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 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #include <sys/dmu_objset.h> 27 #include <sys/dsl_dataset.h> 28 #include <sys/dsl_dir.h> 29 #include <sys/dsl_prop.h> 30 #include <sys/dsl_synctask.h> 31 #include <sys/dmu_traverse.h> 32 #include <sys/dmu_tx.h> 33 #include <sys/arc.h> 34 #include <sys/zio.h> 35 #include <sys/zap.h> 36 #include <sys/unique.h> 37 #include <sys/zfs_context.h> 38 #include <sys/zfs_ioctl.h> 39 #include <sys/spa.h> 40 #include <sys/zfs_znode.h> 41 #include <sys/sunddi.h> 42 43 static char *dsl_reaper = "the grim reaper"; 44 45 static dsl_checkfunc_t dsl_dataset_destroy_begin_check; 46 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync; 47 static dsl_checkfunc_t dsl_dataset_rollback_check; 48 static dsl_syncfunc_t dsl_dataset_rollback_sync; 49 static dsl_syncfunc_t dsl_dataset_set_reservation_sync; 50 51 #define DS_REF_MAX (1ULL << 62) 52 53 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE 54 55 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper) 56 57 58 /* 59 * Figure out how much of this delta should be propogated to the dsl_dir 60 * layer. If there's a refreservation, that space has already been 61 * partially accounted for in our ancestors. 62 */ 63 static int64_t 64 parent_delta(dsl_dataset_t *ds, int64_t delta) 65 { 66 uint64_t old_bytes, new_bytes; 67 68 if (ds->ds_reserved == 0) 69 return (delta); 70 71 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 72 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 73 74 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 75 return (new_bytes - old_bytes); 76 } 77 78 void 79 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 80 { 81 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp); 82 int compressed = BP_GET_PSIZE(bp); 83 int uncompressed = BP_GET_UCSIZE(bp); 84 int64_t delta; 85 86 dprintf_bp(bp, "born, ds=%p\n", ds); 87 88 ASSERT(dmu_tx_is_syncing(tx)); 89 /* It could have been compressed away to nothing */ 90 if (BP_IS_HOLE(bp)) 91 return; 92 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 93 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES); 94 if (ds == NULL) { 95 /* 96 * Account for the meta-objset space in its placeholder 97 * dsl_dir. 98 */ 99 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */ 100 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, 101 used, compressed, uncompressed, tx); 102 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); 103 return; 104 } 105 dmu_buf_will_dirty(ds->ds_dbuf, tx); 106 mutex_enter(&ds->ds_dir->dd_lock); 107 mutex_enter(&ds->ds_lock); 108 delta = parent_delta(ds, used); 109 ds->ds_phys->ds_used_bytes += used; 110 ds->ds_phys->ds_compressed_bytes += compressed; 111 ds->ds_phys->ds_uncompressed_bytes += uncompressed; 112 ds->ds_phys->ds_unique_bytes += used; 113 mutex_exit(&ds->ds_lock); 114 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 115 compressed, uncompressed, tx); 116 dsl_dir_transfer_space(ds->ds_dir, used - delta, 117 DD_USED_REFRSRV, DD_USED_HEAD, tx); 118 mutex_exit(&ds->ds_dir->dd_lock); 119 } 120 121 int 122 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio, 123 dmu_tx_t *tx) 124 { 125 int used = bp_get_dasize(tx->tx_pool->dp_spa, bp); 126 int compressed = BP_GET_PSIZE(bp); 127 int uncompressed = BP_GET_UCSIZE(bp); 128 129 ASSERT(pio != NULL); 130 ASSERT(dmu_tx_is_syncing(tx)); 131 /* No block pointer => nothing to free */ 132 if (BP_IS_HOLE(bp)) 133 return (0); 134 135 ASSERT(used > 0); 136 if (ds == NULL) { 137 int err; 138 /* 139 * Account for the meta-objset space in its placeholder 140 * dataset. 141 */ 142 err = dsl_free(pio, tx->tx_pool, 143 tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT); 144 ASSERT(err == 0); 145 146 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, 147 -used, -compressed, -uncompressed, tx); 148 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); 149 return (used); 150 } 151 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 152 153 ASSERT(!dsl_dataset_is_snapshot(ds)); 154 dmu_buf_will_dirty(ds->ds_dbuf, tx); 155 156 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { 157 int err; 158 int64_t delta; 159 160 dprintf_bp(bp, "freeing: %s", ""); 161 err = dsl_free(pio, tx->tx_pool, 162 tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT); 163 ASSERT(err == 0); 164 165 mutex_enter(&ds->ds_dir->dd_lock); 166 mutex_enter(&ds->ds_lock); 167 ASSERT(ds->ds_phys->ds_unique_bytes >= used || 168 !DS_UNIQUE_IS_ACCURATE(ds)); 169 delta = parent_delta(ds, -used); 170 ds->ds_phys->ds_unique_bytes -= used; 171 mutex_exit(&ds->ds_lock); 172 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 173 delta, -compressed, -uncompressed, tx); 174 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 175 DD_USED_REFRSRV, DD_USED_HEAD, tx); 176 mutex_exit(&ds->ds_dir->dd_lock); 177 } else { 178 dprintf_bp(bp, "putting on dead list: %s", ""); 179 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx)); 180 ASSERT3U(ds->ds_prev->ds_object, ==, 181 ds->ds_phys->ds_prev_snap_obj); 182 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); 183 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 184 if (ds->ds_prev->ds_phys->ds_next_snap_obj == 185 ds->ds_object && bp->blk_birth > 186 ds->ds_prev->ds_phys->ds_prev_snap_txg) { 187 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 188 mutex_enter(&ds->ds_prev->ds_lock); 189 ds->ds_prev->ds_phys->ds_unique_bytes += used; 190 mutex_exit(&ds->ds_prev->ds_lock); 191 } 192 if (bp->blk_birth > ds->ds_origin_txg) { 193 dsl_dir_transfer_space(ds->ds_dir, used, 194 DD_USED_HEAD, DD_USED_SNAP, tx); 195 } 196 } 197 mutex_enter(&ds->ds_lock); 198 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used); 199 ds->ds_phys->ds_used_bytes -= used; 200 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); 201 ds->ds_phys->ds_compressed_bytes -= compressed; 202 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); 203 ds->ds_phys->ds_uncompressed_bytes -= uncompressed; 204 mutex_exit(&ds->ds_lock); 205 206 return (used); 207 } 208 209 uint64_t 210 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 211 { 212 uint64_t trysnap = 0; 213 214 if (ds == NULL) 215 return (0); 216 /* 217 * The snapshot creation could fail, but that would cause an 218 * incorrect FALSE return, which would only result in an 219 * overestimation of the amount of space that an operation would 220 * consume, which is OK. 221 * 222 * There's also a small window where we could miss a pending 223 * snapshot, because we could set the sync task in the quiescing 224 * phase. So this should only be used as a guess. 225 */ 226 if (ds->ds_trysnap_txg > 227 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 228 trysnap = ds->ds_trysnap_txg; 229 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap)); 230 } 231 232 int 233 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth) 234 { 235 return (blk_birth > dsl_dataset_prev_snap_txg(ds)); 236 } 237 238 /* ARGSUSED */ 239 static void 240 dsl_dataset_evict(dmu_buf_t *db, void *dsv) 241 { 242 dsl_dataset_t *ds = dsv; 243 244 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds)); 245 246 dprintf_ds(ds, "evicting %s\n", ""); 247 248 unique_remove(ds->ds_fsid_guid); 249 250 if (ds->ds_user_ptr != NULL) 251 ds->ds_user_evict_func(ds, ds->ds_user_ptr); 252 253 if (ds->ds_prev) { 254 dsl_dataset_drop_ref(ds->ds_prev, ds); 255 ds->ds_prev = NULL; 256 } 257 258 bplist_close(&ds->ds_deadlist); 259 if (ds->ds_dir) 260 dsl_dir_close(ds->ds_dir, ds); 261 262 ASSERT(!list_link_active(&ds->ds_synced_link)); 263 264 mutex_destroy(&ds->ds_lock); 265 mutex_destroy(&ds->ds_opening_lock); 266 mutex_destroy(&ds->ds_deadlist.bpl_lock); 267 rw_destroy(&ds->ds_rwlock); 268 cv_destroy(&ds->ds_exclusive_cv); 269 270 kmem_free(ds, sizeof (dsl_dataset_t)); 271 } 272 273 static int 274 dsl_dataset_get_snapname(dsl_dataset_t *ds) 275 { 276 dsl_dataset_phys_t *headphys; 277 int err; 278 dmu_buf_t *headdbuf; 279 dsl_pool_t *dp = ds->ds_dir->dd_pool; 280 objset_t *mos = dp->dp_meta_objset; 281 282 if (ds->ds_snapname[0]) 283 return (0); 284 if (ds->ds_phys->ds_next_snap_obj == 0) 285 return (0); 286 287 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, 288 FTAG, &headdbuf); 289 if (err) 290 return (err); 291 headphys = headdbuf->db_data; 292 err = zap_value_search(dp->dp_meta_objset, 293 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 294 dmu_buf_rele(headdbuf, FTAG); 295 return (err); 296 } 297 298 static int 299 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 300 { 301 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 302 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 303 matchtype_t mt; 304 int err; 305 306 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 307 mt = MT_FIRST; 308 else 309 mt = MT_EXACT; 310 311 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 312 value, mt, NULL, 0, NULL); 313 if (err == ENOTSUP && mt == MT_FIRST) 314 err = zap_lookup(mos, snapobj, name, 8, 1, value); 315 return (err); 316 } 317 318 static int 319 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx) 320 { 321 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 322 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 323 matchtype_t mt; 324 int err; 325 326 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 327 mt = MT_FIRST; 328 else 329 mt = MT_EXACT; 330 331 err = zap_remove_norm(mos, snapobj, name, mt, tx); 332 if (err == ENOTSUP && mt == MT_FIRST) 333 err = zap_remove(mos, snapobj, name, tx); 334 return (err); 335 } 336 337 static int 338 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, 339 dsl_dataset_t **dsp) 340 { 341 objset_t *mos = dp->dp_meta_objset; 342 dmu_buf_t *dbuf; 343 dsl_dataset_t *ds; 344 int err; 345 346 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 347 dsl_pool_sync_context(dp)); 348 349 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 350 if (err) 351 return (err); 352 ds = dmu_buf_get_user(dbuf); 353 if (ds == NULL) { 354 dsl_dataset_t *winner; 355 356 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 357 ds->ds_dbuf = dbuf; 358 ds->ds_object = dsobj; 359 ds->ds_phys = dbuf->db_data; 360 361 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 362 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 363 mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT, 364 NULL); 365 rw_init(&ds->ds_rwlock, 0, 0, 0); 366 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL); 367 368 err = bplist_open(&ds->ds_deadlist, 369 mos, ds->ds_phys->ds_deadlist_obj); 370 if (err == 0) { 371 err = dsl_dir_open_obj(dp, 372 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); 373 } 374 if (err) { 375 /* 376 * we don't really need to close the blist if we 377 * just opened it. 378 */ 379 mutex_destroy(&ds->ds_lock); 380 mutex_destroy(&ds->ds_opening_lock); 381 mutex_destroy(&ds->ds_deadlist.bpl_lock); 382 rw_destroy(&ds->ds_rwlock); 383 cv_destroy(&ds->ds_exclusive_cv); 384 kmem_free(ds, sizeof (dsl_dataset_t)); 385 dmu_buf_rele(dbuf, tag); 386 return (err); 387 } 388 389 if (!dsl_dataset_is_snapshot(ds)) { 390 ds->ds_snapname[0] = '\0'; 391 if (ds->ds_phys->ds_prev_snap_obj) { 392 err = dsl_dataset_get_ref(dp, 393 ds->ds_phys->ds_prev_snap_obj, 394 ds, &ds->ds_prev); 395 } 396 397 if (err == 0 && dsl_dir_is_clone(ds->ds_dir)) { 398 dsl_dataset_t *origin; 399 400 err = dsl_dataset_hold_obj(dp, 401 ds->ds_dir->dd_phys->dd_origin_obj, 402 FTAG, &origin); 403 if (err == 0) { 404 ds->ds_origin_txg = 405 origin->ds_phys->ds_creation_txg; 406 dsl_dataset_rele(origin, FTAG); 407 } 408 } 409 } else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) { 410 err = dsl_dataset_get_snapname(ds); 411 } 412 413 if (err == 0 && !dsl_dataset_is_snapshot(ds)) { 414 /* 415 * In sync context, we're called with either no lock 416 * or with the write lock. If we're not syncing, 417 * we're always called with the read lock held. 418 */ 419 boolean_t need_lock = 420 !RW_WRITE_HELD(&dp->dp_config_rwlock) && 421 dsl_pool_sync_context(dp); 422 423 if (need_lock) 424 rw_enter(&dp->dp_config_rwlock, RW_READER); 425 426 err = dsl_prop_get_ds(ds, 427 "refreservation", sizeof (uint64_t), 1, 428 &ds->ds_reserved, NULL); 429 if (err == 0) { 430 err = dsl_prop_get_ds(ds, 431 "refquota", sizeof (uint64_t), 1, 432 &ds->ds_quota, NULL); 433 } 434 435 if (need_lock) 436 rw_exit(&dp->dp_config_rwlock); 437 } else { 438 ds->ds_reserved = ds->ds_quota = 0; 439 } 440 441 if (err == 0) { 442 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys, 443 dsl_dataset_evict); 444 } 445 if (err || winner) { 446 bplist_close(&ds->ds_deadlist); 447 if (ds->ds_prev) 448 dsl_dataset_drop_ref(ds->ds_prev, ds); 449 dsl_dir_close(ds->ds_dir, ds); 450 mutex_destroy(&ds->ds_lock); 451 mutex_destroy(&ds->ds_opening_lock); 452 mutex_destroy(&ds->ds_deadlist.bpl_lock); 453 rw_destroy(&ds->ds_rwlock); 454 cv_destroy(&ds->ds_exclusive_cv); 455 kmem_free(ds, sizeof (dsl_dataset_t)); 456 if (err) { 457 dmu_buf_rele(dbuf, tag); 458 return (err); 459 } 460 ds = winner; 461 } else { 462 ds->ds_fsid_guid = 463 unique_insert(ds->ds_phys->ds_fsid_guid); 464 } 465 } 466 ASSERT3P(ds->ds_dbuf, ==, dbuf); 467 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); 468 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 || 469 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 470 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 471 mutex_enter(&ds->ds_lock); 472 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) { 473 mutex_exit(&ds->ds_lock); 474 dmu_buf_rele(ds->ds_dbuf, tag); 475 return (ENOENT); 476 } 477 mutex_exit(&ds->ds_lock); 478 *dsp = ds; 479 return (0); 480 } 481 482 static int 483 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag) 484 { 485 dsl_pool_t *dp = ds->ds_dir->dd_pool; 486 487 /* 488 * In syncing context we don't want the rwlock lock: there 489 * may be an existing writer waiting for sync phase to 490 * finish. We don't need to worry about such writers, since 491 * sync phase is single-threaded, so the writer can't be 492 * doing anything while we are active. 493 */ 494 if (dsl_pool_sync_context(dp)) { 495 ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); 496 return (0); 497 } 498 499 /* 500 * Normal users will hold the ds_rwlock as a READER until they 501 * are finished (i.e., call dsl_dataset_rele()). "Owners" will 502 * drop their READER lock after they set the ds_owner field. 503 * 504 * If the dataset is being destroyed, the destroy thread will 505 * obtain a WRITER lock for exclusive access after it's done its 506 * open-context work and then change the ds_owner to 507 * dsl_reaper once destruction is assured. So threads 508 * may block here temporarily, until the "destructability" of 509 * the dataset is determined. 510 */ 511 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock)); 512 mutex_enter(&ds->ds_lock); 513 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) { 514 rw_exit(&dp->dp_config_rwlock); 515 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock); 516 if (DSL_DATASET_IS_DESTROYED(ds)) { 517 mutex_exit(&ds->ds_lock); 518 dsl_dataset_drop_ref(ds, tag); 519 rw_enter(&dp->dp_config_rwlock, RW_READER); 520 return (ENOENT); 521 } 522 rw_enter(&dp->dp_config_rwlock, RW_READER); 523 } 524 mutex_exit(&ds->ds_lock); 525 return (0); 526 } 527 528 int 529 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 530 dsl_dataset_t **dsp) 531 { 532 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp); 533 534 if (err) 535 return (err); 536 return (dsl_dataset_hold_ref(*dsp, tag)); 537 } 538 539 int 540 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, int flags, void *owner, 541 dsl_dataset_t **dsp) 542 { 543 int err = dsl_dataset_hold_obj(dp, dsobj, owner, dsp); 544 545 ASSERT(DS_MODE_TYPE(flags) != DS_MODE_USER); 546 547 if (err) 548 return (err); 549 if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) { 550 dsl_dataset_rele(*dsp, owner); 551 return (EBUSY); 552 } 553 return (0); 554 } 555 556 int 557 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp) 558 { 559 dsl_dir_t *dd; 560 dsl_pool_t *dp; 561 const char *snapname; 562 uint64_t obj; 563 int err = 0; 564 565 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname); 566 if (err) 567 return (err); 568 569 dp = dd->dd_pool; 570 obj = dd->dd_phys->dd_head_dataset_obj; 571 rw_enter(&dp->dp_config_rwlock, RW_READER); 572 if (obj) 573 err = dsl_dataset_get_ref(dp, obj, tag, dsp); 574 else 575 err = ENOENT; 576 if (err) 577 goto out; 578 579 err = dsl_dataset_hold_ref(*dsp, tag); 580 581 /* we may be looking for a snapshot */ 582 if (err == 0 && snapname != NULL) { 583 dsl_dataset_t *ds = NULL; 584 585 if (*snapname++ != '@') { 586 dsl_dataset_rele(*dsp, tag); 587 err = ENOENT; 588 goto out; 589 } 590 591 dprintf("looking for snapshot '%s'\n", snapname); 592 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); 593 if (err == 0) 594 err = dsl_dataset_get_ref(dp, obj, tag, &ds); 595 dsl_dataset_rele(*dsp, tag); 596 597 ASSERT3U((err == 0), ==, (ds != NULL)); 598 599 if (ds) { 600 mutex_enter(&ds->ds_lock); 601 if (ds->ds_snapname[0] == 0) 602 (void) strlcpy(ds->ds_snapname, snapname, 603 sizeof (ds->ds_snapname)); 604 mutex_exit(&ds->ds_lock); 605 err = dsl_dataset_hold_ref(ds, tag); 606 *dsp = err ? NULL : ds; 607 } 608 } 609 out: 610 rw_exit(&dp->dp_config_rwlock); 611 dsl_dir_close(dd, FTAG); 612 return (err); 613 } 614 615 int 616 dsl_dataset_own(const char *name, int flags, void *owner, dsl_dataset_t **dsp) 617 { 618 int err = dsl_dataset_hold(name, owner, dsp); 619 if (err) 620 return (err); 621 if ((*dsp)->ds_phys->ds_num_children > 0 && 622 !DS_MODE_IS_READONLY(flags)) { 623 dsl_dataset_rele(*dsp, owner); 624 return (EROFS); 625 } 626 if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) { 627 dsl_dataset_rele(*dsp, owner); 628 return (EBUSY); 629 } 630 return (0); 631 } 632 633 void 634 dsl_dataset_name(dsl_dataset_t *ds, char *name) 635 { 636 if (ds == NULL) { 637 (void) strcpy(name, "mos"); 638 } else { 639 dsl_dir_name(ds->ds_dir, name); 640 VERIFY(0 == dsl_dataset_get_snapname(ds)); 641 if (ds->ds_snapname[0]) { 642 (void) strcat(name, "@"); 643 /* 644 * We use a "recursive" mutex so that we 645 * can call dprintf_ds() with ds_lock held. 646 */ 647 if (!MUTEX_HELD(&ds->ds_lock)) { 648 mutex_enter(&ds->ds_lock); 649 (void) strcat(name, ds->ds_snapname); 650 mutex_exit(&ds->ds_lock); 651 } else { 652 (void) strcat(name, ds->ds_snapname); 653 } 654 } 655 } 656 } 657 658 static int 659 dsl_dataset_namelen(dsl_dataset_t *ds) 660 { 661 int result; 662 663 if (ds == NULL) { 664 result = 3; /* "mos" */ 665 } else { 666 result = dsl_dir_namelen(ds->ds_dir); 667 VERIFY(0 == dsl_dataset_get_snapname(ds)); 668 if (ds->ds_snapname[0]) { 669 ++result; /* adding one for the @-sign */ 670 if (!MUTEX_HELD(&ds->ds_lock)) { 671 mutex_enter(&ds->ds_lock); 672 result += strlen(ds->ds_snapname); 673 mutex_exit(&ds->ds_lock); 674 } else { 675 result += strlen(ds->ds_snapname); 676 } 677 } 678 } 679 680 return (result); 681 } 682 683 void 684 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag) 685 { 686 dmu_buf_rele(ds->ds_dbuf, tag); 687 } 688 689 void 690 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 691 { 692 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) { 693 rw_exit(&ds->ds_rwlock); 694 } 695 dsl_dataset_drop_ref(ds, tag); 696 } 697 698 void 699 dsl_dataset_disown(dsl_dataset_t *ds, void *owner) 700 { 701 ASSERT((ds->ds_owner == owner && ds->ds_dbuf) || 702 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL)); 703 704 mutex_enter(&ds->ds_lock); 705 ds->ds_owner = NULL; 706 if (RW_WRITE_HELD(&ds->ds_rwlock)) { 707 rw_exit(&ds->ds_rwlock); 708 cv_broadcast(&ds->ds_exclusive_cv); 709 } 710 mutex_exit(&ds->ds_lock); 711 if (ds->ds_dbuf) 712 dsl_dataset_drop_ref(ds, owner); 713 else 714 dsl_dataset_evict(ds->ds_dbuf, ds); 715 } 716 717 boolean_t 718 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *owner) 719 { 720 boolean_t gotit = FALSE; 721 722 mutex_enter(&ds->ds_lock); 723 if (ds->ds_owner == NULL && 724 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) { 725 ds->ds_owner = owner; 726 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) 727 rw_exit(&ds->ds_rwlock); 728 gotit = TRUE; 729 } 730 mutex_exit(&ds->ds_lock); 731 return (gotit); 732 } 733 734 void 735 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner) 736 { 737 ASSERT3P(owner, ==, ds->ds_owner); 738 if (!RW_WRITE_HELD(&ds->ds_rwlock)) 739 rw_enter(&ds->ds_rwlock, RW_WRITER); 740 } 741 742 uint64_t 743 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 744 uint64_t flags, dmu_tx_t *tx) 745 { 746 dsl_pool_t *dp = dd->dd_pool; 747 dmu_buf_t *dbuf; 748 dsl_dataset_phys_t *dsphys; 749 uint64_t dsobj; 750 objset_t *mos = dp->dp_meta_objset; 751 752 if (origin == NULL) 753 origin = dp->dp_origin_snap; 754 755 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 756 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0); 757 ASSERT(dmu_tx_is_syncing(tx)); 758 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 759 760 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 761 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 762 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 763 dmu_buf_will_dirty(dbuf, tx); 764 dsphys = dbuf->db_data; 765 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 766 dsphys->ds_dir_obj = dd->dd_object; 767 dsphys->ds_flags = flags; 768 dsphys->ds_fsid_guid = unique_create(); 769 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 770 sizeof (dsphys->ds_guid)); 771 dsphys->ds_snapnames_zapobj = 772 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 773 DMU_OT_NONE, 0, tx); 774 dsphys->ds_creation_time = gethrestime_sec(); 775 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 776 dsphys->ds_deadlist_obj = 777 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); 778 779 if (origin) { 780 dsphys->ds_prev_snap_obj = origin->ds_object; 781 dsphys->ds_prev_snap_txg = 782 origin->ds_phys->ds_creation_txg; 783 dsphys->ds_used_bytes = 784 origin->ds_phys->ds_used_bytes; 785 dsphys->ds_compressed_bytes = 786 origin->ds_phys->ds_compressed_bytes; 787 dsphys->ds_uncompressed_bytes = 788 origin->ds_phys->ds_uncompressed_bytes; 789 dsphys->ds_bp = origin->ds_phys->ds_bp; 790 dsphys->ds_flags |= origin->ds_phys->ds_flags; 791 792 dmu_buf_will_dirty(origin->ds_dbuf, tx); 793 origin->ds_phys->ds_num_children++; 794 795 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 796 if (origin->ds_phys->ds_next_clones_obj == 0) { 797 origin->ds_phys->ds_next_clones_obj = 798 zap_create(mos, 799 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 800 } 801 VERIFY(0 == zap_add_int(mos, 802 origin->ds_phys->ds_next_clones_obj, 803 dsobj, tx)); 804 } 805 806 dmu_buf_will_dirty(dd->dd_dbuf, tx); 807 dd->dd_phys->dd_origin_obj = origin->ds_object; 808 } 809 810 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 811 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 812 813 dmu_buf_rele(dbuf, FTAG); 814 815 dmu_buf_will_dirty(dd->dd_dbuf, tx); 816 dd->dd_phys->dd_head_dataset_obj = dsobj; 817 818 return (dsobj); 819 } 820 821 uint64_t 822 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 823 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 824 { 825 dsl_pool_t *dp = pdd->dd_pool; 826 uint64_t dsobj, ddobj; 827 dsl_dir_t *dd; 828 829 ASSERT(lastname[0] != '@'); 830 831 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 832 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd)); 833 834 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx); 835 836 dsl_deleg_set_create_perms(dd, tx, cr); 837 838 dsl_dir_close(dd, FTAG); 839 840 return (dsobj); 841 } 842 843 struct destroyarg { 844 dsl_sync_task_group_t *dstg; 845 char *snapname; 846 char *failed; 847 }; 848 849 static int 850 dsl_snapshot_destroy_one(char *name, void *arg) 851 { 852 struct destroyarg *da = arg; 853 dsl_dataset_t *ds; 854 char *cp; 855 int err; 856 857 (void) strcat(name, "@"); 858 (void) strcat(name, da->snapname); 859 err = dsl_dataset_own(name, DS_MODE_READONLY | DS_MODE_INCONSISTENT, 860 da->dstg, &ds); 861 cp = strchr(name, '@'); 862 *cp = '\0'; 863 if (err == 0) { 864 dsl_dataset_make_exclusive(ds, da->dstg); 865 if (ds->ds_user_ptr) { 866 ds->ds_user_evict_func(ds, ds->ds_user_ptr); 867 ds->ds_user_ptr = NULL; 868 } 869 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check, 870 dsl_dataset_destroy_sync, ds, da->dstg, 0); 871 } else if (err == ENOENT) { 872 err = 0; 873 } else { 874 (void) strcpy(da->failed, name); 875 } 876 return (err); 877 } 878 879 /* 880 * Destroy 'snapname' in all descendants of 'fsname'. 