1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012 by Delphix. All rights reserved. 24 * Copyright (c) 2012, Joyent, Inc. All rights reserved. 25 */ 26 27 #include <sys/dmu_objset.h> 28 #include <sys/dsl_dataset.h> 29 #include <sys/dsl_dir.h> 30 #include <sys/dsl_prop.h> 31 #include <sys/dsl_synctask.h> 32 #include <sys/dmu_traverse.h> 33 #include <sys/dmu_impl.h> 34 #include <sys/dmu_tx.h> 35 #include <sys/arc.h> 36 #include <sys/zio.h> 37 #include <sys/zap.h> 38 #include <sys/zfeature.h> 39 #include <sys/unique.h> 40 #include <sys/zfs_context.h> 41 #include <sys/zfs_ioctl.h> 42 #include <sys/spa.h> 43 #include <sys/zfs_znode.h> 44 #include <sys/zfs_onexit.h> 45 #include <sys/zvol.h> 46 #include <sys/dsl_scan.h> 47 #include <sys/dsl_deadlist.h> 48 49 static char *dsl_reaper = "the grim reaper"; 50 51 static dsl_checkfunc_t dsl_dataset_destroy_begin_check; 52 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync; 53 static dsl_syncfunc_t dsl_dataset_set_reservation_sync; 54 55 #define SWITCH64(x, y) \ 56 { \ 57 uint64_t __tmp = (x); \ 58 (x) = (y); \ 59 (y) = __tmp; \ 60 } 61 62 #define DS_REF_MAX (1ULL << 62) 63 64 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE 65 66 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper) 67 68 69 /* 70 * Figure out how much of this delta should be propogated to the dsl_dir 71 * layer. If there's a refreservation, that space has already been 72 * partially accounted for in our ancestors. 73 */ 74 static int64_t 75 parent_delta(dsl_dataset_t *ds, int64_t delta) 76 { 77 uint64_t old_bytes, new_bytes; 78 79 if (ds->ds_reserved == 0) 80 return (delta); 81 82 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 83 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 84 85 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 86 return (new_bytes - old_bytes); 87 } 88 89 void 90 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 91 { 92 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 93 int compressed = BP_GET_PSIZE(bp); 94 int uncompressed = BP_GET_UCSIZE(bp); 95 int64_t delta; 96 97 dprintf_bp(bp, "ds=%p", ds); 98 99 ASSERT(dmu_tx_is_syncing(tx)); 100 /* It could have been compressed away to nothing */ 101 if (BP_IS_HOLE(bp)) 102 return; 103 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 104 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 105 if (ds == NULL) { 106 dsl_pool_mos_diduse_space(tx->tx_pool, 107 used, compressed, uncompressed); 108 return; 109 } 110 dmu_buf_will_dirty(ds->ds_dbuf, tx); 111 112 mutex_enter(&ds->ds_dir->dd_lock); 113 mutex_enter(&ds->ds_lock); 114 delta = parent_delta(ds, used); 115 ds->ds_phys->ds_referenced_bytes += used; 116 ds->ds_phys->ds_compressed_bytes += compressed; 117 ds->ds_phys->ds_uncompressed_bytes += uncompressed; 118 ds->ds_phys->ds_unique_bytes += used; 119 mutex_exit(&ds->ds_lock); 120 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 121 compressed, uncompressed, tx); 122 dsl_dir_transfer_space(ds->ds_dir, used - delta, 123 DD_USED_REFRSRV, DD_USED_HEAD, tx); 124 mutex_exit(&ds->ds_dir->dd_lock); 125 } 126 127 int 128 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 129 boolean_t async) 130 { 131 if (BP_IS_HOLE(bp)) 132 return (0); 133 134 ASSERT(dmu_tx_is_syncing(tx)); 135 ASSERT(bp->blk_birth <= tx->tx_txg); 136 137 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 138 int compressed = BP_GET_PSIZE(bp); 139 int uncompressed = BP_GET_UCSIZE(bp); 140 141 ASSERT(used > 0); 142 if (ds == NULL) { 143 dsl_free(tx->tx_pool, tx->tx_txg, bp); 144 dsl_pool_mos_diduse_space(tx->tx_pool, 145 -used, -compressed, -uncompressed); 146 return (used); 147 } 148 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 149 150 ASSERT(!dsl_dataset_is_snapshot(ds)); 151 dmu_buf_will_dirty(ds->ds_dbuf, tx); 152 153 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { 154 int64_t delta; 155 156 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 157 dsl_free(tx->tx_pool, tx->tx_txg, bp); 158 159 mutex_enter(&ds->ds_dir->dd_lock); 160 mutex_enter(&ds->ds_lock); 161 ASSERT(ds->ds_phys->ds_unique_bytes >= used || 162 !DS_UNIQUE_IS_ACCURATE(ds)); 163 delta = parent_delta(ds, -used); 164 ds->ds_phys->ds_unique_bytes -= used; 165 mutex_exit(&ds->ds_lock); 166 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 167 delta, -compressed, -uncompressed, tx); 168 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 169 DD_USED_REFRSRV, DD_USED_HEAD, tx); 170 mutex_exit(&ds->ds_dir->dd_lock); 171 } else { 172 dprintf_bp(bp, "putting on dead list: %s", ""); 173 if (async) { 174 /* 175 * We are here as part of zio's write done callback, 176 * which means we're a zio interrupt thread. We can't 177 * call dsl_deadlist_insert() now because it may block 178 * waiting for I/O. Instead, put bp on the deferred 179 * queue and let dsl_pool_sync() finish the job. 180 */ 181 bplist_append(&ds->ds_pending_deadlist, bp); 182 } else { 183 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); 184 } 185 ASSERT3U(ds->ds_prev->ds_object, ==, 186 ds->ds_phys->ds_prev_snap_obj); 187 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); 188 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 189 if (ds->ds_prev->ds_phys->ds_next_snap_obj == 190 ds->ds_object && bp->blk_birth > 191 ds->ds_prev->ds_phys->ds_prev_snap_txg) { 192 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 193 mutex_enter(&ds->ds_prev->ds_lock); 194 ds->ds_prev->ds_phys->ds_unique_bytes += used; 195 mutex_exit(&ds->ds_prev->ds_lock); 196 } 197 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 198 dsl_dir_transfer_space(ds->ds_dir, used, 199 DD_USED_HEAD, DD_USED_SNAP, tx); 200 } 201 } 202 mutex_enter(&ds->ds_lock); 203 ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used); 204 ds->ds_phys->ds_referenced_bytes -= used; 205 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); 206 ds->ds_phys->ds_compressed_bytes -= compressed; 207 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); 208 ds->ds_phys->ds_uncompressed_bytes -= uncompressed; 209 mutex_exit(&ds->ds_lock); 210 211 return (used); 212 } 213 214 uint64_t 215 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 216 { 217 uint64_t trysnap = 0; 218 219 if (ds == NULL) 220 return (0); 221 /* 222 * The snapshot creation could fail, but that would cause an 223 * incorrect FALSE return, which would only result in an 224 * overestimation of the amount of space that an operation would 225 * consume, which is OK. 226 * 227 * There's also a small window where we could miss a pending 228 * snapshot, because we could set the sync task in the quiescing 229 * phase. So this should only be used as a guess. 230 */ 231 if (ds->ds_trysnap_txg > 232 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 233 trysnap = ds->ds_trysnap_txg; 234 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap)); 235 } 236 237 boolean_t 238 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp, 239 uint64_t blk_birth) 240 { 241 if (blk_birth <= dsl_dataset_prev_snap_txg(ds)) 242 return (B_FALSE); 243 244 ddt_prefetch(dsl_dataset_get_spa(ds), bp); 245 246 return (B_TRUE); 247 } 248 249 /* ARGSUSED */ 250 static void 251 dsl_dataset_evict(dmu_buf_t *db, void *dsv) 252 { 253 dsl_dataset_t *ds = dsv; 254 255 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds)); 256 257 unique_remove(ds->ds_fsid_guid); 258 259 if (ds->ds_objset != NULL) 260 dmu_objset_evict(ds->ds_objset); 261 262 if (ds->ds_prev) { 263 dsl_dataset_drop_ref(ds->ds_prev, ds); 264 ds->ds_prev = NULL; 265 } 266 267 bplist_destroy(&ds->ds_pending_deadlist); 268 if (db != NULL) { 269 dsl_deadlist_close(&ds->ds_deadlist); 270 } else { 271 ASSERT(ds->ds_deadlist.dl_dbuf == NULL); 272 ASSERT(!ds->ds_deadlist.dl_oldfmt); 273 } 274 if (ds->ds_dir) 275 dsl_dir_close(ds->ds_dir, ds); 276 277 ASSERT(!list_link_active(&ds->ds_synced_link)); 278 279 mutex_destroy(&ds->ds_lock); 280 mutex_destroy(&ds->ds_recvlock); 281 mutex_destroy(&ds->ds_opening_lock); 282 rw_destroy(&ds->ds_rwlock); 283 cv_destroy(&ds->ds_exclusive_cv); 284 285 kmem_free(ds, sizeof (dsl_dataset_t)); 286 } 287 288 static int 289 dsl_dataset_get_snapname(dsl_dataset_t *ds) 290 { 291 dsl_dataset_phys_t *headphys; 292 int err; 293 dmu_buf_t *headdbuf; 294 dsl_pool_t *dp = ds->ds_dir->dd_pool; 295 objset_t *mos = dp->dp_meta_objset; 296 297 if (ds->ds_snapname[0]) 298 return (0); 299 if (ds->ds_phys->ds_next_snap_obj == 0) 300 return (0); 301 302 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, 303 FTAG, &headdbuf); 304 if (err) 305 return (err); 306 headphys = headdbuf->db_data; 307 err = zap_value_search(dp->dp_meta_objset, 308 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 309 dmu_buf_rele(headdbuf, FTAG); 310 return (err); 311 } 312 313 static int 314 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 315 { 316 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 317 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 318 matchtype_t mt; 319 int err; 320 321 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 322 mt = MT_FIRST; 323 else 324 mt = MT_EXACT; 325 326 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 327 value, mt, NULL, 0, NULL); 328 if (err == ENOTSUP && mt == MT_FIRST) 329 err = zap_lookup(mos, snapobj, name, 8, 1, value); 330 return (err); 331 } 332 333 static int 334 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx) 335 { 336 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 337 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 338 matchtype_t mt; 339 int err; 340 341 dsl_dir_snap_cmtime_update(ds->ds_dir); 342 343 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 344 mt = MT_FIRST; 345 else 346 mt = MT_EXACT; 347 348 err = zap_remove_norm(mos, snapobj, name, mt, tx); 349 if (err == ENOTSUP && mt == MT_FIRST) 350 err = zap_remove(mos, snapobj, name, tx); 351 return (err); 352 } 353 354 static int 355 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, 356 dsl_dataset_t **dsp) 357 { 358 objset_t *mos = dp->dp_meta_objset; 359 dmu_buf_t *dbuf; 360 dsl_dataset_t *ds; 361 int err; 362 dmu_object_info_t doi; 363 364 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 365 dsl_pool_sync_context(dp)); 366 367 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 368 if (err) 369 return (err); 370 371 /* Make sure dsobj has the correct object type. */ 372 dmu_object_info_from_db(dbuf, &doi); 373 if (doi.doi_type != DMU_OT_DSL_DATASET) 374 return (EINVAL); 375 376 ds = dmu_buf_get_user(dbuf); 377 if (ds == NULL) { 378 dsl_dataset_t *winner; 379 380 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 381 ds->ds_dbuf = dbuf; 382 ds->ds_object = dsobj; 383 ds->ds_phys = dbuf->db_data; 384 385 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 386 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL); 387 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 388 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 389 390 rw_init(&ds->ds_rwlock, 0, 0, 0); 391 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL); 392 393 bplist_create(&ds->ds_pending_deadlist); 394 dsl_deadlist_open(&ds->ds_deadlist, 395 mos, ds->ds_phys->ds_deadlist_obj); 396 397 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 398 offsetof(dmu_sendarg_t, dsa_link)); 399 400 if (err == 0) { 401 err = dsl_dir_open_obj(dp, 402 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); 403 } 404 if (err) { 405 mutex_destroy(&ds->ds_lock); 406 mutex_destroy(&ds->ds_recvlock); 407 mutex_destroy(&ds->ds_opening_lock); 408 rw_destroy(&ds->ds_rwlock); 409 cv_destroy(&ds->ds_exclusive_cv); 410 bplist_destroy(&ds->ds_pending_deadlist); 411 dsl_deadlist_close(&ds->ds_deadlist); 412 kmem_free(ds, sizeof (dsl_dataset_t)); 413 dmu_buf_rele(dbuf, tag); 414 return (err); 415 } 416 417 if (!dsl_dataset_is_snapshot(ds)) { 418 ds->ds_snapname[0] = '\0'; 419 if (ds->ds_phys->ds_prev_snap_obj) { 420 err = dsl_dataset_get_ref(dp, 421 ds->ds_phys->ds_prev_snap_obj, 422 ds, &ds->ds_prev); 423 } 424 } else { 425 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 426 err = dsl_dataset_get_snapname(ds); 427 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) { 428 err = zap_count( 429 ds->ds_dir->dd_pool->dp_meta_objset, 430 ds->ds_phys->ds_userrefs_obj, 431 &ds->ds_userrefs); 432 } 433 } 434 435 if (err == 0 && !dsl_dataset_is_snapshot(ds)) { 436 /* 437 * In sync context, we're called with either no lock 438 * or with the write lock. If we're not syncing, 439 * we're always called with the read lock held. 440 */ 441 boolean_t need_lock = 442 !RW_WRITE_HELD(&dp->dp_config_rwlock) && 443 dsl_pool_sync_context(dp); 444 445 if (need_lock) 446 rw_enter(&dp->dp_config_rwlock, RW_READER); 447 448 err = dsl_prop_get_ds(ds, 449 "refreservation", sizeof (uint64_t), 1, 450 &ds->ds_reserved, NULL); 451 if (err == 0) { 452 err = dsl_prop_get_ds(ds, 453 "refquota", sizeof (uint64_t), 1, 454 &ds->ds_quota, NULL); 455 } 456 457 if (need_lock) 458 rw_exit(&dp->dp_config_rwlock); 459 } else { 460 ds->ds_reserved = ds->ds_quota = 0; 461 } 462 463 if (err == 0) { 464 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys, 465 dsl_dataset_evict); 466 } 467 if (err || winner) { 468 bplist_destroy(&ds->ds_pending_deadlist); 469 dsl_deadlist_close(&ds->ds_deadlist); 470 if (ds->ds_prev) 471 dsl_dataset_drop_ref(ds->ds_prev, ds); 472 dsl_dir_close(ds->ds_dir, ds); 473 mutex_destroy(&ds->ds_lock); 474 mutex_destroy(&ds->ds_recvlock); 475 mutex_destroy(&ds->ds_opening_lock); 476 rw_destroy(&ds->ds_rwlock); 477 cv_destroy(&ds->ds_exclusive_cv); 478 kmem_free(ds, sizeof (dsl_dataset_t)); 479 if (err) { 480 dmu_buf_rele(dbuf, tag); 481 return (err); 482 } 483 ds = winner; 484 } else { 485 ds->ds_fsid_guid = 486 unique_insert(ds->ds_phys->ds_fsid_guid); 487 } 488 } 489 ASSERT3P(ds->ds_dbuf, ==, dbuf); 490 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); 491 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 || 492 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 493 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 494 mutex_enter(&ds->ds_lock); 495 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) { 496 mutex_exit(&ds->ds_lock); 497 dmu_buf_rele(ds->ds_dbuf, tag); 498 return (ENOENT); 499 } 500 mutex_exit(&ds->ds_lock); 501 *dsp = ds; 502 return (0); 503 } 504 505 static int 506 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag) 507 { 508 dsl_pool_t *dp = ds->ds_dir->dd_pool; 509 510 /* 511 * In syncing context we don't want the rwlock lock: there 512 * may be an existing writer waiting for sync phase to 513 * finish. We don't need to worry about such writers, since 514 * sync phase is single-threaded, so the writer can't be 515 * doing anything while we are active. 516 */ 517 if (dsl_pool_sync_context(dp)) { 518 ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); 519 return (0); 520 } 521 522 /* 523 * Normal users will hold the ds_rwlock as a READER until they 524 * are finished (i.e., call dsl_dataset_rele()). "Owners" will 525 * drop their READER lock after they set the ds_owner field. 526 * 527 * If the dataset is being destroyed, the destroy thread will 528 * obtain a WRITER lock for exclusive access after it's done its 529 * open-context work and then change the ds_owner to 530 * dsl_reaper once destruction is assured. So threads 531 * may block here temporarily, until the "destructability" of 532 * the dataset is determined. 