1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2013, 2014 by Delphix. All rights reserved. 24 * Copyright (c) 2014, Joyent, Inc. All rights reserved. 25 * Copyright (c) 2014 RackTop Systems. 26 */ 27 28 #include <sys/dmu_objset.h> 29 #include <sys/dsl_dataset.h> 30 #include <sys/dsl_dir.h> 31 #include <sys/dsl_prop.h> 32 #include <sys/dsl_synctask.h> 33 #include <sys/dmu_traverse.h> 34 #include <sys/dmu_impl.h> 35 #include <sys/dmu_tx.h> 36 #include <sys/arc.h> 37 #include <sys/zio.h> 38 #include <sys/zap.h> 39 #include <sys/zfeature.h> 40 #include <sys/unique.h> 41 #include <sys/zfs_context.h> 42 #include <sys/zfs_ioctl.h> 43 #include <sys/spa.h> 44 #include <sys/zfs_znode.h> 45 #include <sys/zfs_onexit.h> 46 #include <sys/zvol.h> 47 #include <sys/dsl_scan.h> 48 #include <sys/dsl_deadlist.h> 49 #include <sys/dsl_destroy.h> 50 #include <sys/dsl_userhold.h> 51 #include <sys/dsl_bookmark.h> 52 53 #define SWITCH64(x, y) \ 54 { \ 55 uint64_t __tmp = (x); \ 56 (x) = (y); \ 57 (y) = __tmp; \ 58 } 59 60 #define DS_REF_MAX (1ULL << 62) 61 62 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE 63 64 /* 65 * Figure out how much of this delta should be propogated to the dsl_dir 66 * layer. If there's a refreservation, that space has already been 67 * partially accounted for in our ancestors. 68 */ 69 static int64_t 70 parent_delta(dsl_dataset_t *ds, int64_t delta) 71 { 72 uint64_t old_bytes, new_bytes; 73 74 if (ds->ds_reserved == 0) 75 return (delta); 76 77 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 78 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 79 80 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 81 return (new_bytes - old_bytes); 82 } 83 84 void 85 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 86 { 87 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 88 int compressed = BP_GET_PSIZE(bp); 89 int uncompressed = BP_GET_UCSIZE(bp); 90 int64_t delta; 91 92 dprintf_bp(bp, "ds=%p", ds); 93 94 ASSERT(dmu_tx_is_syncing(tx)); 95 /* It could have been compressed away to nothing */ 96 if (BP_IS_HOLE(bp)) 97 return; 98 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 99 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 100 if (ds == NULL) { 101 dsl_pool_mos_diduse_space(tx->tx_pool, 102 used, compressed, uncompressed); 103 return; 104 } 105 106 dmu_buf_will_dirty(ds->ds_dbuf, tx); 107 mutex_enter(&ds->ds_lock); 108 delta = parent_delta(ds, used); 109 ds->ds_phys->ds_referenced_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 } 119 120 int 121 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 122 boolean_t async) 123 { 124 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 125 int compressed = BP_GET_PSIZE(bp); 126 int uncompressed = BP_GET_UCSIZE(bp); 127 128 if (BP_IS_HOLE(bp)) 129 return (0); 130 131 ASSERT(dmu_tx_is_syncing(tx)); 132 ASSERT(bp->blk_birth <= tx->tx_txg); 133 134 if (ds == NULL) { 135 dsl_free(tx->tx_pool, tx->tx_txg, bp); 136 dsl_pool_mos_diduse_space(tx->tx_pool, 137 -used, -compressed, -uncompressed); 138 return (used); 139 } 140 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 141 142 ASSERT(!dsl_dataset_is_snapshot(ds)); 143 dmu_buf_will_dirty(ds->ds_dbuf, tx); 144 145 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { 146 int64_t delta; 147 148 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 149 dsl_free(tx->tx_pool, tx->tx_txg, bp); 150 151 mutex_enter(&ds->ds_lock); 152 ASSERT(ds->ds_phys->ds_unique_bytes >= used || 153 !DS_UNIQUE_IS_ACCURATE(ds)); 154 delta = parent_delta(ds, -used); 155 ds->ds_phys->ds_unique_bytes -= used; 156 mutex_exit(&ds->ds_lock); 157 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 158 delta, -compressed, -uncompressed, tx); 159 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 160 DD_USED_REFRSRV, DD_USED_HEAD, tx); 161 } else { 162 dprintf_bp(bp, "putting on dead list: %s", ""); 163 if (async) { 164 /* 165 * We are here as part of zio's write done callback, 166 * which means we're a zio interrupt thread. We can't 167 * call dsl_deadlist_insert() now because it may block 168 * waiting for I/O. Instead, put bp on the deferred 169 * queue and let dsl_pool_sync() finish the job. 170 */ 171 bplist_append(&ds->ds_pending_deadlist, bp); 172 } else { 173 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); 174 } 175 ASSERT3U(ds->ds_prev->ds_object, ==, 176 ds->ds_phys->ds_prev_snap_obj); 177 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); 178 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 179 if (ds->ds_prev->ds_phys->ds_next_snap_obj == 180 ds->ds_object && bp->blk_birth > 181 ds->ds_prev->ds_phys->ds_prev_snap_txg) { 182 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 183 mutex_enter(&ds->ds_prev->ds_lock); 184 ds->ds_prev->ds_phys->ds_unique_bytes += used; 185 mutex_exit(&ds->ds_prev->ds_lock); 186 } 187 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 188 dsl_dir_transfer_space(ds->ds_dir, used, 189 DD_USED_HEAD, DD_USED_SNAP, tx); 190 } 191 } 192 mutex_enter(&ds->ds_lock); 193 ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used); 194 ds->ds_phys->ds_referenced_bytes -= used; 195 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); 196 ds->ds_phys->ds_compressed_bytes -= compressed; 197 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); 198 ds->ds_phys->ds_uncompressed_bytes -= uncompressed; 199 mutex_exit(&ds->ds_lock); 200 201 return (used); 202 } 203 204 uint64_t 205 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 206 { 207 uint64_t trysnap = 0; 208 209 if (ds == NULL) 210 return (0); 211 /* 212 * The snapshot creation could fail, but that would cause an 213 * incorrect FALSE return, which would only result in an 214 * overestimation of the amount of space that an operation would 215 * consume, which is OK. 216 * 217 * There's also a small window where we could miss a pending 218 * snapshot, because we could set the sync task in the quiescing 219 * phase. So this should only be used as a guess. 220 */ 221 if (ds->ds_trysnap_txg > 222 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 223 trysnap = ds->ds_trysnap_txg; 224 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap)); 225 } 226 227 boolean_t 228 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp, 229 uint64_t blk_birth) 230 { 231 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) || 232 (bp != NULL && BP_IS_HOLE(bp))) 233 return (B_FALSE); 234 235 ddt_prefetch(dsl_dataset_get_spa(ds), bp); 236 237 return (B_TRUE); 238 } 239 240 /* ARGSUSED */ 241 static void 242 dsl_dataset_evict(dmu_buf_t *db, void *dsv) 243 { 244 dsl_dataset_t *ds = dsv; 245 246 ASSERT(ds->ds_owner == NULL); 247 248 unique_remove(ds->ds_fsid_guid); 249 250 if (ds->ds_objset != NULL) 251 dmu_objset_evict(ds->ds_objset); 252 253 if (ds->ds_prev) { 254 dsl_dataset_rele(ds->ds_prev, ds); 255 ds->ds_prev = NULL; 256 } 257 258 bplist_destroy(&ds->ds_pending_deadlist); 259 if (ds->ds_phys->ds_deadlist_obj != 0) 260 dsl_deadlist_close(&ds->ds_deadlist); 261 if (ds->ds_dir) 262 dsl_dir_rele(ds->ds_dir, ds); 263 264 ASSERT(!list_link_active(&ds->ds_synced_link)); 265 266 mutex_destroy(&ds->ds_lock); 267 mutex_destroy(&ds->ds_opening_lock); 268 refcount_destroy(&ds->ds_longholds); 269 270 kmem_free(ds, sizeof (dsl_dataset_t)); 271 } 272 273 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 != 0) 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 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 int 319 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 320 boolean_t adj_cnt) 321 { 322 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 323 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 324 matchtype_t mt; 325 int err; 326 327 dsl_dir_snap_cmtime_update(ds->ds_dir); 328 329 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 330 mt = MT_FIRST; 331 else 332 mt = MT_EXACT; 333 334 err = zap_remove_norm(mos, snapobj, name, mt, tx); 335 if (err == ENOTSUP && mt == MT_FIRST) 336 err = zap_remove(mos, snapobj, name, tx); 337 338 if (err == 0 && adj_cnt) 339 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 340 DD_FIELD_SNAPSHOT_COUNT, tx); 341 342 return (err); 343 } 344 345 int 346 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 347 dsl_dataset_t **dsp) 348 { 349 objset_t *mos = dp->dp_meta_objset; 350 dmu_buf_t *dbuf; 351 dsl_dataset_t *ds; 352 int err; 353 dmu_object_info_t doi; 354 355 ASSERT(dsl_pool_config_held(dp)); 356 357 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 358 if (err != 0) 359 return (err); 360 361 /* Make sure dsobj has the correct object type. */ 362 dmu_object_info_from_db(dbuf, &doi); 363 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 364 dmu_buf_rele(dbuf, tag); 365 return (SET_ERROR(EINVAL)); 366 } 367 368 ds = dmu_buf_get_user(dbuf); 369 if (ds == NULL) { 370 dsl_dataset_t *winner = NULL; 371 372 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 373 ds->ds_dbuf = dbuf; 374 ds->ds_object = dsobj; 375 ds->ds_phys = dbuf->db_data; 376 377 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 378 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 379 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 380 refcount_create(&ds->ds_longholds); 381 382 bplist_create(&ds->ds_pending_deadlist); 383 dsl_deadlist_open(&ds->ds_deadlist, 384 mos, ds->ds_phys->ds_deadlist_obj); 385 386 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 387 offsetof(dmu_sendarg_t, dsa_link)); 388 389 if (err == 0) { 390 err = dsl_dir_hold_obj(dp, 391 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); 392 } 393 if (err != 0) { 394 mutex_destroy(&ds->ds_lock); 395 mutex_destroy(&ds->ds_opening_lock); 396 refcount_destroy(&ds->ds_longholds); 397 bplist_destroy(&ds->ds_pending_deadlist); 398 dsl_deadlist_close(&ds->ds_deadlist); 399 kmem_free(ds, sizeof (dsl_dataset_t)); 400 dmu_buf_rele(dbuf, tag); 401 return (err); 402 } 403 404 if (!dsl_dataset_is_snapshot(ds)) { 405 ds->ds_snapname[0] = '\0'; 406 if (ds->ds_phys->ds_prev_snap_obj != 0) { 407 err = dsl_dataset_hold_obj(dp, 408 ds->ds_phys->ds_prev_snap_obj, 409 ds, &ds->ds_prev); 410 } 411 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 412 int zaperr = zap_lookup(mos, ds->ds_object, 413 DS_FIELD_BOOKMARK_NAMES, 414 sizeof (ds->ds_bookmarks), 1, 415 &ds->ds_bookmarks); 416 if (zaperr != ENOENT) 417 VERIFY0(zaperr); 418 } 419 } else { 420 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 421 err = dsl_dataset_get_snapname(ds); 422 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) { 423 err = zap_count( 424 ds->ds_dir->dd_pool->dp_meta_objset, 425 ds->ds_phys->ds_userrefs_obj, 426 &ds->ds_userrefs); 427 } 428 } 429 430 if (err == 0 && !dsl_dataset_is_snapshot(ds)) { 431 err = dsl_prop_get_int_ds(ds, 432 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 433 &ds->ds_reserved); 434 if (err == 0) { 435 err = dsl_prop_get_int_ds(ds, 436 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 437 &ds->ds_quota); 438 } 439 } else { 440 ds->ds_reserved = ds->ds_quota = 0; 441 } 442 443 if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds, 444 &ds->ds_phys, dsl_dataset_evict)) != NULL) { 445 bplist_destroy(&ds->ds_pending_deadlist); 446 dsl_deadlist_close(&ds->ds_deadlist); 447 if (ds->ds_prev) 448 dsl_dataset_rele(ds->ds_prev, ds); 449 dsl_dir_rele(ds->ds_dir, ds); 450 mutex_destroy(&ds->ds_lock); 451 mutex_destroy(&ds->ds_opening_lock); 452 refcount_destroy(&ds->ds_longholds); 453 kmem_free(ds, sizeof (dsl_dataset_t)); 454 if (err != 0) { 455 dmu_buf_rele(dbuf, tag); 456 return (err); 457 } 458 ds = winner; 459 } else { 460 ds->ds_fsid_guid = 461 unique_insert(ds->ds_phys->ds_fsid_guid); 462 } 463 } 464 ASSERT3P(ds->ds_dbuf, ==, dbuf); 465 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); 466 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 || 467 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 468 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 