881 */ 882 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy 883 int 884 dsl_snapshots_destroy(char *fsname, char *snapname) 885 { 886 int err; 887 struct destroyarg da; 888 dsl_sync_task_t *dst; 889 spa_t *spa; 890 891 err = spa_open(fsname, &spa, FTAG); 892 if (err) 893 return (err); 894 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 895 da.snapname = snapname; 896 da.failed = fsname; 897 898 err = dmu_objset_find(fsname, 899 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN); 900 901 if (err == 0) 902 err = dsl_sync_task_group_wait(da.dstg); 903 904 for (dst = list_head(&da.dstg->dstg_tasks); dst; 905 dst = list_next(&da.dstg->dstg_tasks, dst)) { 906 dsl_dataset_t *ds = dst->dst_arg1; 907 /* 908 * Return the file system name that triggered the error 909 */ 910 if (dst->dst_err) { 911 dsl_dataset_name(ds, fsname); 912 *strchr(fsname, '@') = '\0'; 913 } 914 dsl_dataset_disown(ds, da.dstg); 915 } 916 917 dsl_sync_task_group_destroy(da.dstg); 918 spa_close(spa, FTAG); 919 return (err); 920 } 921 922 /* 923 * ds must be opened as OWNER. On return (whether successful or not), 924 * ds will be closed and caller can no longer dereference it. 925 */ 926 int 927 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag) 928 { 929 int err; 930 dsl_sync_task_group_t *dstg; 931 objset_t *os; 932 dsl_dir_t *dd; 933 uint64_t obj; 934 935 if (dsl_dataset_is_snapshot(ds)) { 936 /* Destroying a snapshot is simpler */ 937 dsl_dataset_make_exclusive(ds, tag); 938 939 if (ds->ds_user_ptr) { 940 ds->ds_user_evict_func(ds, ds->ds_user_ptr); 941 ds->ds_user_ptr = NULL; 942 } 943 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 944 dsl_dataset_destroy_check, dsl_dataset_destroy_sync, 945 ds, tag, 0); 946 goto out; 947 } 948 949 dd = ds->ds_dir; 950 951 /* 952 * Check for errors and mark this ds as inconsistent, in 953 * case we crash while freeing the objects. 954 */ 955 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check, 956 dsl_dataset_destroy_begin_sync, ds, NULL, 0); 957 if (err) 958 goto out; 959 960 err = dmu_objset_open_ds(ds, DMU_OST_ANY, &os); 961 if (err) 962 goto out; 963 964 /* 965 * remove the objects in open context, so that we won't 966 * have too much to do in syncing context. 967 */ 968 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 969 ds->ds_phys->ds_prev_snap_txg)) { 970 /* 971 * Ignore errors, if there is not enough disk space 972 * we will deal with it in dsl_dataset_destroy_sync(). 973 */ 974 (void) dmu_free_object(os, obj); 975 } 976 977 dmu_objset_close(os); 978 if (err != ESRCH) 979 goto out; 980 981 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); 982 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd); 983 rw_exit(&dd->dd_pool->dp_config_rwlock); 984 985 if (err) 986 goto out; 987 988 if (ds->ds_user_ptr) { 989 /* 990 * We need to sync out all in-flight IO before we try 991 * to evict (the dataset evict func is trying to clear 992 * the cached entries for this dataset in the ARC). 993 */ 994 txg_wait_synced(dd->dd_pool, 0); 995 } 996 997 /* 998 * Blow away the dsl_dir + head dataset. 999 */ 1000 dsl_dataset_make_exclusive(ds, tag); 1001 if (ds->ds_user_ptr) { 1002 ds->ds_user_evict_func(ds, ds->ds_user_ptr); 1003 ds->ds_user_ptr = NULL; 1004 } 1005 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool); 1006 dsl_sync_task_create(dstg, dsl_dataset_destroy_check, 1007 dsl_dataset_destroy_sync, ds, tag, 0); 1008 dsl_sync_task_create(dstg, dsl_dir_destroy_check, 1009 dsl_dir_destroy_sync, dd, FTAG, 0); 1010 err = dsl_sync_task_group_wait(dstg); 1011 dsl_sync_task_group_destroy(dstg); 1012 /* if it is successful, dsl_dir_destroy_sync will close the dd */ 1013 if (err) 1014 dsl_dir_close(dd, FTAG); 1015 out: 1016 dsl_dataset_disown(ds, tag); 1017 return (err); 1018 } 1019 1020 int 1021 dsl_dataset_rollback(dsl_dataset_t *ds, dmu_objset_type_t ost) 1022 { 1023 int err; 1024 1025 ASSERT(ds->ds_owner); 1026 1027 dsl_dataset_make_exclusive(ds, ds->ds_owner); 1028 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 1029 dsl_dataset_rollback_check, dsl_dataset_rollback_sync, 1030 ds, &ost, 0); 1031 /* drop exclusive access */ 1032 mutex_enter(&ds->ds_lock); 1033 rw_exit(&ds->ds_rwlock); 1034 cv_broadcast(&ds->ds_exclusive_cv); 1035 mutex_exit(&ds->ds_lock); 1036 return (err); 1037 } 1038 1039 void * 1040 dsl_dataset_set_user_ptr(dsl_dataset_t *ds, 1041 void *p, dsl_dataset_evict_func_t func) 1042 { 1043 void *old; 1044 1045 mutex_enter(&ds->ds_lock); 1046 old = ds->ds_user_ptr; 1047 if (old == NULL) { 1048 ds->ds_user_ptr = p; 1049 ds->ds_user_evict_func = func; 1050 } 1051 mutex_exit(&ds->ds_lock); 1052 return (old); 1053 } 1054 1055 void * 1056 dsl_dataset_get_user_ptr(dsl_dataset_t *ds) 1057 { 1058 return (ds->ds_user_ptr); 1059 } 1060 1061 1062 blkptr_t * 1063 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 1064 { 1065 return (&ds->ds_phys->ds_bp); 1066 } 1067 1068 void 1069 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 1070 { 1071 ASSERT(dmu_tx_is_syncing(tx)); 1072 /* If it's the meta-objset, set dp_meta_rootbp */ 1073 if (ds == NULL) { 1074 tx->tx_pool->dp_meta_rootbp = *bp; 1075 } else { 1076 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1077 ds->ds_phys->ds_bp = *bp; 1078 } 1079 } 1080 1081 spa_t * 1082 dsl_dataset_get_spa(dsl_dataset_t *ds) 1083 { 1084 return (ds->ds_dir->dd_pool->dp_spa); 1085 } 1086 1087 void 1088 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1089 { 1090 dsl_pool_t *dp; 1091 1092 if (ds == NULL) /* this is the meta-objset */ 1093 return; 1094 1095 ASSERT(ds->ds_user_ptr != NULL); 1096 1097 if (ds->ds_phys->ds_next_snap_obj != 0) 1098 panic("dirtying snapshot!"); 1099 1100 dp = ds->ds_dir->dd_pool; 1101 1102 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) { 1103 /* up the hold count until we can be written out */ 1104 dmu_buf_add_ref(ds->ds_dbuf, ds); 1105 } 1106 } 1107 1108 /* 1109 * The unique space in the head dataset can be calculated by subtracting 1110 * the space used in the most recent snapshot, that is still being used 1111 * in this file system, from the space currently in use. To figure out 1112 * the space in the most recent snapshot still in use, we need to take 1113 * the total space used in the snapshot and subtract out the space that 1114 * has been freed up since the snapshot was taken. 1115 */ 1116 static void 1117 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 1118 { 1119 uint64_t mrs_used; 1120 uint64_t dlused, dlcomp, dluncomp; 1121 1122 ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj); 1123 1124 if (ds->ds_phys->ds_prev_snap_obj != 0) 1125 mrs_used = ds->ds_prev->ds_phys->ds_used_bytes; 1126 else 1127 mrs_used = 0; 1128 1129 VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp, 1130 &dluncomp)); 1131 1132 ASSERT3U(dlused, <=, mrs_used); 1133 ds->ds_phys->ds_unique_bytes = 1134 ds->ds_phys->ds_used_bytes - (mrs_used - dlused); 1135 1136 if (!DS_UNIQUE_IS_ACCURATE(ds) && 1137 spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1138 SPA_VERSION_UNIQUE_ACCURATE) 1139 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1140 } 1141 1142 static uint64_t 1143 dsl_dataset_unique(dsl_dataset_t *ds) 1144 { 1145 if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds)) 1146 dsl_dataset_recalc_head_uniq(ds); 1147 1148 return (ds->ds_phys->ds_unique_bytes); 1149 } 1150 1151 struct killarg { 1152 dsl_dataset_t *ds; 1153 zio_t *zio; 1154 dmu_tx_t *tx; 1155 }; 1156 1157 /* ARGSUSED */ 1158 static int 1159 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg) 1160 { 1161 struct killarg *ka = arg; 1162 blkptr_t *bp = &bc->bc_blkptr; 1163 1164 ASSERT3U(bc->bc_errno, ==, 0); 1165 1166 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg); 1167 (void) dsl_dataset_block_kill(ka->ds, bp, ka->zio, ka->tx); 1168 1169 return (0); 1170 } 1171 1172 /* ARGSUSED */ 1173 static int 1174 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx) 1175 { 1176 dsl_dataset_t *ds = arg1; 1177 dmu_objset_type_t *ost = arg2; 1178 1179 /* 1180 * We can only roll back to emptyness if it is a ZPL objset. 1181 */ 1182 if (*ost != DMU_OST_ZFS && ds->ds_phys->ds_prev_snap_txg == 0) 1183 return (EINVAL); 1184 1185 /* 1186 * This must not be a snapshot. 1187 */ 1188 if (ds->ds_phys->ds_next_snap_obj != 0) 1189 return (EINVAL); 1190 1191 /* 1192 * If we made changes this txg, traverse_dsl_dataset won't find 1193 * them. Try again. 1194 */ 1195 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) 1196 return (EAGAIN); 1197 1198 return (0); 1199 } 1200 1201 /* ARGSUSED */ 1202 static void 1203 dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 1204 { 1205 dsl_dataset_t *ds = arg1; 1206 dmu_objset_type_t *ost = arg2; 1207 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1208 1209 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1210 1211 /* 1212 * Before the roll back destroy the zil. 1213 */ 1214 if (ds->ds_user_ptr != NULL) { 1215 zil_rollback_destroy( 1216 ((objset_impl_t *)ds->ds_user_ptr)->os_zil, tx); 1217 1218 /* 1219 * We need to make sure that the objset_impl_t is reopened after 1220 * we do the rollback, otherwise it will have the wrong 1221 * objset_phys_t. Normally this would happen when this 1222 * dataset-open is closed, thus causing the 1223 * dataset to be immediately evicted. But when doing "zfs recv 1224 * -F", we reopen the objset before that, so that there is no 1225 * window where the dataset is closed and inconsistent. 1226 */ 1227 ds->ds_user_evict_func(ds, ds->ds_user_ptr); 1228 ds->ds_user_ptr = NULL; 1229 } 1230 1231 /* Transfer space that was freed since last snap back to the head. */ 1232 { 1233 uint64_t used; 1234 1235 VERIFY(0 == bplist_space_birthrange(&ds->ds_deadlist, 1236 ds->ds_origin_txg, UINT64_MAX, &used)); 1237 dsl_dir_transfer_space(ds->ds_dir, used, 1238 DD_USED_SNAP, DD_USED_HEAD, tx); 1239 } 1240 1241 /* Zero out the deadlist. */ 1242 bplist_close(&ds->ds_deadlist); 1243 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); 1244 ds->ds_phys->ds_deadlist_obj = 1245 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); 1246 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos, 1247 ds->ds_phys->ds_deadlist_obj)); 1248 1249 { 1250 /* Free blkptrs that we gave birth to */ 1251 zio_t *zio; 1252 struct killarg ka; 1253 1254 zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL, 1255 ZIO_FLAG_MUSTSUCCEED); 1256 ka.ds = ds; 1257 ka.zio = zio; 1258 ka.tx = tx; 1259 (void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg, 1260 ADVANCE_POST, kill_blkptr, &ka); 1261 (void) zio_wait(zio); 1262 } 1263 1264 ASSERT(!