533 */ 534 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock)); 535 mutex_enter(&ds->ds_lock); 536 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) { 537 rw_exit(&dp->dp_config_rwlock); 538 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock); 539 if (DSL_DATASET_IS_DESTROYED(ds)) { 540 mutex_exit(&ds->ds_lock); 541 dsl_dataset_drop_ref(ds, tag); 542 rw_enter(&dp->dp_config_rwlock, RW_READER); 543 return (ENOENT); 544 } 545 /* 546 * The dp_config_rwlock lives above the ds_lock. And 547 * we need to check DSL_DATASET_IS_DESTROYED() while 548 * holding the ds_lock, so we have to drop and reacquire 549 * the ds_lock here. 550 */ 551 mutex_exit(&ds->ds_lock); 552 rw_enter(&dp->dp_config_rwlock, RW_READER); 553 mutex_enter(&ds->ds_lock); 554 } 555 mutex_exit(&ds->ds_lock); 556 return (0); 557 } 558 559 int 560 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 561 dsl_dataset_t **dsp) 562 { 563 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp); 564 565 if (err) 566 return (err); 567 return (dsl_dataset_hold_ref(*dsp, tag)); 568 } 569 570 int 571 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok, 572 void *tag, dsl_dataset_t **dsp) 573 { 574 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 575 if (err) 576 return (err); 577 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { 578 dsl_dataset_rele(*dsp, tag); 579 *dsp = NULL; 580 return (EBUSY); 581 } 582 return (0); 583 } 584 585 int 586 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp) 587 { 588 dsl_dir_t *dd; 589 dsl_pool_t *dp; 590 const char *snapname; 591 uint64_t obj; 592 int err = 0; 593 594 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname); 595 if (err) 596 return (err); 597 598 dp = dd->dd_pool; 599 obj = dd->dd_phys->dd_head_dataset_obj; 600 rw_enter(&dp->dp_config_rwlock, RW_READER); 601 if (obj) 602 err = dsl_dataset_get_ref(dp, obj, tag, dsp); 603 else 604 err = ENOENT; 605 if (err) 606 goto out; 607 608 err = dsl_dataset_hold_ref(*dsp, tag); 609 610 /* we may be looking for a snapshot */ 611 if (err == 0 && snapname != NULL) { 612 dsl_dataset_t *ds = NULL; 613 614 if (*snapname++ != '@') { 615 dsl_dataset_rele(*dsp, tag); 616 err = ENOENT; 617 goto out; 618 } 619 620 dprintf("looking for snapshot '%s'\n", snapname); 621 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); 622 if (err == 0) 623 err = dsl_dataset_get_ref(dp, obj, tag, &ds); 624 dsl_dataset_rele(*dsp, tag); 625 626 ASSERT3U((err == 0), ==, (ds != NULL)); 627 628 if (ds) { 629 mutex_enter(&ds->ds_lock); 630 if (ds->ds_snapname[0] == 0) 631 (void) strlcpy(ds->ds_snapname, snapname, 632 sizeof (ds->ds_snapname)); 633 mutex_exit(&ds->ds_lock); 634 err = dsl_dataset_hold_ref(ds, tag); 635 *dsp = err ? NULL : ds; 636 } 637 } 638 out: 639 rw_exit(&dp->dp_config_rwlock); 640 dsl_dir_close(dd, FTAG); 641 return (err); 642 } 643 644 int 645 dsl_dataset_own(const char *name, boolean_t inconsistentok, 646 void *tag, dsl_dataset_t **dsp) 647 { 648 int err = dsl_dataset_hold(name, tag, dsp); 649 if (err) 650 return (err); 651 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { 652 dsl_dataset_rele(*dsp, tag); 653 return (EBUSY); 654 } 655 return (0); 656 } 657 658 void 659 dsl_dataset_name(dsl_dataset_t *ds, char *name) 660 { 661 if (ds == NULL) { 662 (void) strcpy(name, "mos"); 663 } else { 664 dsl_dir_name(ds->ds_dir, name); 665 VERIFY(0 == dsl_dataset_get_snapname(ds)); 666 if (ds->ds_snapname[0]) { 667 (void) strcat(name, "@"); 668 /* 669 * We use a "recursive" mutex so that we 670 * can call dprintf_ds() with ds_lock held. 671 */ 672 if (!MUTEX_HELD(&ds->ds_lock)) { 673 mutex_enter(&ds->ds_lock); 674 (void) strcat(name, ds->ds_snapname); 675 mutex_exit(&ds->ds_lock); 676 } else { 677 (void) strcat(name, ds->ds_snapname); 678 } 679 } 680 } 681 } 682 683 static int 684 dsl_dataset_namelen(dsl_dataset_t *ds) 685 { 686 int result; 687 688 if (ds == NULL) { 689 result = 3; /* "mos" */ 690 } else { 691 result = dsl_dir_namelen(ds->ds_dir); 692 VERIFY(0 == dsl_dataset_get_snapname(ds)); 693 if (ds->ds_snapname[0]) { 694 ++result; /* adding one for the @-sign */ 695 if (!MUTEX_HELD(&ds->ds_lock)) { 696 mutex_enter(&ds->ds_lock); 697 result += strlen(ds->ds_snapname); 698 mutex_exit(&ds->ds_lock); 699 } else { 700 result += strlen(ds->ds_snapname); 701 } 702 } 703 } 704 705 return (result); 706 } 707 708 void 709 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag) 710 { 711 dmu_buf_rele(ds->ds_dbuf, tag); 712 } 713 714 void 715 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 716 { 717 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) { 718 rw_exit(&ds->ds_rwlock); 719 } 720 dsl_dataset_drop_ref(ds, tag); 721 } 722 723 void 724 dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 725 { 726 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) || 727 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL)); 728 729 mutex_enter(&ds->ds_lock); 730 ds->ds_owner = NULL; 731 if (RW_WRITE_HELD(&ds->ds_rwlock)) { 732 rw_exit(&ds->ds_rwlock); 733 cv_broadcast(&ds->ds_exclusive_cv); 734 } 735 mutex_exit(&ds->ds_lock); 736 if (ds->ds_dbuf) 737 dsl_dataset_drop_ref(ds, tag); 738 else 739 dsl_dataset_evict(NULL, ds); 740 } 741 742 boolean_t 743 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag) 744 { 745 boolean_t gotit = FALSE; 746 747 mutex_enter(&ds->ds_lock); 748 if (ds->ds_owner == NULL && 749 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) { 750 ds->ds_owner = tag; 751 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) 752 rw_exit(&ds->ds_rwlock); 753 gotit = TRUE; 754 } 755 mutex_exit(&ds->ds_lock); 756 return (gotit); 757 } 758 759 void 760 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner) 761 { 762 ASSERT3P(owner, ==, ds->ds_owner); 763 if (!RW_WRITE_HELD(&ds->ds_rwlock)) 764 rw_enter(&ds->ds_rwlock, RW_WRITER); 765 } 766 767 uint64_t 768 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 769 uint64_t flags, dmu_tx_t *tx) 770 { 771 dsl_pool_t *dp = dd->dd_pool; 772 dmu_buf_t *dbuf; 773 dsl_dataset_phys_t *dsphys; 774 uint64_t dsobj; 775 objset_t *mos = dp->dp_meta_objset; 776 777 if (origin == NULL) 778 origin = dp->dp_origin_snap; 779 780 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 781 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0); 782 ASSERT(dmu_tx_is_syncing(tx)); 783 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 784 785 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 786 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 787 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 788 dmu_buf_will_dirty(dbuf, tx); 789 dsphys = dbuf->db_data; 790 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 791 dsphys->ds_dir_obj = dd->dd_object; 792 dsphys->ds_flags = flags; 793 dsphys->ds_fsid_guid = unique_create(); 794 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 795 sizeof (dsphys->ds_guid)); 796 dsphys->ds_snapnames_zapobj = 797 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 798 DMU_OT_NONE, 0, tx); 799 dsphys->ds_creation_time = gethrestime_sec(); 800 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 801 802 if (origin == NULL) { 803 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 804 } else { 805 dsl_dataset_t *ohds; 806 807 dsphys->ds_prev_snap_obj = origin->ds_object; 808 dsphys->ds_prev_snap_txg = 809 origin->ds_phys->ds_creation_txg; 810 dsphys->ds_referenced_bytes = 811 origin->ds_phys->ds_referenced_bytes; 812 dsphys->ds_compressed_bytes = 813 origin->ds_phys->ds_compressed_bytes; 814 dsphys->ds_uncompressed_bytes = 815 origin->ds_phys->ds_uncompressed_bytes; 816 dsphys->ds_bp = origin->ds_phys->ds_bp; 817 dsphys->ds_flags |= origin->ds_phys->ds_flags; 818 819 dmu_buf_will_dirty(origin->ds_dbuf, tx); 820 origin->ds_phys->ds_num_children++; 821 822 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, 823 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds)); 824 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 825 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 826 dsl_dataset_rele(ohds, FTAG); 827 828 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 829 if (origin->ds_phys->ds_next_clones_obj == 0) { 830 origin->ds_phys->ds_next_clones_obj = 831 zap_create(mos, 832 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 833 } 834 VERIFY(0 == zap_add_int(mos, 835 origin->ds_phys->ds_next_clones_obj, 836 dsobj, tx)); 837 } 838 839 dmu_buf_will_dirty(dd->dd_dbuf, tx); 840 dd->dd_phys->dd_origin_obj = origin->ds_object; 841 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 842 if (origin->ds_dir->dd_phys->dd_clones == 0) { 843 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 844 origin->ds_dir->dd_phys->dd_clones = 845 zap_create(mos, 846 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 847 } 848 VERIFY3U(0, ==, zap_add_int(mos, 849 origin->ds_dir->dd_phys->dd_clones, dsobj, tx)); 850 } 851 } 852 853 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 854 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 855 856 dmu_buf_rele(dbuf, FTAG); 857 858 dmu_buf_will_dirty(dd->dd_dbuf, tx); 859 dd->dd_phys->dd_head_dataset_obj = dsobj; 860 861 return (dsobj); 862 } 863 864 uint64_t 865 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 866 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 867 { 868 dsl_pool_t *dp = pdd->dd_pool; 869 uint64_t dsobj, ddobj; 870 dsl_dir_t *dd; 871 872 ASSERT(lastname[0] != '@'); 873 874 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 875 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd)); 876 877 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx); 878 879 dsl_deleg_set_create_perms(dd, tx, cr); 880 881 dsl_dir_close(dd, FTAG); 882 883 /* 884 * If we are creating a clone, make sure we zero out any stale 885 * data from the origin snapshots zil header. 886 */ 887 if (origin != NULL) { 888 dsl_dataset_t *ds; 889 objset_t *os; 890 891 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 892 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os)); 893 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 894 dsl_dataset_dirty(ds, tx); 895 dsl_dataset_rele(ds, FTAG); 896 } 897 898 return (dsobj); 899 } 900 901 /* 902 * The snapshots must all be in the same pool. 903 */ 904 int 905 dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer, 906 nvlist_t *errlist) 907 { 908 int err; 909 dsl_sync_task_t *dst; 910 spa_t *spa; 911 nvpair_t *pair; 912 dsl_sync_task_group_t *dstg; 913 914 pair = nvlist_next_nvpair(snaps, NULL); 915 if (pair == NULL) 916 return (0); 917 918 err = spa_open(nvpair_name(pair), &spa, FTAG); 919 if (err) 920 return (err); 921 dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 922 923 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 924 pair = nvlist_next_nvpair(snaps, pair)) { 925 dsl_dataset_t *ds; 926 927 err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds); 928 if (err == 0) { 929 struct dsl_ds_destroyarg *dsda; 930 931 dsl_dataset_make_exclusive(ds, dstg); 932 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), 933 KM_SLEEP); 934 dsda->ds = ds; 935 dsda->defer = defer; 936 dsl_sync_task_create(dstg, dsl_dataset_destroy_check, 937 dsl_dataset_destroy_sync, dsda, dstg, 0); 938 } else if (err == ENOENT) { 939 err = 0; 940 } else { 941 fnvlist_add_int32(errlist, nvpair_name(pair), err); 942 break; 943 } 944 } 945 946 if (err == 0) 947 err = dsl_sync_task_group_wait(dstg); 948 949 for (dst = list_head(&dstg->dstg_tasks); dst; 950 dst = list_next(&dstg->dstg_tasks, dst)) { 951 struct dsl_ds_destroyarg *dsda = dst->dst_arg1; 952 dsl_dataset_t *ds = dsda->ds; 953 954 /* 955 * Return the snapshots that triggered the error. 956 */ 957 if (dst->dst_err != 0) { 958 char name[ZFS_MAXNAMELEN]; 959 dsl_dataset_name(ds, name); 960 fnvlist_add_int32(errlist, name, dst->dst_err); 961 } 962 ASSERT3P(dsda->rm_origin, ==, NULL); 963 dsl_dataset_disown(ds, dstg); 964 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg)); 965 } 966 967 dsl_sync_task_group_destroy(dstg); 968 spa_close(spa, FTAG); 969 return (err); 970 971 } 972 973 static boolean_t 974 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds) 975 { 976 boolean_t might_destroy = B_FALSE; 977 978 mutex_enter(&ds->ds_lock); 979 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 && 980 DS_IS_DEFER_DESTROY(ds)) 981 might_destroy = B_TRUE; 982 mutex_exit(&ds->ds_lock); 983 984 return (might_destroy); 985 } 986 987 /* 988 * If we're removing a clone, and these three conditions are true: 989 * 1) the clone's origin has no other children 990 * 2) the clone's origin has no user references 991 * 3) the clone's origin has been marked for deferred destruction 992 * Then, prepare to remove the origin as part of this sync task group. 993 */ 994 static int 995 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag) 996 { 997 dsl_dataset_t *ds = dsda->ds; 998 dsl_dataset_t *origin = ds->ds_prev; 999 1000 if (dsl_dataset_might_destroy_origin(origin)) { 1001 char *name; 1002 int namelen; 1003 int error; 1004 1005 namelen = dsl_dataset_namelen(origin) + 1; 1006 name = kmem_alloc(namelen, KM_SLEEP); 1007 dsl_dataset_name(origin, name); 1008 #ifdef _KERNEL 1009 error = zfs_unmount_snap(name, NULL); 1010 if (error) { 1011 kmem_free(name, namelen); 1012 return (error); 1013 } 1014 #endif 1015 error = dsl_dataset_own(name, B_TRUE, tag, &origin); 1016 kmem_free(name, namelen); 1017 if (error) 1018 return (error); 1019 dsda->rm_origin = origin; 1020 dsl_dataset_make_exclusive(origin, tag); 1021 } 1022 1023 return (0); 1024 } 1025 1026 /* 1027 * ds must be opened as OWNER. On return (whether successful or not), 1028 * ds will be closed and caller can no longer dereference it. 1029 */ 1030 int 1031 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer) 1032 { 1033 int err; 1034 dsl_sync_task_group_t *dstg; 1035 objset_t *os; 1036 dsl_dir_t *dd; 1037 uint64_t obj; 1038 struct dsl_ds_destroyarg dsda = { 0 }; 1039 1040 dsda.ds = ds; 1041 1042 if (dsl_dataset_is_snapshot(ds)) { 1043 /* Destroying a snapshot is simpler */ 1044 dsl_dataset_make_exclusive(ds, tag); 1045 1046 dsda.defer = defer; 1047 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 1048 dsl_dataset_destroy_check, dsl_dataset_destroy_sync, 1049 &dsda, tag, 0); 1050 ASSERT3P(dsda.rm_origin, ==, NULL); 1051 goto out; 1052 } else if (defer) { 1053 err = EINVAL; 1054 goto out; 1055 } 1056 1057 dd = ds->ds_dir; 1058 1059 if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds), 1060 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) { 1061 /* 1062 * Check for errors and mark this ds as inconsistent, in 1063 * case we crash while freeing the objects. 1064 */ 1065 err = dsl_sync_task_do(dd->dd_pool, 1066 dsl_dataset_destroy_begin_check, 1067 dsl_dataset_destroy_begin_sync, ds, NULL, 0); 1068 if (err) 1069 goto out; 1070 1071 err = dmu_objset_from_ds(ds, &os); 1072 if (err) 1073 goto out; 1074 1075 /* 1076 * Remove all objects while in the open context so that 1077 * there is less work to do in the syncing context. 1078 */ 1079 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 1080 ds->ds_phys->ds_prev_snap_txg)) { 1081 /* 1082 * Ignore errors, if there is not enough disk space 1083 * we will deal with it in dsl_dataset_destroy_sync(). 1084 */ 1085 (void) dmu_free_object(os, obj); 1086 } 1087 if (err != ESRCH) 1088 goto out; 1089 1090 /* 1091 * Sync out all in-flight IO. 1092 */ 1093 txg_wait_synced(dd->dd_pool, 0); 1094 1095 /* 1096 * If we managed to free all the objects in open 1097 * context, the user space accounting should be zero. 1098 */ 1099 if (ds->ds_phys->ds_bp.blk_fill == 0 && 1100 dmu_objset_userused_enabled(os)) { 1101 uint64_t count; 1102 1103 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, 1104 &count) != 0 || count == 0); 1105 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, 1106 &count) != 0 || count == 0); 1107 } 1108 } 1109 1110 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); 1111 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd); 1112 rw_exit(&dd->dd_pool->dp_config_rwlock); 1113 1114 if (err) 1115 goto out; 1116 1117 /* 1118 * Blow away the dsl_dir + head dataset. 1119 */ 1120 dsl_dataset_make_exclusive(ds, tag); 1121 /* 1122 * If we're removing a clone, we might also need to remove its 1123 * origin. 