx); 2866 2867 if (csa->cds->ds_objset != NULL) { 2868 dmu_objset_evict(csa->cds->ds_objset); 2869 csa->cds->ds_objset = NULL; 2870 } 2871 2872 if (csa->ohds->ds_objset != NULL) { 2873 dmu_objset_evict(csa->ohds->ds_objset); 2874 csa->ohds->ds_objset = NULL; 2875 } 2876 2877 /* 2878 * Reset origin's unique bytes, if it exists. 2879 */ 2880 if (csa->cds->ds_prev) { 2881 dsl_dataset_t *origin = csa->cds->ds_prev; 2882 dmu_buf_will_dirty(origin->ds_dbuf, tx); 2883 VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist, 2884 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2885 &origin->ds_phys->ds_unique_bytes)); 2886 } 2887 2888 /* swap blkptrs */ 2889 { 2890 blkptr_t tmp; 2891 tmp = csa->ohds->ds_phys->ds_bp; 2892 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp; 2893 csa->cds->ds_phys->ds_bp = tmp; 2894 } 2895 2896 /* set dd_*_bytes */ 2897 { 2898 int64_t dused, dcomp, duncomp; 2899 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2900 uint64_t odl_used, odl_comp, odl_uncomp; 2901 2902 ASSERT3U(csa->cds->ds_dir->dd_phys-> 2903 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2904 2905 VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used, 2906 &cdl_comp, &cdl_uncomp)); 2907 VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used, 2908 &odl_comp, &odl_uncomp)); 2909 2910 dused = csa->cds->ds_phys->ds_used_bytes + cdl_used - 2911 (csa->ohds->ds_phys->ds_used_bytes + odl_used); 2912 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp - 2913 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp); 2914 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes + 2915 cdl_uncomp - 2916 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp); 2917 2918 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD, 2919 dused, dcomp, duncomp, tx); 2920 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD, 2921 -dused, -dcomp, -duncomp, tx); 2922 2923 /* 2924 * The difference in the space used by snapshots is the 2925 * difference in snapshot space due to the head's 2926 * deadlist (since that's the only thing that's 2927 * changing that affects the snapused). 2928 */ 2929 VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist, 2930 csa->ohds->ds_origin_txg, UINT64_MAX, &cdl_used)); 2931 VERIFY(0 == bplist_space_birthrange(&csa->ohds->ds_deadlist, 2932 csa->ohds->ds_origin_txg, UINT64_MAX, &odl_used)); 2933 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used, 2934 DD_USED_HEAD, DD_USED_SNAP, tx); 2935 } 2936 2937 #define SWITCH64(x, y) \ 2938 { \ 2939 uint64_t __tmp = (x); \ 2940 (x) = (y); \ 2941 (y) = __tmp; \ 2942 } 2943 2944 /* swap ds_*_bytes */ 2945 SWITCH64(csa->ohds->ds_phys->ds_used_bytes, 2946 csa->cds->ds_phys->ds_used_bytes); 2947 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes, 2948 csa->cds->ds_phys->ds_compressed_bytes); 2949 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes, 2950 csa->cds->ds_phys->ds_uncompressed_bytes); 2951 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes, 2952 csa->cds->ds_phys->ds_unique_bytes); 2953 2954 /* apply any parent delta for change in unconsumed refreservation */ 2955 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV, 2956 csa->unused_refres_delta, 0, 0, tx); 2957 2958 /* swap deadlists */ 2959 bplist_close(&csa->cds->ds_deadlist); 2960 bplist_close(&csa->ohds->ds_deadlist); 2961 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj, 2962 csa->cds->ds_phys->ds_deadlist_obj); 2963 VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset, 2964 csa->cds->ds_phys->ds_deadlist_obj)); 2965 VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset, 2966 csa->ohds->ds_phys->ds_deadlist_obj)); 2967 2968 dsl_pool_ds_clone_swapped(csa->ohds, csa->cds, tx); 2969 } 2970 2971 /* 2972 * Swap 'clone' with its origin head datasets. Used at the end of "zfs 2973 * recv" into an existing fs to swizzle the file system to the new 2974 * version, and by "zfs rollback". Can also be used to swap two 2975 * independent head datasets if neither has any snapshots. 2976 */ 2977 int 2978 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head, 2979 boolean_t force) 2980 { 2981 struct cloneswaparg csa; 2982 int error; 2983 2984 ASSERT(clone->ds_owner); 2985 ASSERT(origin_head->ds_owner); 2986 retry: 2987 /* Need exclusive access for the swap */ 2988 rw_enter(&clone->ds_rwlock, RW_WRITER); 2989 if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) { 2990 rw_exit(&clone->ds_rwlock); 2991 rw_enter(&origin_head->ds_rwlock, RW_WRITER); 2992 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) { 2993 rw_exit(&origin_head->ds_rwlock); 2994 goto retry; 2995 } 2996 } 2997 csa.cds = clone; 2998 csa.ohds = origin_head; 2999 csa.force = force; 3000 error = dsl_sync_task_do(clone->ds_dir->dd_pool, 3001 dsl_dataset_clone_swap_check, 3002 dsl_dataset_clone_swap_sync, &csa, NULL, 9); 3003 return (error); 3004 } 3005 3006 /* 3007 * Given a pool name and a dataset object number in that pool, 3008 * return the name of that dataset. 3009 */ 3010 int 3011 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 3012 { 3013 spa_t *spa; 3014 dsl_pool_t *dp; 3015 dsl_dataset_t *ds; 3016 int error; 3017 3018 if ((error = spa_open(pname, &spa, FTAG)) != 0) 3019 return (error); 3020 dp = spa_get_dsl(spa); 3021 rw_enter(&dp->dp_config_rwlock, RW_READER); 3022 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) { 3023 dsl_dataset_name(ds, buf); 3024 dsl_dataset_rele(ds, FTAG); 3025 } 3026 rw_exit(&dp->dp_config_rwlock); 3027 spa_close(spa, FTAG); 3028 3029 return (error); 3030 } 3031 3032 int 3033 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 3034 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 3035 { 3036 int error = 0; 3037 3038 ASSERT3S(asize, >, 0); 3039 3040 /* 3041 * *ref_rsrv is the portion of asize that will come from any 3042 * unconsumed refreservation space. 3043 */ 3044 *ref_rsrv = 0; 3045 3046 mutex_enter(&ds->ds_lock); 3047 /* 3048 * Make a space adjustment for reserved bytes. 3049 */ 3050 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) { 3051 ASSERT3U(*used, >=, 3052 ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 3053 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 3054 *ref_rsrv = 3055 asize - MIN(asize, parent_delta(ds, asize + inflight)); 3056 } 3057 3058 if (!check_quota || ds->ds_quota == 0) { 3059 mutex_exit(&ds->ds_lock); 3060 return (0); 3061 } 3062 /* 3063 * If they are requesting more space, and our current estimate 3064 * is over quota, they get to try again unless the actual 3065 * on-disk is over quota and there are no pending changes (which 3066 * may free up space for us). 