(ds->ds_phys->ds_flags & DS_FLAG_UNIQUE_ACCURATE) || 1265 ds->ds_phys->ds_unique_bytes == 0); 1266 1267 if (ds->ds_prev && ds->ds_prev != ds->ds_dir->dd_pool->dp_origin_snap) { 1268 /* Change our contents to that of the prev snapshot */ 1269 1270 ASSERT3U(ds->ds_prev->ds_object, ==, 1271 ds->ds_phys->ds_prev_snap_obj); 1272 ASSERT3U(ds->ds_phys->ds_used_bytes, <=, 1273 ds->ds_prev->ds_phys->ds_used_bytes); 1274 1275 ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp; 1276 ds->ds_phys->ds_used_bytes = 1277 ds->ds_prev->ds_phys->ds_used_bytes; 1278 ds->ds_phys->ds_compressed_bytes = 1279 ds->ds_prev->ds_phys->ds_compressed_bytes; 1280 ds->ds_phys->ds_uncompressed_bytes = 1281 ds->ds_prev->ds_phys->ds_uncompressed_bytes; 1282 ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags; 1283 1284 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { 1285 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1286 ds->ds_prev->ds_phys->ds_unique_bytes = 0; 1287 } 1288 } else { 1289 objset_impl_t *osi; 1290 1291 ASSERT3U(ds->ds_phys->ds_used_bytes, ==, 0); 1292 ASSERT3U(ds->ds_phys->ds_compressed_bytes, ==, 0); 1293 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, ==, 0); 1294 1295 bzero(&ds->ds_phys->ds_bp, sizeof (blkptr_t)); 1296 ds->ds_phys->ds_flags = 0; 1297 ds->ds_phys->ds_unique_bytes = 0; 1298 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 1299 SPA_VERSION_UNIQUE_ACCURATE) 1300 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1301 1302 osi = dmu_objset_create_impl(ds->ds_dir->dd_pool->dp_spa, ds, 1303 &ds->ds_phys->ds_bp, *ost, tx); 1304 #ifdef _KERNEL 1305 zfs_create_fs(&osi->os, kcred, NULL, tx); 1306 #endif 1307 } 1308 1309 spa_history_internal_log(LOG_DS_ROLLBACK, ds->ds_dir->dd_pool->dp_spa, 1310 tx, cr, "dataset = %llu", ds->ds_object); 1311 } 1312 1313 /* ARGSUSED */ 1314 static int 1315 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx) 1316 { 1317 dsl_dataset_t *ds = arg1; 1318 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1319 uint64_t count; 1320 int err; 1321 1322 /* 1323 * Can't delete a head dataset if there are snapshots of it. 1324 * (Except if the only snapshots are from the branch we cloned 1325 * from.) 1326 */ 1327 if (ds->ds_prev != NULL && 1328 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) 1329 return (EINVAL); 1330 1331 /* 1332 * This is really a dsl_dir thing, but check it here so that 1333 * we'll be less likely to leave this dataset inconsistent & 1334 * nearly destroyed. 1335 */ 1336 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count); 1337 if (err) 1338 return (err); 1339 if (count != 0) 1340 return (EEXIST); 1341 1342 return (0); 1343 } 1344 1345 /* ARGSUSED */ 1346 static void 1347 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 1348 { 1349 dsl_dataset_t *ds = arg1; 1350 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1351 1352 /* Mark it as inconsistent on-disk, in case we crash */ 1353 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1354 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; 1355 1356 spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx, 1357 cr, "dataset = %llu", ds->ds_object); 1358 } 1359 1360 /* ARGSUSED */ 1361 int 1362 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) 1363 { 1364 dsl_dataset_t *ds = arg1; 1365 1366 /* we have an owner hold, so noone else can destroy us */ 1367 ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); 1368 1369 /* Can't delete a branch point. */ 1370 if (ds->ds_phys->ds_num_children > 1) 1371 return (EEXIST); 1372 1373 /* 1374 * Can't delete a head dataset if there are snapshots of it. 1375 * (Except if the only snapshots are from the branch we cloned 1376 * from.) 1377 */ 1378 if (ds->ds_prev != NULL && 1379 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) 1380 return (EINVAL); 1381 1382 /* 1383 * If we made changes this txg, traverse_dsl_dataset won't find 1384 * them. Try again. 1385 */ 1386 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) 1387 return (EAGAIN); 1388 1389 /* XXX we should do some i/o error checking... */ 1390 return (0); 1391 } 1392 1393 struct refsarg { 1394 kmutex_t lock; 1395 boolean_t gone; 1396 kcondvar_t cv; 1397 }; 1398 1399 /* ARGSUSED */ 1400 static void 1401 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv) 1402 { 1403 struct refsarg *arg = argv; 1404 1405 mutex_enter(&arg->lock); 1406 arg->gone = TRUE; 1407 cv_signal(&arg->cv); 1408 mutex_exit(&arg->lock); 1409 } 1410 1411 static void 1412 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag) 1413 { 1414 struct refsarg arg; 1415 1416 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL); 1417 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL); 1418 arg.gone = FALSE; 1419 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys, 1420 dsl_dataset_refs_gone); 1421 dmu_buf_rele(ds->ds_dbuf, tag); 1422 mutex_enter(&arg.lock); 1423 while (!arg.gone) 1424 cv_wait(&arg.cv, &arg.lock); 1425 ASSERT(arg.gone); 1426 mutex_exit(&arg.lock); 1427 ds->ds_dbuf = NULL; 1428 ds->ds_phys = NULL; 1429 mutex_destroy(&arg.lock); 1430 cv_destroy(&arg.cv); 1431 } 1432 1433 void 1434 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx) 1435 { 1436 dsl_dataset_t *ds = arg1; 1437 zio_t *zio; 1438 int err; 1439 int after_branch_point = FALSE; 1440 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1441 objset_t *mos = dp->dp_meta_objset; 1442 dsl_dataset_t *ds_prev = NULL; 1443 uint64_t obj; 1444 1445 ASSERT(ds->ds_owner); 1446 ASSERT3U(ds->ds_phys->ds_num_children, <=, 1); 1447 ASSERT(ds->ds_prev == NULL || 1448 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object); 1449 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); 1450 1451 /* signal any waiters that this dataset is going away */ 1452 mutex_enter(&ds->ds_lock); 1453 ds->ds_owner = dsl_reaper; 1454 cv_broadcast(&ds->ds_exclusive_cv); 1455 mutex_exit(&ds->ds_lock); 1456 1457 /* Remove our reservation */ 1458 if (ds->ds_reserved != 0) { 1459 uint64_t val = 0; 1460 dsl_dataset_set_reservation_sync(ds, &val, cr, tx); 1461 ASSERT3U(ds->ds_reserved, ==, 0); 1462 } 1463 1464 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); 1465 1466 dsl_pool_ds_destroyed(ds, tx); 1467 1468 obj = ds->ds_object; 1469 1470 if (ds->ds_phys->ds_prev_snap_obj != 0) { 1471 if (ds->ds_prev) { 1472 ds_prev = ds->ds_prev; 1473 } else { 1474 VERIFY(0 == dsl_dataset_hold_obj(dp, 1475 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev)); 1476 } 1477 after_branch_point = 1478 (ds_prev->ds_phys->ds_next_snap_obj != obj); 1479 1480 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); 1481 if (after_branch_point && 1482 ds_prev->ds_phys->ds_next_clones_obj != 0) { 1483 VERIFY(0 == zap_remove_int(mos, 1484 ds_prev->ds_phys->ds_next_clones_obj, obj, tx)); 1485 if (ds->ds_phys->ds_next_snap_obj != 0) { 1486 VERIFY(0 == zap_add_int(mos, 1487 ds_prev->ds_phys->ds_next_clones_obj, 1488 ds->ds_phys->ds_next_snap_obj, tx)); 1489 } 1490 } 1491 if (after_branch_point && 1492 ds->ds_phys->ds_next_snap_obj == 0) { 1493 /* This clone is toast. */ 1494 ASSERT(ds_prev->ds_phys->ds_num_children > 1); 1495 ds_prev->ds_phys->ds_num_children--; 1496 } else if (!after_branch_point) { 1497 ds_prev->ds_phys->ds_next_snap_obj = 1498 ds->ds_phys->ds_next_snap_obj; 1499 } 1500 } 1501 1502 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1503 1504 if (ds->ds_phys->ds_next_snap_obj != 0) { 1505 blkptr_t bp; 1506 dsl_dataset_t *ds_next; 1507 uint64_t itor = 0; 1508 uint64_t old_unique; 1509 int64_t used = 0, compressed = 0, uncompressed = 0; 1510 1511 VERIFY(0 == dsl_dataset_hold_obj(dp, 1512 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next)); 1513 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); 1514 1515 old_unique = dsl_dataset_unique(ds_next); 1516 1517 dmu_buf_will_dirty(ds_next->ds_dbuf, tx); 1518 ds_next->ds_phys->ds_prev_snap_obj = 1519 ds->ds_phys->ds_prev_snap_obj; 1520 ds_next->ds_phys->ds_prev_snap_txg = 1521 ds->ds_phys->ds_prev_snap_txg; 1522 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1523 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); 1524 1525 /* 1526 * Transfer to our deadlist (which will become next's 1527 * new deadlist) any entries from next's current 1528 * deadlist which were born before prev, and free the 1529 * other entries. 1530 * 1531 * XXX we're doing this long task with the config lock held 1532 */ 1533 while (bplist_iterate(&ds_next->ds_deadlist, &itor, &bp) == 0) { 1534 if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) { 1535 VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, 1536 &bp, tx)); 1537 if (ds_prev && !after_branch_point && 1538 bp.blk_birth > 1539 ds_prev->ds_phys->ds_prev_snap_txg) { 1540 ds_prev->ds_phys->ds_unique_bytes += 1541 bp_get_dasize(dp->dp_spa, &bp); 1542 } 1543 } else { 1544 used += bp_get_dasize(dp->dp_spa, &bp); 1545 compressed += BP_GET_PSIZE(&bp); 1546 uncompressed += BP_GET_UCSIZE(&bp); 1547 /* XXX check return value? */ 1548 (void) dsl_free(zio, dp, tx->tx_txg, 1549 &bp, NULL, NULL, ARC_NOWAIT); 1550 } 1551 } 1552 1553 ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes); 1554 1555 /* change snapused */ 1556 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, 1557 -used, -compressed, -uncompressed, tx); 1558 1559 /* free next's deadlist */ 1560 bplist_close(&ds_next->ds_deadlist); 1561 bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx); 1562 1563 /* set next's deadlist to our deadlist */ 1564 bplist_close(&ds->ds_deadlist); 1565 ds_next->ds_phys->ds_deadlist_obj = 1566 ds->ds_phys->ds_deadlist_obj; 1567 VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos, 1568 ds_next->ds_phys->ds_deadlist_obj)); 1569 ds->ds_phys->ds_deadlist_obj = 0; 1570 1571 if (ds_next->ds_phys->ds_next_snap_obj != 0) { 1572 /* 1573 * Update next's unique to include blocks which 1574 * were previously shared by only this snapshot 1575 * and it. Those blocks will be born after the 1576 * prev snap and before this snap, and will have 1577 * died after the next snap and before the one 1578 * after that (ie. be on the snap after next's 1579 * deadlist). 1580 * 1581 * XXX we're doing this long task with the 1582 * config lock held 1583 */ 1584 dsl_dataset_t *ds_after_next; 1585 uint64_t space; 1586 1587 VERIFY(0 == dsl_dataset_hold_obj(dp, 1588 ds_next->ds_phys->ds_next_snap_obj, 1589 FTAG, &ds_after_next)); 1590 1591 VERIFY(0 == 1592 bplist_space_birthrange(&ds_after_next->ds_deadlist, 1593 ds->ds_phys->ds_prev_snap_txg, 1594 ds->ds_phys->ds_creation_txg, &space)); 1595 ds_next->ds_phys->ds_unique_bytes += space; 1596 1597 dsl_dataset_rele(ds_after_next, FTAG); 1598 ASSERT3P(ds_next->ds_prev, ==, NULL); 1599 } else { 1600 ASSERT3P(ds_next->ds_prev, ==, ds); 1601 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next); 1602 ds_next->ds_prev = NULL; 1603 if (ds_prev) { 1604 VERIFY(0 == dsl_dataset_get_ref(dp, 1605 ds->ds_phys->ds_prev_snap_obj, 1606 ds_next, &ds_next->ds_prev)); 1607 } 1608 1609 dsl_dataset_recalc_head_uniq(ds_next); 1610 1611 /* 1612 * Reduce the amount of our unconsmed refreservation 1613 * being charged to our parent by the amount of 1614 * new unique data we have gained. 