1124 */ 1125 do { 1126 dsda.need_prep = B_FALSE; 1127 if (dsl_dir_is_clone(dd)) { 1128 err = dsl_dataset_origin_rm_prep(&dsda, tag); 1129 if (err) { 1130 dsl_dir_close(dd, FTAG); 1131 goto out; 1132 } 1133 } 1134 1135 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool); 1136 dsl_sync_task_create(dstg, dsl_dataset_destroy_check, 1137 dsl_dataset_destroy_sync, &dsda, tag, 0); 1138 dsl_sync_task_create(dstg, dsl_dir_destroy_check, 1139 dsl_dir_destroy_sync, dd, FTAG, 0); 1140 err = dsl_sync_task_group_wait(dstg); 1141 dsl_sync_task_group_destroy(dstg); 1142 1143 /* 1144 * We could be racing against 'zfs release' or 'zfs destroy -d' 1145 * on the origin snap, in which case we can get EBUSY if we 1146 * needed to destroy the origin snap but were not ready to 1147 * do so. 1148 */ 1149 if (dsda.need_prep) { 1150 ASSERT(err == EBUSY); 1151 ASSERT(dsl_dir_is_clone(dd)); 1152 ASSERT(dsda.rm_origin == NULL); 1153 } 1154 } while (dsda.need_prep); 1155 1156 if (dsda.rm_origin != NULL) 1157 dsl_dataset_disown(dsda.rm_origin, tag); 1158 1159 /* if it is successful, dsl_dir_destroy_sync will close the dd */ 1160 if (err) 1161 dsl_dir_close(dd, FTAG); 1162 out: 1163 dsl_dataset_disown(ds, tag); 1164 return (err); 1165 } 1166 1167 blkptr_t * 1168 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1169 { 1170 return (&ds->ds_phys->ds_bp); 1171 } 1172 1173 void 1174 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 1175 { 1176 ASSERT(dmu_tx_is_syncing(tx)); 1177 /* If it's the meta-objset, set dp_meta_rootbp */ 1178 if (ds == NULL) { 1179 tx->tx_pool->dp_meta_rootbp = *bp; 1180 } else { 1181 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1182 ds->ds_phys->ds_bp = *bp; 1183 } 1184 } 1185 1186 spa_t * 1187 dsl_dataset_get_spa(dsl_dataset_t *ds) 1188 { 1189 return (ds->ds_dir->dd_pool->dp_spa); 1190 } 1191 1192 void 1193 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1194 { 1195 dsl_pool_t *dp; 1196 1197 if (ds == NULL) /* this is the meta-objset */ 1198 return; 1199 1200 ASSERT(ds->ds_objset != NULL); 1201 1202 if (ds->ds_phys->ds_next_snap_obj != 0) 1203 panic("dirtying snapshot!"); 1204 1205 dp = ds->ds_dir->dd_pool; 1206 1207 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) { 1208 /* up the hold count until we can be written out */ 1209 dmu_buf_add_ref(ds->ds_dbuf, ds); 1210 } 1211 } 1212 1213 /* 1214 * The unique space in the head dataset can be calculated by subtracting 1215 * the space used in the most recent snapshot, that is still being used 1216 * in this file system, from the space currently in use. To figure out 1217 * the space in the most recent snapshot still in use, we need to take 1218 * the total space used in the snapshot and subtract out the space that 1219 * has been freed up since the snapshot was taken. 1220 */ 1221 static void 1222 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 1223 { 1224 uint64_t mrs_used; 1225 uint64_t dlused, dlcomp, dluncomp; 1226 1227 ASSERT(!dsl_dataset_is_snapshot(ds)); 1228 1229 if (ds->ds_phys->ds_prev_snap_obj != 0) 1230 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes; 1231 else 1232 mrs_used = 0; 1233 1234 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 1235 1236 ASSERT3U(dlused, <=, mrs_used); 1237 ds->ds_phys->ds_unique_bytes = 1238 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused); 1239 1240 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1241 SPA_VERSION_UNIQUE_ACCURATE) 1242 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1243 } 1244 1245 struct killarg { 1246 dsl_dataset_t *ds; 1247 dmu_tx_t *tx; 1248 }; 1249 1250 /* ARGSUSED */ 1251 static int 1252 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf, 1253 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 1254 { 1255 struct killarg *ka = arg; 1256 dmu_tx_t *tx = ka->tx; 1257 1258 if (bp == NULL) 1259 return (0); 1260 1261 if (zb->zb_level == ZB_ZIL_LEVEL) { 1262 ASSERT(zilog != NULL); 1263 /* 1264 * It's a block in the intent log. It has no 1265 * accounting, so just free it. 1266 */ 1267 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp); 1268 } else { 1269 ASSERT(zilog == NULL); 1270 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg); 1271 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE); 1272 } 1273 1274 return (0); 1275 } 1276 1277 /* ARGSUSED */ 1278 static int 1279 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx) 1280 { 1281 dsl_dataset_t *ds = arg1; 1282 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1283 uint64_t count; 1284 int err; 1285 1286 /* 1287 * Can't delete a head dataset if there are snapshots of it. 1288 * (Except if the only snapshots are from the branch we cloned 1289 * from.) 1290 */ 1291 if (ds->ds_prev != NULL && 1292 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) 1293 return (EBUSY); 1294 1295 /* 1296 * This is really a dsl_dir thing, but check it here so that 1297 * we'll be less likely to leave this dataset inconsistent & 1298 * nearly destroyed. 1299 */ 1300 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count); 1301 if (err) 1302 return (err); 1303 if (count != 0) 1304 return (EEXIST); 1305 1306 return (0); 1307 } 1308 1309 /* ARGSUSED */ 1310 static void 1311 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx) 1312 { 1313 dsl_dataset_t *ds = arg1; 1314 1315 /* Mark it as inconsistent on-disk, in case we crash */ 1316 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1317 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; 1318 1319 spa_history_log_internal_ds(ds, "destroy begin", tx, ""); 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 /* 1359 * If you add new checks here, you may need to add 1360 * additional checks to the "temporary" case in 1361 * snapshot_check() in dmu_objset.c. 1362 */ 1363 /* ARGSUSED */ 1364 int 1365 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) 1366 { 1367 struct dsl_ds_destroyarg *dsda = arg1; 1368 dsl_dataset_t *ds = dsda->ds; 1369 1370 /* we have an owner hold, so noone else can destroy us */ 1371 ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); 1372 1373 /* 1374 * Only allow deferred destroy on pools that support it. 1375 * NOTE: deferred destroy is only supported on snapshots. 1376 */ 1377 if (dsda->defer) { 1378 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < 1379 SPA_VERSION_USERREFS) 1380 return (ENOTSUP); 1381 ASSERT(dsl_dataset_is_snapshot(ds)); 1382 return (0); 1383 } 1384 1385 /* 1386 * Can't delete a head dataset if there are snapshots of it. 1387 * (Except if the only snapshots are from the branch we cloned 1388 * from.) 1389 */ 1390 if (ds->ds_prev != NULL && 1391 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) 1392 return (EBUSY); 1393 1394 /* 1395 * If we made changes this txg, traverse_dsl_dataset won't find 1396 * them. Try again. 1397 */ 1398 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) 1399 return (EAGAIN); 1400 1401 if (dsl_dataset_is_snapshot(ds)) { 1402 /* 1403 * If this snapshot has an elevated user reference count, 1404 * we can't destroy it yet. 1405 */ 1406 if (ds->ds_userrefs > 0 && !dsda->releasing) 1407 return (EBUSY); 1408 1409 mutex_enter(&ds->ds_lock); 1410 /* 1411 * Can't delete a branch point. However, if we're destroying 1412 * a clone and removing its origin due to it having a user 1413 * hold count of 0 and having been marked for deferred destroy, 1414 * it's OK for the origin to have a single clone. 1415 */ 1416 if (ds->ds_phys->ds_num_children > 1417 (dsda->is_origin_rm ? 2 : 1)) { 1418 mutex_exit(&ds->ds_lock); 1419 return (EEXIST); 1420 } 1421 mutex_exit(&ds->ds_lock); 1422 } else if (dsl_dir_is_clone(ds->ds_dir)) { 1423 return (dsl_dataset_origin_check(dsda, arg2, tx)); 1424 } 1425 1426 /* XXX we should do some i/o error checking... */ 1427 return (0); 1428 } 1429 1430 struct refsarg { 1431 kmutex_t lock; 1432 boolean_t gone; 1433 kcondvar_t cv; 1434 }; 1435 1436 /* ARGSUSED */ 1437 static void 1438 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv) 1439 { 1440 struct refsarg *arg = argv; 1441 1442 mutex_enter(&arg->lock); 1443 arg->gone = TRUE; 1444 cv_signal(&arg->cv); 1445 mutex_exit(&arg->lock); 1446 } 1447 1448 static void 1449 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag) 1450 { 1451 struct refsarg arg; 1452 1453 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL); 1454 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL); 1455 arg.gone = FALSE; 1456 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys, 1457 dsl_dataset_refs_gone); 1458 dmu_buf_rele(ds->ds_dbuf, tag); 1459 mutex_enter(&arg.lock); 1460 while (!arg.gone) 1461 cv_wait(&arg.cv, &arg.lock); 1462 ASSERT(arg.gone); 1463 mutex_exit(&arg.lock); 1464 ds->ds_dbuf = NULL; 1465 ds->ds_phys = NULL; 1466 mutex_destroy(&arg.lock); 1467 cv_destroy(&arg.cv); 1468 } 1469 1470 static void 1471 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx) 1472 { 1473 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1474 uint64_t count; 1475 int err; 1476 1477 ASSERT(ds->ds_phys->ds_num_children >= 2); 1478 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx); 1479 /* 1480 * The err should not be ENOENT, but a bug in a previous version 1481 * of the code could cause upgrade_clones_cb() to not set 1482 * ds_next_snap_obj when it should, leading to a missing entry. 1483 * If we knew that the pool was created after 1484 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 1485 * ENOENT. However, at least we can check that we don't have 1486 * too many entries in the next_clones_obj even after failing to 1487 * remove this one. 1488 */ 1489 if (err != ENOENT) { 1490 VERIFY3U(err, ==, 0); 1491 } 1492 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj, 1493 &count)); 1494 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2); 1495 } 1496 1497 static void 1498 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx) 1499 { 1500 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1501 zap_cursor_t zc; 1502 zap_attribute_t za; 1503 1504 /* 1505 * If it is the old version, dd_clones doesn't exist so we can't 1506 * find the clones, but deadlist_remove_key() is a no-op so it 1507 * doesn't matter. 1508 */ 1509 if (ds->ds_dir->dd_phys->dd_clones == 0) 1510 return; 1511 1512 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones); 1513 zap_cursor_retrieve(&zc, &za) == 0; 1514 zap_cursor_advance(&zc)) { 1515 dsl_dataset_t *clone; 1516 1517 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1518 za.za_first_integer, FTAG, &clone)); 1519 if (clone->ds_dir->dd_origin_txg > mintxg) { 1520 dsl_deadlist_remove_key(&clone->ds_deadlist, 1521 mintxg, tx); 1522 dsl_dataset_remove_clones_key(clone, mintxg, tx); 1523 } 1524 dsl_dataset_rele(clone, FTAG); 1525 } 1526 zap_cursor_fini(&zc); 1527 } 1528 1529 struct process_old_arg { 1530 dsl_dataset_t *ds; 1531 dsl_dataset_t *ds_prev; 1532 boolean_t after_branch_point; 1533 zio_t *pio; 1534 uint64_t used, comp, uncomp; 1535 }; 1536 1537 static int 1538 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 1539 { 1540 struct process_old_arg *poa = arg; 1541 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool; 1542 1543 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) { 1544 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx); 1545 if (poa->ds_prev && !poa->after_branch_point && 1546 bp->blk_birth > 1547 poa->ds_prev->ds_phys->ds_prev_snap_txg) { 1548 poa->ds_prev->ds_phys->ds_unique_bytes += 1549 bp_get_dsize_sync(dp->dp_spa, bp); 1550 } 1551 } else { 1552 poa->used += bp_get_dsize_sync(dp->dp_spa, bp); 1553 poa->comp += BP_GET_PSIZE(bp); 1554 poa->uncomp += BP_GET_UCSIZE(bp); 1555 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp); 1556 } 1557 return (0); 1558 } 1559 1560 static void 1561 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev, 1562 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx) 1563 { 1564 struct process_old_arg poa = { 0 }; 1565 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1566 objset_t *mos = dp->dp_meta_objset; 1567 1568 ASSERT(ds->ds_deadlist.dl_oldfmt); 1569 ASSERT(ds_next->ds_deadlist.dl_oldfmt); 1570 1571 poa.ds = ds; 1572 poa.ds_prev = ds_prev; 1573 poa.after_branch_point = after_branch_point; 1574 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1575 VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj, 1576 process_old_cb, &poa, tx)); 1577 VERIFY3U(zio_wait(poa.pio), ==, 0); 1578 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes); 1579 1580 /* change snapused */ 1581 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, 1582 -poa.used, -poa.comp, -poa.uncomp, tx); 1583 1584 /* swap next's deadlist to our deadlist */ 1585 dsl_deadlist_close(&ds->ds_deadlist); 1586 dsl_deadlist_close(&ds_next->ds_deadlist); 1587 SWITCH64(ds_next->ds_phys->ds_deadlist_obj, 1588 ds->ds_phys->ds_deadlist_obj); 1589 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); 1590 dsl_deadlist_open(&ds_next->ds_deadlist, mos, 1591 ds_next->ds_phys->ds_deadlist_obj); 1592 } 1593 1594 static int 1595 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx) 1596 { 1597 int err; 1598 struct killarg ka; 1599 1600 /* 1601 * Free everything that we point to (that's born after 1602 * the previous snapshot, if we are a clone) 1603 * 1604 * NB: this should be very quick, because we already 1605 * freed all the objects in open context. 1606 */ 1607 ka.ds = ds; 1608 ka.tx = tx; 1609 err = traverse_dataset(ds, 1610 ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST, 1611 kill_blkptr, &ka); 1612 ASSERT3U(err, ==, 0); 1613 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0); 1614 1615 return (err); 1616 } 1617 1618 void 1619 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) 1620 { 1621 struct dsl_ds_destroyarg *dsda = arg1; 1622 dsl_dataset_t *ds = dsda->ds; 1623 int err; 1624 int after_branch_point = FALSE; 1625 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1626 objset_t *mos = dp->dp_meta_objset; 1627 dsl_dataset_t *ds_prev = NULL; 1628 boolean_t wont_destroy; 1629 uint64_t obj; 1630 1631 wont_destroy = (dsda->defer && 1632 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1)); 1633 1634 ASSERT(ds->ds_owner || wont_destroy); 1635 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1); 1636 ASSERT(ds->ds_prev == NULL || 1637 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object); 1638 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); 1639 1640 if (wont_destroy) { 1641 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 1642 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1643 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY; 1644 spa_history_log_internal_ds(ds, "defer_destroy", tx, ""); 1645 return; 1646 } 1647 1648 /* We need to log before removing it from the namespace. */ 1649 spa_history_log_internal_ds(ds, "destroy", tx, ""); 1650 1651 /* signal any waiters that this dataset is going away */ 1652 mutex_enter(&ds->ds_lock); 1653 ds->ds_owner = dsl_reaper; 1654 cv_broadcast(&ds->ds_exclusive_cv); 1655 mutex_exit(&ds->ds_lock); 1656 1657 /* Remove our reservation */ 1658 if (ds->ds_reserved != 0) { 1659 dsl_prop_setarg_t psa; 1660 uint64_t value = 0; 1661 1662 dsl_prop_setarg_init_uint64(&psa, "refreservation", 1663 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED), 1664 &value); 1665 psa.psa_effective_value = 0; /* predict default value */ 1666 1667 dsl_dataset_set_reservation_sync(ds, &psa, tx); 1668 ASSERT3U(ds->ds_reserved, ==, 0); 1669 } 1670 1671 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); 1672 1673 dsl_scan_ds_destroyed(ds, tx); 1674 1675 obj = ds->ds_object; 1676 1677 if (ds->ds_phys->ds_prev_snap_obj != 0) { 1678 if (ds->ds_prev) { 1679 ds_prev = ds->ds_prev; 1680 } else { 1681 VERIFY(0 == dsl_dataset_hold_obj(dp, 1682 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev)); 1683 } 1684 after_branch_point = 1685 (ds_prev->ds_phys->ds_next_snap_obj != obj); 1686 1687 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); 1688 if (after_branch_point && 1689 ds_prev->ds_phys->ds_next_clones_obj != 0) { 1690 remove_from_next_clones(ds_prev, obj, tx); 1691 if (ds->ds_phys->ds_next_snap_obj != 0) { 1692 VERIFY(0 == zap_add_int(mos, 1693 ds_prev->ds_phys->ds_next_clones_obj, 1694 ds->ds_phys->ds_next_snap_obj, tx)); 1695 } 1696 } 1697 if (after_branch_point && 1698 ds->ds_phys->ds_next_snap_obj == 0) { 1699 /* This clone is toast. */ 1700 ASSERT(ds_prev->ds_phys->ds_num_children > 1); 1701 ds_prev->ds_phys->ds_num_children--; 1702 1703 /* 1704 * If the clone's origin has no other clones, no 1705 * user holds, and has been marked for deferred 1706 * deletion, then we should have done the necessary 1707 * destroy setup for it. 1708 */ 1709 if (ds_prev->ds_phys->ds_num_children == 1 && 1710 ds_prev->ds_userrefs == 0 && 1711 DS_IS_DEFER_DESTROY(ds_prev)) { 1712 ASSERT3P(dsda->rm_origin, !