3067 */ 3068 if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) { 3069 if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota) 3070 error = ERESTART; 3071 else 3072 error = EDQUOT; 3073 } 3074 mutex_exit(&ds->ds_lock); 3075 3076 return (error); 3077 } 3078 3079 /* ARGSUSED */ 3080 static int 3081 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) 3082 { 3083 dsl_dataset_t *ds = arg1; 3084 uint64_t *quotap = arg2; 3085 uint64_t new_quota = *quotap; 3086 3087 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA) 3088 return (ENOTSUP); 3089 3090 if (new_quota == 0) 3091 return (0); 3092 3093 if (new_quota < ds->ds_phys->ds_used_bytes || 3094 new_quota < ds->ds_reserved) 3095 return (ENOSPC); 3096 3097 return (0); 3098 } 3099 3100 /* ARGSUSED */ 3101 void 3102 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 3103 { 3104 dsl_dataset_t *ds = arg1; 3105 uint64_t *quotap = arg2; 3106 uint64_t new_quota = *quotap; 3107 3108 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3109 3110 ds->ds_quota = new_quota; 3111 3112 dsl_dir_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx); 3113 3114 spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa, 3115 tx, cr, "%lld dataset = %llu ", 3116 (longlong_t)new_quota, ds->ds_object); 3117 } 3118 3119 int 3120 dsl_dataset_set_quota(const char *dsname, uint64_t quota) 3121 { 3122 dsl_dataset_t *ds; 3123 int err; 3124 3125 err = dsl_dataset_hold(dsname, FTAG, &ds); 3126 if (err) 3127 return (err); 3128 3129 if (quota != ds->ds_quota) { 3130 /* 3131 * If someone removes a file, then tries to set the quota, we 3132 * want to make sure the file freeing takes effect. 3133 */ 3134 txg_wait_open(ds->ds_dir->dd_pool, 0); 3135 3136 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 3137 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync, 3138 ds, "a, 0); 3139 } 3140 dsl_dataset_rele(ds, FTAG); 3141 return (err); 3142 } 3143 3144 static int 3145 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) 3146 { 3147 dsl_dataset_t *ds = arg1; 3148 uint64_t *reservationp = arg2; 3149 uint64_t new_reservation = *reservationp; 3150 uint64_t unique; 3151 3152 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < 3153 SPA_VERSION_REFRESERVATION) 3154 return (ENOTSUP); 3155 3156 if (dsl_dataset_is_snapshot(ds)) 3157 return (EINVAL); 3158 3159 /* 3160 * If we are doing the preliminary check in open context, the 3161 * space estimates may be inaccurate. 3162 */ 3163 if (!dmu_tx_is_syncing(tx)) 3164 return (0); 3165 3166 mutex_enter(&ds->ds_lock); 3167 unique = dsl_dataset_unique(ds); 3168 mutex_exit(&ds->ds_lock); 3169 3170 if (MAX(unique, new_reservation) > MAX(unique, ds->ds_reserved)) { 3171 uint64_t delta = MAX(unique, new_reservation) - 3172 MAX(unique, ds->ds_reserved); 3173 3174 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 3175 return (ENOSPC); 3176 if (ds->ds_quota > 0 && 3177 new_reservation > ds->ds_quota) 3178 return (ENOSPC); 3179 } 3180 3181 return (0); 3182 } 3183 3184 /* ARGSUSED */ 3185 static void 3186 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr, 3187 dmu_tx_t *tx) 3188 { 3189 dsl_dataset_t *ds = arg1; 3190 uint64_t *reservationp = arg2; 3191 uint64_t new_reservation = *reservationp; 3192 uint64_t unique; 3193 int64_t delta; 3194 3195 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3196 3197 mutex_enter(&ds->ds_dir->dd_lock); 3198 mutex_enter(&ds->ds_lock); 3199 unique = dsl_dataset_unique(ds); 3200 delta = MAX(0, (int64_t)(new_reservation - unique)) - 3201 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3202 ds->ds_reserved = new_reservation; 3203 mutex_exit(&ds->ds_lock); 3204 3205 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3206 mutex_exit(&ds->ds_dir->dd_lock); 3207 dsl_dir_prop_set_uint64_sync(ds->ds_dir, "refreservation", 3208 