1615 */ 1616 if (old_unique < ds_next->ds_reserved) { 1617 int64_t mrsdelta; 1618 uint64_t new_unique = 1619 ds_next->ds_phys->ds_unique_bytes; 1620 1621 ASSERT(old_unique <= new_unique); 1622 mrsdelta = MIN(new_unique - old_unique, 1623 ds_next->ds_reserved - old_unique); 1624 dsl_dir_diduse_space(ds->ds_dir, 1625 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx); 1626 } 1627 } 1628 dsl_dataset_rele(ds_next, FTAG); 1629 } else { 1630 /* 1631 * There's no next snapshot, so this is a head dataset. 1632 * Destroy the deadlist. Unless it's a clone, the 1633 * deadlist should be empty. (If it's a clone, it's 1634 * safe to ignore the deadlist contents.) 1635 */ 1636 struct killarg ka; 1637 1638 ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist)); 1639 bplist_close(&ds->ds_deadlist); 1640 bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx); 1641 ds->ds_phys->ds_deadlist_obj = 0; 1642 1643 /* 1644 * Free everything that we point to (that's born after 1645 * the previous snapshot, if we are a clone) 1646 * 1647 * NB: this should be very quick, because we already 1648 * freed all the objects in open context. 1649 */ 1650 ka.ds = ds; 1651 ka.zio = zio; 1652 ka.tx = tx; 1653 err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg, 1654 ADVANCE_POST, kill_blkptr, &ka); 1655 ASSERT3U(err, ==, 0); 1656 ASSERT(spa_version(dp->dp_spa) < SPA_VERSION_UNIQUE_ACCURATE || 1657 ds->ds_phys->ds_unique_bytes == 0); 1658 } 1659 1660 err = zio_wait(zio); 1661 ASSERT3U(err, ==, 0); 1662 1663 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) { 1664 /* Erase the link in the dir */ 1665 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 1666 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0; 1667 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0); 1668 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx); 1669 ASSERT(err == 0); 1670 } else { 1671 /* remove from snapshot namespace */ 1672 dsl_dataset_t *ds_head; 1673 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0); 1674 VERIFY(0 == dsl_dataset_hold_obj(dp, 1675 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head)); 1676 VERIFY(0 == dsl_dataset_get_snapname(ds)); 1677 #ifdef ZFS_DEBUG 1678 { 1679 uint64_t val; 1680 1681 err = dsl_dataset_snap_lookup(ds_head, 1682 ds->ds_snapname, &val); 1683 ASSERT3U(err, ==, 0); 1684 ASSERT3U(val, ==, obj); 1685 } 1686 #endif 1687 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx); 1688 ASSERT(err == 0); 1689 dsl_dataset_rele(ds_head, FTAG); 1690 } 1691 1692 if (ds_prev && ds->ds_prev != ds_prev) 1693 dsl_dataset_rele(ds_prev, FTAG); 1694 1695 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx); 1696 spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx, 1697 cr, "dataset = %llu", ds->ds_object); 1698 1699 if (ds->ds_phys->ds_next_clones_obj != 0) { 1700 uint64_t count; 1701 ASSERT(0 == zap_count(mos, 1702 ds->ds_phys->ds_next_clones_obj, &count) && count == 0); 1703 VERIFY(0 == dmu_object_free(mos, 1704 ds->ds_phys->ds_next_clones_obj, tx)); 1705 } 1706 if (ds->ds_phys->ds_props_obj != 0) 1707 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx)); 1708 dsl_dir_close(ds->ds_dir, ds); 1709 ds->ds_dir = NULL; 1710 dsl_dataset_drain_refs(ds, tag); 1711 VERIFY(0 == dmu_object_free(mos, obj, tx)); 1712 } 1713 1714 static int 1715 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1716 { 1717 uint64_t asize; 1718 1719 if (!dmu_tx_is_syncing(tx)) 1720 return (0); 1721 1722 /* 1723 * If there's an fs-only reservation, any blocks that might become 1724 * owned by the snapshot dataset must be accommodated by space 1725 * outside of the reservation. 1726 */ 1727 asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved); 1728 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE)) 1729 return (ENOSPC); 1730 1731 /* 1732 * Propogate any reserved space for this snapshot to other 1733 * snapshot checks in this sync group. 1734 */ 1735 if (asize > 0) 1736 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1737 1738 return (0); 1739 } 1740 1741 /* ARGSUSED */ 1742 int 1743 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx) 1744 { 1745 dsl_dataset_t *ds = arg1; 1746 const char *snapname = arg2; 1747 int err; 1748 uint64_t value; 1749 1750 /* 1751 * We don't allow multiple snapshots of the same txg. If there 1752 * is already one, try again. 1753 */ 1754 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) 1755 return (EAGAIN); 1756 1757 /* 1758 * Check for conflicting name snapshot name. 1759 */ 1760 err = dsl_dataset_snap_lookup(ds, snapname, &value); 1761 if (err == 0) 1762 return (EEXIST); 1763 if (err != ENOENT) 1764 return (err); 1765 1766 /* 1767 * Check that the dataset's name is not too long. Name consists 1768 * of the dataset's length + 1 for the @-sign + snapshot name's length 1769 */ 1770 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN) 1771 return (ENAMETOOLONG); 1772 1773 err = dsl_dataset_snapshot_reserve_space(ds, tx); 1774 if (err) 1775 return (err); 1776 1777 ds->ds_trysnap_txg = tx->tx_txg; 1778 return (0); 1779 } 1780 1781 void 1782 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 1783 { 1784 dsl_dataset_t *ds = arg1; 1785 const char *snapname = arg2; 1786 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1787 dmu_buf_t *dbuf; 1788 dsl_dataset_phys_t *dsphys; 1789 uint64_t dsobj, crtxg; 1790 objset_t *mos = dp->dp_meta_objset; 1791 int err; 1792 1793 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); 1794 1795 /* 1796 * The origin's ds_creation_txg has to be < TXG_INITIAL 1797 */ 1798 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1799 crtxg = 1; 1800 else 1801 crtxg = tx->tx_txg; 1802 1803 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1804 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1805 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1806 dmu_buf_will_dirty(dbuf, tx); 1807 dsphys = dbuf->db_data; 1808 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1809 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1810 dsphys->ds_fsid_guid = unique_create(); 1811 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1812 sizeof (dsphys->ds_guid)); 1813 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; 1814 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; 1815 dsphys->ds_next_snap_obj = ds->ds_object; 1816 dsphys->ds_num_children = 1; 1817 dsphys->ds_creation_time = gethrestime_sec(); 1818 dsphys->ds_creation_txg = crtxg; 1819 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; 1820 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes; 1821 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; 1822 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; 1823 dsphys->ds_flags = ds->ds_phys->ds_flags; 1824 dsphys->ds_bp = ds->ds_phys->ds_bp; 1825 dmu_buf_rele(dbuf, FTAG); 1826 1827 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); 1828 if (ds->ds_prev) { 1829 uint64_t next_clones_obj = 1830 ds->ds_prev->ds_phys->ds_next_clones_obj; 1831 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == 1832 ds->ds_object || 1833 ds->ds_prev->ds_phys->ds_num_children > 1); 1834 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { 1835 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1836 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1837 ds->ds_prev->ds_phys->ds_creation_txg); 1838 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; 1839 } else if (next_clones_obj != 0) { 1840 VERIFY3U(0, ==, zap_remove_int(mos, 1841 next_clones_obj, dsphys->ds_next_snap_obj, tx)); 1842 VERIFY3U(0, ==, zap_add_int(mos, 1843 next_clones_obj, dsobj, tx)); 1844 } 1845 } 1846 1847 /* 1848 * If we have a reference-reservation on this dataset, we will 1849 * need to increase the amount of refreservation being charged 1850 * since our unique space is going to zero. 1851 */ 1852 if (ds->ds_reserved) { 1853 int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved); 1854 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1855 add, 0, 0, tx); 1856 } 1857 1858 bplist_close(&ds->ds_deadlist); 1859 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1860 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg); 1861 ds->ds_phys->ds_prev_snap_obj = dsobj; 1862 ds->ds_phys->ds_prev_snap_txg = crtxg; 1863 ds->ds_phys->ds_unique_bytes = 0; 1864 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1865 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1866 ds->ds_phys->ds_deadlist_obj = 1867 bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx); 1868 VERIFY(0 == bplist_open(&ds->ds_deadlist, mos, 1869 ds->ds_phys->ds_deadlist_obj)); 1870 1871 dprintf("snap '%s' -> obj %llu\n", snapname, dsobj); 1872 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, 1873 snapname, 8, 1, &dsobj, tx); 1874 ASSERT(err == 0); 1875 1876 if (ds->ds_prev) 1877 dsl_dataset_drop_ref(ds->ds_prev, ds); 1878 VERIFY(0 == dsl_dataset_get_ref(dp, 1879 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 1880 1881 dsl_pool_ds_snapshotted(ds, tx); 1882 1883 spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr, 1884 "dataset = %llu", dsobj); 1885 } 1886 1887 void 1888 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1889 { 1890 ASSERT(dmu_tx_is_syncing(tx)); 1891 ASSERT(ds->ds_user_ptr != NULL); 1892 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); 1893 1894 /* 1895 * in case we had to change ds_fsid_guid when we opened it, 1896 * sync it out now. 1897 */ 1898 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1899 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; 1900 1901 dsl_dir_dirty(ds->ds_dir, tx); 1902 dmu_objset_sync(ds->ds_user_ptr, zio, tx); 1903 } 1904 1905 void 1906 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1907 { 1908 uint64_t refd, avail, uobjs, aobjs; 1909 1910 dsl_dir_stats(ds->ds_dir, nv); 1911 1912 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 1913 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 1914 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 1915 1916 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 1917 ds->ds_phys->ds_creation_time); 1918 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 1919 ds->ds_phys->ds_creation_txg); 1920 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 1921 ds->ds_quota); 1922 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 1923 ds->ds_reserved); 1924 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 1925 ds->ds_phys->ds_guid); 1926 1927 if (ds->ds_phys->ds_next_snap_obj) { 1928 /* 1929 * This is a snapshot; override the dd's space used with 1930 * our unique space and compression ratio. 1931 */ 1932 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 1933 ds->ds_phys->ds_unique_bytes); 1934 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 1935 ds->ds_phys->ds_compressed_bytes == 0 ? 