=, NULL); 1713 } else { 1714 ASSERT3P(dsda->rm_origin, ==, NULL); 1715 } 1716 } else if (!after_branch_point) { 1717 ds_prev->ds_phys->ds_next_snap_obj = 1718 ds->ds_phys->ds_next_snap_obj; 1719 } 1720 } 1721 1722 if (dsl_dataset_is_snapshot(ds)) { 1723 dsl_dataset_t *ds_next; 1724 uint64_t old_unique; 1725 uint64_t used = 0, comp = 0, uncomp = 0; 1726 1727 VERIFY(0 == dsl_dataset_hold_obj(dp, 1728 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next)); 1729 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); 1730 1731 old_unique = ds_next->ds_phys->ds_unique_bytes; 1732 1733 dmu_buf_will_dirty(ds_next->ds_dbuf, tx); 1734 ds_next->ds_phys->ds_prev_snap_obj = 1735 ds->ds_phys->ds_prev_snap_obj; 1736 ds_next->ds_phys->ds_prev_snap_txg = 1737 ds->ds_phys->ds_prev_snap_txg; 1738 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1739 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); 1740 1741 1742 if (ds_next->ds_deadlist.dl_oldfmt) { 1743 process_old_deadlist(ds, ds_prev, ds_next, 1744 after_branch_point, tx); 1745 } else { 1746 /* Adjust prev's unique space. */ 1747 if (ds_prev && !after_branch_point) { 1748 dsl_deadlist_space_range(&ds_next->ds_deadlist, 1749 ds_prev->ds_phys->ds_prev_snap_txg, 1750 ds->ds_phys->ds_prev_snap_txg, 1751 &used, &comp, &uncomp); 1752 ds_prev->ds_phys->ds_unique_bytes += used; 1753 } 1754 1755 /* Adjust snapused. */ 1756 dsl_deadlist_space_range(&ds_next->ds_deadlist, 1757 ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, 1758 &used, &comp, &uncomp); 1759 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, 1760 -used, -comp, -uncomp, tx); 1761 1762 /* Move blocks to be freed to pool's free list. */ 1763 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist, 1764 &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg, 1765 tx); 1766 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, 1767 DD_USED_HEAD, used, comp, uncomp, tx); 1768 1769 /* Merge our deadlist into next's and free it. */ 1770 dsl_deadlist_merge(&ds_next->ds_deadlist, 1771 ds->ds_phys->ds_deadlist_obj, tx); 1772 } 1773 dsl_deadlist_close(&ds->ds_deadlist); 1774 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx); 1775 1776 /* Collapse range in clone heads */ 1777 dsl_dataset_remove_clones_key(ds, 1778 ds->ds_phys->ds_creation_txg, tx); 1779 1780 if (dsl_dataset_is_snapshot(ds_next)) { 1781 dsl_dataset_t *ds_nextnext; 1782 1783 /* 1784 * Update next's unique to include blocks which 1785 * were previously shared by only this snapshot 1786 * and it. Those blocks will be born after the 1787 * prev snap and before this snap, and will have 1788 * died after the next snap and before the one 1789 * after that (ie. be on the snap after next's 1790 * deadlist). 1791 */ 1792 VERIFY(0 == dsl_dataset_hold_obj(dp, 1793 ds_next->ds_phys->ds_next_snap_obj, 1794 FTAG, &ds_nextnext)); 1795 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist, 1796 ds->ds_phys->ds_prev_snap_txg, 1797 ds->ds_phys->ds_creation_txg, 1798 &used, &comp, &uncomp); 1799 ds_next->ds_phys->ds_unique_bytes += used; 1800 dsl_dataset_rele(ds_nextnext, FTAG); 1801 ASSERT3P(ds_next->ds_prev, ==, NULL); 1802 1803 /* Collapse range in this head. */ 1804 dsl_dataset_t *hds; 1805 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, 1806 ds->ds_dir->dd_phys->dd_head_dataset_obj, 1807 FTAG, &hds)); 1808 dsl_deadlist_remove_key(&hds->ds_deadlist, 1809 ds->ds_phys->ds_creation_txg, tx); 1810 dsl_dataset_rele(hds, FTAG); 1811 1812 } else { 1813 ASSERT3P(ds_next->ds_prev, ==, ds); 1814 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next); 1815 ds_next->ds_prev = NULL; 1816 if (ds_prev) { 1817 VERIFY(0 == dsl_dataset_get_ref(dp, 1818 ds->ds_phys->ds_prev_snap_obj, 1819 ds_next, &ds_next->ds_prev)); 1820 } 1821 1822 dsl_dataset_recalc_head_uniq(ds_next); 1823 1824 /* 1825 * Reduce the amount of our unconsmed refreservation 1826 * being charged to our parent by the amount of 1827 * new unique data we have gained. 1828 */ 1829 if (old_unique < ds_next->ds_reserved) { 1830 int64_t mrsdelta; 1831 uint64_t new_unique = 1832 ds_next->ds_phys->ds_unique_bytes; 1833 1834 ASSERT(old_unique <= new_unique); 1835 mrsdelta = MIN(new_unique - old_unique, 1836 ds_next->ds_reserved - old_unique); 1837 dsl_dir_diduse_space(ds->ds_dir, 1838 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx); 1839 } 1840 } 1841 dsl_dataset_rele(ds_next, FTAG); 1842 } else { 1843 zfeature_info_t *async_destroy = 1844 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY]; 1845 objset_t *os; 1846 1847 /* 1848 * There's no next snapshot, so this is a head dataset. 1849 * Destroy the deadlist. Unless it's a clone, the 1850 * deadlist should be empty. (If it's a clone, it's 1851 * safe to ignore the deadlist contents.) 1852 */ 1853 dsl_deadlist_close(&ds->ds_deadlist); 1854 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx); 1855 ds->ds_phys->ds_deadlist_obj = 0; 1856 1857 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os)); 1858 1859 if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) { 1860 err = old_synchronous_dataset_destroy(ds, tx); 1861 } else { 1862 /* 1863 * Move the bptree into the pool's list of trees to 1864 * clean up and update space accounting information. 1865 */ 1866 uint64_t used, comp, uncomp; 1867 1868 zil_destroy_sync(dmu_objset_zil(os), tx); 1869 1870 if (!spa_feature_is_active(dp->dp_spa, async_destroy)) { 1871 spa_feature_incr(dp->dp_spa, async_destroy, tx); 1872 dp->dp_bptree_obj = bptree_alloc(mos, tx); 1873 VERIFY(zap_add(mos, 1874 DMU_POOL_DIRECTORY_OBJECT, 1875 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, 1876 &dp->dp_bptree_obj, tx) == 0); 1877 } 1878 1879 used = ds->ds_dir->dd_phys->dd_used_bytes; 1880 comp = ds->ds_dir->dd_phys->dd_compressed_bytes; 1881 uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes; 1882 1883 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || 1884 ds->ds_phys->ds_unique_bytes == used); 1885 1886 bptree_add(mos, dp->dp_bptree_obj, 1887 &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg, 1888 used, comp, uncomp, tx); 1889 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 1890 -used, -comp, -uncomp, tx); 1891 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD, 1892 used, comp, uncomp, tx); 1893 } 1894 1895 if (ds->ds_prev != NULL) { 1896 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 1897 VERIFY3U(0, ==, zap_remove_int(mos, 1898 ds->ds_prev->ds_dir->dd_phys->dd_clones, 1899 ds->ds_object, tx)); 1900 } 1901 dsl_dataset_rele(ds->ds_prev, ds); 1902 ds->ds_prev = ds_prev = NULL; 1903 } 1904 } 1905 1906 /* 1907 * This must be done after the dsl_traverse(), because it will 1908 * re-open the objset. 1909 */ 1910 if (ds->ds_objset) { 1911 dmu_objset_evict(ds->ds_objset); 1912 ds->ds_objset = NULL; 1913 } 1914 1915 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) { 1916 /* Erase the link in the dir */ 1917 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 1918 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0; 1919 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0); 1920 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx); 1921 ASSERT(err == 0); 1922 } else { 1923 /* remove from snapshot namespace */ 1924 dsl_dataset_t *ds_head; 1925 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0); 1926 VERIFY(0 == dsl_dataset_hold_obj(dp, 1927 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head)); 1928 VERIFY(0 == dsl_dataset_get_snapname(ds)); 1929 #ifdef ZFS_DEBUG 1930 { 1931 uint64_t val; 1932 1933 err = dsl_dataset_snap_lookup(ds_head, 1934 ds->ds_snapname, &val); 1935 ASSERT3U(err, ==, 0); 1936 ASSERT3U(val, ==, obj); 1937 } 1938 #endif 1939 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx); 1940 ASSERT(err == 0); 1941 dsl_dataset_rele(ds_head, FTAG); 1942 } 1943 1944 if (ds_prev && ds->ds_prev != ds_prev) 1945 dsl_dataset_rele(ds_prev, FTAG); 1946 1947 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx); 1948 1949 if (ds->ds_phys->ds_next_clones_obj != 0) { 1950 uint64_t count; 1951 ASSERT(0 == zap_count(mos, 1952 ds->ds_phys->ds_next_clones_obj, &count) && count == 0); 1953 VERIFY(0 == dmu_object_free(mos, 1954 ds->ds_phys->ds_next_clones_obj, tx)); 1955 } 1956 if (ds->ds_phys->ds_props_obj != 0) 1957 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx)); 1958 if (ds->ds_phys->ds_userrefs_obj != 0) 1959 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx)); 1960 dsl_dir_close(ds->ds_dir, ds); 1961 ds->ds_dir = NULL; 1962 dsl_dataset_drain_refs(ds, tag); 1963 VERIFY(0 == dmu_object_free(mos, obj, tx)); 1964 1965 if (dsda->rm_origin) { 1966 /* 1967 * Remove the origin of the clone we just destroyed. 1968 */ 1969 struct dsl_ds_destroyarg ndsda = {0}; 1970 1971 ndsda.ds = dsda->rm_origin; 1972 dsl_dataset_destroy_sync(&ndsda, tag, tx); 1973 } 1974 } 1975 1976 static int 1977 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1978 { 1979 uint64_t asize; 1980 1981 if (!dmu_tx_is_syncing(tx)) 1982 return (0); 1983 1984 /* 1985 * If there's an fs-only reservation, any blocks that might become 1986 * owned by the snapshot dataset must be accommodated by space 1987 * outside of the reservation. 1988 */ 1989 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1990 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 1991 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1992 return (ENOSPC); 1993 1994 /* 1995 * Propagate any reserved space for this snapshot to other 1996 * snapshot checks in this sync group. 1997 */ 1998 if (asize > 0) 1999 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 2000 2001 return (0); 2002 } 2003 2004 int 2005 dsl_dataset_snapshot_check(dsl_dataset_t *ds, const char *snapname, 2006 dmu_tx_t *tx) 2007 { 2008 int err; 2009 uint64_t value; 2010 2011 /* 2012 * We don't allow multiple snapshots of the same txg. If there 2013 * is already one, try again. 2014 */ 2015 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) 2016 return (EAGAIN); 2017 2018 /* 2019 * Check for conflicting snapshot name. 2020 */ 2021 err = dsl_dataset_snap_lookup(ds, snapname, &value); 2022 if (err == 0) 2023 return (EEXIST); 2024 if (err != ENOENT) 2025 return (err); 2026 2027 /* 2028 * Check that the dataset's name is not too long. Name consists 2029 * of the dataset's length + 1 for the @-sign + snapshot name's length 2030 */ 2031 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN) 2032 return (ENAMETOOLONG); 2033 2034 err = dsl_dataset_snapshot_reserve_space(ds, tx); 2035 if (err) 2036 return (err); 2037 2038 ds->ds_trysnap_txg = tx->tx_txg; 2039 return (0); 2040 } 2041 2042 void 2043 dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *snapname, 2044 dmu_tx_t *tx) 2045 { 2046 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2047 dmu_buf_t *dbuf; 2048 dsl_dataset_phys_t *dsphys; 2049 uint64_t dsobj, crtxg; 2050 objset_t *mos = dp->dp_meta_objset; 2051 int err; 2052 2053 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); 2054 2055 /* 2056 * The origin's ds_creation_txg has to be < TXG_INITIAL 2057 */ 2058 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 2059 crtxg = 1; 2060 else 2061 crtxg = tx->tx_txg; 2062 2063 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 2064 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 2065 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 2066 dmu_buf_will_dirty(dbuf, tx); 2067 dsphys = dbuf->db_data; 2068 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 2069 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 2070 dsphys->ds_fsid_guid = unique_create(); 2071 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 2072 sizeof (dsphys->ds_guid)); 2073 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; 2074 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; 2075 dsphys->ds_next_snap_obj = ds->ds_object; 2076 dsphys->ds_num_children = 1; 2077 dsphys->ds_creation_time = gethrestime_sec(); 2078 dsphys->ds_creation_txg = crtxg; 2079 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; 2080 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes; 2081 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; 2082 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; 2083 dsphys->ds_flags = ds->ds_phys->ds_flags; 2084 dsphys->ds_bp = ds->ds_phys->ds_bp; 2085 dmu_buf_rele(dbuf, FTAG); 2086 2087 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); 2088 if (ds->ds_prev) { 2089 uint64_t next_clones_obj = 2090 ds->ds_prev->ds_phys->ds_next_clones_obj; 2091 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == 2092 ds->ds_object || 2093 ds->ds_prev->ds_phys->ds_num_children > 1); 2094 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { 2095 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 2096 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 2097 ds->ds_prev->ds_phys->ds_creation_txg); 2098 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; 2099 } else if (next_clones_obj != 0) { 2100 remove_from_next_clones(ds->ds_prev, 2101 dsphys->ds_next_snap_obj, tx); 2102 VERIFY3U(0, ==, zap_add_int(mos, 2103 next_clones_obj, dsobj, tx)); 2104 } 2105 } 2106 2107 /* 2108 * If we have a reference-reservation on this dataset, we will 2109 * need to increase the amount of refreservation being charged 2110 * since our unique space is going to zero. 2111 */ 2112 if (ds->ds_reserved) { 2113 int64_t delta; 2114 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 2115 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 2116 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 2117 delta, 0, 0, tx); 2118 } 2119 2120 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2121 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu", 2122 ds->ds_dir->dd_myname, snapname, dsobj, 2123 ds->ds_phys->ds_prev_snap_txg); 2124 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist, 2125 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx); 2126 dsl_deadlist_close(&ds->ds_deadlist); 2127 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); 2128 dsl_deadlist_add_key(&ds->ds_deadlist, 2129 ds->ds_phys->ds_prev_snap_txg, tx); 2130 2131 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg); 2132 ds->ds_phys->ds_prev_snap_obj = dsobj; 2133 ds->ds_phys->ds_prev_snap_txg = crtxg; 2134 ds->ds_phys->ds_unique_bytes = 0; 2135 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 2136 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 2137 2138 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, 2139 snapname, 8, 1, &dsobj, tx); 2140 ASSERT(err == 0); 2141 2142 if (ds->ds_prev) 2143 dsl_dataset_drop_ref(ds->ds_prev, ds); 2144 VERIFY(0 == dsl_dataset_get_ref(dp, 2145 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 2146 2147 dsl_scan_ds_snapshotted(ds, tx); 2148 2149 dsl_dir_snap_cmtime_update(ds->ds_dir); 2150 2151 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 2152 } 2153 2154 void 2155 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 2156 { 2157 ASSERT(dmu_tx_is_syncing(tx)); 2158 ASSERT(ds->ds_objset != NULL); 2159 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); 2160 2161 /* 2162 * in case we had to change ds_fsid_guid when we opened it, 2163 * sync it out now. 2164 */ 2165 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2166 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; 2167 2168 dmu_objset_sync(ds->ds_objset, zio, tx); 2169 } 2170 2171 static void 2172 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 2173 { 2174 uint64_t count = 0; 2175 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 2176 zap_cursor_t zc; 2177 zap_attribute_t za; 2178 nvlist_t *propval; 2179 nvlist_t *val; 2180 2181 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); 2182 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2183 VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2184 2185 /* 2186 * There may me missing entries in ds_next_clones_obj 2187 * due to a bug in a previous version of the code. 2188 * Only trust it if it has the right number of entries. 