469 *dsp = ds; 470 return (0); 471 } 472 473 int 474 dsl_dataset_hold(dsl_pool_t *dp, const char *name, 475 void *tag, dsl_dataset_t **dsp) 476 { 477 dsl_dir_t *dd; 478 const char *snapname; 479 uint64_t obj; 480 int err = 0; 481 482 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 483 if (err != 0) 484 return (err); 485 486 ASSERT(dsl_pool_config_held(dp)); 487 obj = dd->dd_phys->dd_head_dataset_obj; 488 if (obj != 0) 489 err = dsl_dataset_hold_obj(dp, obj, tag, dsp); 490 else 491 err = SET_ERROR(ENOENT); 492 493 /* we may be looking for a snapshot */ 494 if (err == 0 && snapname != NULL) { 495 dsl_dataset_t *ds; 496 497 if (*snapname++ != '@') { 498 dsl_dataset_rele(*dsp, tag); 499 dsl_dir_rele(dd, FTAG); 500 return (SET_ERROR(ENOENT)); 501 } 502 503 dprintf("looking for snapshot '%s'\n", snapname); 504 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); 505 if (err == 0) 506 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 507 dsl_dataset_rele(*dsp, tag); 508 509 if (err == 0) { 510 mutex_enter(&ds->ds_lock); 511 if (ds->ds_snapname[0] == 0) 512 (void) strlcpy(ds->ds_snapname, snapname, 513 sizeof (ds->ds_snapname)); 514 mutex_exit(&ds->ds_lock); 515 *dsp = ds; 516 } 517 } 518 519 dsl_dir_rele(dd, FTAG); 520 return (err); 521 } 522 523 int 524 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 525 void *tag, dsl_dataset_t **dsp) 526 { 527 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 528 if (err != 0) 529 return (err); 530 if (!dsl_dataset_tryown(*dsp, tag)) { 531 dsl_dataset_rele(*dsp, tag); 532 *dsp = NULL; 533 return (SET_ERROR(EBUSY)); 534 } 535 return (0); 536 } 537 538 int 539 dsl_dataset_own(dsl_pool_t *dp, const char *name, 540 void *tag, dsl_dataset_t **dsp) 541 { 542 int err = dsl_dataset_hold(dp, name, tag, dsp); 543 if (err != 0) 544 return (err); 545 if (!dsl_dataset_tryown(*dsp, tag)) { 546 dsl_dataset_rele(*dsp, tag); 547 return (SET_ERROR(EBUSY)); 548 } 549 return (0); 550 } 551 552 /* 553 * See the comment above dsl_pool_hold() for details. In summary, a long 554 * hold is used to prevent destruction of a dataset while the pool hold 555 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 556 * 557 * The dataset and pool must be held when this function is called. After it 558 * is called, the pool hold may be released while the dataset is still held 559 * and accessed. 560 */ 561 void 562 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 563 { 564 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 565 (void) refcount_add(&ds->ds_longholds, tag); 566 } 567 568 void 569 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 570 { 571 (void) refcount_remove(&ds->ds_longholds, tag); 572 } 573 574 /* Return B_TRUE if there are any long holds on this dataset. */ 575 boolean_t 576 dsl_dataset_long_held(dsl_dataset_t *ds) 577 { 578 return (!refcount_is_zero(&ds->ds_longholds)); 579 } 580 581 void 582 dsl_dataset_name(dsl_dataset_t *ds, char *name) 583 { 584 if (ds == NULL) { 585 (void) strcpy(name, "mos"); 586 } else { 587 dsl_dir_name(ds->ds_dir, name); 588 VERIFY0(dsl_dataset_get_snapname(ds)); 589 if (ds->ds_snapname[0]) { 590 (void) strcat(name, "@"); 591 /* 592 * We use a "recursive" mutex so that we 593 * can call dprintf_ds() with ds_lock held. 594 */ 595 if (!MUTEX_HELD(&ds->ds_lock)) { 596 mutex_enter(&ds->ds_lock); 597 (void) strcat(name, ds->ds_snapname); 598 mutex_exit(&ds->ds_lock); 599 } else { 600 (void) strcat(name, ds->ds_snapname); 601 } 602 } 603 } 604 } 605 606 void 607 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 608 { 609 dmu_buf_rele(ds->ds_dbuf, tag); 610 } 611 612 void 613 dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 614 { 615 ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL); 616 617 mutex_enter(&ds->ds_lock); 618 ds->ds_owner = NULL; 619 mutex_exit(&ds->ds_lock); 620 dsl_dataset_long_rele(ds, tag); 621 if (ds->ds_dbuf != NULL) 622 dsl_dataset_rele(ds, tag); 623 else 624 dsl_dataset_evict(NULL, ds); 625 } 626 627 boolean_t 628 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 629 { 630 boolean_t gotit = FALSE; 631 632 mutex_enter(&ds->ds_lock); 633 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 634 ds->ds_owner = tag; 635 dsl_dataset_long_hold(ds, tag); 636 gotit = TRUE; 637 } 638 mutex_exit(&ds->ds_lock); 639 return (gotit); 640 } 641 642 uint64_t 643 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 644 uint64_t flags, dmu_tx_t *tx) 645 { 646 dsl_pool_t *dp = dd->dd_pool; 647 dmu_buf_t *dbuf; 648 dsl_dataset_phys_t *dsphys; 649 uint64_t dsobj; 650 objset_t *mos = dp->dp_meta_objset; 651 652 if (origin == NULL) 653 origin = dp->dp_origin_snap; 654 655 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 656 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0); 657 ASSERT(dmu_tx_is_syncing(tx)); 658 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 659 660 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 661 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 662 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 663 dmu_buf_will_dirty(dbuf, tx); 664 dsphys = dbuf->db_data; 665 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 666 dsphys->ds_dir_obj = dd->dd_object; 667 dsphys->ds_flags = flags; 668 dsphys->ds_fsid_guid = unique_create(); 669 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 670 sizeof (dsphys->ds_guid)); 671 dsphys->ds_snapnames_zapobj = 672 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 673 DMU_OT_NONE, 0, tx); 674 dsphys->ds_creation_time = gethrestime_sec(); 675 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 676 677 if (origin == NULL) { 678 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 679 } else { 680 dsl_dataset_t *ohds; /* head of the origin snapshot */ 681 682 dsphys->ds_prev_snap_obj = origin->ds_object; 683 dsphys->ds_prev_snap_txg = 684 origin->ds_phys->ds_creation_txg; 685 dsphys->ds_referenced_bytes = 686 origin->ds_phys->ds_referenced_bytes; 687 dsphys->ds_compressed_bytes = 688 origin->ds_phys->ds_compressed_bytes; 689 dsphys->ds_uncompressed_bytes = 690 origin->ds_phys->ds_uncompressed_bytes; 691 dsphys->ds_bp = origin->ds_phys->ds_bp; 692 dsphys->ds_flags |= origin->ds_phys->ds_flags; 693 694 dmu_buf_will_dirty(origin->ds_dbuf, tx); 695 origin->ds_phys->ds_num_children++; 696 697 VERIFY0(dsl_dataset_hold_obj(dp, 698 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds)); 699 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 700 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 701 dsl_dataset_rele(ohds, FTAG); 702 703 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 704 if (origin->ds_phys->ds_next_clones_obj == 0) { 705 origin->ds_phys->ds_next_clones_obj = 706 zap_create(mos, 707 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 708 } 709 VERIFY0(zap_add_int(mos, 710 origin->ds_phys->ds_next_clones_obj, dsobj, tx)); 711 } 712 713 dmu_buf_will_dirty(dd->dd_dbuf, tx); 714 dd->dd_phys->dd_origin_obj = origin->ds_object; 715 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 716 if (origin->ds_dir->dd_phys->dd_clones == 0) { 717 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 718 origin->ds_dir->dd_phys->dd_clones = 719 zap_create(mos, 720 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 721 } 722 VERIFY0(zap_add_int(mos, 723 origin->ds_dir->dd_phys->dd_clones, dsobj, tx)); 724 } 725 } 726 727 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 728 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 729 730 dmu_buf_rele(dbuf, FTAG); 731 732 dmu_buf_will_dirty(dd->dd_dbuf, tx); 733 dd->dd_phys->dd_head_dataset_obj = dsobj; 734 735 return (dsobj); 736 } 737 738 static void 739 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 740 { 741 objset_t *os; 742 743 VERIFY0(dmu_objset_from_ds(ds, &os)); 744 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 745 dsl_dataset_dirty(ds, tx); 746 } 747 748 uint64_t 749 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 750 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 751 { 752 dsl_pool_t *dp = pdd->dd_pool; 753 uint64_t dsobj, ddobj; 754 dsl_dir_t *dd; 755 756 ASSERT(dmu_tx_is_syncing(tx)); 757 ASSERT(lastname[0] != '@'); 758 759 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 760 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 761 762 dsobj = dsl_dataset_create_sync_dd(dd, origin, 763 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 764 765 dsl_deleg_set_create_perms(dd, tx, cr); 766 767 /* 768 * Since we're creating a new node we know it's a leaf, so we can 769 * initialize the counts if the limit feature is active. 770 */ 771 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 772 uint64_t cnt = 0; 773 objset_t *os = dd->dd_pool->dp_meta_objset; 774 775 dsl_dir_zapify(dd, tx); 776 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 777 sizeof (cnt), 1, &cnt, tx)); 778 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 779 sizeof (cnt), 1, &cnt, tx)); 780 } 781 782 dsl_dir_rele(dd, FTAG); 783 784 /* 785 * If we are creating a clone, make sure we zero out any stale 786 * data from the origin snapshots zil header. 787 */ 788 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 789 dsl_dataset_t *ds; 790 791 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 792 dsl_dataset_zero_zil(ds, tx); 793 dsl_dataset_rele(ds, FTAG); 794 } 795 796 return (dsobj); 797 } 798 799 /* 800 * The unique space in the head dataset can be calculated by subtracting 801 * the space used in the most recent snapshot, that is still being used 802 * in this file system, from the space currently in use. To figure out 803 * the space in the most recent snapshot still in use, we need to take 804 * the total space used in the snapshot and subtract out the space that 805 * has been freed up since the snapshot was taken. 806 */ 807 void 808 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 809 { 810 uint64_t mrs_used; 811 uint64_t dlused, dlcomp, dluncomp; 812 813 ASSERT(!dsl_dataset_is_snapshot(ds)); 814 815 if (ds->ds_phys->ds_prev_snap_obj != 0) 816 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes; 817 else 818 mrs_used = 0; 819 820 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 821 822 ASSERT3U(dlused, <=, mrs_used); 823 ds->ds_phys->ds_unique_bytes = 824 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused); 825 826 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 827 SPA_VERSION_UNIQUE_ACCURATE) 828 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 829 } 830 831 void 832 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 833 dmu_tx_t *tx) 834 { 835 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 836 uint64_t count; 837 int err; 838 839 ASSERT(ds->ds_phys->ds_num_children >= 2); 840 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx); 841 /* 842 * The err should not be ENOENT, but a bug in a previous version 843 * of the code could cause upgrade_clones_cb() to not set 844 * ds_next_snap_obj when it should, leading to a missing entry. 