100 : 1936 (ds->ds_phys->ds_uncompressed_bytes * 100 / 1937 ds->ds_phys->ds_compressed_bytes)); 1938 } 1939 } 1940 1941 void 1942 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 1943 { 1944 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; 1945 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; 1946 stat->dds_guid = ds->ds_phys->ds_guid; 1947 if (ds->ds_phys->ds_next_snap_obj) { 1948 stat->dds_is_snapshot = B_TRUE; 1949 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; 1950 } 1951 1952 /* clone origin is really a dsl_dir thing... */ 1953 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); 1954 if (dsl_dir_is_clone(ds->ds_dir)) { 1955 dsl_dataset_t *ods; 1956 1957 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool, 1958 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods)); 1959 dsl_dataset_name(ods, stat->dds_origin); 1960 dsl_dataset_drop_ref(ods, FTAG); 1961 } 1962 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); 1963 } 1964 1965 uint64_t 1966 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 1967 { 1968 return (ds->ds_fsid_guid); 1969 } 1970 1971 void 1972 dsl_dataset_space(dsl_dataset_t *ds, 1973 uint64_t *refdbytesp, uint64_t *availbytesp, 1974 uint64_t *usedobjsp, uint64_t *availobjsp) 1975 { 1976 *refdbytesp = ds->ds_phys->ds_used_bytes; 1977 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 1978 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) 1979 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes; 1980 if (ds->ds_quota != 0) { 1981 /* 1982 * Adjust available bytes according to refquota 1983 */ 1984 if (*refdbytesp < ds->ds_quota) 1985 *availbytesp = MIN(*availbytesp, 1986 ds->ds_quota - *refdbytesp); 1987 else 1988 *availbytesp = 0; 1989 } 1990 *usedobjsp = ds->ds_phys->ds_bp.blk_fill; 1991 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 1992 } 1993 1994 boolean_t 1995 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds) 1996 { 1997 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1998 1999 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || 2000 dsl_pool_sync_context(dp)); 2001 if (ds->ds_prev == NULL) 2002 return (B_FALSE); 2003 if (ds->ds_phys->ds_bp.blk_birth > 2004 ds->ds_prev->ds_phys->ds_creation_txg) 2005 return (B_TRUE); 2006 return (B_FALSE); 2007 } 2008 2009 /* ARGSUSED */ 2010 static int 2011 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) 2012 { 2013 dsl_dataset_t *ds = arg1; 2014 char *newsnapname = arg2; 2015 dsl_dir_t *dd = ds->ds_dir; 2016 dsl_dataset_t *hds; 2017 uint64_t val; 2018 int err; 2019 2020 err = dsl_dataset_hold_obj(dd->dd_pool, 2021 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds); 2022 if (err) 2023 return (err); 2024 2025 /* new name better not be in use */ 2026 err = dsl_dataset_snap_lookup(hds, newsnapname, &val); 2027 dsl_dataset_rele(hds, FTAG); 2028 2029 if (err == 0) 2030 err = EEXIST; 2031 else if (err == ENOENT) 2032 err = 0; 2033 2034 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 2035 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN) 2036 err = ENAMETOOLONG; 2037 2038 return (err); 2039 } 2040 2041 static void 2042 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, 2043 cred_t *cr, dmu_tx_t *tx) 2044 { 2045 dsl_dataset_t *ds = arg1; 2046 const char *newsnapname = arg2; 2047 dsl_dir_t *dd = ds->ds_dir; 2048 objset_t *mos = dd->dd_pool->dp_meta_objset; 2049 dsl_dataset_t *hds; 2050 int err; 2051 2052 ASSERT(ds->ds_phys->ds_next_snap_obj != 0); 2053 2054 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, 2055 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds)); 2056 2057 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2058 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx); 2059 ASSERT3U(err, ==, 0); 2060 mutex_enter(&ds->ds_lock); 2061 (void) strcpy(ds->ds_snapname, newsnapname); 2062 mutex_exit(&ds->ds_lock); 2063 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj, 2064 ds->ds_snapname, 8, 1, &ds->ds_object, tx); 2065 ASSERT3U(err, ==, 0); 2066 2067 spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx, 2068 cr, "dataset = %llu", ds->ds_object); 2069 dsl_dataset_rele(hds, FTAG); 2070 } 2071 2072 struct renamesnaparg { 2073 dsl_sync_task_group_t *dstg; 2074 char failed[MAXPATHLEN]; 2075 char *oldsnap; 2076 char *newsnap; 2077 }; 2078 2079 static int 2080 dsl_snapshot_rename_one(char *name, void *arg) 2081 { 2082 struct renamesnaparg *ra = arg; 2083 dsl_dataset_t *ds = NULL; 2084 char *cp; 2085 int err; 2086 2087 cp = name + strlen(name); 2088 *cp = '@'; 2089 (void) strcpy(cp + 1, ra->oldsnap); 2090 2091 /* 2092 * For recursive snapshot renames the parent won't be changing 2093 * so we just pass name for both the to/from argument. 2094 */ 2095 err = zfs_secpolicy_rename_perms(name, name, CRED()); 2096 if (err == ENOENT) { 2097 return (0); 2098 } else if (err) { 2099 (void) strcpy(ra->failed, name); 2100 return (err); 2101 } 2102 2103 #ifdef _KERNEL 2104 /* 2105 * For all filesystems undergoing rename, we'll need to unmount it. 2106 */ 2107 (void) zfs_unmount_snap(name, NULL); 2108 #endif 2109 err = dsl_dataset_hold(name, ra->dstg, &ds); 2110 *cp = '\0'; 2111 if (err == ENOENT) { 2112 return (0); 2113 } else if (err) { 2114 (void) strcpy(ra->failed, name); 2115 return (err); 2116 } 2117 2118 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check, 2119 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0); 2120 2121 return (0); 2122 } 2123 2124 static int 2125 dsl_recursive_rename(char *oldname, const char *newname) 2126 { 2127 int err; 2128 struct renamesnaparg *ra; 2129 dsl_sync_task_t *dst; 2130 spa_t *spa; 2131 char *cp, *fsname = spa_strdup(oldname); 2132 int len = strlen(oldname); 2133 2134 /* truncate the snapshot name to get the fsname */ 2135 cp = strchr(fsname, '@'); 2136 *cp = '\0'; 2137 2138 err = spa_open(fsname, &spa, FTAG); 2139 if (err) { 2140 kmem_free(fsname, len + 1); 2141 return (err); 2142 } 2143 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP); 2144 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); 2145 2146 ra->oldsnap = strchr(oldname, '@') + 1; 2147 ra->newsnap = strchr(newname, '@') + 1; 2148 *ra->failed = '\0'; 2149 2150 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra, 2151 DS_FIND_CHILDREN); 2152 kmem_free(fsname, len + 1); 2153 2154 if (err == 0) { 2155 err = dsl_sync_task_group_wait(ra->dstg); 2156 } 2157 2158 for (dst = list_head(&ra->dstg->dstg_tasks); dst; 2159 dst = list_next(&ra->dstg->dstg_tasks, dst)) { 2160 dsl_dataset_t *ds = dst->dst_arg1; 2161 if (dst->dst_err) { 2162 dsl_dir_name(ds->ds_dir, ra->failed); 2163 (void) strcat(ra->failed, "@"); 2164 (void) strcat(ra->failed, ra->newsnap); 2165 } 2166 dsl_dataset_rele(ds, ra->dstg); 2167 } 2168 2169 if (err) 2170 (void) strcpy(oldname, ra->failed); 2171 2172 dsl_sync_task_group_destroy(ra->dstg); 2173 kmem_free(ra, sizeof (struct renamesnaparg)); 2174 spa_close(spa, FTAG); 2175 return (err); 2176 } 2177 2178 static int 2179 dsl_valid_rename(char *oldname, void *arg) 2180 { 2181 int delta = *(int *)arg; 2182 2183 if (strlen(oldname) + delta >= MAXNAMELEN) 2184 return (ENAMETOOLONG); 2185 2186 return (0); 2187 } 2188 2189 #pragma weak dmu_objset_rename = dsl_dataset_rename 2190 int 2191 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive) 2192 { 2193 dsl_dir_t *dd; 2194 dsl_dataset_t *ds; 2195 const char *tail; 2196 int err; 2197 2198 err = dsl_dir_open(oldname, FTAG, &dd, &tail); 2199 if (err) 2200 return (err); 2201 if (tail == NULL) { 2202 int delta = strlen(newname) - strlen(oldname); 2203 2204 /* if we're growing, validate child name lengths */ 2205 if (delta > 0) 2206 err = dmu_objset_find(oldname, dsl_valid_rename, 2207 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 2208 2209 if (!err) 2210 err = dsl_dir_rename(dd, newname); 2211 dsl_dir_close(dd, FTAG); 2212 return (err); 2213 } 2214 if (tail[0] != '@') { 2215 /* the name ended in a nonexistant component */ 2216 dsl_dir_close(dd, FTAG); 2217 return (ENOENT); 2218 } 2219 2220 dsl_dir_close(dd, FTAG); 2221 2222 /* new name must be snapshot in same filesystem */ 2223 tail = strchr(newname, '@'); 2224 if (tail == NULL) 2225 return (EINVAL); 2226 tail++; 2227 if (strncmp(oldname, newname, tail - newname) != 0) 2228 return (EXDEV); 2229 2230 if (recursive) { 2231 err = dsl_recursive_rename(oldname, newname); 2232 } else { 2233 err = dsl_dataset_hold(oldname, FTAG, &ds); 2234 if (err) 2235 return (err); 2236 2237 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 2238 dsl_dataset_snapshot_rename_check, 2239 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1); 2240 2241 dsl_dataset_rele(ds, FTAG); 2242 } 2243 2244 return (err); 2245 } 2246 2247 struct promotenode { 2248 list_node_t link; 2249 dsl_dataset_t *ds; 2250 }; 2251 2252 struct promotearg { 2253 list_t shared_snaps, origin_snaps, clone_snaps; 2254 dsl_dataset_t *origin_origin, *origin_head; 2255 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2256 }; 2257 2258 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2259 2260 /* ARGSUSED */ 2261 static int 2262 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) 2263 { 2264 dsl_dataset_t *hds = arg1; 2265 struct promotearg *pa = arg2; 2266 struct promotenode *snap = list_head(&pa->shared_snaps); 2267 dsl_dataset_t *origin_ds = snap->ds; 2268 int err; 2269 2270 /* Check that it is a real clone */ 2271 if (!dsl_dir_is_clone(hds->ds_dir)) 2272 return (EINVAL); 2273 2274 /* Since this is so expensive, don't do the preliminary check */ 2275 if (!dmu_tx_is_syncing(tx)) 2276 return (0); 2277 2278 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) 2279 return (EXDEV); 2280 2281 /* compute origin's new unique space */ 2282 snap = list_tail(&pa->clone_snaps); 2283 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2284 err = bplist_space_birthrange(&snap->ds->ds_deadlist, 2285 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, &pa->unique); 2286 if (err) 2287 return (err); 2288 2289 /* 2290 * Walk the snapshots that we are moving 2291 * 2292 * Compute space to transfer. Consider the incremental changes 2293 * to used for each snapshot: 2294 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2295 * So each snapshot gave birth to: 2296 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2297 * So a sequence would look like: 2298 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2299 * Which simplifies to: 2300 * uN + kN + kN-1 + ... + k1 + k0 2301 * Note however, if we stop before we reach the ORIGIN we get: 2302 * uN + kN + kN-1 + ... + kM - uM-1 2303 */ 2304 pa->used = origin_ds->ds_phys->ds_used_bytes; 2305 pa->comp = origin_ds->ds_phys->ds_compressed_bytes; 2306 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; 2307 for (snap = list_head(&pa->shared_snaps); snap; 2308 snap = list_next(&pa->shared_snaps, snap)) { 2309 uint64_t val, dlused, dlcomp, dluncomp; 2310 dsl_dataset_t *ds = snap->ds; 2311 2312 /* Check that the snapshot name does not conflict */ 2313 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2314 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2315 if (err == 0) 2316 return (EEXIST); 2317 if (err != ENOENT) 2318 return (err); 2319 2320 /* The very first snapshot does not have a deadlist */ 2321 if (ds->ds_phys->ds_prev_snap_obj == 0) 2322 continue; 2323 2324 if (err = bplist_space(&ds->ds_deadlist, 2325 &dlused, &dlcomp, &dluncomp)) 2326 return (err); 2327 pa->used += dlused; 2328 pa->comp += dlcomp; 2329 pa->uncomp += dluncomp; 2330 } 2331 2332 /* 2333 * If we are a clone of a clone then we never reached ORIGIN, 2334 * so we need to subtract out the clone origin's used space. 2335 */ 2336 if (pa->origin_origin) { 2337 pa->used -= pa->origin_origin->ds_phys->ds_used_bytes; 2338 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes; 2339 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes; 2340 } 2341 2342 /* Check that there is enough space here */ 2343 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2344 pa->used); 2345 if (err) 2346 return (err); 2347 2348 /* 2349 * Compute the amounts of space that will be used by snapshots 2350 * after the promotion (for both origin and clone). For each, 2351 * it is the amount of space that will be on all of their 2352 * deadlists (that was not born before their new origin). 