2189 */ 2190 if (ds->ds_phys->ds_next_clones_obj != 0) { 2191 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj, 2192 &count)); 2193 } 2194 if (count != ds->ds_phys->ds_num_children - 1) { 2195 goto fail; 2196 } 2197 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj); 2198 zap_cursor_retrieve(&zc, &za) == 0; 2199 zap_cursor_advance(&zc)) { 2200 dsl_dataset_t *clone; 2201 char buf[ZFS_MAXNAMELEN]; 2202 /* 2203 * Even though we hold the dp_config_rwlock, the dataset 2204 * may fail to open, returning ENOENT. If there is a 2205 * thread concurrently attempting to destroy this 2206 * dataset, it will have the ds_rwlock held for 2207 * RW_WRITER. Our call to dsl_dataset_hold_obj() -> 2208 * dsl_dataset_hold_ref() will fail its 2209 * rw_tryenter(&ds->ds_rwlock, RW_READER), drop the 2210 * dp_config_rwlock, and wait for the destroy progress 2211 * and signal ds_exclusive_cv. If the destroy was 2212 * successful, we will see that 2213 * DSL_DATASET_IS_DESTROYED(), and return ENOENT. 2214 */ 2215 if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 2216 za.za_first_integer, FTAG, &clone) != 0) 2217 continue; 2218 dsl_dir_name(clone->ds_dir, buf); 2219 VERIFY(nvlist_add_boolean(val, buf) == 0); 2220 dsl_dataset_rele(clone, FTAG); 2221 } 2222 zap_cursor_fini(&zc); 2223 VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0); 2224 VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), 2225 propval) == 0); 2226 fail: 2227 nvlist_free(val); 2228 nvlist_free(propval); 2229 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); 2230 } 2231 2232 void 2233 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 2234 { 2235 uint64_t refd, avail, uobjs, aobjs, ratio; 2236 2237 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 : 2238 (ds->ds_phys->ds_uncompressed_bytes * 100 / 2239 ds->ds_phys->ds_compressed_bytes); 2240 2241 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); 2242 2243 if (dsl_dataset_is_snapshot(ds)) { 2244 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); 2245 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 2246 ds->ds_phys->ds_unique_bytes); 2247 get_clones_stat(ds, nv); 2248 } else { 2249 dsl_dir_stats(ds->ds_dir, nv); 2250 } 2251 2252 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 2253 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 2254 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 2255 2256 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 2257 ds->ds_phys->ds_creation_time); 2258 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 2259 ds->ds_phys->ds_creation_txg); 2260 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 2261 ds->ds_quota); 2262 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 2263 ds->ds_reserved); 2264 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 2265 ds->ds_phys->ds_guid); 2266 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 2267 ds->ds_phys->ds_unique_bytes); 2268 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 2269 ds->ds_object); 2270 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 2271 ds->ds_userrefs); 2272 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 2273 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 2274 2275 if (ds->ds_phys->ds_prev_snap_obj != 0) { 2276 uint64_t written, comp, uncomp; 2277 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2278 dsl_dataset_t *prev; 2279 2280 rw_enter(&dp->dp_config_rwlock, RW_READER); 2281 int err = dsl_dataset_hold_obj(dp, 2282 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev); 2283 rw_exit(&dp->dp_config_rwlock); 2284 if (err == 0) { 2285 err = dsl_dataset_space_written(prev, ds, &written, 2286 &comp, &uncomp); 2287 dsl_dataset_rele(prev, FTAG); 2288 if (err == 0) { 2289 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 2290 written); 2291 } 2292 } 2293 } 2294 2295 } 2296 2297 void 2298 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 2299 { 2300 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; 2301 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; 2302 stat->dds_guid = ds->ds_phys->ds_guid; 2303 stat->dds_origin[0] = '\0'; 2304 if (dsl_dataset_is_snapshot(ds)) { 2305 stat->dds_is_snapshot = B_TRUE; 2306 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; 2307 } else { 2308 stat->dds_is_snapshot = B_FALSE; 2309 stat->dds_num_clones = 0; 2310 2311 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); 2312 if (dsl_dir_is_clone(ds->ds_dir)) { 2313 dsl_dataset_t *ods; 2314 2315 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool, 2316 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods)); 2317 dsl_dataset_name(ods, stat->dds_origin); 2318 dsl_dataset_drop_ref(ods, FTAG); 2319 } 2320 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); 2321 } 2322 } 2323 2324 uint64_t 2325 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 2326 { 2327 return (ds->ds_fsid_guid); 2328 } 2329 2330 void 2331 dsl_dataset_space(dsl_dataset_t *ds, 2332 uint64_t *refdbytesp, uint64_t *availbytesp, 2333 uint64_t *usedobjsp, uint64_t *availobjsp) 2334 { 2335 *refdbytesp = ds->ds_phys->ds_referenced_bytes; 2336 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 2337 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) 2338 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes; 2339 if (ds->ds_quota != 0) { 2340 /* 2341 * Adjust available bytes according to refquota 2342 */ 2343 if (*refdbytesp < ds->ds_quota) 2344 *availbytesp = MIN(*availbytesp, 2345 ds->ds_quota - *refdbytesp); 2346 else 2347 *availbytesp = 0; 2348 } 2349 *usedobjsp = ds->ds_phys->ds_bp.blk_fill; 2350 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 2351 } 2352 2353 boolean_t 2354 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds) 2355 { 2356 dsl_pool_t *dp = ds->ds_dir->dd_pool; 2357 2358 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 2359 dsl_pool_sync_context(dp)); 2360 if (ds->ds_prev == NULL) 2361 return (B_FALSE); 2362 if (ds->ds_phys->ds_bp.blk_birth > 2363 ds->ds_prev->ds_phys->ds_creation_txg) { 2364 objset_t *os, *os_prev; 2365 /* 2366 * It may be that only the ZIL differs, because it was 2367 * reset in the head. Don't count that as being 2368 * modified. 2369 */ 2370 if (dmu_objset_from_ds(ds, &os) != 0) 2371 return (B_TRUE); 2372 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0) 2373 return (B_TRUE); 2374 return (bcmp(&os->os_phys->os_meta_dnode, 2375 &os_prev->os_phys->os_meta_dnode, 2376 sizeof (os->os_phys->os_meta_dnode)) != 0); 2377 } 2378 return (B_FALSE); 2379 } 2380 2381 /* ARGSUSED */ 2382 static int 2383 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) 2384 { 2385 dsl_dataset_t *ds = arg1; 2386 char *newsnapname = arg2; 2387 dsl_dir_t *dd = ds->ds_dir; 2388 dsl_dataset_t *hds; 2389 uint64_t val; 2390 int err; 2391 2392 err = dsl_dataset_hold_obj(dd->dd_pool, 2393 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds); 2394 if (err) 2395 return (err); 2396 2397 /* new name better not be in use */ 2398 err = dsl_dataset_snap_lookup(hds, newsnapname, &val); 2399 dsl_dataset_rele(hds, FTAG); 2400 2401 if (err == 0) 2402 err = EEXIST; 2403 else if (err == ENOENT) 2404 err = 0; 2405 2406 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2407 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN) 2408 err = ENAMETOOLONG; 2409 2410 return (err); 2411 } 2412 2413 static void 2414 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx) 2415 { 2416 dsl_dataset_t *ds = arg1; 2417 const char *newsnapname = arg2; 2418 dsl_dir_t *dd = ds->ds_dir; 2419 objset_t *mos = dd->dd_pool->dp_meta_objset; 2420 dsl_dataset_t *hds; 2421 int err; 2422 2423 ASSERT(ds->ds_phys->ds_next_snap_obj != 0); 2424 2425 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, 2426 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds)); 2427 2428 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2429 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx); 2430 ASSERT3U(err, ==, 0); 2431 mutex_enter(&ds->ds_lock); 2432 (void) strcpy(ds->ds_snapname, newsnapname); 2433 mutex_exit(&ds->ds_lock); 2434 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj, 2435 ds->ds_snapname, 8, 1, &ds->ds_object, tx); 2436 ASSERT3U(err, ==, 0); 2437 2438 spa_history_log_internal_ds(ds, "rename", tx, 2439 "-> @%s", newsnapname); 2440 dsl_dataset_rele(hds, FTAG); 2441 } 2442 2443 struct renamesnaparg { 2444 dsl_sync_task_group_t *dstg; 2445 char failed[MAXPATHLEN]; 2446 char *oldsnap; 2447 char *newsnap; 2448 }; 2449 2450 static int 2451 dsl_snapshot_rename_one(const char *name, void *arg) 2452 { 2453 struct renamesnaparg *ra = arg; 2454 dsl_dataset_t *ds = NULL; 2455 char *snapname; 2456 int err; 2457 2458 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap); 2459 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed)); 2460 2461 /* 2462 * For recursive snapshot renames the parent won't be changing 2463 * so we just pass name for both the to/from argument. 2464 */ 2465 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED()); 2466 if (err != 0) { 2467 strfree(snapname); 2468 return (err == ENOENT ? 0 : err); 2469 } 2470 2471 #ifdef _KERNEL 2472 /* 2473 * For all filesystems undergoing rename, we'll need to unmount it. 2474 */ 2475 (void) zfs_unmount_snap(snapname, NULL); 2476 #endif 2477 err = dsl_dataset_hold(snapname, ra->dstg, &ds); 2478 strfree(snapname); 2479 if (err != 0) 2480 return (err == ENOENT ? 0 : err); 2481 2482 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check, 2483 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0); 2484 2485 return (0); 2486 } 2487 2488 static int 2489 dsl_recursive_rename(char *oldname, const char *newname) 2490 { 2491 int err; 2492 struct renamesnaparg *ra; 2493 dsl_sync_task_t *dst; 2494 spa_t *spa; 2495 char *cp, *fsname = spa_strdup(oldname); 2496 int len = strlen(oldname) + 1; 2497 2498 /* truncate the snapshot name to get the fsname */ 2499 cp = strchr(fsname, '@'); 2500 *cp = '\0'; 2501 2502 err = spa_open(fsname, &spa, FTAG); 2503 if (err) { 2504 kmem_free(fsname, len); 2505 return (err); 2506 } 2507 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP); 2508 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 2509 2510 ra->oldsnap = strchr(oldname, '@') + 1; 2511 ra->newsnap = strchr(newname, '@') + 1; 2512 *ra->failed = '\0'; 2513 2514 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra, 2515 DS_FIND_CHILDREN); 2516 kmem_free(fsname, len); 2517 2518 if (err == 0) { 2519 err = dsl_sync_task_group_wait(ra->dstg); 2520 } 2521 2522 for (dst = list_head(&ra->dstg->dstg_tasks); dst; 2523 dst = list_next(&ra->dstg->dstg_tasks, dst)) { 2524 dsl_dataset_t *ds = dst->dst_arg1; 2525 if (dst->dst_err) { 2526 dsl_dir_name(ds->ds_dir, ra->failed); 2527 (void) strlcat(ra->failed, "@", sizeof (ra->failed)); 2528 (void) strlcat(ra->failed, ra->newsnap, 2529 sizeof (ra->failed)); 2530 } 2531 dsl_dataset_rele(ds, ra->dstg); 2532 } 2533 2534 if (err) 2535 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed)); 2536 2537 dsl_sync_task_group_destroy(ra->dstg); 2538 kmem_free(ra, sizeof (struct renamesnaparg)); 2539 spa_close(spa, FTAG); 2540 return (err); 2541 } 2542 2543 static int 2544 dsl_valid_rename(const char *oldname, void *arg) 2545 { 2546 int delta = *(int *)arg; 2547 2548 if (strlen(oldname) + delta >= MAXNAMELEN) 2549 return (ENAMETOOLONG); 2550 2551 return (0); 2552 } 2553 2554 #pragma weak dmu_objset_rename = dsl_dataset_rename 2555 int 2556 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive) 2557 { 2558 dsl_dir_t *dd; 2559 dsl_dataset_t *ds; 2560 const char *tail; 2561 int err; 2562 2563 err = dsl_dir_open(oldname, FTAG, &dd, &tail); 2564 if (err) 2565 return (err); 2566 2567 if (tail == NULL) { 2568 int delta = strlen(newname) - strlen(oldname); 2569 2570 /* if we're growing, validate child name lengths */ 2571 if (delta > 0) 2572 err = dmu_objset_find(oldname, dsl_valid_rename, 2573 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 2574 2575 if (err == 0) 2576 err = dsl_dir_rename(dd, newname); 2577 dsl_dir_close(dd, FTAG); 2578 return (err); 2579 } 2580 2581 if (tail[0] != '@') { 2582 /* the name ended in a nonexistent component */ 2583 dsl_dir_close(dd, FTAG); 2584 return (ENOENT); 2585 } 2586 2587 dsl_dir_close(dd, FTAG); 2588 2589 /* new name must be snapshot in same filesystem */ 2590 tail = strchr(newname, '@'); 2591 if (tail == NULL) 2592 return (EINVAL); 2593 tail++; 2594 if (strncmp(oldname, newname, tail - newname) != 0) 2595 return (EXDEV); 2596 2597 if (recursive) { 2598 err = dsl_recursive_rename(oldname, newname); 2599 } else { 2600 err = dsl_dataset_hold(oldname, FTAG, &ds); 2601 if (err) 2602 return (err); 2603 2604 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 2605 dsl_dataset_snapshot_rename_check, 2606 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1); 2607 2608 dsl_dataset_rele(ds, FTAG); 2609 } 2610 2611 return (err); 2612 } 2613 2614 struct promotenode { 2615 list_node_t link; 2616 dsl_dataset_t *ds; 2617 }; 2618 2619 struct promotearg { 2620 list_t shared_snaps, origin_snaps, clone_snaps; 2621 dsl_dataset_t *origin_origin; 2622 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2623 char *err_ds; 2624 }; 2625 2626 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2627 static boolean_t snaplist_unstable(list_t *l); 2628 2629 static int 2630 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) 2631 { 2632 dsl_dataset_t *hds = arg1; 2633 struct promotearg *pa = arg2; 2634 struct promotenode *snap = list_head(&pa->shared_snaps); 2635 dsl_dataset_t *origin_ds = snap->ds; 2636 int err; 2637 uint64_t unused; 2638 2639 /* Check that it is a real clone */ 2640 if (!dsl_dir_is_clone(hds->ds_dir)) 2641 return (EINVAL); 2642 2643 /* Since this is so expensive, don't do the preliminary check */ 2644 if (!dmu_tx_is_syncing(tx)) 2645 return (0); 2646 2647 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) 2648 return (EXDEV); 2649 2650 /* compute origin's new unique space */ 2651 snap = list_tail(&pa->clone_snaps); 2652 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2653 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2654 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2655 &pa->unique, &unused, &unused); 2656 2657 /* 2658 * Walk the snapshots that we are moving 2659 * 2660 * Compute space to transfer. Consider the incremental changes 2661 * to used for each snapshot: 2662 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2663 * So each snapshot gave birth to: 2664 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2665 * So a sequence would look like: 2666 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2667 * Which simplifies to: 2668 * uN + kN + kN-1 + ... + k1 + k0 2669 * Note however, if we stop before we reach the ORIGIN we get: 2670 * uN + kN + kN-1 + ... + kM - uM-1 2671 */ 2672 pa->used = origin_ds->ds_phys->ds_referenced_bytes; 2673 pa->comp = origin_ds->ds_phys->ds_compressed_bytes; 2674 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; 2675 for (snap = list_head(&pa->shared_snaps); snap; 2676 snap = list_next(&pa->shared_snaps, snap)) { 2677 uint64_t val, dlused, dlcomp, dluncomp; 2678 dsl_dataset_t *ds = snap->ds; 2679 2680 /* Check that the snapshot name does not conflict */ 2681 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2682 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2683 if (err == 0) { 2684 err = EEXIST; 2685 goto out; 2686 } 2687 if (err != ENOENT) 2688 goto out; 2689 2690 /* The very first snapshot does not have a deadlist */ 2691 if (ds->ds_phys->ds_prev_snap_obj == 0) 2692 continue; 2693 2694 dsl_deadlist_space(&ds->ds_deadlist, 2695 &dlused, &dlcomp, &dluncomp); 2696 pa->used += dlused; 2697 pa->comp += dlcomp; 2698 pa->uncomp += dluncomp; 2699 } 2700 2701 /* 2702 * If we are a clone of a clone then we never reached ORIGIN, 2703 * so we need to subtract out the clone origin's used space. 