new_reservation, cr, tx); 3209 3210 spa_history_internal_log(LOG_DS_REFRESERV, 3211 ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu", 3212 (longlong_t)new_reservation, ds->ds_object); 3213 } 3214 3215 int 3216 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation) 3217 { 3218 dsl_dataset_t *ds; 3219 int err; 3220 3221 err = dsl_dataset_hold(dsname, FTAG, &ds); 3222 if (err) 3223 return (err); 3224 3225 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 3226 dsl_dataset_set_reservation_check, 3227 dsl_dataset_set_reservation_sync, ds, &reservation, 0); 3228 dsl_dataset_rele(ds, FTAG); 3229 return (err); 3230 } 3231 3232 struct dsl_ds_holdarg { 3233 dsl_sync_task_group_t *dstg; 3234 char *htag; 3235 char *snapname; 3236 boolean_t recursive; 3237 boolean_t gotone; 3238 boolean_t temphold; 3239 char failed[MAXPATHLEN]; 3240 }; 3241 3242 /* 3243 * The max length of a temporary tag prefix is the number of hex digits 3244 * required to express UINT64_MAX plus one for the hyphen. 3245 */ 3246 #define MAX_TAG_PREFIX_LEN 17 3247 3248 static int 3249 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx) 3250 { 3251 dsl_dataset_t *ds = arg1; 3252 struct dsl_ds_holdarg *ha = arg2; 3253 char *htag = ha->htag; 3254 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3255 int error = 0; 3256 3257 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) 3258 return (ENOTSUP); 3259 3260 if (!dsl_dataset_is_snapshot(ds)) 3261 return (EINVAL); 3262 3263 /* tags must be unique */ 3264 mutex_enter(&ds->ds_lock); 3265 if (ds->ds_phys->ds_userrefs_obj) { 3266 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag, 3267 8, 1, tx); 3268 if (error == 0) 3269 error = EEXIST; 3270 else if (error == ENOENT) 3271 error = 0; 3272 } 3273 mutex_exit(&ds->ds_lock); 3274 3275 if (error == 0 && ha->temphold && 3276 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN) 3277 error = E2BIG; 3278 3279 return (error); 3280 } 3281 3282 static void 3283 dsl_dataset_user_hold_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 3284 { 3285 dsl_dataset_t *ds = arg1; 3286 struct dsl_ds_holdarg *ha = arg2; 3287 char *htag = ha->htag; 3288 dsl_pool_t *dp = ds->ds_dir->dd_pool; 3289 objset_t *mos = dp->dp_meta_objset; 3290 uint64_t now = gethrestime_sec(); 3291 uint64_t zapobj; 3292 3293 mutex_enter(&ds->ds_lock); 3294 if (ds->ds_phys->ds_userrefs_obj == 0) { 3295 /* 3296 * This is the first user hold for this dataset. Create 3297 * the userrefs zap object. 3298 */ 3299 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3300 zapobj = ds->ds_phys->ds_userrefs_obj = 3301 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx); 3302 } else { 3303 zapobj = ds->ds_phys->ds_userrefs_obj; 3304 } 3305 ds->ds_userrefs++; 3306 mutex_exit(&ds->ds_lock); 3307 3308 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx)); 3309 3310 if (ha->temphold) { 3311 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object, 3312 htag, &now, tx)); 3313 } 3314 3315 spa_history_internal_log(LOG_DS_USER_HOLD, 3316 dp->dp_spa, tx, cr, "<%s> temp = %d dataset = %llu", htag, 3317 (int)ha->temphold, ds->ds_object); 3318 } 3319 3320 static int 3321 dsl_dataset_user_hold_one(char *dsname, void *arg) 3322 { 3323 struct dsl_ds_holdarg *ha = arg; 3324 dsl_dataset_t *ds; 3325 int error; 3326 char *name; 3327 3328 /* alloc a buffer to hold dsname@snapname plus terminating NULL */ 3329 name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3330 error = dsl_dataset_hold(name, ha->dstg, &ds); 3331 strfree(name); 3332 if (error == 0) { 3333 ha->gotone = B_TRUE; 3334 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check, 3335 dsl_dataset_user_hold_sync, ds, ha, 0); 3336 } else if (error == ENOENT && ha->recursive) { 3337 error = 0; 3338 } else { 3339 (void) strcpy(ha->failed, dsname); 3340 } 3341 return (error); 3342 } 3343 3344 int 3345 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag, 3346 boolean_t recursive, boolean_t temphold) 3347 { 3348 struct dsl_ds_holdarg *ha; 3349 dsl_sync_task_t *dst; 3350 spa_t *spa; 3351 int error; 3352 3353 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); 3354 3355 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3356 3357 error = spa_open(dsname, &spa, FTAG); 3358 if (error) { 3359 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3360 return (error); 3361 } 3362 3363 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 3364 ha->htag = htag; 3365 ha->snapname = snapname; 3366 ha->recursive = recursive; 3367 ha->temphold = temphold; 3368 if (recursive) { 3369 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one, 3370 ha, DS_FIND_CHILDREN); 3371 } else { 3372 error = dsl_dataset_user_hold_one(dsname, ha); 3373 } 3374 if (error == 0) 3375 error = dsl_sync_task_group_wait(ha->dstg); 3376 3377 for (dst = list_head(&ha->dstg->dstg_tasks); dst; 3378 dst = list_next(&ha->dstg->dstg_tasks, dst)) { 3379 dsl_dataset_t *ds = dst->dst_arg1; 3380 3381 if (dst->dst_err) { 3382 dsl_dataset_name(ds, ha->failed); 3383 *strchr(ha->failed, '@') = '\0'; 3384 } 3385 dsl_dataset_rele(ds, ha->dstg); 3386 } 3387 3388 if (error == 0 && recursive && !ha->gotone) 3389 error = ENOENT; 3390 3391 if (error) 3392 (void) strcpy(dsname, ha->failed); 3393 3394 dsl_sync_task_group_destroy(ha->dstg); 3395 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3396 spa_close(spa, FTAG); 3397 return (error); 3398 } 3399 3400 struct dsl_ds_releasearg { 3401 dsl_dataset_t *ds; 3402 const char *htag; 3403 boolean_t own; /* do we own or just hold ds? */ 3404 }; 3405 3406 static int 3407 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag, 3408 boolean_t *might_destroy) 3409 { 3410 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3411 uint64_t zapobj; 3412 uint64_t tmp; 3413 int error; 3414 3415 *might_destroy = B_FALSE; 3416 3417 mutex_enter(&ds->ds_lock); 3418 zapobj = ds->ds_phys->ds_userrefs_obj; 3419 if (zapobj == 0) { 3420 /* The tag can't possibly exist */ 3421 mutex_exit(&ds->ds_lock); 3422 return (ESRCH); 3423 } 3424 3425 /* Make sure the tag exists */ 3426 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp); 3427 if (error) { 3428 mutex_exit(&ds->ds_lock); 3429 if (error == ENOENT) 3430 error = ESRCH; 3431 return (error); 3432 } 3433 3434 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 && 3435 DS_IS_DEFER_DESTROY(ds)) 3436 *might_destroy = B_TRUE; 3437 3438 mutex_exit(&ds->ds_lock); 3439 return (0); 3440 } 3441 3442 static int 3443 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx) 3444 { 3445 struct dsl_ds_releasearg *ra = arg1; 3446 dsl_dataset_t *ds = ra->ds; 3447 boolean_t might_destroy; 3448 int error; 3449 3450 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) 3451 return (ENOTSUP); 3452 3453 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy); 3454 if (error) 3455 return (error); 3456 3457 if (might_destroy) { 3458 struct dsl_ds_destroyarg dsda = {0}; 3459 3460 if (dmu_tx_is_syncing(tx)) { 3461 /* 3462 * If