2353 */ 2354 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2355 uint64_t space; 2356 2357 /* 2358 * Note, typically this will not be a clone of a clone, 2359 * so snap->ds->ds_origin_txg will be < TXG_INITIAL, so 2360 * these snaplist_space() -> bplist_space_birthrange() 2361 * calls will be fast because they do not have to 2362 * iterate over all bps. 2363 */ 2364 snap = list_head(&pa->origin_snaps); 2365 err = snaplist_space(&pa->shared_snaps, 2366 snap->ds->ds_origin_txg, &pa->cloneusedsnap); 2367 if (err) 2368 return (err); 2369 2370 err = snaplist_space(&pa->clone_snaps, 2371 snap->ds->ds_origin_txg, &space); 2372 if (err) 2373 return (err); 2374 pa->cloneusedsnap += space; 2375 } 2376 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2377 err = snaplist_space(&pa->origin_snaps, 2378 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap); 2379 if (err) 2380 return (err); 2381 } 2382 2383 return (0); 2384 } 2385 2386 static void 2387 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 2388 { 2389 dsl_dataset_t *hds = arg1; 2390 struct promotearg *pa = arg2; 2391 struct promotenode *snap = list_head(&pa->shared_snaps); 2392 dsl_dataset_t *origin_ds = snap->ds; 2393 dsl_dataset_t *origin_head; 2394 dsl_dir_t *dd = hds->ds_dir; 2395 dsl_pool_t *dp = hds->ds_dir->dd_pool; 2396 dsl_dir_t *odd = NULL; 2397 uint64_t oldnext_obj; 2398 int64_t delta; 2399 2400 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)); 2401 2402 snap = list_head(&pa->origin_snaps); 2403 origin_head = snap->ds; 2404 2405 /* 2406 * We need to explicitly open odd, since origin_ds's dd will be 2407 * changing. 2408 */ 2409 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object, 2410 NULL, FTAG, &odd)); 2411 2412 /* change origin's next snap */ 2413 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2414 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj; 2415 snap = list_tail(&pa->clone_snaps); 2416 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2417 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object; 2418 2419 /* change the origin's next clone */ 2420 if (origin_ds->ds_phys->ds_next_clones_obj) { 2421 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, 2422 origin_ds->ds_phys->ds_next_clones_obj, 2423 origin_ds->ds_phys->ds_next_snap_obj, tx)); 2424 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, 2425 origin_ds->ds_phys->ds_next_clones_obj, 2426 oldnext_obj, tx)); 2427 } 2428 2429 /* change origin */ 2430 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2431 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object); 2432 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj; 2433 hds->ds_origin_txg = origin_head->ds_origin_txg; 2434 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2435 odd->dd_phys->dd_origin_obj = origin_ds->ds_object; 2436 origin_head->ds_origin_txg = origin_ds->ds_phys->ds_creation_txg; 2437 2438 /* move snapshots to this dir */ 2439 for (snap = list_head(&pa->shared_snaps); snap; 2440 snap = list_next(&pa->shared_snaps, snap)) { 2441 dsl_dataset_t *ds = snap->ds; 2442 2443 /* unregister props as dsl_dir is changing */ 2444 if (ds->ds_user_ptr) { 2445 ds->ds_user_evict_func(ds, ds->ds_user_ptr); 2446 ds->ds_user_ptr = NULL; 2447 } 2448 /* move snap name entry */ 2449 VERIFY(0 == dsl_dataset_get_snapname(ds)); 2450 VERIFY(0 == dsl_dataset_snap_remove(origin_head, 2451 ds->ds_snapname, tx)); 2452 VERIFY(0 == zap_add(dp->dp_meta_objset, 2453 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, 2454 8, 1, &ds->ds_object, tx)); 2455 /* change containing dsl_dir */ 2456 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2457 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object); 2458 ds->ds_phys->ds_dir_obj = dd->dd_object; 2459 ASSERT3P(ds->ds_dir, ==, odd); 2460 dsl_dir_close(ds->ds_dir, ds); 2461 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object, 2462 NULL, ds, &ds->ds_dir)); 2463 2464 ASSERT3U(dsl_prop_numcb(ds), ==, 0); 2465 } 2466 2467 /* 2468 * Change space accounting. 2469 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2470 * both be valid, or both be 0 (resulting in delta == 0). This 2471 * is true for each of {clone,origin} independently. 2472 */ 2473 2474 delta = pa->cloneusedsnap - 2475 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2476 ASSERT3S(delta, >=, 0); 2477 ASSERT3U(pa->used, >=, delta); 2478 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2479 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2480 pa->used - delta, pa->comp, pa->uncomp, tx); 2481 2482 delta = pa->originusedsnap - 2483 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2484 ASSERT3S(delta, <=, 0); 2485 ASSERT3U(pa->used, >=, -delta); 2486 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2487 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2488 -pa->used - delta, -pa->comp, -pa->uncomp, tx); 2489 2490 origin_ds->ds_phys->ds_unique_bytes = pa->unique; 2491 2492 /* log history record */ 2493 spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx, 2494 cr, "dataset = %llu", hds->ds_object); 2495 2496 dsl_dir_close(odd, FTAG); 2497 } 2498 2499 static char *snaplist_tag = "snaplist"; 2500 /* 2501 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2502 * (exclusive) and last_obj (inclusive). The list will be in reverse 2503 * order (last_obj will be the list_head()). If first_obj == 0, do all 2504 * snapshots back to this dataset's origin. 2505 */ 2506 static int 2507 snaplist_make(dsl_pool_t *dp, boolean_t own, 2508 uint64_t first_obj, uint64_t last_obj, list_t *l) 2509 { 2510 uint64_t obj = last_obj; 2511 2512 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock)); 2513 2514 list_create(l, sizeof (struct promotenode), 2515 offsetof(struct promotenode, link)); 2516 2517 while (obj != first_obj) { 2518 dsl_dataset_t *ds; 2519 struct promotenode *snap; 2520 int err; 2521 2522 if (own) { 2523 err = dsl_dataset_own_obj(dp, obj, 2524 0, snaplist_tag, &ds); 2525 if (err == 0) 2526 dsl_dataset_make_exclusive(ds, snaplist_tag); 2527 } else { 2528 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds); 2529 } 2530 if (err == ENOENT) { 2531 /* lost race with snapshot destroy */ 2532 struct promotenode *last = list_tail(l); 2533 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj); 2534 obj = last->ds->ds_phys->ds_prev_snap_obj; 2535 continue; 2536 } else if (err) { 2537 return (err); 2538 } 2539 2540 if (first_obj == 0) 2541 first_obj = ds->ds_dir->dd_phys->dd_origin_obj; 2542 2543 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP); 2544 snap->ds = ds; 2545 list_insert_tail(l, snap); 2546 obj = ds->ds_phys->ds_prev_snap_obj; 2547 } 2548 2549 return (0); 2550 } 2551 2552 static int 2553 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2554 { 2555 struct promotenode *snap; 2556 2557 *spacep = 0; 2558 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2559 uint64_t used; 2560 int err = bplist_space_birthrange(&snap->ds->ds_deadlist, 2561 mintxg, UINT64_MAX, &used); 2562 if (err) 2563 return (err); 2564 *spacep += used; 2565 } 2566 return (0); 2567 } 2568 2569 static void 2570 snaplist_destroy(list_t *l, boolean_t own) 2571 { 2572 struct promotenode *snap; 2573 2574 if (!list_link_active(&l->list_head)) 2575 return; 2576 2577 while ((snap = list_tail(l)) != NULL) { 2578 list_remove(l, snap); 2579 if (own) 2580 dsl_dataset_disown(snap->ds, snaplist_tag); 2581 else 2582 dsl_dataset_rele(snap->ds, snaplist_tag); 2583 kmem_free(snap, sizeof (struct promotenode)); 2584 } 2585 list_destroy(l); 2586 } 2587 2588 /* 2589 * Promote a clone. Nomenclature note: 2590 * "clone" or "cds": the original clone which is being promoted 2591 * "origin" or "ods": the snapshot which is originally clone's origin 2592 * "origin head" or "ohds": the dataset which is the head 2593 * (filesystem/volume) for the origin 2594 * "origin origin": the origin of the origin's filesystem (typically 2595 * NULL, indicating that the clone is not a clone of a clone). 2596 */ 2597 int 2598 dsl_dataset_promote(const char *name) 2599 { 2600 dsl_dataset_t *ds; 2601 dsl_dir_t *dd; 2602 dsl_pool_t *dp; 2603 dmu_object_info_t doi; 2604 struct promotearg pa = { 0 }; 2605 struct promotenode *snap; 2606 int err; 2607 2608 err = dsl_dataset_hold(name, FTAG, &ds); 2609 if (err) 2610 return (err); 2611 dd = ds->ds_dir; 2612 dp = dd->dd_pool; 2613 2614 err = dmu_object_info(dp->dp_meta_objset, 2615 ds->ds_phys->ds_snapnames_zapobj, &doi); 2616 if (err) { 2617 dsl_dataset_rele(ds, FTAG); 2618 return (err); 2619 } 2620 2621 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) { 2622 dsl_dataset_rele(ds, FTAG); 2623 return (EINVAL); 2624 } 2625 2626 /* 2627 * We are going to inherit all the snapshots taken before our 2628 * origin (i.e., our new origin will be our parent's origin). 2629 * Take ownership of them so that we can rename them into our 2630 * namespace. 2631 */ 2632 rw_enter(&dp->dp_config_rwlock, RW_READER); 2633 2634 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj, 2635 &pa.shared_snaps); 2636 if (err != 0) 2637 goto out; 2638 2639 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps); 2640 if (err != 0) 2641 goto out; 2642 2643 snap = list_head(&pa.shared_snaps); 2644 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); 2645 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj, 2646 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps); 2647 if (err != 0) 2648 goto out; 2649 2650 if (dsl_dir_is_clone(snap->ds->ds_dir)) { 2651 err = dsl_dataset_own_obj(dp, 2652 snap->ds->ds_dir->dd_phys->dd_origin_obj, 2653 0, FTAG, &pa.origin_origin); 2654 if (err != 0) 2655 goto out; 2656 } 2657 2658 out: 2659 rw_exit(&dp->dp_config_rwlock); 2660 2661 /* 2662 * Add in 128x the snapnames zapobj size, since we will be moving 2663 * a bunch of snapnames to the promoted ds, and dirtying their 2664 * bonus buffers. 