2704 */ 2705 if (pa->origin_origin) { 2706 pa->used -= pa->origin_origin->ds_phys->ds_referenced_bytes; 2707 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes; 2708 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes; 2709 } 2710 2711 /* Check that there is enough space here */ 2712 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2713 pa->used); 2714 if (err) 2715 return (err); 2716 2717 /* 2718 * Compute the amounts of space that will be used by snapshots 2719 * after the promotion (for both origin and clone). For each, 2720 * it is the amount of space that will be on all of their 2721 * deadlists (that was not born before their new origin). 2722 */ 2723 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2724 uint64_t space; 2725 2726 /* 2727 * Note, typically this will not be a clone of a clone, 2728 * so dd_origin_txg will be < TXG_INITIAL, so 2729 * these snaplist_space() -> dsl_deadlist_space_range() 2730 * calls will be fast because they do not have to 2731 * iterate over all bps. 2732 */ 2733 snap = list_head(&pa->origin_snaps); 2734 err = snaplist_space(&pa->shared_snaps, 2735 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap); 2736 if (err) 2737 return (err); 2738 2739 err = snaplist_space(&pa->clone_snaps, 2740 snap->ds->ds_dir->dd_origin_txg, &space); 2741 if (err) 2742 return (err); 2743 pa->cloneusedsnap += space; 2744 } 2745 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2746 err = snaplist_space(&pa->origin_snaps, 2747 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap); 2748 if (err) 2749 return (err); 2750 } 2751 2752 return (0); 2753 out: 2754 pa->err_ds = snap->ds->ds_snapname; 2755 return (err); 2756 } 2757 2758 static void 2759 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx) 2760 { 2761 dsl_dataset_t *hds = arg1; 2762 struct promotearg *pa = arg2; 2763 struct promotenode *snap = list_head(&pa->shared_snaps); 2764 dsl_dataset_t *origin_ds = snap->ds; 2765 dsl_dataset_t *origin_head; 2766 dsl_dir_t *dd = hds->ds_dir; 2767 dsl_pool_t *dp = hds->ds_dir->dd_pool; 2768 dsl_dir_t *odd = NULL; 2769 uint64_t oldnext_obj; 2770 int64_t delta; 2771 2772 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)); 2773 2774 snap = list_head(&pa->origin_snaps); 2775 origin_head = snap->ds; 2776 2777 /* 2778 * We need to explicitly open odd, since origin_ds's dd will be 2779 * changing. 2780 */ 2781 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object, 2782 NULL, FTAG, &odd)); 2783 2784 /* change origin's next snap */ 2785 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2786 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj; 2787 snap = list_tail(&pa->clone_snaps); 2788 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2789 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object; 2790 2791 /* change the origin's next clone */ 2792 if (origin_ds->ds_phys->ds_next_clones_obj) { 2793 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx); 2794 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, 2795 origin_ds->ds_phys->ds_next_clones_obj, 2796 oldnext_obj, tx)); 2797 } 2798 2799 /* change origin */ 2800 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2801 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object); 2802 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj; 2803 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2804 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2805 odd->dd_phys->dd_origin_obj = origin_ds->ds_object; 2806 origin_head->ds_dir->dd_origin_txg = 2807 origin_ds->ds_phys->ds_creation_txg; 2808 2809 /* change dd_clone entries */ 2810 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2811 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, 2812 odd->dd_phys->dd_clones, hds->ds_object, tx)); 2813 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, 2814 pa->origin_origin->ds_dir->dd_phys->dd_clones, 2815 hds->ds_object, tx)); 2816 2817 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, 2818 pa->origin_origin->ds_dir->dd_phys->dd_clones, 2819 origin_head->ds_object, tx)); 2820 if (dd->dd_phys->dd_clones == 0) { 2821 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset, 2822 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 2823 } 2824 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, 2825 dd->dd_phys->dd_clones, origin_head->ds_object, tx)); 2826 2827 } 2828 2829 /* move snapshots to this dir */ 2830 for (snap = list_head(&pa->shared_snaps); snap; 2831 snap = list_next(&pa->shared_snaps, snap)) { 2832 dsl_dataset_t *ds = snap->ds; 2833 2834 /* unregister props as dsl_dir is changing */ 2835 if (ds->ds_objset) { 2836 dmu_objset_evict(ds->ds_objset); 2837 ds->ds_objset = NULL; 2838 } 2839 /* move snap name entry */ 2840 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2841 VERIFY(0 == dsl_dataset_snap_remove(origin_head, 2842 ds->ds_snapname, tx)); 2843 VERIFY(0 == zap_add(dp->dp_meta_objset, 2844 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, 2845 8, 1, &ds->ds_object, tx)); 2846 2847 /* change containing dsl_dir */ 2848 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2849 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object); 2850 ds->ds_phys->ds_dir_obj = dd->dd_object; 2851 ASSERT3P(ds->ds_dir, ==, odd); 2852 dsl_dir_close(ds->ds_dir, ds); 2853 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object, 2854 NULL, ds, &ds->ds_dir)); 2855 2856 /* move any clone references */ 2857 if (ds->ds_phys->ds_next_clones_obj && 2858 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2859 zap_cursor_t zc; 2860 zap_attribute_t za; 2861 2862 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2863 ds->ds_phys->ds_next_clones_obj); 2864 zap_cursor_retrieve(&zc, &za) == 0; 2865 zap_cursor_advance(&zc)) { 2866 dsl_dataset_t *cnds; 2867 uint64_t o; 2868 2869 if (za.za_first_integer == oldnext_obj) { 2870 /* 2871 * We've already moved the 2872 * origin's reference. 2873 */ 2874 continue; 2875 } 2876 2877 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, 2878 za.za_first_integer, FTAG, &cnds)); 2879 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj; 2880 2881 VERIFY3U(zap_remove_int(dp->dp_meta_objset, 2882 odd->dd_phys->dd_clones, o, tx), ==, 0); 2883 VERIFY3U(zap_add_int(dp->dp_meta_objset, 2884 dd->dd_phys->dd_clones, o, tx), ==, 0); 2885 dsl_dataset_rele(cnds, FTAG); 2886 } 2887 zap_cursor_fini(&zc); 2888 } 2889 2890 ASSERT3U(dsl_prop_numcb(ds), ==, 0); 2891 } 2892 2893 /* 2894 * Change space accounting. 2895 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2896 * both be valid, or both be 0 (resulting in delta == 0). This 2897 * is true for each of {clone,origin} independently. 2898 */ 2899 2900 delta = pa->cloneusedsnap - 2901 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2902 ASSERT3S(delta, >=, 0); 2903 ASSERT3U(pa->used, >=, delta); 2904 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2905 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2906 pa->used - delta, pa->comp, pa->uncomp, tx); 2907 2908 delta = pa->originusedsnap - 2909 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2910 ASSERT3S(delta, <=, 0); 2911 ASSERT3U(pa->used, >=, -delta); 2912 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2913 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2914 -pa->used - delta, -pa->comp, -pa->uncomp, tx); 2915 2916 origin_ds->ds_phys->ds_unique_bytes = pa->unique; 2917 2918 /* log history record */ 2919 spa_history_log_internal_ds(hds, "promote", tx, ""); 2920 2921 dsl_dir_close(odd, FTAG); 2922 } 2923 2924 static char *snaplist_tag = "snaplist"; 2925 /* 2926 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2927 * (exclusive) and last_obj (inclusive). The list will be in reverse 2928 * order (last_obj will be the list_head()). If first_obj == 0, do all 2929 * snapshots back to this dataset's origin. 2930 */ 2931 static int 2932 snaplist_make(dsl_pool_t *dp, boolean_t own, 2933 uint64_t first_obj, uint64_t last_obj, list_t *l) 2934 { 2935 uint64_t obj = last_obj; 2936 2937 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock)); 2938 2939 list_create(l, sizeof (struct promotenode), 2940 offsetof(struct promotenode, link)); 2941 2942 while (obj != first_obj) { 2943 dsl_dataset_t *ds; 2944 struct promotenode *snap; 2945 int err; 2946 2947 if (own) { 2948 err = dsl_dataset_own_obj(dp, obj, 2949 0, snaplist_tag, &ds); 2950 if (err == 0) 2951 dsl_dataset_make_exclusive(ds, snaplist_tag); 2952 } else { 2953 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds); 2954 } 2955 if (err == ENOENT) { 2956 /* lost race with snapshot destroy */ 2957 struct promotenode *last = list_tail(l); 2958 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj); 2959 obj = last->ds->ds_phys->ds_prev_snap_obj; 2960 continue; 2961 } else if (err) { 2962 return (err); 2963 } 2964 2965 if (first_obj == 0) 2966 first_obj = ds->ds_dir->dd_phys->dd_origin_obj; 2967 2968 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP); 2969 snap->ds = ds; 2970 list_insert_tail(l, snap); 2971 obj = ds->ds_phys->ds_prev_snap_obj; 2972 } 2973 2974 return (0); 2975 } 2976 2977 static int 2978 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2979 { 2980 struct promotenode *snap; 2981 2982 *spacep = 0; 2983 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2984 uint64_t used, comp, uncomp; 2985 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2986 mintxg, UINT64_MAX, &used, &comp, &uncomp); 2987 *spacep += used; 2988 } 2989 return (0); 2990 } 2991 2992 static void 2993 snaplist_destroy(list_t *l, boolean_t own) 2994 { 2995 struct promotenode *snap; 2996 2997 if (!l || !list_link_active(&l->list_head)) 2998 return; 2999 3000 while ((snap = list_tail(l)) != NULL) { 3001 list_remove(l, snap); 3002 if (own) 3003 dsl_dataset_disown(snap->ds, snaplist_tag); 3004 else 3005 dsl_dataset_rele(snap->ds, snaplist_tag); 3006 kmem_free(snap, sizeof (struct promotenode)); 3007 } 3008 list_destroy(l); 3009 } 3010 3011 /* 3012 * Promote a clone. Nomenclature note: 3013 * "clone" or "cds": the original clone which is being promoted 3014 * "origin" or "ods": the snapshot which is originally clone's origin 3015 * "origin head" or "ohds": the dataset which is the head 3016 * (filesystem/volume) for the origin 3017 * "origin origin": the origin of the origin's filesystem (typically 3018 * NULL, indicating that the clone is not a clone of a clone). 3019 */ 3020 int 3021 dsl_dataset_promote(const char *name, char *conflsnap) 3022 { 3023 dsl_dataset_t *ds; 3024 dsl_dir_t *dd; 3025 dsl_pool_t *dp; 3026 dmu_object_info_t doi; 3027 struct promotearg pa = { 0 }; 3028 struct promotenode *snap; 3029 int err; 3030 3031 err = dsl_dataset_hold(name, FTAG, &ds); 3032 if (err) 3033 return (err); 3034 dd = ds->ds_dir; 3035 dp = dd->dd_pool; 3036 3037 err = dmu_object_info(dp->dp_meta_objset, 3038 ds->ds_phys->ds_snapnames_zapobj, &doi); 3039 if (err) { 3040 dsl_dataset_rele(ds, FTAG); 3041 return (err); 3042 } 3043 3044 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) { 3045 dsl_dataset_rele(ds, FTAG); 3046 return (EINVAL); 3047 } 3048 3049 /* 3050 * We are going to inherit all the snapshots taken before our 3051 * origin (i.e., our new origin will be our parent's origin). 3052 * Take ownership of them so that we can rename them into our 3053 * namespace. 3054 */ 3055 rw_enter(&dp->dp_config_rwlock, RW_READER); 3056 3057 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj, 3058 &pa.shared_snaps); 3059 if (err != 0) 3060 goto out; 3061 3062 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps); 3063 if (err != 0) 3064 goto out; 3065 3066 snap = list_head(&pa.shared_snaps); 3067 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); 3068 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj, 3069 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps); 3070 if (err != 0) 3071 goto out; 3072 3073 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) { 3074 err = dsl_dataset_hold_obj(dp, 3075 snap->ds->ds_dir->dd_phys->dd_origin_obj, 3076 FTAG, &pa.origin_origin); 3077 if (err != 0) 3078 goto out; 3079 } 3080 3081 out: 3082 rw_exit(&dp->dp_config_rwlock); 3083 3084 /* 3085 * Add in 128x the snapnames zapobj size, since we will be moving 3086 * a bunch of snapnames to the promoted ds, and dirtying their 3087 * bonus buffers. 3088 */ 3089 if (err == 0) { 3090 err = dsl_sync_task_do(dp, dsl_dataset_promote_check, 3091 dsl_dataset_promote_sync, ds, &pa, 3092 2 + 2 * doi.doi_physical_blocks_512); 3093 if (err && pa.err_ds && conflsnap) 3094 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN); 3095 } 3096 3097 snaplist_destroy(&pa.shared_snaps, B_TRUE); 3098 snaplist_destroy(&pa.clone_snaps, B_FALSE); 3099 snaplist_destroy(&pa.origin_snaps, B_FALSE); 3100 if (pa.origin_origin) 3101 dsl_dataset_rele(pa.origin_origin, FTAG); 3102 dsl_dataset_rele(ds, FTAG); 3103 return (err); 3104 } 3105 3106 struct cloneswaparg { 3107 dsl_dataset_t *cds; /* clone dataset */ 3108 dsl_dataset_t *ohds; /* origin's head dataset */ 3109 boolean_t force; 3110 int64_t unused_refres_delta; /* change in unconsumed refreservation */ 3111 }; 3112 3113 /* ARGSUSED */ 3114 static int 3115 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx) 3116 { 3117 struct cloneswaparg *csa = arg1; 3118 3119 /* they should both be heads */ 3120 if (dsl_dataset_is_snapshot(csa->cds) || 3121 dsl_dataset_is_snapshot(csa->ohds)) 3122 return (EINVAL); 3123 3124 /* the branch point should be just before them */ 3125 if (csa->cds->ds_prev != csa->ohds->ds_prev) 3126 return (EINVAL); 3127 3128 /* cds should be the clone (unless they are unrelated) */ 3129 if (csa->cds->ds_prev != NULL && 3130 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap && 3131 csa->ohds->ds_object != 3132 csa->cds->ds_prev->ds_phys->ds_next_snap_obj) 3133 return (EINVAL); 3134 3135 /* the clone should be a child of the origin */ 3136 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir) 3137 return (EINVAL); 3138 3139 /* ohds shouldn't be modified unless 'force' */ 3140 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds)) 3141 return (ETXTBSY); 3142 3143 /* adjust amount of any unconsumed refreservation */ 3144 csa->unused_refres_delta = 3145 (int64_t)MIN(csa->ohds->ds_reserved, 3146 csa->ohds->ds_phys->ds_unique_bytes) - 3147 (int64_t)MIN(csa->ohds->ds_reserved, 3148 csa->cds->ds_phys->ds_unique_bytes); 3149 3150 if (csa->unused_refres_delta > 0 && 3151 csa->unused_refres_delta > 3152 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE)) 3153 return (ENOSPC); 3154 3155 if (csa->ohds->ds_quota != 0 && 3156 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota) 3157 return (EDQUOT); 3158 3159 return (0); 3160 } 3161 3162 /* ARGSUSED */ 3163 static void 3164 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx) 3165 { 3166 struct cloneswaparg *csa = arg1; 3167 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool; 3168 3169 ASSERT(csa->cds->ds_reserved == 0); 3170 ASSERT(csa->ohds->ds_quota == 0 || 3171 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota); 3172 3173 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx); 3174 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx); 3175 3176 if (csa->cds->ds_objset != NULL) { 3177 dmu_objset_evict(csa->cds->ds_objset); 3178 csa->cds->ds_objset = NULL; 3179 } 3180 3181 if (csa->ohds->ds_objset != NULL) { 3182 dmu_objset_evict(csa->ohds->ds_objset); 3183 csa->ohds->ds_objset = NULL; 3184 } 3185 3186 /* 3187 * Reset origin's unique bytes, if it exists. 