845 * If we knew that the pool was created after 846 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 847 * ENOENT. However, at least we can check that we don't have 848 * too many entries in the next_clones_obj even after failing to 849 * remove this one. 850 */ 851 if (err != ENOENT) 852 VERIFY0(err); 853 ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj, 854 &count)); 855 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2); 856 } 857 858 859 blkptr_t * 860 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 861 { 862 return (&ds->ds_phys->ds_bp); 863 } 864 865 void 866 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 867 { 868 ASSERT(dmu_tx_is_syncing(tx)); 869 /* If it's the meta-objset, set dp_meta_rootbp */ 870 if (ds == NULL) { 871 tx->tx_pool->dp_meta_rootbp = *bp; 872 } else { 873 dmu_buf_will_dirty(ds->ds_dbuf, tx); 874 ds->ds_phys->ds_bp = *bp; 875 } 876 } 877 878 spa_t * 879 dsl_dataset_get_spa(dsl_dataset_t *ds) 880 { 881 return (ds->ds_dir->dd_pool->dp_spa); 882 } 883 884 void 885 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 886 { 887 dsl_pool_t *dp; 888 889 if (ds == NULL) /* this is the meta-objset */ 890 return; 891 892 ASSERT(ds->ds_objset != NULL); 893 894 if (ds->ds_phys->ds_next_snap_obj != 0) 895 panic("dirtying snapshot!"); 896 897 dp = ds->ds_dir->dd_pool; 898 899 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 900 /* up the hold count until we can be written out */ 901 dmu_buf_add_ref(ds->ds_dbuf, ds); 902 } 903 } 904 905 boolean_t 906 dsl_dataset_is_dirty(dsl_dataset_t *ds) 907 { 908 for (int t = 0; t < TXG_SIZE; t++) { 909 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 910 ds, t)) 911 return (B_TRUE); 912 } 913 return (B_FALSE); 914 } 915 916 static int 917 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 918 { 919 uint64_t asize; 920 921 if (!dmu_tx_is_syncing(tx)) 922 return (0); 923 924 /* 925 * If there's an fs-only reservation, any blocks that might become 926 * owned by the snapshot dataset must be accommodated by space 927 * outside of the reservation. 928 */ 929 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 930 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 931 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 932 return (SET_ERROR(ENOSPC)); 933 934 /* 935 * Propagate any reserved space for this snapshot to other 936 * snapshot checks in this sync group. 937 */ 938 if (asize > 0) 939 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 940 941 return (0); 942 } 943 944 typedef struct dsl_dataset_snapshot_arg { 945 nvlist_t *ddsa_snaps; 946 nvlist_t *ddsa_props; 947 nvlist_t *ddsa_errors; 948 cred_t *ddsa_cr; 949 } dsl_dataset_snapshot_arg_t; 950 951 int 952 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 953 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 954 { 955 int error; 956 uint64_t value; 957 958 ds->ds_trysnap_txg = tx->tx_txg; 959 960 if (!dmu_tx_is_syncing(tx)) 961 return (0); 962 963 /* 964 * We don't allow multiple snapshots of the same txg. If there 965 * is already one, try again. 966 */ 967 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) 968 return (SET_ERROR(EAGAIN)); 969 970 /* 971 * Check for conflicting snapshot name. 972 */ 973 error = dsl_dataset_snap_lookup(ds, snapname, &value); 974 if (error == 0) 975 return (SET_ERROR(EEXIST)); 976 if (error != ENOENT) 977 return (error); 978 979 /* 980 * We don't allow taking snapshots of inconsistent datasets, such as 981 * those into which we are currently receiving. However, if we are 982 * creating this snapshot as part of a receive, this check will be 983 * executed atomically with respect to the completion of the receive 984 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 985 * case we ignore this, knowing it will be fixed up for us shortly in 986 * dmu_recv_end_sync(). 987 */ 988 if (!recv && DS_IS_INCONSISTENT(ds)) 989 return (SET_ERROR(EBUSY)); 990 991 /* 992 * Skip the check for temporary snapshots or if we have already checked 993 * the counts in dsl_dataset_snapshot_check. This means we really only 994 * check the count here when we're receiving a stream. 995 */ 996 if (cnt != 0 && cr != NULL) { 997 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 998 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 999 if (error != 0) 1000 return (error); 1001 } 1002 1003 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1004 if (error != 0) 1005 return (error); 1006 1007 return (0); 1008 } 1009 1010 static int 1011 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1012 { 1013 dsl_dataset_snapshot_arg_t *ddsa = arg; 1014 dsl_pool_t *dp = dmu_tx_pool(tx); 1015 nvpair_t *pair; 1016 int rv = 0; 1017 1018 /* 1019 * Pre-compute how many total new snapshots will be created for each 1020 * level in the tree and below. This is needed for validating the 1021 * snapshot limit when either taking a recursive snapshot or when 1022 * taking multiple snapshots. 1023 * 1024 * The problem is that the counts are not actually adjusted when 1025 * we are checking, only when we finally sync. For a single snapshot, 1026 * this is easy, the count will increase by 1 at each node up the tree, 1027 * but its more complicated for the recursive/multiple snapshot case. 1028 * 1029 * The dsl_fs_ss_limit_check function does recursively check the count 1030 * at each level up the tree but since it is validating each snapshot 1031 * independently we need to be sure that we are validating the complete 1032 * count for the entire set of snapshots. We do this by rolling up the 1033 * counts for each component of the name into an nvlist and then 1034 * checking each of those cases with the aggregated count. 1035 * 1036 * This approach properly handles not only the recursive snapshot 1037 * case (where we get all of those on the ddsa_snaps list) but also 1038 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1039 * validate the limit on 'a' using a count of 2). 1040 * 1041 * We validate the snapshot names in the third loop and only report 1042 * name errors once. 1043 */ 1044 if (dmu_tx_is_syncing(tx)) { 1045 nvlist_t *cnt_track = NULL; 1046 cnt_track = fnvlist_alloc(); 1047 1048 /* Rollup aggregated counts into the cnt_track list */ 1049 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1050 pair != NULL; 1051 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1052 char *pdelim; 1053 uint64_t val; 1054 char nm[MAXPATHLEN]; 1055 1056 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1057 pdelim = strchr(nm, '@'); 1058 if (pdelim == NULL) 1059 continue; 1060 *pdelim = '\0'; 1061 1062 do { 1063 if (nvlist_lookup_uint64(cnt_track, nm, 1064 &val) == 0) { 1065 /* update existing entry */ 1066 fnvlist_add_uint64(cnt_track, nm, 1067 val + 1); 1068 } else { 1069 /* add to list */ 1070 fnvlist_add_uint64(cnt_track, nm, 1); 1071 } 1072 1073 pdelim = strrchr(nm, '/'); 1074 if (pdelim != NULL) 1075 *pdelim = '\0'; 1076 } while (pdelim != NULL); 1077 } 1078 1079 /* Check aggregated counts at each level */ 1080 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1081 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1082 int error = 0; 1083 char *name; 1084 uint64_t cnt = 0; 1085 dsl_dataset_t *ds; 1086 1087 name = nvpair_name(pair); 1088 cnt = fnvpair_value_uint64(pair); 1089 ASSERT(cnt > 0); 1090 1091 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1092 if (error == 0) { 1093 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1094 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1095 ddsa->ddsa_cr); 1096 dsl_dataset_rele(ds, FTAG); 1097 } 1098 1099 if (error != 0) { 1100 if (ddsa->ddsa_errors != NULL) 1101 fnvlist_add_int32(ddsa->ddsa_errors, 1102 name, error); 1103 rv = error; 1104 /* only report one error for this check */ 1105 break; 1106 } 1107 } 1108 nvlist_free(cnt_track); 1109 } 1110 1111 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1112 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1113 int error = 0; 1114 dsl_dataset_t *ds; 1115 char *name, *atp; 1116 char dsname[MAXNAMELEN]; 1117 1118 name = nvpair_name(pair); 1119 if (strlen(name) >= MAXNAMELEN) 1120 error = SET_ERROR(ENAMETOOLONG); 1121 if (error == 0) { 1122 atp = strchr(name, '@'); 1123 if (atp == NULL) 1124 error = SET_ERROR(EINVAL); 1125 if (error == 0) 1126 (void) strlcpy(dsname, name, atp - name + 1); 1127 } 1128 if (error == 0) 1129 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1130 if (error == 0) { 1131 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1132 error = dsl_dataset_snapshot_check_impl(ds, 1133 atp + 1, tx, B_FALSE, 0, NULL); 1134 dsl_dataset_rele(ds, FTAG); 1135 } 1136 1137 if (error != 0) { 1138 if (ddsa->ddsa_errors != NULL) { 1139 fnvlist_add_int32(ddsa->ddsa_errors, 1140 name, error); 1141 } 1142 rv = error; 1143 } 1144 } 1145 1146 return (rv); 1147 } 1148 1149 void 1150 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1151 dmu_tx_t *tx) 1152 { 1153 static zil_header_t zero_zil; 1154 1155 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1156 dmu_buf_t *dbuf; 1157 dsl_dataset_phys_t *dsphys; 1158 uint64_t dsobj, crtxg; 1159 objset_t *mos = dp->dp_meta_objset; 1160 objset_t *os; 1161 1162 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1163 1164 /* 1165 * If we are on an old pool, the zil must not be active, in which 1166 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1167 */ 1168 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1169 dmu_objset_from_ds(ds, &os) != 0 || 1170 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1171 sizeof (zero_zil)) == 0); 1172 1173 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1174 1175 /* 1176 * The origin's ds_creation_txg has to be < TXG_INITIAL 1177 */ 1178 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1179 crtxg = 1; 1180 else 1181 crtxg = tx->tx_txg; 1182 1183 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1184 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1185 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1186 dmu_buf_will_dirty(dbuf, tx); 1187 dsphys = dbuf->db_data; 1188 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1189 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1190 dsphys->ds_fsid_guid = unique_create(); 1191 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1192 sizeof (dsphys->ds_guid)); 1193 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; 1194 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; 1195 dsphys->ds_next_snap_obj = ds->ds_object; 1196 dsphys->ds_num_children = 1; 1197 dsphys->ds_creation_time = gethrestime_sec(); 1198 dsphys->ds_creation_txg = crtxg; 1199 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; 1200 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes; 1201 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; 1202 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; 1203 dsphys->ds_flags = ds->ds_phys->ds_flags; 1204 dsphys->ds_bp = ds->ds_phys->ds_bp; 1205 dmu_buf_rele(dbuf, FTAG); 1206 1207 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); 1208 if (ds->ds_prev) { 1209 uint64_t next_clones_obj = 1210 ds->ds_prev->ds_phys->ds_next_clones_obj; 1211 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == 1212 ds->ds_object || 1213 ds->ds_prev->ds_phys->ds_num_children > 1); 1214 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { 1215 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1216 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1217 ds->ds_prev->ds_phys->ds_creation_txg); 1218 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; 1219 } else if (next_clones_obj != 0) { 1220 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1221 dsphys->ds_next_snap_obj, tx); 1222 VERIFY0(zap_add_int(mos, 1223 next_clones_obj, dsobj, tx)); 1224 } 1225 } 1226 1227 /* 1228 * If we have a reference-reservation on this dataset, we will 1229 * need to increase the amount of refreservation being charged 1230 * since our unique space is going to zero. 1231 */ 1232 if (ds->ds_reserved) { 1233 int64_t delta; 1234 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1235 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 1236 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1237 delta, 0, 0, tx); 1238 } 1239 1240 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1241 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist, 1242 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx); 1243 dsl_deadlist_close(&ds->ds_deadlist); 1244 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); 1245 dsl_deadlist_add_key(&ds->ds_deadlist, 1246 ds->ds_phys->ds_prev_snap_txg, tx); 1247 1248 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg); 1249 ds->ds_phys->ds_prev_snap_obj = dsobj; 1250 ds->ds_phys->ds_prev_snap_txg = crtxg; 1251 ds->ds_phys->ds_unique_bytes = 0; 1252 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1253 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1254 1255 VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, 1256 snapname, 8, 1, &dsobj, tx)); 1257 1258 if (ds->ds_prev) 1259 dsl_dataset_rele(ds->ds_prev, ds); 1260 VERIFY0(dsl_dataset_hold_obj(dp, 1261 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 1262 1263 dsl_scan_ds_snapshotted(ds, tx); 1264 1265 dsl_dir_snap_cmtime_update(ds->ds_dir); 1266 1267 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1268 } 1269 1270 static void 1271 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1272 { 1273 dsl_dataset_snapshot_arg_t *ddsa = arg; 1274 dsl_pool_t *dp = dmu_tx_pool(tx); 1275 nvpair_t *pair; 1276 1277 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1278 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1279 dsl_dataset_t *ds; 1280 char *name, *atp; 1281 char dsname[MAXNAMELEN]; 1282 1283 name = nvpair_name(pair); 1284 atp = strchr(name, '@'); 1285 (void) strlcpy(dsname, name, atp - name + 1); 1286 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1287 1288 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1289 if (ddsa->ddsa_props != NULL) { 1290 dsl_props_set_sync_impl(ds->ds_prev, 1291 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1292 } 1293 dsl_dataset_rele(ds, FTAG); 1294 } 1295 } 1296 1297 /* 1298 * The snapshots must all be in the same pool. 1299 * All-or-nothing: if there are any failures, nothing will be modified. 