we're not prepared to remove the snapshot, 3463 * we can't allow the release to happen right now. 3464 */ 3465 if (!ra->own) 3466 return (EBUSY); 3467 if (ds->ds_objset) { 3468 dmu_objset_evict(ds->ds_objset); 3469 ds->ds_objset = NULL; 3470 } 3471 } 3472 dsda.ds = ds; 3473 dsda.releasing = B_TRUE; 3474 return (dsl_dataset_destroy_check(&dsda, tag, tx)); 3475 } 3476 3477 return (0); 3478 } 3479 3480 static void 3481 dsl_dataset_user_release_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx) 3482 { 3483 struct dsl_ds_releasearg *ra = arg1; 3484 dsl_dataset_t *ds = ra->ds; 3485 dsl_pool_t *dp = ds->ds_dir->dd_pool; 3486 objset_t *mos = dp->dp_meta_objset; 3487 uint64_t zapobj; 3488 uint64_t dsobj = ds->ds_object; 3489 uint64_t refs; 3490 int error; 3491 3492 mutex_enter(&ds->ds_lock); 3493 ds->ds_userrefs--; 3494 refs = ds->ds_userrefs; 3495 mutex_exit(&ds->ds_lock); 3496 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx); 3497 VERIFY(error == 0 || error == ENOENT); 3498 zapobj = ds->ds_phys->ds_userrefs_obj; 3499 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx)); 3500 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 && 3501 DS_IS_DEFER_DESTROY(ds)) { 3502 struct dsl_ds_destroyarg dsda = {0}; 3503 3504 ASSERT(ra->own); 3505 dsda.ds = ds; 3506 dsda.releasing = B_TRUE; 3507 /* We already did the destroy_check */ 3508 dsl_dataset_destroy_sync(&dsda, tag, cr, tx); 3509 } 3510 3511 spa_history_internal_log(LOG_DS_USER_RELEASE, 3512 dp->dp_spa, tx, cr, "<%s> %lld dataset = %llu", 3513 ra->htag, (longlong_t)refs, dsobj); 3514 } 3515 3516 static int 3517 dsl_dataset_user_release_one(char *dsname, void *arg) 3518 { 3519 struct dsl_ds_holdarg *ha = arg; 3520 struct dsl_ds_releasearg *ra; 3521 dsl_dataset_t *ds; 3522 int error; 3523 void *dtag = ha->dstg; 3524 char *name; 3525 boolean_t own = B_FALSE; 3526 boolean_t might_destroy; 3527 3528 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */ 3529 name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3530 error = dsl_dataset_hold(name, dtag, &ds); 3531 strfree(name); 3532 if (error == ENOENT && ha->recursive) 3533 return (0); 3534 (void) strcpy(ha->failed, dsname); 3535 if (error) 3536 return (error); 3537 3538 ha->gotone = B_TRUE; 3539 3540 ASSERT(dsl_dataset_is_snapshot(ds)); 3541 3542 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy); 3543 if (error) { 3544 dsl_dataset_rele(ds, dtag); 3545 return (error); 3546 } 3547 3548 if (might_destroy) { 3549 #ifdef _KERNEL 3550 error = zfs_unmount_snap(name, NULL); 3551 if (error) { 3552 dsl_dataset_rele(ds, dtag); 3553 return (error); 3554 } 3555 #endif 3556 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) { 3557 dsl_dataset_rele(ds, dtag); 3558 return (EBUSY); 3559 } else { 3560 own = B_TRUE; 3561 dsl_dataset_make_exclusive(ds, dtag); 3562 } 3563 } 3564 3565 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP); 3566 ra->ds = ds; 3567 ra->htag = ha->htag; 3568 ra->own = own; 3569 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check, 3570 dsl_dataset_user_release_sync, ra, dtag, 0); 3571 3572 return (0); 3573 } 3574 3575 int 3576 dsl_dataset_user_release(char *dsname, char *snapname, char *htag, 3577 boolean_t recursive) 3578 { 3579 struct dsl_ds_holdarg *ha; 3580 dsl_sync_task_t *dst; 3581 spa_t *spa; 3582 int error; 3583 3584 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); 3585 3586 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3587 3588 error = spa_open(dsname, &spa, FTAG); 