3188 */ 3189 if (csa->cds->ds_prev) { 3190 dsl_dataset_t *origin = csa->cds->ds_prev; 3191 uint64_t comp, uncomp; 3192 3193 dmu_buf_will_dirty(origin->ds_dbuf, tx); 3194 dsl_deadlist_space_range(&csa->cds->ds_deadlist, 3195 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX, 3196 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp); 3197 } 3198 3199 /* swap blkptrs */ 3200 { 3201 blkptr_t tmp; 3202 tmp = csa->ohds->ds_phys->ds_bp; 3203 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp; 3204 csa->cds->ds_phys->ds_bp = tmp; 3205 } 3206 3207 /* set dd_*_bytes */ 3208 { 3209 int64_t dused, dcomp, duncomp; 3210 uint64_t cdl_used, cdl_comp, cdl_uncomp; 3211 uint64_t odl_used, odl_comp, odl_uncomp; 3212 3213 ASSERT3U(csa->cds->ds_dir->dd_phys-> 3214 dd_used_breakdown[DD_USED_SNAP], ==, 0); 3215 3216 dsl_deadlist_space(&csa->cds->ds_deadlist, 3217 &cdl_used, &cdl_comp, &cdl_uncomp); 3218 dsl_deadlist_space(&csa->ohds->ds_deadlist, 3219 &odl_used, &odl_comp, &odl_uncomp); 3220 3221 dused = csa->cds->ds_phys->ds_referenced_bytes + cdl_used - 3222 (csa->ohds->ds_phys->ds_referenced_bytes + odl_used); 3223 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp - 3224 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp); 3225 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes + 3226 cdl_uncomp - 3227 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp); 3228 3229 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD, 3230 dused, dcomp, duncomp, tx); 3231 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD, 3232 -dused, -dcomp, -duncomp, tx); 3233 3234 /* 3235 * The difference in the space used by snapshots is the 3236 * difference in snapshot space due to the head's 3237 * deadlist (since that's the only thing that's 3238 * changing that affects the snapused). 3239 */ 3240 dsl_deadlist_space_range(&csa->cds->ds_deadlist, 3241 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX, 3242 &cdl_used, &cdl_comp, &cdl_uncomp); 3243 dsl_deadlist_space_range(&csa->ohds->ds_deadlist, 3244 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX, 3245 &odl_used, &odl_comp, &odl_uncomp); 3246 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used, 3247 DD_USED_HEAD, DD_USED_SNAP, tx); 3248 } 3249 3250 /* swap ds_*_bytes */ 3251 SWITCH64(csa->ohds->ds_phys->ds_referenced_bytes, 3252 csa->cds->ds_phys->ds_referenced_bytes); 3253 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes, 3254 csa->cds->ds_phys->ds_compressed_bytes); 3255 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes, 3256 csa->cds->ds_phys->ds_uncompressed_bytes); 3257 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes, 3258 csa->cds->ds_phys->ds_unique_bytes); 3259 3260 /* apply any parent delta for change in unconsumed refreservation */ 3261 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV, 3262 csa->unused_refres_delta, 0, 0, tx); 3263 3264 /* 3265 * Swap deadlists. 3266 */ 3267 dsl_deadlist_close(&csa->cds->ds_deadlist); 3268 dsl_deadlist_close(&csa->ohds->ds_deadlist); 3269 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj, 3270 csa->cds->ds_phys->ds_deadlist_obj); 3271 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset, 3272 csa->cds->ds_phys->ds_deadlist_obj); 3273 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset, 3274 csa->ohds->ds_phys->ds_deadlist_obj); 3275 3276 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx); 3277 3278 spa_history_log_internal_ds(csa->cds, "clone swap", tx, 3279 "parent=%s", csa->ohds->ds_dir->dd_myname); 3280 } 3281 3282 /* 3283 * Swap 'clone' with its origin head datasets. Used at the end of "zfs 3284 * recv" into an existing fs to swizzle the file system to the new 3285 * version, and by "zfs rollback". Can also be used to swap two 3286 * independent head datasets if neither has any snapshots. 3287 */ 3288 int 3289 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head, 3290 boolean_t force) 3291 { 3292 struct cloneswaparg csa; 3293 int error; 3294 3295 ASSERT(clone->ds_owner); 3296 ASSERT(origin_head->ds_owner); 3297 retry: 3298 /* 3299 * Need exclusive access for the swap. If we're swapping these 3300 * datasets back after an error, we already hold the locks. 3301 */ 3302 if (!RW_WRITE_HELD(&clone->ds_rwlock)) 3303 rw_enter(&clone->ds_rwlock, RW_WRITER); 3304 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) && 3305 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) { 3306 rw_exit(&clone->ds_rwlock); 3307 rw_enter(&origin_head->ds_rwlock, RW_WRITER); 3308 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) { 3309 rw_exit(&origin_head->ds_rwlock); 3310 goto retry; 3311 } 3312 } 3313 csa.cds = clone; 3314 csa.ohds = origin_head; 3315 csa.force = force; 3316 error = dsl_sync_task_do(clone->ds_dir->dd_pool, 3317 dsl_dataset_clone_swap_check, 3318 dsl_dataset_clone_swap_sync, &csa, NULL, 9); 3319 return (error); 3320 } 3321 3322 /* 3323 * Given a pool name and a dataset object number in that pool, 3324 * return the name of that dataset. 3325 */ 3326 int 3327 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 3328 { 3329 spa_t *spa; 3330 dsl_pool_t *dp; 3331 dsl_dataset_t *ds; 3332 int error; 3333 3334 if ((error = spa_open(pname, &spa, FTAG)) != 0) 3335 return (error); 3336 dp = spa_get_dsl(spa); 3337 rw_enter(&dp->dp_config_rwlock, RW_READER); 3338 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) { 3339 dsl_dataset_name(ds, buf); 3340 dsl_dataset_rele(ds, FTAG); 3341 } 3342 rw_exit(&dp->dp_config_rwlock); 3343 spa_close(spa, FTAG); 3344 3345 return (error); 3346 } 3347 3348 int 3349 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 3350 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 3351 { 3352 int error = 0; 3353 3354 ASSERT3S(asize, >, 0); 3355 3356 /* 3357 * *ref_rsrv is the portion of asize that will come from any 3358 * unconsumed refreservation space. 3359 */ 3360 *ref_rsrv = 0; 3361 3362 mutex_enter(&ds->ds_lock); 3363 /* 3364 * Make a space adjustment for reserved bytes. 3365 */ 3366 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) { 3367 ASSERT3U(*used, >=, 3368 ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 3369 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 3370 *ref_rsrv = 3371 asize - MIN(asize, parent_delta(ds, asize + inflight)); 3372 } 3373 3374 if (!check_quota || ds->ds_quota == 0) { 3375 mutex_exit(&ds->ds_lock); 3376 return (0); 3377 } 3378 /* 3379 * If they are requesting more space, and our current estimate 3380 * is over quota, they get to try again unless the actual 3381 * on-disk is over quota and there are no pending changes (which 3382 * may free up space for us). 3383 */ 3384 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) { 3385 if (inflight > 0 || 3386 ds->ds_phys->ds_referenced_bytes < ds->ds_quota) 3387 error = ERESTART; 3388 else 3389 error = EDQUOT; 3390 } 3391 mutex_exit(&ds->ds_lock); 3392 3393 return (error); 3394 } 3395 3396 /* ARGSUSED */ 3397 static int 3398 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) 3399 { 3400 dsl_dataset_t *ds = arg1; 3401 dsl_prop_setarg_t *psa = arg2; 3402 int err; 3403 3404 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA) 3405 return (ENOTSUP); 3406 3407 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) 3408 return (err); 3409 3410 if (psa->psa_effective_value == 0) 3411 return (0); 3412 3413 if (psa->psa_effective_value < ds->ds_phys->ds_referenced_bytes || 3414 psa->psa_effective_value < ds->ds_reserved) 3415 return (ENOSPC); 3416 3417 return (0); 3418 } 3419 3420 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *); 3421 3422 void 3423 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx) 3424 { 3425 dsl_dataset_t *ds = arg1; 3426 dsl_prop_setarg_t *psa = arg2; 3427 uint64_t effective_value = psa->psa_effective_value; 3428 3429 dsl_prop_set_sync(ds, psa, tx); 3430 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa); 3431 3432 if (ds->ds_quota != effective_value) { 3433 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3434 ds->ds_quota = effective_value; 3435 3436 spa_history_log_internal_ds(ds, "set refquota", tx, 3437 "refquota=%lld", (longlong_t)ds->ds_quota); 3438 } 3439 } 3440 3441 int 3442 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota) 3443 { 3444 dsl_dataset_t *ds; 3445 dsl_prop_setarg_t psa; 3446 int err; 3447 3448 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a); 3449 3450 err = dsl_dataset_hold(dsname, FTAG, &ds); 3451 if (err) 3452 return (err); 3453 3454 /* 3455 * If someone removes a file, then tries to set the quota, we 3456 * want to make sure the file freeing takes effect. 3457 */ 3458 txg_wait_open(ds->ds_dir->dd_pool, 0); 3459 3460 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 3461 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync, 3462 ds, &psa, 0); 3463 3464 dsl_dataset_rele(ds, FTAG); 3465 return (err); 3466 } 3467 3468 static int 3469 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) 3470 { 3471 dsl_dataset_t *ds = arg1; 3472 dsl_prop_setarg_t *psa = arg2; 3473 uint64_t effective_value; 3474 uint64_t unique; 3475 int err; 3476 3477 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < 3478 SPA_VERSION_REFRESERVATION) 3479 return (ENOTSUP); 3480 3481 if (dsl_dataset_is_snapshot(ds)) 3482 return (EINVAL); 3483 3484 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) 3485 return (err); 3486 3487 effective_value = psa->psa_effective_value; 3488 3489 /* 3490 * If we are doing the preliminary check in open context, the 3491 * space estimates may be inaccurate. 3492 */ 3493 if (!dmu_tx_is_syncing(tx)) 3494 return (0); 3495 3496 mutex_enter(&ds->ds_lock); 3497 if (!DS_UNIQUE_IS_ACCURATE(ds)) 3498 dsl_dataset_recalc_head_uniq(ds); 3499 unique = ds->ds_phys->ds_unique_bytes; 3500 mutex_exit(&ds->ds_lock); 3501 3502 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) { 3503 uint64_t delta = MAX(unique, effective_value) - 3504 MAX(unique, ds->ds_reserved); 3505 3506 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 3507 return (ENOSPC); 3508 if (ds->ds_quota > 0 && 3509 effective_value > ds->ds_quota) 3510 return (ENOSPC); 3511 } 3512 3513 return (0); 3514 } 3515 3516 static void 3517 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx) 3518 { 3519 dsl_dataset_t *ds = arg1; 3520 dsl_prop_setarg_t *psa = arg2; 3521 uint64_t effective_value = psa->psa_effective_value; 3522 uint64_t unique; 3523 int64_t delta; 3524 3525 dsl_prop_set_sync(ds, psa, tx); 3526 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa); 3527 3528 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3529 3530 mutex_enter(&ds->ds_dir->dd_lock); 3531 mutex_enter(&ds->ds_lock); 3532 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3533 unique = ds->ds_phys->ds_unique_bytes; 3534 delta = MAX(0, (int64_t)(effective_value - unique)) - 3535 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3536 ds->ds_reserved = effective_value; 3537 mutex_exit(&ds->ds_lock); 3538 3539 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3540 mutex_exit(&ds->ds_dir->dd_lock); 3541 3542 spa_history_log_internal_ds(ds, "set refreservation", tx, 3543 "refreservation=%lld", (longlong_t)effective_value); 3544 } 3545 3546 int 3547 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source, 3548 uint64_t reservation) 3549 { 3550 dsl_dataset_t *ds; 3551 dsl_prop_setarg_t psa; 3552 int err; 3553 3554 dsl_prop_setarg_init_uint64(&psa, "refreservation", source, 3555 &reservation); 3556 3557 err = dsl_dataset_hold(dsname, FTAG, &ds); 3558 if (err) 3559 return (err); 3560 3561 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 3562 dsl_dataset_set_reservation_check, 3563 dsl_dataset_set_reservation_sync, ds, &psa, 0); 3564 3565 dsl_dataset_rele(ds, FTAG); 3566 return (err); 3567 } 3568 3569 typedef struct zfs_hold_cleanup_arg { 3570 dsl_pool_t *dp; 3571 uint64_t dsobj; 3572 char htag[MAXNAMELEN]; 3573 } zfs_hold_cleanup_arg_t; 3574 3575 static void 3576 dsl_dataset_user_release_onexit(void *arg) 3577 { 3578 zfs_hold_cleanup_arg_t *ca = arg; 3579 3580 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag, 3581 B_TRUE); 3582 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t)); 3583 } 3584 3585 void 3586 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag, 3587 minor_t minor) 3588 { 3589 zfs_hold_cleanup_arg_t *ca; 3590 3591 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP); 3592 ca->dp = ds->ds_dir->dd_pool; 3593 ca->dsobj = ds->ds_object; 3594 (void) strlcpy(ca->htag, htag, sizeof (ca->htag)); 3595 VERIFY3U(0, ==, zfs_onexit_add_cb(minor, 3596 dsl_dataset_user_release_onexit, ca, NULL)); 3597 } 3598 3599 /* 3600 * If you add new checks here, you may need to add 3601 * additional checks to the "temporary" case in 3602 * snapshot_check() in dmu_objset.c. 3603 */ 3604 static int 3605 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx) 3606 { 3607 dsl_dataset_t *ds = arg1; 3608 struct dsl_ds_holdarg *ha = arg2; 3609 const char *htag = ha->htag; 3610 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3611 int error = 0; 3612 3613 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) 3614 return (ENOTSUP); 3615 3616 if (!dsl_dataset_is_snapshot(ds)) 3617 return (EINVAL); 3618 3619 /* tags must be unique */ 3620 mutex_enter(&ds->ds_lock); 3621 if (ds->ds_phys->ds_userrefs_obj) { 3622 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag, 3623 8, 1, tx); 3624 if (error == 0) 3625 error = EEXIST; 3626 else if (error == ENOENT) 3627 error = 0; 3628 } 3629 mutex_exit(&ds->ds_lock); 3630 3631 if (error == 0 && ha->temphold && 3632 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN) 3633 error = E2BIG; 3634 3635 return (error); 3636 } 3637 3638 void 3639 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx) 3640 { 3641 dsl_dataset_t *ds = arg1; 3642 struct dsl_ds_holdarg *ha = arg2; 3643 const char *htag = ha->htag; 3644 dsl_pool_t *dp = ds->ds_dir->dd_pool; 3645 objset_t *mos = dp->dp_meta_objset; 3646 uint64_t now = gethrestime_sec(); 3647 uint64_t zapobj; 3648 3649 mutex_enter(&ds->ds_lock); 3650 if (ds->ds_phys->ds_userrefs_obj == 0) { 3651 /* 3652 * This is the first user hold for this dataset. Create 3653 * the userrefs zap object. 3654 */ 3655 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3656 zapobj = ds->ds_phys->ds_userrefs_obj = 3657 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx); 3658 } else { 3659 zapobj = ds->ds_phys->ds_userrefs_obj; 3660 } 3661 ds->ds_userrefs++; 3662 mutex_exit(&ds->ds_lock); 3663 3664 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx)); 3665 3666 if (ha->temphold) { 3667 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object, 3668 htag, &now, tx)); 3669 } 3670 3671 spa_history_log_internal_ds(ds, "hold", tx, 3672 "tag = %s temp = %d holds now = %llu", 3673 htag, (int)ha->temphold, ds->ds_userrefs); 3674 } 3675 3676 static int 3677 dsl_dataset_user_hold_one(const char *dsname, void *arg) 3678 { 3679 struct dsl_ds_holdarg *ha = arg; 3680 dsl_dataset_t *ds; 3681 int error; 3682 char *name; 3683 3684 /* alloc a buffer to hold dsname@snapname plus terminating NULL */ 3685 name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3686 error = dsl_dataset_hold(name, ha->dstg, &ds); 3687 strfree(name); 3688 if (error == 0) { 3689 ha->gotone = B_TRUE; 3690 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check, 3691 dsl_dataset_user_hold_sync, ds, ha, 0); 3692 } else if (error == ENOENT && ha->recursive) { 3693 error = 0; 3694 } else { 3695 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3696 } 3697 return (error); 3698 } 3699 3700 int 3701 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag, 3702 boolean_t temphold) 3703 { 3704 struct dsl_ds_holdarg *ha; 3705 int error; 3706 3707 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); 3708 ha->htag = htag; 3709 ha->temphold = temphold; 3710 error = dsl_sync_task_do(ds->ds_dir->dd_pool, 3711 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync, 3712 ds, ha, 0); 3713 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3714 3715 return (error); 3716 } 3717 3718 int 3719 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag, 3720 boolean_t recursive, boolean_t temphold, int cleanup_fd) 3721 { 3722 struct dsl_ds_holdarg *ha; 3723 dsl_sync_task_t *dst; 3724 spa_t *spa; 3725 int error; 3726 minor_t minor = 0; 3727 3728 if (cleanup_fd != -1) { 3729 /* Currently we only support cleanup-on-exit of tempholds. */ 3730 if (!temphold) 3731 return (EINVAL); 3732 error = zfs_onexit_fd_hold(cleanup_fd, &minor); 3733 if (error) 3734 return (error); 3735 } 3736 3737 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); 3738 3739 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3740 3741 error = spa_open(dsname, &spa, FTAG); 3742 if (error) { 3743 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3744 if (cleanup_fd != -1) 