1300 */ 1301 int 1302 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1303 { 1304 dsl_dataset_snapshot_arg_t ddsa; 1305 nvpair_t *pair; 1306 boolean_t needsuspend; 1307 int error; 1308 spa_t *spa; 1309 char *firstname; 1310 nvlist_t *suspended = NULL; 1311 1312 pair = nvlist_next_nvpair(snaps, NULL); 1313 if (pair == NULL) 1314 return (0); 1315 firstname = nvpair_name(pair); 1316 1317 error = spa_open(firstname, &spa, FTAG); 1318 if (error != 0) 1319 return (error); 1320 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1321 spa_close(spa, FTAG); 1322 1323 if (needsuspend) { 1324 suspended = fnvlist_alloc(); 1325 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1326 pair = nvlist_next_nvpair(snaps, pair)) { 1327 char fsname[MAXNAMELEN]; 1328 char *snapname = nvpair_name(pair); 1329 char *atp; 1330 void *cookie; 1331 1332 atp = strchr(snapname, '@'); 1333 if (atp == NULL) { 1334 error = SET_ERROR(EINVAL); 1335 break; 1336 } 1337 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1338 1339 error = zil_suspend(fsname, &cookie); 1340 if (error != 0) 1341 break; 1342 fnvlist_add_uint64(suspended, fsname, 1343 (uintptr_t)cookie); 1344 } 1345 } 1346 1347 ddsa.ddsa_snaps = snaps; 1348 ddsa.ddsa_props = props; 1349 ddsa.ddsa_errors = errors; 1350 ddsa.ddsa_cr = CRED(); 1351 1352 if (error == 0) { 1353 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1354 dsl_dataset_snapshot_sync, &ddsa, 1355 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1356 } 1357 1358 if (suspended != NULL) { 1359 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1360 pair = nvlist_next_nvpair(suspended, pair)) { 1361 zil_resume((void *)(uintptr_t) 1362 fnvpair_value_uint64(pair)); 1363 } 1364 fnvlist_free(suspended); 1365 } 1366 1367 return (error); 1368 } 1369 1370 typedef struct dsl_dataset_snapshot_tmp_arg { 1371 const char *ddsta_fsname; 1372 const char *ddsta_snapname; 1373 minor_t ddsta_cleanup_minor; 1374 const char *ddsta_htag; 1375 } dsl_dataset_snapshot_tmp_arg_t; 1376 1377 static int 1378 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1379 { 1380 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1381 dsl_pool_t *dp = dmu_tx_pool(tx); 1382 dsl_dataset_t *ds; 1383 int error; 1384 1385 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1386 if (error != 0) 1387 return (error); 1388 1389 /* NULL cred means no limit check for tmp snapshot */ 1390 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1391 tx, B_FALSE, 0, NULL); 1392 if (error != 0) { 1393 dsl_dataset_rele(ds, FTAG); 1394 return (error); 1395 } 1396 1397 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1398 dsl_dataset_rele(ds, FTAG); 1399 return (SET_ERROR(ENOTSUP)); 1400 } 1401 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1402 B_TRUE, tx); 1403 if (error != 0) { 1404 dsl_dataset_rele(ds, FTAG); 1405 return (error); 1406 } 1407 1408 dsl_dataset_rele(ds, FTAG); 1409 return (0); 1410 } 1411 1412 static void 1413 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1414 { 1415 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1416 dsl_pool_t *dp = dmu_tx_pool(tx); 1417 dsl_dataset_t *ds; 1418 1419 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1420 1421 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1422 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1423 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1424 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1425 1426 dsl_dataset_rele(ds, FTAG); 1427 } 1428 1429 int 1430 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1431 minor_t cleanup_minor, const char *htag) 1432 { 1433 dsl_dataset_snapshot_tmp_arg_t ddsta; 1434 int error; 1435 spa_t *spa; 1436 boolean_t needsuspend; 1437 void *cookie; 1438 1439 ddsta.ddsta_fsname = fsname; 1440 ddsta.ddsta_snapname = snapname; 1441 ddsta.ddsta_cleanup_minor = cleanup_minor; 1442 ddsta.ddsta_htag = htag; 1443 1444 error = spa_open(fsname, &spa, FTAG); 1445 if (error != 0) 1446 return (error); 1447 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1448 spa_close(spa, FTAG); 1449 1450 if (needsuspend) { 1451 error = zil_suspend(fsname, &cookie); 1452 if (error != 0) 1453 return (error); 1454 } 1455 1456 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1457 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 1458 1459 if (needsuspend) 1460 zil_resume(cookie); 1461 return (error); 1462 } 1463 1464 1465 void 1466 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1467 { 1468 ASSERT(dmu_tx_is_syncing(tx)); 1469 ASSERT(ds->ds_objset != NULL); 1470 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); 1471 1472 /* 1473 * in case we had to change ds_fsid_guid when we opened it, 1474 * sync it out now. 1475 */ 1476 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1477 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; 1478 1479 dmu_objset_sync(ds->ds_objset, zio, tx); 1480 } 1481 1482 static void 1483 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1484 { 1485 uint64_t count = 0; 1486 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1487 zap_cursor_t zc; 1488 zap_attribute_t za; 1489 nvlist_t *propval = fnvlist_alloc(); 1490 nvlist_t *val = fnvlist_alloc(); 1491 1492 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1493 1494 /* 1495 * There may be missing entries in ds_next_clones_obj 1496 * due to a bug in a previous version of the code. 1497 * Only trust it if it has the right number of entries. 1498 */ 1499 if (ds->ds_phys->ds_next_clones_obj != 0) { 1500 VERIFY0(zap_count(mos, ds->ds_phys->ds_next_clones_obj, 1501 &count)); 1502 } 1503 if (count != ds->ds_phys->ds_num_children - 1) 1504 goto fail; 1505 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj); 1506 zap_cursor_retrieve(&zc, &za) == 0; 1507 zap_cursor_advance(&zc)) { 1508 dsl_dataset_t *clone; 1509 char buf[ZFS_MAXNAMELEN]; 1510 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1511 za.za_first_integer, FTAG, &clone)); 1512 dsl_dir_name(clone->ds_dir, buf); 1513 fnvlist_add_boolean(val, buf); 1514 dsl_dataset_rele(clone, FTAG); 1515 } 1516 zap_cursor_fini(&zc); 1517 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1518 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); 1519 fail: 1520 nvlist_free(val); 1521 nvlist_free(propval); 1522 } 1523 1524 void 1525 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1526 { 1527 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1528 uint64_t refd, avail, uobjs, aobjs, ratio; 1529 1530 ASSERT(dsl_pool_config_held(dp)); 1531 1532 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 : 1533 (ds->ds_phys->ds_uncompressed_bytes * 100 / 1534 ds->ds_phys->ds_compressed_bytes); 1535 1536 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); 1537 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 1538 ds->ds_phys->ds_uncompressed_bytes); 1539 1540 if (dsl_dataset_is_snapshot(ds)) { 1541 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); 1542 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 1543 ds->ds_phys->ds_unique_bytes); 1544 get_clones_stat(ds, nv); 1545 } else { 1546 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 1547 char buf[MAXNAMELEN]; 1548 dsl_dataset_name(ds->ds_prev, buf); 1549 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf); 1550 } 1551 1552 dsl_dir_stats(ds->ds_dir, nv); 1553 } 1554 1555 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 1556 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 1557 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 1558 1559 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 1560 ds->ds_phys->ds_creation_time); 1561 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 1562 ds->ds_phys->ds_creation_txg); 1563 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 1564 ds->ds_quota); 1565 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 1566 ds->ds_reserved); 1567 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 1568 ds->ds_phys->ds_guid); 1569 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 1570 ds->ds_phys->ds_unique_bytes); 1571 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 1572 ds->ds_object); 1573 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 1574 ds->ds_userrefs); 1575 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 1576 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 1577 1578 if (ds->ds_phys->ds_prev_snap_obj != 0) { 1579 uint64_t written, comp, uncomp; 1580 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1581 dsl_dataset_t *prev; 1582 1583 int err = dsl_dataset_hold_obj(dp, 1584 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev); 1585 if (err == 0) { 1586 err = dsl_dataset_space_written(prev, ds, &written, 1587 &comp, &uncomp); 1588 dsl_dataset_rele(prev, FTAG); 1589 if (err == 0) { 1590 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 1591 written); 1592 } 1593 } 1594 } 1595 } 1596 1597 void 1598 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 1599 { 1600 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1601 ASSERT(dsl_pool_config_held(dp)); 1602 1603 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; 1604 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; 1605 stat->dds_guid = ds->ds_phys->ds_guid; 1606 stat->dds_origin[0] = '\0'; 1607 if (dsl_dataset_is_snapshot(ds)) { 1608 stat->dds_is_snapshot = B_TRUE; 1609 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; 1610 } else { 1611 stat->dds_is_snapshot = B_FALSE; 1612 stat->dds_num_clones = 0; 1613 1614 if (dsl_dir_is_clone(ds->ds_dir)) { 1615 dsl_dataset_t *ods; 1616 1617 VERIFY0(dsl_dataset_hold_obj(dp, 1618 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods)); 1619 dsl_dataset_name(ods, stat->dds_origin); 1620 dsl_dataset_rele(ods, FTAG); 1621 } 1622 } 1623 } 1624 1625 uint64_t 1626 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 1627 { 1628 return (ds->ds_fsid_guid); 1629 } 1630 1631 void 1632 dsl_dataset_space(dsl_dataset_t *ds, 1633 uint64_t *refdbytesp, uint64_t *availbytesp, 1634 uint64_t *usedobjsp, uint64_t *availobjsp) 1635 { 1636 *refdbytesp = ds->ds_phys->ds_referenced_bytes; 1637 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 1638 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) 1639 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes; 1640 if (ds->ds_quota != 0) { 1641 /* 1642 * Adjust available bytes according to refquota 1643 */ 1644 if (*refdbytesp < ds->ds_quota) 1645 *availbytesp = MIN(*availbytesp, 1646 ds->ds_quota - *refdbytesp); 1647 else 1648 *availbytesp = 0; 1649 } 1650 *usedobjsp = BP_GET_FILL(&ds->ds_phys->ds_bp); 1651 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 1652 } 1653 1654 boolean_t 1655 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 1656 { 1657 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1658 1659 ASSERT(dsl_pool_config_held(dp)); 1660 if (snap == NULL) 1661 return (B_FALSE); 1662 if (ds->ds_phys->ds_bp.blk_birth > 1663 snap->ds_phys->ds_creation_txg) { 1664 objset_t *os, *os_snap; 1665 /* 1666 * It may be that only the ZIL differs, because it was 1667 * reset in the head. Don't count that as being 1668 * modified. 