2665 */ 2666 if (err == 0) { 2667 err = dsl_sync_task_do(dp, dsl_dataset_promote_check, 2668 dsl_dataset_promote_sync, ds, &pa, 2669 2 + 2 * doi.doi_physical_blks); 2670 } 2671 2672 snaplist_destroy(&pa.shared_snaps, B_TRUE); 2673 snaplist_destroy(&pa.clone_snaps, B_FALSE); 2674 snaplist_destroy(&pa.origin_snaps, B_FALSE); 2675 if (pa.origin_origin) 2676 dsl_dataset_disown(pa.origin_origin, FTAG); 2677 dsl_dataset_rele(ds, FTAG); 2678 return (err); 2679 } 2680 2681 struct cloneswaparg { 2682 dsl_dataset_t *cds; /* clone dataset */ 2683 dsl_dataset_t *ohds; /* origin's head dataset */ 2684 boolean_t force; 2685 int64_t unused_refres_delta; /* change in unconsumed refreservation */ 2686 }; 2687 2688 /* ARGSUSED */ 2689 static int 2690 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx) 2691 { 2692 struct cloneswaparg *csa = arg1; 2693 2694 /* they should both be heads */ 2695 if (dsl_dataset_is_snapshot(csa->cds) || 2696 dsl_dataset_is_snapshot(csa->ohds)) 2697 return (EINVAL); 2698 2699 /* the branch point should be just before them */ 2700 if (csa->cds->ds_prev != csa->ohds->ds_prev) 2701 return (EINVAL); 2702 2703 /* cds should be the clone */ 2704 if (csa->cds->ds_prev->ds_phys->ds_next_snap_obj != 2705 csa->ohds->ds_object) 2706 return (EINVAL); 2707 2708 /* the clone should be a child of the origin */ 2709 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir) 2710 return (EINVAL); 2711 2712 /* ohds shouldn't be modified unless 'force' */ 2713 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds)) 2714 return (ETXTBSY); 2715 2716 /* adjust amount of any unconsumed refreservation */ 2717 csa->unused_refres_delta = 2718 (int64_t)MIN(csa->ohds->ds_reserved, 2719 csa->ohds->ds_phys->ds_unique_bytes) - 2720 (int64_t)MIN(csa->ohds->ds_reserved, 2721 csa->cds->ds_phys->ds_unique_bytes); 2722 2723 if (csa->unused_refres_delta > 0 && 2724 csa->unused_refres_delta > 2725 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE)) 2726 return (ENOSPC); 2727 2728 return (0); 2729 } 2730 2731 /* ARGSUSED */ 2732 static void 2733 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 2734 { 2735 struct cloneswaparg *csa = arg1; 2736 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool; 2737 2738 ASSERT(csa->cds->ds_reserved == 0); 2739 ASSERT(csa->cds->ds_quota == csa->ohds->ds_quota); 2740 2741 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx); 2742 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx); 2743 dmu_buf_will_dirty(csa->cds->ds_prev->ds_dbuf, tx); 2744 2745 if (csa->cds->ds_user_ptr != NULL) { 2746 csa->cds->ds_user_evict_func(csa->cds, csa->cds->ds_user_ptr); 2747 csa->cds->ds_user_ptr = NULL; 2748 } 2749 2750 if (csa->ohds->ds_user_ptr != NULL) { 2751 csa->ohds->ds_user_evict_func(csa->ohds, 2752 csa->ohds->ds_user_ptr); 2753 csa->ohds->ds_user_ptr = NULL; 2754 } 2755 2756 /* reset origin's unique bytes */ 2757 VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist, 2758 csa->cds->ds_prev->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2759 &csa->cds->ds_prev->ds_phys->ds_unique_bytes)); 2760 2761 /* swap blkptrs */ 2762 { 2763 blkptr_t tmp; 2764 tmp = csa->ohds->ds_phys->ds_bp; 2765 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp; 2766 csa->cds->ds_phys->ds_bp = tmp; 2767 } 2768 2769 /* set dd_*_bytes */ 2770 { 2771 int64_t dused, dcomp, duncomp; 2772 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2773 uint64_t odl_used, odl_comp, odl_uncomp; 2774 2775 ASSERT3U(csa->cds->ds_dir->dd_phys-> 2776 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2777 2778 VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used, 2779 &cdl_comp, &cdl_uncomp)); 2780 VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used, 2781 &odl_comp, &odl_uncomp)); 2782 2783 dused = csa->cds->ds_phys->ds_used_bytes + cdl_used - 2784 (csa->ohds->ds_phys->ds_used_bytes + odl_used); 2785 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp - 2786 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp); 2787 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes + 2788 cdl_uncomp - 2789 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp); 2790 2791 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD, 2792 dused, dcomp, duncomp, tx); 2793 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD, 2794 -dused, -dcomp, -duncomp, tx); 2795 2796 /* 2797 * The difference in the space used by snapshots is the 2798 * difference in snapshot space due to the head's 2799 * deadlist (since that's the only thing that's 2800 * changing that affects the snapused). 2801 */ 2802 VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist, 2803 csa->ohds->ds_origin_txg, UINT64_MAX, &cdl_used)); 2804 VERIFY(0 == bplist_space_birthrange(&csa->ohds->ds_deadlist, 2805 csa->ohds->ds_origin_txg, UINT64_MAX, &odl_used)); 2806 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used, 2807 DD_USED_HEAD, DD_USED_SNAP, tx); 2808 } 2809 2810 #define SWITCH64(x, y) \ 2811 { \ 2812 uint64_t __tmp = (x); \ 2813 (x) = (y); \ 2814 (y) = __tmp; \ 2815 } 2816 2817 /* swap ds_*_bytes */ 2818 SWITCH64(csa->ohds->ds_phys->ds_used_bytes, 2819 csa->cds->ds_phys->ds_used_bytes); 2820 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes, 2821 csa->cds->ds_phys->ds_compressed_bytes); 2822 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes, 2823 csa->cds->ds_phys->ds_uncompressed_bytes); 2824 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes, 2825 csa->cds->ds_phys->ds_unique_bytes); 2826 2827 /* apply any parent delta for change in unconsumed refreservation */ 2828 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV, 2829 csa->unused_refres_delta, 0, 0, tx); 2830 2831 /* swap deadlists */ 2832 bplist_close(&csa->cds->ds_deadlist); 2833 bplist_close(&csa->ohds->ds_deadlist); 2834 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj, 2835 csa->cds->ds_phys->ds_deadlist_obj); 2836 VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset, 2837 csa->cds->ds_phys->ds_deadlist_obj)); 2838 VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset, 2839 csa->ohds->ds_phys->ds_deadlist_obj)); 2840 } 2841 2842 /* 2843 * Swap 'clone' with its origin head file system. Used at the end 2844 * of "online recv" to swizzle the file system to the new version. 2845 */ 2846 int 2847 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head, 2848 boolean_t force) 2849 { 2850 struct cloneswaparg csa; 2851 int error; 2852 2853 ASSERT(clone->ds_owner); 2854 ASSERT(origin_head->ds_owner); 2855 retry: 2856 /* Need exclusive access for the swap */ 2857 rw_enter(&clone->ds_rwlock, RW_WRITER); 2858 if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) { 2859 rw_exit(&clone->ds_rwlock); 2860 rw_enter(&origin_head->ds_rwlock, RW_WRITER); 2861 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) { 2862 rw_exit(&origin_head->ds_rwlock); 2863 goto retry; 2864 } 2865 } 2866 csa.cds = clone; 2867 csa.ohds = origin_head; 2868 csa.force = force; 2869 error = dsl_sync_task_do(clone->ds_dir->dd_pool, 2870 dsl_dataset_clone_swap_check, 2871 dsl_dataset_clone_swap_sync, &csa, NULL, 9); 2872 return (error); 2873 } 2874 2875 /* 2876 * Given a pool name and a dataset object number in that pool, 2877 * return the name of that dataset. 2878 */ 2879 int 2880 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 2881 { 2882 spa_t *spa; 2883 dsl_pool_t *dp; 2884 dsl_dataset_t *ds; 2885 int error; 2886 2887 if ((error = spa_open(pname, &spa, FTAG)) != 0) 2888 return (error); 2889 dp = spa_get_dsl(spa); 2890 rw_enter(&dp->dp_config_rwlock, RW_READER); 2891 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) { 2892 dsl_dataset_name(ds, buf); 2893 dsl_dataset_rele(ds, FTAG); 2894 } 2895 rw_exit(&dp->dp_config_rwlock); 2896 spa_close(spa, FTAG); 2897 2898 return (error); 2899 } 2900 2901 int 2902 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 2903 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 2904 { 2905 int error = 0; 2906 2907 ASSERT3S(asize, >, 0); 2908 2909 /* 2910 * *ref_rsrv is the portion of asize that will come from any 2911 * unconsumed refreservation space. 2912 */ 2913 *ref_rsrv = 0; 2914 2915 mutex_enter(&ds->ds_lock); 2916 /* 2917 * Make a space adjustment for reserved bytes. 2918 */ 2919 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) { 2920 ASSERT3U(*used, >=, 2921 ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 2922 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 2923 *ref_rsrv = 2924 asize - MIN(asize, parent_delta(ds, asize + inflight)); 2925 } 2926 2927 if (!check_quota || ds->ds_quota == 0) { 2928 mutex_exit(&ds->ds_lock); 2929 return (0); 2930 } 2931 /* 2932 * If they are requesting more space, and our current estimate 2933 * is over quota, they get to try again unless the actual 2934 * on-disk is over quota and there are no pending changes (which 2935 * may free up space for us). 2936 */ 2937 if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) { 2938 if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota) 2939 error = ERESTART; 2940 else 2941 error = EDQUOT; 2942 } 2943 mutex_exit(&ds->ds_lock); 2944 2945 return (error); 2946 } 2947 2948 /* ARGSUSED */ 2949 static int 2950 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) 2951 { 2952 dsl_dataset_t *ds = arg1; 2953 uint64_t *quotap = arg2; 2954 uint64_t new_quota = *quotap; 2955 2956 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA) 2957 return (ENOTSUP); 2958 2959 if (new_quota == 0) 2960 return (0); 2961 2962 if (new_quota < ds->ds_phys->ds_used_bytes || 2963 new_quota < ds->ds_reserved) 2964 return (ENOSPC); 2965 2966 return (0); 2967 } 2968 2969 /* ARGSUSED */ 2970 void 2971 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 2972 { 2973 dsl_dataset_t *ds = arg1; 2974 uint64_t *quotap = arg2; 2975 uint64_t new_quota = *quotap; 2976 2977 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2978 2979 ds->ds_quota = new_quota; 2980 2981 dsl_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx); 2982 2983 spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa, 2984 tx, cr, "%lld dataset = %llu ", 2985 (longlong_t)new_quota, ds->ds_object); 2986 } 2987 2988 int 2989 dsl_dataset_set_quota(const char *dsname, uint64_t quota) 2990 { 2991 dsl_dataset_t *ds; 2992 int err; 2993 2994 err = dsl_dataset_hold(dsname, FTAG, &ds); 2995 if (err) 2996 return (err); 2997 2998 if (quota != ds->ds_quota) { 2999 /* 3000 * If someone removes a file, then tries to set the quota, we 3001 * want to make sure the file freeing takes effect. 3002 */ 3003 txg_wait_open(ds->ds_dir->dd_pool, 0); 3004 3005 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 3006 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync, 3007 ds, "a, 0); 3008 } 3009 dsl_dataset_rele(ds, FTAG); 3010 return (err); 3011 } 3012 3013 static int 3014 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) 3015 { 3016 dsl_dataset_t *ds = arg1; 3017 uint64_t *reservationp = arg2; 3018 uint64_t new_reservation = *reservationp; 3019 int64_t delta; 3020 uint64_t unique; 3021 3022 if (new_reservation > INT64_MAX) 3023 return (EOVERFLOW); 3024 3025 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < 3026 SPA_VERSION_REFRESERVATION) 3027 return (ENOTSUP); 3028 3029 if (dsl_dataset_is_snapshot(ds)) 3030 return (EINVAL); 3031 3032 /* 3033 * If we are doing the preliminary check in open context, the 3034 * space estimates may be inaccurate. 3035 */ 3036 if (!dmu_tx_is_syncing(tx)) 3037 return (0); 3038 3039 mutex_enter(&ds->ds_lock); 3040 unique = dsl_dataset_unique(ds); 3041 delta = MAX(unique, new_reservation) - MAX(unique, ds->ds_reserved); 3042 mutex_exit(&ds->ds_lock); 3043 3044 if (delta > 0 && 3045 delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 3046 return (ENOSPC); 3047 if (delta > 0 && ds->ds_quota > 0 && 3048 new_reservation > ds->ds_quota) 3049 return (ENOSPC); 3050 3051 return (0); 3052 } 3053 3054 /* ARGSUSED */ 3055 static void 3056 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr, 3057 dmu_tx_t *tx) 3058 { 3059 dsl_dataset_t *ds = arg1; 3060 uint64_t *reservationp = arg2; 3061 uint64_t new_reservation = *reservationp; 3062 uint64_t unique; 3063 int64_t delta; 3064 3065 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3066 3067 mutex_enter(&ds->ds_dir->dd_lock); 3068 mutex_enter(&ds->ds_lock); 3069 unique = dsl_dataset_unique(ds); 3070 delta = MAX(0, (int64_t)(new_reservation - unique)) - 3071 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3072 ds->ds_reserved = new_reservation; 3073 mutex_exit(&ds->ds_lock); 3074 3075 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3076 mutex_exit(&ds->ds_dir->dd_lock); 3077 dsl_prop_set_uint64_sync(ds->ds_dir, "refreservation", 3078 new_reservation, cr, tx); 3079 3080 spa_history_internal_log(LOG_DS_REFRESERV, 3081 ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu", 3082 (longlong_t)new_reservation, ds->ds_object); 3083 } 3084 3085 int 3086 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation) 3087 { 3088 dsl_dataset_t *ds; 3089 int err; 3090 3091 err = dsl_dataset_hold(dsname, FTAG, &ds); 3092 if (err) 3093 return (err); 3094 3095 err = dsl_sync_task_do(ds->ds_dir->dd_pool, 3096 dsl_dataset_set_reservation_check, 3097 dsl_dataset_set_reservation_sync, ds, &reservation, 0); 3098 dsl_dataset_rele(ds, FTAG); 3099 return (err); 3100 } 3101