3745 zfs_onexit_fd_rele(cleanup_fd); 3746 return (error); 3747 } 3748 3749 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 3750 ha->htag = htag; 3751 ha->snapname = snapname; 3752 ha->recursive = recursive; 3753 ha->temphold = temphold; 3754 3755 if (recursive) { 3756 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one, 3757 ha, DS_FIND_CHILDREN); 3758 } else { 3759 error = dsl_dataset_user_hold_one(dsname, ha); 3760 } 3761 if (error == 0) 3762 error = dsl_sync_task_group_wait(ha->dstg); 3763 3764 for (dst = list_head(&ha->dstg->dstg_tasks); dst; 3765 dst = list_next(&ha->dstg->dstg_tasks, dst)) { 3766 dsl_dataset_t *ds = dst->dst_arg1; 3767 3768 if (dst->dst_err) { 3769 dsl_dataset_name(ds, ha->failed); 3770 *strchr(ha->failed, '@') = '\0'; 3771 } else if (error == 0 && minor != 0 && temphold) { 3772 /* 3773 * If this hold is to be released upon process exit, 3774 * register that action now. 3775 */ 3776 dsl_register_onexit_hold_cleanup(ds, htag, minor); 3777 } 3778 dsl_dataset_rele(ds, ha->dstg); 3779 } 3780 3781 if (error == 0 && recursive && !ha->gotone) 3782 error = ENOENT; 3783 3784 if (error) 3785 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed)); 3786 3787 dsl_sync_task_group_destroy(ha->dstg); 3788 3789 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3790 spa_close(spa, FTAG); 3791 if (cleanup_fd != -1) 3792 zfs_onexit_fd_rele(cleanup_fd); 3793 return (error); 3794 } 3795 3796 struct dsl_ds_releasearg { 3797 dsl_dataset_t *ds; 3798 const char *htag; 3799 boolean_t own; /* do we own or just hold ds? */ 3800 }; 3801 3802 static int 3803 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag, 3804 boolean_t *might_destroy) 3805 { 3806 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3807 uint64_t zapobj; 3808 uint64_t tmp; 3809 int error; 3810 3811 *might_destroy = B_FALSE; 3812 3813 mutex_enter(&ds->ds_lock); 3814 zapobj = ds->ds_phys->ds_userrefs_obj; 3815 if (zapobj == 0) { 3816 /* The tag can't possibly exist */ 3817 mutex_exit(&ds->ds_lock); 3818 return (ESRCH); 3819 } 3820 3821 /* Make sure the tag exists */ 3822 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp); 3823 if (error) { 3824 mutex_exit(&ds->ds_lock); 3825 if (error == ENOENT) 3826 error = ESRCH; 3827 return (error); 3828 } 3829 3830 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 && 3831 DS_IS_DEFER_DESTROY(ds)) 3832 *might_destroy = B_TRUE; 3833 3834 mutex_exit(&ds->ds_lock); 3835 return (0); 3836 } 3837 3838 static int 3839 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx) 3840 { 3841 struct dsl_ds_releasearg *ra = arg1; 3842 dsl_dataset_t *ds = ra->ds; 3843 boolean_t might_destroy; 3844 int error; 3845 3846 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) 3847 return (ENOTSUP); 3848 3849 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy); 3850 if (error) 3851 return (error); 3852 3853 if (might_destroy) { 3854 struct dsl_ds_destroyarg dsda = {0}; 3855 3856 if (dmu_tx_is_syncing(tx)) { 3857 /* 3858 * If we're not prepared to remove the snapshot, 3859 * we can't allow the release to happen right now. 3860 */ 3861 if (!ra->own) 3862 return (EBUSY); 3863 } 3864 dsda.ds = ds; 3865 dsda.releasing = B_TRUE; 3866 return (dsl_dataset_destroy_check(&dsda, tag, tx)); 3867 } 3868 3869 return (0); 3870 } 3871 3872 static void 3873 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx) 3874 { 3875 struct dsl_ds_releasearg *ra = arg1; 3876 dsl_dataset_t *ds = ra->ds; 3877 dsl_pool_t *dp = ds->ds_dir->dd_pool; 3878 objset_t *mos = dp->dp_meta_objset; 3879 uint64_t zapobj; 3880 uint64_t refs; 3881 int error; 3882 3883 mutex_enter(&ds->ds_lock); 3884 ds->ds_userrefs--; 3885 refs = ds->ds_userrefs; 3886 mutex_exit(&ds->ds_lock); 3887 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx); 3888 VERIFY(error == 0 || error == ENOENT); 3889 zapobj = ds->ds_phys->ds_userrefs_obj; 3890 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx)); 3891 3892 spa_history_log_internal_ds(ds, "release", tx, 3893 "tag = %s refs now = %lld", ra->htag, (longlong_t)refs); 3894 3895 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 && 3896 DS_IS_DEFER_DESTROY(ds)) { 3897 struct dsl_ds_destroyarg dsda = {0}; 3898 3899 ASSERT(ra->own); 3900 dsda.ds = ds; 3901 dsda.releasing = B_TRUE; 3902 /* We already did the destroy_check */ 3903 dsl_dataset_destroy_sync(&dsda, tag, tx); 3904 } 3905 } 3906 3907 static int 3908 dsl_dataset_user_release_one(const char *dsname, void *arg) 3909 { 3910 struct dsl_ds_holdarg *ha = arg; 3911 struct dsl_ds_releasearg *ra; 3912 dsl_dataset_t *ds; 3913 int error; 3914 void *dtag = ha->dstg; 3915 char *name; 3916 boolean_t own = B_FALSE; 3917 boolean_t might_destroy; 3918 3919 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */ 3920 name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3921 error = dsl_dataset_hold(name, dtag, &ds); 3922 strfree(name); 3923 if (error == ENOENT && ha->recursive) 3924 return (0); 3925 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3926 if (error) 3927 return (error); 3928 3929 ha->gotone = B_TRUE; 3930 3931 ASSERT(dsl_dataset_is_snapshot(ds)); 3932 3933 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy); 3934 if (error) { 3935 dsl_dataset_rele(ds, dtag); 3936 return (error); 3937 } 3938 3939 if (might_destroy) { 3940 #ifdef _KERNEL 3941 name = kmem_asprintf("%s@%s", dsname, ha->snapname); 3942 error = zfs_unmount_snap(name, NULL); 3943 strfree(name); 3944 if (error) { 3945 dsl_dataset_rele(ds, dtag); 3946 return (error); 3947 } 3948 #endif 3949 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) { 3950 dsl_dataset_rele(ds, dtag); 3951 return (EBUSY); 3952 } else { 3953 own = B_TRUE; 3954 dsl_dataset_make_exclusive(ds, dtag); 3955 } 3956 } 3957 3958 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP); 3959 ra->ds = ds; 3960 ra->htag = ha->htag; 3961 ra->own = own; 3962 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check, 3963 dsl_dataset_user_release_sync, ra, dtag, 0); 3964 3965 return (0); 3966 } 3967 3968 int 3969 dsl_dataset_user_release(char *dsname, char *snapname, char *htag, 3970 boolean_t recursive) 3971 { 3972 struct dsl_ds_holdarg *ha; 3973 dsl_sync_task_t *dst; 3974 spa_t *spa; 3975 int error; 3976 3977 top: 3978 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); 3979 3980 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); 3981 3982 error = spa_open(dsname, &spa, FTAG); 3983 if (error) { 3984 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 3985 return (error); 3986 } 3987 3988 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 3989 ha->htag = htag; 3990 ha->snapname = snapname; 3991 ha->recursive = recursive; 3992 if (recursive) { 3993 error = dmu_objset_find(dsname, dsl_dataset_user_release_one, 3994 ha, DS_FIND_CHILDREN); 3995 } else { 3996 error = dsl_dataset_user_release_one(dsname, ha); 3997 } 3998 if (error == 0) 3999 error = dsl_sync_task_group_wait(ha->dstg); 4000 4001 for (dst = list_head(&ha->dstg->dstg_tasks); dst; 4002 dst = list_next(&ha->dstg->dstg_tasks, dst)) { 4003 struct dsl_ds_releasearg *ra = dst->dst_arg1; 4004 dsl_dataset_t *ds = ra->ds; 4005 4006 if (dst->dst_err) 4007 dsl_dataset_name(ds, ha->failed); 4008 4009 if (ra->own) 4010 dsl_dataset_disown(ds, ha->dstg); 4011 else 4012 dsl_dataset_rele(ds, ha->dstg); 4013 4014 kmem_free(ra, sizeof (struct dsl_ds_releasearg)); 4015 } 4016 4017 if (error == 0 && recursive && !ha->gotone) 4018 error = ENOENT; 4019 4020 if (error && error != EBUSY) 4021 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed)); 4022 4023 dsl_sync_task_group_destroy(ha->dstg); 4024 kmem_free(ha, sizeof (struct dsl_ds_holdarg)); 4025 spa_close(spa, FTAG); 4026 4027 /* 4028 * We can get EBUSY if we were racing with deferred destroy and 4029 * dsl_dataset_user_release_check() hadn't done the necessary 4030 * open context setup. We can also get EBUSY if we're racing 4031 * with destroy and that thread is the ds_owner. Either way 4032 * the busy condition should be transient, and we should retry 4033 * the release operation. 4034 */ 4035 if (error == EBUSY) 4036 goto top; 4037 4038 return (error); 4039 } 4040 4041 /* 4042 * Called at spa_load time (with retry == B_FALSE) to release a stale 4043 * temporary user hold. Also called by the onexit code (with retry == B_TRUE). 4044 */ 4045 int 4046 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag, 4047 boolean_t retry) 4048 { 4049 dsl_dataset_t *ds; 4050 char *snap; 4051 char *name; 4052 int namelen; 4053 int error; 4054 4055 do { 4056 rw_enter(&dp->dp_config_rwlock, RW_READER); 4057 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); 4058 rw_exit(&dp->dp_config_rwlock); 4059 if (error) 4060 return (error); 4061 namelen = dsl_dataset_namelen(ds)+1; 4062 name = kmem_alloc(namelen, KM_SLEEP); 4063 dsl_dataset_name(ds, name); 4064 dsl_dataset_rele(ds, FTAG); 4065 4066 snap = strchr(name, '@'); 4067 *snap = '\0'; 4068 ++snap; 4069 error = dsl_dataset_user_release(name, snap, htag, B_FALSE); 4070 kmem_free(name, namelen); 4071 4072 /* 4073 * The object can't have been destroyed because we have a hold, 4074 * but it might have been renamed, resulting in ENOENT. Retry 4075 * if we've been requested to do so. 4076 * 4077 * It would be nice if we could use the dsobj all the way 4078 * through and avoid ENOENT entirely. But we might need to 4079 * unmount the snapshot, and there's currently no way to lookup 4080 * a vfsp using a ZFS object id. 4081 */ 4082 } while ((error == ENOENT) && retry); 4083 4084 return (error); 4085 } 4086 4087 int 4088 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp) 4089 { 4090 dsl_dataset_t *ds; 4091 int err; 4092 4093 err = dsl_dataset_hold(dsname, FTAG, &ds); 4094 if (err) 4095 return (err); 4096 4097 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP)); 4098 if (ds->ds_phys->ds_userrefs_obj != 0) { 4099 zap_attribute_t *za; 4100 zap_cursor_t zc; 4101 4102 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 4103 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset, 4104 ds->ds_phys->ds_userrefs_obj); 4105 zap_cursor_retrieve(&zc, za) == 0; 4106 zap_cursor_advance(&zc)) { 4107 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name, 4108 za->za_first_integer)); 4109 } 4110 zap_cursor_fini(&zc); 4111 kmem_free(za, sizeof (zap_attribute_t)); 4112 } 4113 dsl_dataset_rele(ds, FTAG); 4114 return (0); 4115 } 4116 4117 /* 4118 * Note, this function is used as the callback for dmu_objset_find(). We 4119 * always return 0 so that we will continue to find and process 4120 * inconsistent datasets, even if we encounter an error trying to 4121 * process one of them. 4122 */ 4123 /* ARGSUSED */ 4124 int 4125 dsl_destroy_inconsistent(const char *dsname, void *arg) 4126 { 4127 dsl_dataset_t *ds; 4128 4129 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) { 4130 if (DS_IS_INCONSISTENT(ds)) 4131 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE); 4132 else 4133 dsl_dataset_disown(ds, FTAG); 4134 } 4135 return (0); 4136 } 4137 4138 /* 4139 * Return (in *usedp) the amount of space written in new that is not 4140 * present in oldsnap. New may be a snapshot or the head. Old must be 4141 * a snapshot before new, in new's filesystem (or its origin). If not then 4142 * fail and return EINVAL. 4143 * 4144 * The written space is calculated by considering two components: First, we 4145 * ignore any freed space, and calculate the written as new's used space 4146 * minus old's used space. Next, we add in the amount of space that was freed 4147 * between the two snapshots, thus reducing new's used space relative to old's. 4148 * Specifically, this is the space that was born before old->ds_creation_txg, 4149 * and freed before new (ie. on new's deadlist or a previous deadlist). 4150 * 4151 * space freed [---------------------] 4152 * snapshots ---O-------O--------O-------O------ 4153 * oldsnap new 4154 */ 4155 int 4156 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 4157 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4158 { 4159 int err = 0; 4160 uint64_t snapobj; 4161 dsl_pool_t *dp = new->ds_dir->dd_pool; 4162 4163 *usedp = 0; 4164 *usedp += new->ds_phys->ds_referenced_bytes; 4165 *usedp -= oldsnap->ds_phys->ds_referenced_bytes; 4166 4167 *compp = 0; 4168 *compp += new->ds_phys->ds_compressed_bytes; 4169 *compp -= oldsnap->ds_phys->ds_compressed_bytes; 4170 4171 *uncompp = 0; 4172 *uncompp += new->ds_phys->ds_uncompressed_bytes; 4173 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes; 4174 4175 rw_enter(&dp->dp_config_rwlock, RW_READER); 4176 snapobj = new->ds_object; 4177 while (snapobj != oldsnap->ds_object) { 4178 dsl_dataset_t *snap; 4179 uint64_t used, comp, uncomp; 4180 4181 if (snapobj == new->ds_object) { 4182 snap = new; 4183 } else { 4184 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 4185 if (err != 0) 4186 break; 4187 } 4188 4189 if (snap->ds_phys->ds_prev_snap_txg == 4190 oldsnap->ds_phys->ds_creation_txg) { 4191 /* 4192 * The blocks in the deadlist can not be born after 4193 * ds_prev_snap_txg, so get the whole deadlist space, 4194 * which is more efficient (especially for old-format 4195 * deadlists). Unfortunately the deadlist code 4196 * doesn't have enough information to make this 4197 * optimization itself. 4198 */ 4199 dsl_deadlist_space(&snap->ds_deadlist, 4200 &used, &comp, &uncomp); 4201 } else { 4202 dsl_deadlist_space_range(&snap->ds_deadlist, 4203 0, oldsnap->ds_phys->ds_creation_txg, 4204 &used, &comp, &uncomp); 4205 } 4206 *usedp += used; 4207 *compp += comp; 4208 *uncompp += uncomp; 4209 4210 /* 4211 * If we get to the beginning of the chain of snapshots 4212 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 4213 * was not a snapshot of/before new. 4214 */ 4215 snapobj = snap->ds_phys->ds_prev_snap_obj; 4216 if (snap != new) 4217 dsl_dataset_rele(snap, FTAG); 4218 if (snapobj == 0) { 4219 err = EINVAL; 4220 break; 4221 } 4222 4223 } 4224 rw_exit(&dp->dp_config_rwlock); 4225 return (err); 4226 } 4227 4228 /* 4229 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 4230 * lastsnap, and all snapshots in between are deleted. 4231 * 4232 * blocks that would be freed [---------------------------] 4233 * snapshots ---O-------O--------O-------O--------O 4234 * firstsnap lastsnap 4235 * 4236 * This is the set of blocks that were born after the snap before firstsnap, 4237 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 4238 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 4239 * We calculate this by iterating over the relevant deadlists (from the snap 4240 * after lastsnap, backward to the snap after firstsnap), summing up the 4241 * space on the deadlist that was born after the snap before firstsnap. 4242 */ 4243 int 4244 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 4245 dsl_dataset_t *lastsnap, 4246 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 4247 { 4248 int err = 0; 4249 uint64_t snapobj; 4250 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 4251 4252 ASSERT(dsl_dataset_is_snapshot(firstsnap)); 4253 ASSERT(dsl_dataset_is_snapshot(lastsnap)); 4254 4255 /* 4256 * Check that the snapshots are in the same dsl_dir, and firstsnap 4257 * is before lastsnap. 4258 */ 4259 if (firstsnap->ds_dir != lastsnap->ds_dir || 4260 firstsnap->ds_phys->ds_creation_txg > 4261 lastsnap->ds_phys->ds_creation_txg) 4262 return (EINVAL); 4263 4264 *usedp = *compp = *uncompp = 0; 4265 4266 rw_enter(&dp->dp_config_rwlock, RW_READER); 4267 snapobj = lastsnap->ds_phys->ds_next_snap_obj; 4268 while (snapobj != firstsnap->ds_object) { 4269 dsl_dataset_t *ds; 4270 uint64_t used, comp, uncomp; 4271 4272 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 4273 if (err != 0) 4274 break; 4275 4276 dsl_deadlist_space_range(&ds->ds_deadlist, 4277 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX, 4278 &used, &comp, &uncomp); 4279 *usedp += used; 4280 *compp += comp; 4281 *uncompp += uncomp; 4282 4283 snapobj = ds->ds_phys->ds_prev_snap_obj; 4284 ASSERT3U(snapobj, !=, 0); 4285 dsl_dataset_rele(ds, FTAG); 4286 } 4287 rw_exit(&dp->dp_config_rwlock); 4288 return (err); 4289 } 4290