3589 if (error) { 3590 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3591 return (error); 3592 } 3593 3594 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 3595 ha->htag = htag; 3596 ha->snapname = snapname; 3597 ha->recursive = recursive; 3598 if (recursive) { 3599 error = dmu_objset_find(dsname, dsl_dataset_user_release_one, 3600 ha, DS_FIND_CHILDREN); 3601 } else { 3602 error = dsl_dataset_user_release_one(dsname, ha); 3603 } 3604 if (error == 0) 3605 error = dsl_sync_task_group_wait(ha->dstg); 3606 3607 for (dst = list_head(&ha->dstg->dstg_tasks); dst; 3608 dst = list_next(&ha->dstg->dstg_tasks, dst)) { 3609 struct dsl_ds_releasearg *ra = dst->dst_arg1; 3610 dsl_dataset_t *ds = ra->ds; 3611 3612 if (dst->dst_err) 3613 dsl_dataset_name(ds, ha->failed); 3614 3615 if (ra->own) 3616 dsl_dataset_disown(ds, ha->dstg); 3617 else 3618 dsl_dataset_rele(ds, ha->dstg); 3619 3620 kmem_free(ra, sizeof (struct dsl_ds_releasearg)); 3621 } 3622 3623 if (error == 0 && recursive && !ha->gotone) 3624 error = ENOENT; 3625 3626 if (error) 3627 (void) strcpy(dsname, ha->failed); 3628 3629 dsl_sync_task_group_destroy(ha->dstg); 3630 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3631 spa_close(spa, FTAG); 3632 return (error); 3633 } 3634 3635 /* 3636 * Called at spa_load time to release a stale temporary user hold. 3637 */ 3638 int 3639 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag) 3640 { 3641 dsl_dataset_t *ds; 3642 char *snap; 3643 char *name; 3644 int namelen; 3645 int error; 3646 3647 rw_enter(&dp->dp_config_rwlock, RW_READER); 3648 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); 3649 rw_exit(&dp->dp_config_rwlock); 3650 if (error) 3651 return (error); 3652 namelen = dsl_dataset_namelen(ds)+1; 3653 name = kmem_alloc(namelen, KM_SLEEP); 3654 dsl_dataset_name(ds, name); 3655 dsl_dataset_rele(ds, FTAG); 3656 3657 snap = strchr(name, '@'); 3658 *snap = '\0'; 3659 ++snap; 3660 return (dsl_dataset_user_release(name, snap, htag, B_FALSE)); 3661 } 3662 3663 int 3664 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp) 3665 { 3666 dsl_dataset_t *ds; 3667 int err; 3668 3669 err = dsl_dataset_hold(dsname, FTAG, &ds); 3670 if (err) 3671 return (err); 3672 3673 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP)); 3674 if (ds->ds_phys->ds_userrefs_obj != 0) { 3675 zap_attribute_t *za; 3676 zap_cursor_t zc; 3677 3678 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 3679 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset, 3680 ds->ds_phys->ds_userrefs_obj); 3681 zap_cursor_retrieve(&zc, za) == 0; 3682 zap_cursor_advance(&zc)) { 3683 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name, 3684 za->za_first_integer)); 3685 } 3686 zap_cursor_fini(&zc); 3687 kmem_free(za, sizeof (zap_attribute_t)); 3688 } 3689 dsl_dataset_rele(ds, FTAG); 3690 return (0); 3691 } 3692 3693 /* 3694 * Note, this fuction is used as the callback for dmu_objset_find(). We 3695 * always return 0 so that we will continue to find and process 3696 * inconsistent datasets, even if we encounter an error trying to 3697 * process one of them. 3698 */ 3699 /* ARGSUSED */ 3700 int 3701 dsl_destroy_inconsistent(char *dsname, void *arg) 3702 { 3703 dsl_dataset_t *ds; 3704 3705 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) { 3706 if (DS_IS_INCONSISTENT(ds)) 3707 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE); 3708 else 3709 dsl_dataset_disown(ds, FTAG); 3710 } 3711 return (0); 3712 } 3713