1669 */ 1670 if (dmu_objset_from_ds(ds, &os) != 0) 1671 return (B_TRUE); 1672 if (dmu_objset_from_ds(snap, &os_snap) != 0) 1673 return (B_TRUE); 1674 return (bcmp(&os->os_phys->os_meta_dnode, 1675 &os_snap->os_phys->os_meta_dnode, 1676 sizeof (os->os_phys->os_meta_dnode)) != 0); 1677 } 1678 return (B_FALSE); 1679 } 1680 1681 typedef struct dsl_dataset_rename_snapshot_arg { 1682 const char *ddrsa_fsname; 1683 const char *ddrsa_oldsnapname; 1684 const char *ddrsa_newsnapname; 1685 boolean_t ddrsa_recursive; 1686 dmu_tx_t *ddrsa_tx; 1687 } dsl_dataset_rename_snapshot_arg_t; 1688 1689 /* ARGSUSED */ 1690 static int 1691 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 1692 dsl_dataset_t *hds, void *arg) 1693 { 1694 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1695 int error; 1696 uint64_t val; 1697 1698 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1699 if (error != 0) { 1700 /* ignore nonexistent snapshots */ 1701 return (error == ENOENT ? 0 : error); 1702 } 1703 1704 /* new name should not exist */ 1705 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 1706 if (error == 0) 1707 error = SET_ERROR(EEXIST); 1708 else if (error == ENOENT) 1709 error = 0; 1710 1711 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 1712 if (dsl_dir_namelen(hds->ds_dir) + 1 + 1713 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN) 1714 error = SET_ERROR(ENAMETOOLONG); 1715 1716 return (error); 1717 } 1718 1719 static int 1720 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 1721 { 1722 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1723 dsl_pool_t *dp = dmu_tx_pool(tx); 1724 dsl_dataset_t *hds; 1725 int error; 1726 1727 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 1728 if (error != 0) 1729 return (error); 1730 1731 if (ddrsa->ddrsa_recursive) { 1732 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1733 dsl_dataset_rename_snapshot_check_impl, ddrsa, 1734 DS_FIND_CHILDREN); 1735 } else { 1736 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 1737 } 1738 dsl_dataset_rele(hds, FTAG); 1739 return (error); 1740 } 1741 1742 static int 1743 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 1744 dsl_dataset_t *hds, void *arg) 1745 { 1746 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1747 dsl_dataset_t *ds; 1748 uint64_t val; 1749 dmu_tx_t *tx = ddrsa->ddrsa_tx; 1750 int error; 1751 1752 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1753 ASSERT(error == 0 || error == ENOENT); 1754 if (error == ENOENT) { 1755 /* ignore nonexistent snapshots */ 1756 return (0); 1757 } 1758 1759 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 1760 1761 /* log before we change the name */ 1762 spa_history_log_internal_ds(ds, "rename", tx, 1763 "-> @%s", ddrsa->ddrsa_newsnapname); 1764 1765 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 1766 B_FALSE)); 1767 mutex_enter(&ds->ds_lock); 1768 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 1769 mutex_exit(&ds->ds_lock); 1770 VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj, 1771 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 1772 1773 dsl_dataset_rele(ds, FTAG); 1774 return (0); 1775 } 1776 1777 static void 1778 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 1779 { 1780 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1781 dsl_pool_t *dp = dmu_tx_pool(tx); 1782 dsl_dataset_t *hds; 1783 1784 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 1785 ddrsa->ddrsa_tx = tx; 1786 if (ddrsa->ddrsa_recursive) { 1787 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1788 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 1789 DS_FIND_CHILDREN)); 1790 } else { 1791 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 1792 } 1793 dsl_dataset_rele(hds, FTAG); 1794 } 1795 1796 int 1797 dsl_dataset_rename_snapshot(const char *fsname, 1798 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 1799 { 1800 dsl_dataset_rename_snapshot_arg_t ddrsa; 1801 1802 ddrsa.ddrsa_fsname = fsname; 1803 ddrsa.ddrsa_oldsnapname = oldsnapname; 1804 ddrsa.ddrsa_newsnapname = newsnapname; 1805 ddrsa.ddrsa_recursive = recursive; 1806 1807 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 1808 dsl_dataset_rename_snapshot_sync, &ddrsa, 1809 1, ZFS_SPACE_CHECK_RESERVED)); 1810 } 1811 1812 /* 1813 * If we're doing an ownership handoff, we need to make sure that there is 1814 * only one long hold on the dataset. We're not allowed to change anything here 1815 * so we don't permanently release the long hold or regular hold here. We want 1816 * to do this only when syncing to avoid the dataset unexpectedly going away 1817 * when we release the long hold. 1818 */ 1819 static int 1820 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 1821 { 1822 boolean_t held; 1823 1824 if (!dmu_tx_is_syncing(tx)) 1825 return (0); 1826 1827 if (owner != NULL) { 1828 VERIFY3P(ds->ds_owner, ==, owner); 1829 dsl_dataset_long_rele(ds, owner); 1830 } 1831 1832 held = dsl_dataset_long_held(ds); 1833 1834 if (owner != NULL) 1835 dsl_dataset_long_hold(ds, owner); 1836 1837 if (held) 1838 return (SET_ERROR(EBUSY)); 1839 1840 return (0); 1841 } 1842 1843 typedef struct dsl_dataset_rollback_arg { 1844 const char *ddra_fsname; 1845 void *ddra_owner; 1846 nvlist_t *ddra_result; 1847 } dsl_dataset_rollback_arg_t; 1848 1849 static int 1850 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 1851 { 1852 dsl_dataset_rollback_arg_t *ddra = arg; 1853 dsl_pool_t *dp = dmu_tx_pool(tx); 1854 dsl_dataset_t *ds; 1855 int64_t unused_refres_delta; 1856 int error; 1857 1858 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 1859 if (error != 0) 1860 return (error); 1861 1862 /* must not be a snapshot */ 1863 if (dsl_dataset_is_snapshot(ds)) { 1864 dsl_dataset_rele(ds, FTAG); 1865 return (SET_ERROR(EINVAL)); 1866 } 1867 1868 /* must have a most recent snapshot */ 1869 if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) { 1870 dsl_dataset_rele(ds, FTAG); 1871 return (SET_ERROR(EINVAL)); 1872 } 1873 1874 /* must not have any bookmarks after the most recent snapshot */ 1875 nvlist_t *proprequest = fnvlist_alloc(); 1876 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 1877 nvlist_t *bookmarks = fnvlist_alloc(); 1878 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 1879 fnvlist_free(proprequest); 1880 if (error != 0) 1881 return (error); 1882 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 1883 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 1884 nvlist_t *valuenv = 1885 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 1886 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 1887 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 1888 if (createtxg > ds->ds_phys->ds_prev_snap_txg) { 1889 fnvlist_free(bookmarks); 1890 dsl_dataset_rele(ds, FTAG); 1891 return (SET_ERROR(EEXIST)); 1892 } 1893 } 1894 fnvlist_free(bookmarks); 1895 1896 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 1897 if (error != 0) { 1898 dsl_dataset_rele(ds, FTAG); 1899 return (error); 1900 } 1901 1902 /* 1903 * Check if the snap we are rolling back to uses more than 1904 * the refquota. 1905 */ 1906 if (ds->ds_quota != 0 && 1907 ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) { 1908 dsl_dataset_rele(ds, FTAG); 1909 return (SET_ERROR(EDQUOT)); 1910 } 1911 1912 /* 1913 * When we do the clone swap, we will temporarily use more space 1914 * due to the refreservation (the head will no longer have any 1915 * unique space, so the entire amount of the refreservation will need 1916 * to be free). We will immediately destroy the clone, freeing 1917 * this space, but the freeing happens over many txg's. 1918 */ 1919 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 1920 ds->ds_phys->ds_unique_bytes); 1921 1922 if (unused_refres_delta > 0 && 1923 unused_refres_delta > 1924 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 1925 dsl_dataset_rele(ds, FTAG); 1926 return (SET_ERROR(ENOSPC)); 1927 } 1928 1929 dsl_dataset_rele(ds, FTAG); 1930 return (0); 1931 } 1932 1933 static void 1934 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 1935 { 1936 dsl_dataset_rollback_arg_t *ddra = arg; 1937 dsl_pool_t *dp = dmu_tx_pool(tx); 1938 dsl_dataset_t *ds, *clone; 1939 uint64_t cloneobj; 1940 char namebuf[ZFS_MAXNAMELEN]; 1941 1942 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 1943 1944 dsl_dataset_name(ds->ds_prev, namebuf); 1945 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 1946 1947 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 1948 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 1949 1950 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 1951 1952 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 1953 dsl_dataset_zero_zil(ds, tx); 1954 1955 dsl_destroy_head_sync_impl(clone, tx); 1956 1957 dsl_dataset_rele(clone, FTAG); 1958 dsl_dataset_rele(ds, FTAG); 1959 } 1960 1961 /* 1962 * Rolls back the given filesystem or volume to the most recent snapshot. 1963 * The name of the most recent snapshot will be returned under key "target" 1964 * in the result nvlist. 1965 * 1966 * If owner != NULL: 1967 * - The existing dataset MUST be owned by the specified owner at entry 1968 * - Upon return, dataset will still be held by the same owner, whether we 1969 * succeed or not. 1970 * 1971 * This mode is required any time the existing filesystem is mounted. See 1972 * notes above zfs_suspend_fs() for further details. 1973 */ 1974 int 1975 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result) 1976 { 1977 dsl_dataset_rollback_arg_t ddra; 1978 1979 ddra.ddra_fsname = fsname; 1980 ddra.ddra_owner = owner; 1981 ddra.ddra_result = result; 1982 1983 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 1984 dsl_dataset_rollback_sync, &ddra, 1985 1, ZFS_SPACE_CHECK_RESERVED)); 1986 } 1987 1988 struct promotenode { 1989 list_node_t link; 1990 dsl_dataset_t *ds; 1991 }; 1992 1993 typedef struct dsl_dataset_promote_arg { 1994 const char *ddpa_clonename; 1995 dsl_dataset_t *ddpa_clone; 1996 list_t shared_snaps, origin_snaps, clone_snaps; 1997 dsl_dataset_t *origin_origin; /* origin of the origin */ 1998 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 1999 char *err_ds; 2000 cred_t *cr; 2001 } dsl_dataset_promote_arg_t; 2002 2003 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2004 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2005 void *tag); 2006 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2007 2008 static int 2009 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2010 { 2011 dsl_dataset_promote_arg_t *ddpa = arg; 2012 dsl_pool_t *dp = dmu_tx_pool(tx); 2013 dsl_dataset_t *hds; 2014 struct promotenode *snap; 2015 dsl_dataset_t *origin_ds; 2016 int err; 2017 uint64_t unused; 2018 uint64_t ss_mv_cnt; 2019 2020 err = promote_hold(ddpa, dp, FTAG); 2021 if (err != 0) 2022 return (err); 2023 2024 hds = ddpa->ddpa_clone; 2025 2026 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) { 2027 promote_rele(ddpa, FTAG); 2028 return (SET_ERROR(EXDEV)); 2029 } 2030 2031 /* 2032 * Compute and check the amount of space to transfer. Since this is 2033 * so expensive, don't do the preliminary check. 2034 */ 2035 if (!dmu_tx_is_syncing(tx)) { 2036 promote_rele(ddpa, FTAG); 2037 return (0); 2038 } 2039 2040 snap = list_head(&ddpa->shared_snaps); 2041 origin_ds = snap->ds; 2042 2043 /* compute origin's new unique space */ 2044 snap = list_tail(&ddpa->clone_snaps); 2045 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2046 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2047 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2048 &ddpa->unique, &unused, &unused); 2049 2050 /* 2051 * Walk the snapshots that we are moving 2052 * 2053 * Compute space to transfer. Consider the incremental changes 2054 * to used by each snapshot: 2055 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2056 * So each snapshot gave birth to: 2057 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2058 * So a sequence would look like: 2059 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2060 * Which simplifies to: 2061 * uN + kN + kN-1 + ... + k1 + k0 2062 * Note however, if we stop before we reach the ORIGIN we get: 2063 * uN + kN + kN-1 + ... + kM - uM-1 2064 */ 2065 ss_mv_cnt = 0; 2066 ddpa->used = origin_ds->ds_phys->ds_referenced_bytes; 2067 ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes; 2068 ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; 2069 for (snap = list_head(&ddpa->shared_snaps); snap; 2070 snap = list_next(&ddpa->shared_snaps, snap)) { 2071 uint64_t val, dlused, dlcomp, dluncomp; 2072 dsl_dataset_t *ds = snap->ds; 2073 2074 ss_mv_cnt++; 2075 2076 /* 2077 * If there are long holds, we won't be able to evict 2078 * the objset. 2079 */ 2080 if (dsl_dataset_long_held(ds)) { 2081 err = SET_ERROR(EBUSY); 2082 goto out; 2083 } 2084 2085 /* Check that the snapshot name does not conflict */ 2086 VERIFY0(dsl_dataset_get_snapname(ds)); 2087 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2088 if (err == 0) { 2089 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname); 2090 err = SET_ERROR(EEXIST); 2091 goto out; 2092 } 2093 if (err != ENOENT) 2094 goto out; 2095 2096 /* The very first snapshot does not have a deadlist */ 2097 if (ds->ds_phys->ds_prev_snap_obj == 0) 2098 continue; 2099 2100 dsl_deadlist_space(&ds->ds_deadlist, 2101 &dlused, &dlcomp, &dluncomp); 2102 ddpa->used += dlused; 2103 ddpa->comp += dlcomp; 2104 ddpa->uncomp += dluncomp; 2105 } 2106 2107 /* 2108 * If we are a clone of a clone then we never reached ORIGIN, 2109 * so we need to subtract out the clone origin's used space. 2110 */ 2111 if (ddpa->origin_origin) { 2112 ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes; 2113 ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes; 2114 ddpa->uncomp -= 2115 ddpa->origin_origin->ds_phys->ds_uncompressed_bytes; 2116 } 2117 2118 /* Check that there is enough space and limit headroom here */ 2119 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2120 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2121 if (err != 0) 2122 goto out; 2123 2124 /* 2125 * Compute the amounts of space that will be used by snapshots 2126 * after the promotion (for both origin and clone). For each, 2127 * it is the amount of space that will be on all of their 2128 * deadlists (that was not born before their new origin). 2129 */ 2130 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2131 uint64_t space; 2132 2133 /* 2134 * Note, typically this will not be a clone of a clone, 2135 * so dd_origin_txg will be < TXG_INITIAL, so 2136 * these snaplist_space() -> dsl_deadlist_space_range() 2137 * calls will be fast because they do not have to 2138 * iterate over all bps. 2139 */ 2140 snap = list_head(&ddpa->origin_snaps); 2141 err = snaplist_space(&ddpa->shared_snaps, 2142 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2143 if (err != 0) 2144 goto out; 2145 2146 err = snaplist_space(&ddpa->clone_snaps, 2147 snap->ds->ds_dir->dd_origin_txg, &space); 2148 if (err != 0) 2149 goto out; 2150 ddpa->cloneusedsnap += space; 2151 } 2152 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2153 err = snaplist_space(&ddpa->origin_snaps, 2154 origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap); 2155 if (err != 0) 2156 goto out; 2157 } 2158 2159 out: 2160 promote_rele(ddpa, FTAG); 2161 return (err); 2162 } 2163 2164 static void 2165 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2166 { 2167 dsl_dataset_promote_arg_t *ddpa = arg; 2168 dsl_pool_t *dp = dmu_tx_pool(tx); 2169 dsl_dataset_t *hds; 2170 struct promotenode *snap; 2171 dsl_dataset_t *origin_ds; 2172 dsl_dataset_t *origin_head; 2173 dsl_dir_t *dd; 2174 dsl_dir_t *odd = NULL; 2175 uint64_t oldnext_obj; 2176 int64_t delta; 2177 2178 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2179 hds = ddpa->ddpa_clone; 2180 2181 ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE); 2182 2183 snap = list_head(&ddpa->shared_snaps); 2184 origin_ds = snap->ds; 2185 dd = hds->ds_dir; 2186 2187 snap = list_head(&ddpa->origin_snaps); 2188 origin_head = snap->ds; 2189 2190 /* 2191 * We need to explicitly open odd, since origin_ds's dd will be 2192 * changing. 2193 */ 2194 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2195 NULL, FTAG, &odd)); 2196 2197 /* change origin's next snap */ 2198 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2199 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj; 2200 snap = list_tail(&ddpa->clone_snaps); 2201 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2202 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object; 2203 2204 /* change the origin's next clone */ 2205 if (origin_ds->ds_phys->ds_next_clones_obj) { 2206 dsl_dataset_remove_from_next_clones(origin_ds, 2207 snap->ds->ds_object, tx); 2208 VERIFY0(zap_add_int(dp->dp_meta_objset, 2209 origin_ds->ds_phys->ds_next_clones_obj, 2210 oldnext_obj, tx)); 2211 } 2212 2213 /* change origin */ 2214 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2215 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object); 2216 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj; 2217 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2218 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2219 odd->dd_phys->dd_origin_obj = origin_ds->ds_object; 2220 origin_head->ds_dir->dd_origin_txg = 2221 origin_ds->ds_phys->ds_creation_txg; 2222 2223 /* change dd_clone entries */ 2224 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2225 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2226 odd->dd_phys->dd_clones, hds->ds_object, tx)); 2227 VERIFY0(zap_add_int(dp->dp_meta_objset, 2228 ddpa->origin_origin->ds_dir->dd_phys->dd_clones, 2229 hds->ds_object, tx)); 2230 2231 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2232 ddpa->origin_origin->ds_dir->dd_phys->dd_clones, 2233 origin_head->ds_object, tx)); 2234 if (dd->dd_phys->dd_clones == 0) { 2235 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset, 2236 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 2237 } 2238 VERIFY0(zap_add_int(dp->dp_meta_objset, 2239 dd->dd_phys->dd_clones, origin_head->ds_object, tx)); 2240 } 2241 2242 /* move snapshots to this dir */ 2243 for (snap = list_head(&ddpa->shared_snaps); snap; 2244 snap = list_next(&ddpa->shared_snaps, snap)) { 2245 dsl_dataset_t *ds = snap->ds; 2246 2247 /* 2248 * Property callbacks are registered to a particular 2249 * dsl_dir. Since ours is changing, evict the objset 2250 * so that they will be unregistered from the old dsl_dir. 2251 */ 2252 if (ds->ds_objset) { 2253 dmu_objset_evict(ds->ds_objset); 2254 ds->ds_objset = NULL; 2255 } 2256 2257 /* move snap name entry */ 2258 VERIFY0(dsl_dataset_get_snapname(ds)); 2259 VERIFY0(dsl_dataset_snap_remove(origin_head, 2260 ds->ds_snapname, tx, B_TRUE)); 2261 VERIFY0(zap_add(dp->dp_meta_objset, 2262 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, 2263 8, 1, &ds->ds_object, tx)); 2264 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 2265 DD_FIELD_SNAPSHOT_COUNT, tx); 2266 2267 /* change containing dsl_dir */ 2268 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2269 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object); 2270 ds->ds_phys->ds_dir_obj = dd->dd_object; 2271 ASSERT3P(ds->ds_dir, ==, odd); 2272 dsl_dir_rele(ds->ds_dir, ds); 2273 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 2274 NULL, ds, &ds->ds_dir)); 2275 2276 /* move any clone references */ 2277 if (ds->ds_phys->ds_next_clones_obj && 2278 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2279 zap_cursor_t zc; 2280 zap_attribute_t za; 2281 2282 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2283 ds->ds_phys->ds_next_clones_obj); 2284 zap_cursor_retrieve(&zc, &za) == 0; 2285 zap_cursor_advance(&zc)) { 2286 dsl_dataset_t *cnds; 2287 uint64_t o; 2288 2289 if (za.za_first_integer == oldnext_obj) { 2290 /* 2291 * We've already moved the 2292 * origin's reference. 2293 */ 2294 continue; 2295 } 2296 2297 VERIFY0(dsl_dataset_hold_obj(dp, 2298 za.za_first_integer, FTAG, &cnds)); 2299 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj; 2300 2301 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2302 odd->dd_phys->dd_clones, o, tx)); 2303 VERIFY0(zap_add_int(dp->dp_meta_objset, 2304 dd->dd_phys->dd_clones, o, tx)); 2305 dsl_dataset_rele(cnds, FTAG); 2306 } 2307 zap_cursor_fini(&zc); 2308 } 2309 2310 ASSERT(!dsl_prop_hascb(ds)); 2311 } 2312 2313 /* 2314 * Change space accounting. 2315 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2316 * both be valid, or both be 0 (resulting in delta == 0). This 2317 * is true for each of {clone,origin} independently. 2318 */ 2319 2320 delta = ddpa->cloneusedsnap - 2321 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2322 ASSERT3S(delta, >=, 0); 2323 ASSERT3U(ddpa->used, >=, delta); 2324 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2325 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2326 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 2327 2328 delta = ddpa->originusedsnap - 2329 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2330 ASSERT3S(delta, <=, 0); 2331 ASSERT3U(ddpa->used, >=, -delta); 2332 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2333 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2334 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 2335 2336 origin_ds->ds_phys->ds_unique_bytes = ddpa->unique; 2337 2338 /* log history record */ 2339 spa_history_log_internal_ds(hds, "promote", tx, ""); 2340 2341 dsl_dir_rele(odd, FTAG); 2342 promote_rele(ddpa, FTAG); 2343 } 2344 2345 /* 2346 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2347 * (exclusive) and last_obj (inclusive). The list will be in reverse 2348 * order (last_obj will be the list_head()). If first_obj == 0, do all 2349 * snapshots back to this dataset's origin. 2350 */ 2351 static int 2352 snaplist_make(dsl_pool_t *dp, 2353 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 2354 { 2355 uint64_t obj = last_obj; 2356 2357 list_create(l, sizeof (struct promotenode), 2358 offsetof(struct promotenode, link)); 2359 2360 while (obj != first_obj) { 2361 dsl_dataset_t *ds; 2362 struct promotenode *snap; 2363 int err; 2364 2365 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 2366 ASSERT(err != ENOENT); 2367 if (err != 0) 2368 return (err); 2369 2370 if (first_obj == 0) 2371 first_obj = ds->ds_dir->dd_phys->dd_origin_obj; 2372 2373 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 2374 snap->ds = ds; 2375 list_insert_tail(l, snap); 2376 obj = ds->ds_phys->ds_prev_snap_obj; 2377 } 2378 2379 return (0); 2380 } 2381 2382 static int 2383 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2384 { 2385 struct promotenode *snap; 2386 2387 *spacep = 0; 2388 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2389 uint64_t used, comp, uncomp; 2390 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2391 mintxg, UINT64_MAX, &used, &comp, &uncomp); 2392 *spacep += used; 2393 } 2394 return (0); 2395 } 2396 2397 static void 2398 snaplist_destroy(list_t *l, void *tag) 2399 { 2400 struct promotenode *snap; 2401 2402 if (l == NULL || !list_link_active(&l->list_head)) 2403 return; 2404 2405 while ((snap = list_tail(l)) != NULL) { 2406 list_remove(l, snap); 2407 dsl_dataset_rele(snap->ds, tag); 2408 kmem_free(snap, sizeof (*snap)); 2409 } 2410 list_destroy(l); 2411 } 2412 2413 static int 2414 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 2415 { 2416 int error; 2417 dsl_dir_t *dd; 2418 struct promotenode *snap; 2419 2420 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 2421 &ddpa->ddpa_clone); 2422 if (error != 0) 2423 return (error); 2424 dd = ddpa->ddpa_clone->ds_dir; 2425 2426 if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) || 2427 !dsl_dir_is_clone(dd)) { 2428 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2429 return (SET_ERROR(EINVAL)); 2430 } 2431 2432 error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj, 2433 &ddpa->shared_snaps, tag); 2434 if (error != 0) 2435 goto out; 2436 2437 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 2438 &ddpa->clone_snaps, tag); 2439 if (error != 0) 2440 goto out; 2441 2442 snap = list_head(&ddpa->shared_snaps); 2443 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); 2444 error = snaplist_make(dp, dd->dd_phys->dd_origin_obj, 2445 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, 2446 &ddpa->origin_snaps, tag); 2447 if (error != 0) 2448 goto out; 2449 2450 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) { 2451 error = dsl_dataset_hold_obj(dp, 2452 snap->ds->ds_dir->dd_phys->dd_origin_obj, 2453 tag, &ddpa->origin_origin); 2454 if (error != 0) 2455 goto out; 2456 } 2457 out: 2458 if (error != 0) 2459 promote_rele(ddpa, tag); 2460 return (error); 2461 } 2462 2463 static void 2464 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 2465 { 2466 snaplist_destroy(&ddpa->shared_snaps, tag); 2467 snaplist_destroy(&ddpa->clone_snaps, tag); 2468 snaplist_destroy(&ddpa->origin_snaps, tag); 2469 if (ddpa->origin_origin != NULL) 2470 dsl_dataset_rele(ddpa->origin_origin, tag); 2471 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2472 } 2473 2474 /* 2475 * Promote a clone. 2476 * 2477 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 2478 * in with the name. (It must be at least MAXNAMELEN bytes long.) 2479 */ 2480 int 2481 dsl_dataset_promote(const char *name, char *conflsnap) 2482 { 2483 dsl_dataset_promote_arg_t ddpa = { 0 }; 2484 uint64_t numsnaps; 2485 int error; 2486 objset_t *os; 2487 2488 /* 2489 * We will modify space proportional to the number of 2490 * snapshots. Compute numsnaps. 2491 */ 2492 error = dmu_objset_hold(name, FTAG, &os); 2493 if (error != 0) 2494 return (error); 2495 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 2496 dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps); 2497 dmu_objset_rele(os, FTAG); 2498 if (error != 0) 2499 return (error); 2500 2501 ddpa.ddpa_clonename = name; 2502 ddpa.err_ds = conflsnap; 2503 ddpa.cr = CRED(); 2504 2505 return (dsl_sync_task(name, dsl_dataset_promote_check, 2506 dsl_dataset_promote_sync, &ddpa, 2507 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED)); 2508 } 2509 2510 int 2511 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 2512 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 2513 { 2514 int64_t unused_refres_delta; 2515 2516 /* they should both be heads */ 2517 if (dsl_dataset_is_snapshot(clone) || 2518 dsl_dataset_is_snapshot(origin_head)) 2519 return (SET_ERROR(EINVAL)); 2520 2521 /* if we are not forcing, the branch point should be just before them */ 2522 if (!force && clone->ds_prev != origin_head->ds_prev) 2523 return (SET_ERROR(EINVAL)); 2524 2525 /* clone should be the clone (unless they are unrelated) */ 2526 if (clone->ds_prev != NULL && 2527 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 2528 origin_head->ds_dir != clone->ds_prev->ds_dir) 2529 return (SET_ERROR(EINVAL)); 2530 2531 /* the clone should be a child of the origin */ 2532 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 2533 return (SET_ERROR(EINVAL)); 2534 2535 /* origin_head shouldn't be modified unless 'force' */ 2536 if (!force && 2537 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 2538 return (SET_ERROR(ETXTBSY)); 2539 2540 /* origin_head should have no long holds (e.g. is not mounted) */ 2541 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 2542 return (SET_ERROR(EBUSY)); 2543 2544 /* check amount of any unconsumed refreservation */ 2545 unused_refres_delta = 2546 (int64_t)MIN(origin_head->ds_reserved, 2547 origin_head->ds_phys->ds_unique_bytes) - 2548 (int64_t)MIN(origin_head->ds_reserved, 2549 clone->ds_phys->ds_unique_bytes); 2550 2551 if (unused_refres_delta > 0 && 2552 unused_refres_delta > 2553 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 2554 return (SET_ERROR(ENOSPC)); 2555 2556 /* clone can't be over the head's refquota */ 2557 if (origin_head->ds_quota != 0 && 2558 clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota) 2559 return (SET_ERROR(EDQUOT)); 2560 2561 return (0); 2562 } 2563 2564 void 2565 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 2566 dsl_dataset_t *origin_head, dmu_tx_t *tx) 2567 { 2568 dsl_pool_t *dp = dmu_tx_pool(tx); 2569 int64_t unused_refres_delta; 2570 2571 ASSERT(clone->ds_reserved == 0); 2572 ASSERT(origin_head->ds_quota == 0 || 2573 clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota); 2574 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 2575 2576 dmu_buf_will_dirty(clone->ds_dbuf, tx); 2577 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 2578 2579 if (clone->ds_objset != NULL) { 2580 dmu_objset_evict(clone->ds_objset); 2581 clone->ds_objset = NULL; 2582 } 2583 2584 if (origin_head->ds_objset != NULL) { 2585 dmu_objset_evict(origin_head->ds_objset); 2586 origin_head->ds_objset = NULL; 2587 } 2588 2589 unused_refres_delta = 2590 (int64_t)MIN(origin_head->ds_reserved, 2591 origin_head->ds_phys->ds_unique_bytes) - 2592 (int64_t)MIN(origin_head->ds_reserved, 2593 clone->ds_phys->ds_unique_bytes); 2594 2595 /* 2596 * Reset origin's unique bytes, if it exists. 2597 */ 2598 if (clone->ds_prev) { 2599 dsl_dataset_t *origin = clone->ds_prev; 2600 uint64_t comp, uncomp; 2601 2602 dmu_buf_will_dirty(origin->ds_dbuf, tx); 2603 dsl_deadlist_space_range(&clone->ds_deadlist, 2604 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2605 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp); 2606 } 2607 2608 /* swap blkptrs */ 2609 { 2610 blkptr_t tmp; 2611 tmp = origin_head->ds_phys->ds_bp; 2612 origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp; 2613 clone->ds_phys->ds_bp = tmp; 2614 } 2615 2616 /* set dd_*_bytes */ 2617 { 2618 int64_t dused, dcomp, duncomp; 2619 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2620 uint64_t odl_used, odl_comp, odl_uncomp; 2621 2622 ASSERT3U(clone->ds_dir->dd_phys-> 2623 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2624 2625 dsl_deadlist_space(&clone->ds_deadlist, 2626 &cdl_used, &cdl_comp, &cdl_uncomp); 2627 dsl_deadlist_space(&origin_head->ds_deadlist, 2628 &odl_used, &odl_comp, &odl_uncomp); 2629 2630 dused = clone->ds_phys->ds_referenced_bytes + cdl_used - 2631 (origin_head->ds_phys->ds_referenced_bytes + odl_used); 2632 dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp - 2633 (origin_head->ds_phys->ds_compressed_bytes + odl_comp); 2634 duncomp = clone->ds_phys->ds_uncompressed_bytes + 2635 cdl_uncomp - 2636 (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp); 2637 2638 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 2639 dused, dcomp, duncomp, tx); 2640 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 2641 -dused, -dcomp, -duncomp, tx); 2642 2643 /* 2644 * The difference in the space used by snapshots is the 2645 * difference in snapshot space due to the head's 2646 * deadlist (since that's the only thing that's 2647 * changing that affects the snapused). 2648 */ 2649 dsl_deadlist_space_range(&clone->ds_deadlist, 2650 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2651 &cdl_used, &cdl_comp, &cdl_uncomp); 2652 dsl_deadlist_space_range(&origin_head->ds_deadlist, 2653 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2654 &odl_used, &odl_comp, &odl_uncomp); 2655 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 2656 DD_USED_HEAD, DD_USED_SNAP, tx); 2657 } 2658 2659 /* swap ds_*_bytes */ 2660 SWITCH64(origin_head->ds_phys->ds_referenced_bytes, 2661 clone->ds_phys->ds_referenced_bytes); 2662 SWITCH64(origin_head->ds_phys->ds_compressed_bytes, 2663 clone->ds_phys->ds_compressed_bytes); 2664 SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes, 2665 clone->ds_phys->ds_uncompressed_bytes); 2666 SWITCH64(origin_head->ds_phys->ds_unique_bytes, 2667 clone->ds_phys->ds_unique_bytes); 2668 2669 /* apply any parent delta for change in unconsumed refreservation */ 2670 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 2671 unused_refres_delta, 0, 0, tx); 2672 2673 /* 2674 * Swap deadlists. 2675 */ 2676 dsl_deadlist_close(&clone->ds_deadlist); 2677 dsl_deadlist_close(&origin_head->ds_deadlist); 2678 SWITCH64(origin_head->ds_phys->ds_deadlist_obj, 2679 clone->ds_phys->ds_deadlist_obj); 2680 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 2681 clone->ds_phys->ds_deadlist_obj); 2682 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 2683 origin_head->ds_phys->ds_deadlist_obj); 2684 2685 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 2686 2687 spa_history_log_internal_ds(clone, "clone swap", tx, 2688 "parent=%s", origin_head->ds_dir->dd_myname); 2689 } 2690 2691 /* 2692 * Given a pool name and a dataset object number in that pool, 2693 * return the name of that dataset. 2694 */ 2695 int 2696 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 2697 { 2698 dsl_pool_t *dp; 2699 dsl_dataset_t *ds; 2700 int error; 2701 2702 error = dsl_pool_hold(pname, FTAG, &dp); 2703 if (error != 0) 2704 return (error); 2705 2706 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 2707 if (error == 0) { 2708 dsl_dataset_name(ds, buf); 2709 dsl_dataset_rele(ds, FTAG); 2710 } 2711 dsl_pool_rele(dp, FTAG); 2712 2713 return (error); 2714 } 2715 2716 int 2717 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 2718 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 2719 { 2720 int error = 0; 2721 2722 ASSERT3S(asize, >, 0); 2723 2724 /* 2725 * *ref_rsrv is the portion of asize that will come from any 2726 * unconsumed refreservation space. 2727 */ 2728 *ref_rsrv = 0; 2729 2730 mutex_enter(&ds->ds_lock); 2731 /* 2732 * Make a space adjustment for reserved bytes. 2733 */ 2734 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) { 2735 ASSERT3U(*used, >=, 2736 ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 2737 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 2738 *ref_rsrv = 2739 asize - MIN(asize, parent_delta(ds, asize + inflight)); 2740 } 2741 2742 if (!check_quota || ds->ds_quota == 0) { 2743 mutex_exit(&ds->ds_lock); 2744 return (0); 2745 } 2746 /* 2747 * If they are requesting more space, and our current estimate 2748 * is over quota, they get to try again unless the actual 2749 * on-disk is over quota and there are no pending changes (which 2750 * may free up space for us). 2751 */ 2752 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) { 2753 if (inflight > 0 || 2754 ds->ds_phys->ds_referenced_bytes < ds->ds_quota) 2755 error = SET_ERROR(ERESTART); 2756 else 2757 error = SET_ERROR(EDQUOT); 2758 } 2759 mutex_exit(&ds->ds_lock); 2760 2761 return (error); 2762 } 2763 2764 typedef struct dsl_dataset_set_qr_arg { 2765 const char *ddsqra_name; 2766 zprop_source_t ddsqra_source; 2767 uint64_t ddsqra_value; 2768 } dsl_dataset_set_qr_arg_t; 2769 2770 2771 /* ARGSUSED */ 2772 static int 2773 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 2774 { 2775 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2776 dsl_pool_t *dp = dmu_tx_pool(tx); 2777 dsl_dataset_t *ds; 2778 int error; 2779 uint64_t newval; 2780 2781 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 2782 return (SET_ERROR(ENOTSUP)); 2783 2784 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 2785 if (error != 0) 2786 return (error); 2787 2788 if (dsl_dataset_is_snapshot(ds)) { 2789 dsl_dataset_rele(ds, FTAG); 2790 return (SET_ERROR(EINVAL)); 2791 } 2792 2793 error = dsl_prop_predict(ds->ds_dir, 2794 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 2795 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 2796 if (error != 0) { 2797 dsl_dataset_rele(ds, FTAG); 2798 return (error); 2799 } 2800 2801 if (newval == 0) { 2802 dsl_dataset_rele(ds, FTAG); 2803 return (0); 2804 } 2805 2806 if (newval < ds->ds_phys->ds_referenced_bytes || 2807 newval < ds->ds_reserved) { 2808 dsl_dataset_rele(ds, FTAG); 2809 return (SET_ERROR(ENOSPC)); 2810 } 2811 2812 dsl_dataset_rele(ds, FTAG); 2813 return (0); 2814 } 2815 2816 static void 2817 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 2818 { 2819 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2820 dsl_pool_t *dp = dmu_tx_pool(tx); 2821 dsl_dataset_t *ds; 2822 uint64_t newval; 2823 2824 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 2825 2826 dsl_prop_set_sync_impl(ds, 2827 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 2828 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 2829 &ddsqra->ddsqra_value, tx); 2830 2831 VERIFY0(dsl_prop_get_int_ds(ds, 2832 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 2833 2834 if (ds->ds_quota != newval) { 2835 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2836 ds->ds_quota = newval; 2837 } 2838 dsl_dataset_rele(ds, FTAG); 2839 } 2840 2841 int 2842 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 2843 uint64_t refquota) 2844 { 2845 dsl_dataset_set_qr_arg_t ddsqra; 2846 2847 ddsqra.ddsqra_name = dsname; 2848 ddsqra.ddsqra_source = source; 2849 ddsqra.ddsqra_value = refquota; 2850 2851 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 2852 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 2853 } 2854 2855 static int 2856 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 2857 { 2858 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2859 dsl_pool_t *dp = dmu_tx_pool(tx); 2860 dsl_dataset_t *ds; 2861 int error; 2862 uint64_t newval, unique; 2863 2864 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 2865 return (SET_ERROR(ENOTSUP)); 2866 2867 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 2868 if (error != 0) 2869 return (error); 2870 2871 if (dsl_dataset_is_snapshot(ds)) { 2872 dsl_dataset_rele(ds, FTAG); 2873 return (SET_ERROR(EINVAL)); 2874 } 2875 2876 error = dsl_prop_predict(ds->ds_dir, 2877 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 2878 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 2879 if (error != 0) { 2880 dsl_dataset_rele(ds, FTAG); 2881 return (error); 2882 } 2883 2884 /* 2885 * If we are doing the preliminary check in open context, the 2886 * space estimates may be inaccurate. 2887 */ 2888 if (!dmu_tx_is_syncing(tx)) { 2889 dsl_dataset_rele(ds, FTAG); 2890 return (0); 2891 } 2892 2893 mutex_enter(&ds->ds_lock); 2894 if (!DS_UNIQUE_IS_ACCURATE(ds)) 2895 dsl_dataset_recalc_head_uniq(ds); 2896 unique = ds->ds_phys->ds_unique_bytes; 2897 mutex_exit(&ds->ds_lock); 2898 2899 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 2900 uint64_t delta = MAX(unique, newval) - 2901 MAX(unique, ds->ds_reserved); 2902 2903 if (delta > 2904 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 2905 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 2906 dsl_dataset_rele(ds, FTAG); 2907 return (SET_ERROR(ENOSPC)); 2908 } 2909 } 2910 2911 dsl_dataset_rele(ds, FTAG); 2912 return (0); 2913 } 2914 2915 void 2916 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 2917 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 2918 { 2919 uint64_t newval; 2920 uint64_t unique; 2921 int64_t delta; 2922 2923 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 2924 source, sizeof (value), 1, &value, tx); 2925 2926 VERIFY0(dsl_prop_get_int_ds(ds, 2927 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 2928 2929 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2930 mutex_enter(&ds->ds_dir->dd_lock); 2931 mutex_enter(&ds->ds_lock); 2932 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 2933 unique = ds->ds_phys->ds_unique_bytes; 2934 delta = MAX(0, (int64_t)(newval - unique)) - 2935 MAX(0, (int64_t)(ds->ds_reserved - unique)); 2936 ds->ds_reserved = newval; 2937 mutex_exit(&ds->ds_lock); 2938 2939 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 2940 mutex_exit(&ds->ds_dir->dd_lock); 2941 } 2942 2943 static void 2944 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 2945 { 2946 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2947 dsl_pool_t *dp = dmu_tx_pool(tx); 2948 dsl_dataset_t *ds; 2949 2950 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 2951 dsl_dataset_set_refreservation_sync_impl(ds, 2952 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 2953 dsl_dataset_rele(ds, FTAG); 2954 } 2955 2956 int 2957 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 2958 uint64_t refreservation) 2959 { 2960 dsl_dataset_set_qr_arg_t ddsqra; 2961 2962 ddsqra.ddsqra_name = dsname; 2963 ddsqra.ddsqra_source = source; 2964 ddsqra.ddsqra_value = refreservation; 2965 2966 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 2967 dsl_dataset_set_refreservation_sync, &ddsqra, 2968 0, ZFS_SPACE_CHECK_NONE)); 2969 } 2970 2971 /* 2972 * Return (in *usedp) the amount of space written in new that is not 2973 * present in oldsnap. New may be a snapshot or the head. Old must be 2974 * a snapshot before new, in new's filesystem (or its origin). If not then 2975 * fail and return EINVAL. 2976 * 2977 * The written space is calculated by considering two components: First, we 2978 * ignore any freed space, and calculate the written as new's used space 2979 * minus old's used space. Next, we add in the amount of space that was freed 2980 * between the two snapshots, thus reducing new's used space relative to old's. 2981 * Specifically, this is the space that was born before old->ds_creation_txg, 2982 * and freed before new (ie. on new's deadlist or a previous deadlist). 2983 * 2984 * space freed [---------------------] 2985 * snapshots ---O-------O--------O-------O------ 2986 * oldsnap new 2987 */ 2988 int 2989 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 2990 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 2991 { 2992 int err = 0; 2993 uint64_t snapobj; 2994 dsl_pool_t *dp = new->ds_dir->dd_pool; 2995 2996 ASSERT(dsl_pool_config_held(dp)); 2997 2998 *usedp = 0; 2999 *usedp += new->ds_phys->ds_referenced_bytes; 3000 *usedp -= oldsnap->ds_phys->ds_referenced_bytes; 3001 3002 *compp = 0; 3003 *compp += new->ds_phys->ds_compressed_bytes; 3004 *compp -= oldsnap->ds_phys->ds_compressed_bytes; 3005 3006 *uncompp = 0; 3007 *uncompp += new->ds_phys->ds_uncompressed_bytes; 3008 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes; 3009 3010 snapobj = new->ds_object; 3011 while (snapobj != oldsnap->ds_object) { 3012 dsl_dataset_t *snap; 3013 uint64_t used, comp, uncomp; 3014 3015 if (snapobj == new->ds_object) { 3016 snap = new; 3017 } else { 3018 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3019 if (err != 0) 3020 break; 3021 } 3022 3023 if (snap->ds_phys->ds_prev_snap_txg == 3024 oldsnap->ds_phys->ds_creation_txg) { 3025 /* 3026 * The blocks in the deadlist can not be born after 3027 * ds_prev_snap_txg, so get the whole deadlist space, 3028 * which is more efficient (especially for old-format 3029 * deadlists). Unfortunately the deadlist code 3030 * doesn't have enough information to make this 3031 * optimization itself. 3032 */ 3033 dsl_deadlist_space(&snap->ds_deadlist, 3034 &used, &comp, &uncomp); 3035 } else { 3036 dsl_deadlist_space_range(&snap->ds_deadlist, 3037 0, oldsnap->ds_phys->ds_creation_txg, 3038 &used, &comp, &uncomp); 3039 } 3040 *usedp += used; 3041 *compp += comp; 3042 *uncompp += uncomp; 3043 3044 /* 3045 * If we get to the beginning of the chain of snapshots 3046 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3047 * was not a snapshot of/before new. 3048 */ 3049 snapobj = snap->ds_phys->ds_prev_snap_obj; 3050 if (snap != new) 3051 dsl_dataset_rele(snap, FTAG); 3052 if (snapobj == 0) { 3053 err = SET_ERROR(EINVAL); 3054 break; 3055 } 3056 3057 } 3058 return (err); 3059 } 3060 3061 /* 3062 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3063 * lastsnap, and all snapshots in between are deleted. 3064 * 3065 * blocks that would be freed [---------------------------] 3066 * snapshots ---O-------O--------O-------O--------O 3067 * firstsnap lastsnap 3068 * 3069 * This is the set of blocks that were born after the snap before firstsnap, 3070 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3071 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3072 * We calculate this by iterating over the relevant deadlists (from the snap 3073 * after lastsnap, backward to the snap after firstsnap), summing up the 3074 * space on the deadlist that was born after the snap before firstsnap. 3075 */ 3076 int 3077 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3078 dsl_dataset_t *lastsnap, 3079 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3080 { 3081 int err = 0; 3082 uint64_t snapobj; 3083 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3084 3085 ASSERT(dsl_dataset_is_snapshot(firstsnap)); 3086 ASSERT(dsl_dataset_is_snapshot(lastsnap)); 3087 3088 /* 3089 * Check that the snapshots are in the same dsl_dir, and firstsnap 3090 * is before lastsnap. 3091 */ 3092 if (firstsnap->ds_dir != lastsnap->ds_dir || 3093 firstsnap->ds_phys->ds_creation_txg > 3094 lastsnap->ds_phys->ds_creation_txg) 3095 return (SET_ERROR(EINVAL)); 3096 3097 *usedp = *compp = *uncompp = 0; 3098 3099 snapobj = lastsnap->ds_phys->ds_next_snap_obj; 3100 while (snapobj != firstsnap->ds_object) { 3101 dsl_dataset_t *ds; 3102 uint64_t used, comp, uncomp; 3103 3104 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3105 if (err != 0) 3106 break; 3107 3108 dsl_deadlist_space_range(&ds->ds_deadlist, 3109 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX, 3110 &used, &comp, &uncomp); 3111 *usedp += used; 3112 *compp += comp; 3113 *uncompp += uncomp; 3114 3115 snapobj = ds->ds_phys->ds_prev_snap_obj; 3116 ASSERT3U(snapobj, !=, 0); 3117 dsl_dataset_rele(ds, FTAG); 3118 } 3119 return (err); 3120 } 3121 3122 /* 3123 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 3124 * For example, they could both be snapshots of the same filesystem, and 3125 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 3126 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 3127 * filesystem. Or 'earlier' could be the origin's origin. 3128 * 3129 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 3130 */ 3131 boolean_t 3132 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 3133 uint64_t earlier_txg) 3134 { 3135 dsl_pool_t *dp = later->ds_dir->dd_pool; 3136 int error; 3137 boolean_t ret; 3138 3139 ASSERT(dsl_pool_config_held(dp)); 3140 ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0); 3141 3142 if (earlier_txg == 0) 3143 earlier_txg = earlier->ds_phys->ds_creation_txg; 3144 3145 if (dsl_dataset_is_snapshot(later) && 3146 earlier_txg >= later->ds_phys->ds_creation_txg) 3147 return (B_FALSE); 3148 3149 if (later->ds_dir == earlier->ds_dir) 3150 return (B_TRUE); 3151 if (!dsl_dir_is_clone(later->ds_dir)) 3152 return (B_FALSE); 3153 3154 if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object) 3155 return (B_TRUE); 3156 dsl_dataset_t *origin; 3157 error = dsl_dataset_hold_obj(dp, 3158 later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin); 3159 if (error != 0) 3160 return (B_FALSE); 3161 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 3162 dsl_dataset_rele(origin, FTAG); 3163 return (ret); 3164 } 3165 3166 3167 void 3168 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 3169 { 3170 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3171 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 3172 } 3173