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) 2011, 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 /* 54 * The SPA supports block sizes up to 16MB. However, very large blocks 55 * can have an impact on i/o latency (e.g. tying up a spinning disk for 56 * ~300ms), and also potentially on the memory allocator. Therefore, 57 * we do not allow the recordsize to be set larger than zfs_max_recordsize 58 * (default 1MB). Larger blocks can be created by changing this tunable, 59 * and pools with larger blocks can always be imported and used, regardless 60 * of this setting. 61 */ 62 int zfs_max_recordsize = 1 * 1024 * 1024; 63 64 #define SWITCH64(x, y) \ 65 { \ 66 uint64_t __tmp = (x); \ 67 (x) = (y); \ 68 (y) = __tmp; \ 69 } 70 71 #define DS_REF_MAX (1ULL << 62) 72 73 /* 74 * Figure out how much of this delta should be propogated to the dsl_dir 75 * layer. If there's a refreservation, that space has already been 76 * partially accounted for in our ancestors. 77 */ 78 static int64_t 79 parent_delta(dsl_dataset_t *ds, int64_t delta) 80 { 81 uint64_t old_bytes, new_bytes; 82 83 if (ds->ds_reserved == 0) 84 return (delta); 85 86 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 87 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 88 89 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 90 return (new_bytes - old_bytes); 91 } 92 93 void 94 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 95 { 96 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 97 int compressed = BP_GET_PSIZE(bp); 98 int uncompressed = BP_GET_UCSIZE(bp); 99 int64_t delta; 100 101 dprintf_bp(bp, "ds=%p", ds); 102 103 ASSERT(dmu_tx_is_syncing(tx)); 104 /* It could have been compressed away to nothing */ 105 if (BP_IS_HOLE(bp)) 106 return; 107 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 108 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 109 if (ds == NULL) { 110 dsl_pool_mos_diduse_space(tx->tx_pool, 111 used, compressed, uncompressed); 112 return; 113 } 114 115 dmu_buf_will_dirty(ds->ds_dbuf, tx); 116 mutex_enter(&ds->ds_lock); 117 delta = parent_delta(ds, used); 118 ds->ds_phys->ds_referenced_bytes += used; 119 ds->ds_phys->ds_compressed_bytes += compressed; 120 ds->ds_phys->ds_uncompressed_bytes += uncompressed; 121 ds->ds_phys->ds_unique_bytes += used; 122 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) 123 ds->ds_need_large_blocks = B_TRUE; 124 mutex_exit(&ds->ds_lock); 125 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 126 compressed, uncompressed, tx); 127 dsl_dir_transfer_space(ds->ds_dir, used - delta, 128 DD_USED_REFRSRV, DD_USED_HEAD, tx); 129 } 130 131 int 132 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 133 boolean_t async) 134 { 135 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 136 int compressed = BP_GET_PSIZE(bp); 137 int uncompressed = BP_GET_UCSIZE(bp); 138 139 if (BP_IS_HOLE(bp)) 140 return (0); 141 142 ASSERT(dmu_tx_is_syncing(tx)); 143 ASSERT(bp->blk_birth <= tx->tx_txg); 144 145 if (ds == NULL) { 146 dsl_free(tx->tx_pool, tx->tx_txg, bp); 147 dsl_pool_mos_diduse_space(tx->tx_pool, 148 -used, -compressed, -uncompressed); 149 return (used); 150 } 151 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 152 153 ASSERT(!dsl_dataset_is_snapshot(ds)); 154 dmu_buf_will_dirty(ds->ds_dbuf, tx); 155 156 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { 157 int64_t delta; 158 159 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 160 dsl_free(tx->tx_pool, tx->tx_txg, bp); 161 162 mutex_enter(&ds->ds_lock); 163 ASSERT(ds->ds_phys->ds_unique_bytes >= used || 164 !DS_UNIQUE_IS_ACCURATE(ds)); 165 delta = parent_delta(ds, -used); 166 ds->ds_phys->ds_unique_bytes -= used; 167 mutex_exit(&ds->ds_lock); 168 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 169 delta, -compressed, -uncompressed, tx); 170 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 171 DD_USED_REFRSRV, DD_USED_HEAD, tx); 172 } else { 173 dprintf_bp(bp, "putting on dead list: %s", ""); 174 if (async) { 175 /* 176 * We are here as part of zio's write done callback, 177 * which means we're a zio interrupt thread. We can't 178 * call dsl_deadlist_insert() now because it may block 179 * waiting for I/O. Instead, put bp on the deferred 180 * queue and let dsl_pool_sync() finish the job. 181 */ 182 bplist_append(&ds->ds_pending_deadlist, bp); 183 } else { 184 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); 185 } 186 ASSERT3U(ds->ds_prev->ds_object, ==, 187 ds->ds_phys->ds_prev_snap_obj); 188 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); 189 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 190 if (ds->ds_prev->ds_phys->ds_next_snap_obj == 191 ds->ds_object && bp->blk_birth > 192 ds->ds_prev->ds_phys->ds_prev_snap_txg) { 193 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 194 mutex_enter(&ds->ds_prev->ds_lock); 195 ds->ds_prev->ds_phys->ds_unique_bytes += used; 196 mutex_exit(&ds->ds_prev->ds_lock); 197 } 198 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 199 dsl_dir_transfer_space(ds->ds_dir, used, 200 DD_USED_HEAD, DD_USED_SNAP, tx); 201 } 202 } 203 mutex_enter(&ds->ds_lock); 204 ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used); 205 ds->ds_phys->ds_referenced_bytes -= used; 206 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); 207 ds->ds_phys->ds_compressed_bytes -= compressed; 208 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); 209 ds->ds_phys->ds_uncompressed_bytes -= uncompressed; 210 mutex_exit(&ds->ds_lock); 211 212 return (used); 213 } 214 215 uint64_t 216 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 217 { 218 uint64_t trysnap = 0; 219 220 if (ds == NULL) 221 return (0); 222 /* 223 * The snapshot creation could fail, but that would cause an 224 * incorrect FALSE return, which would only result in an 225 * overestimation of the amount of space that an operation would 226 * consume, which is OK. 227 * 228 * There's also a small window where we could miss a pending 229 * snapshot, because we could set the sync task in the quiescing 230 * phase. So this should only be used as a guess. 231 */ 232 if (ds->ds_trysnap_txg > 233 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 234 trysnap = ds->ds_trysnap_txg; 235 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap)); 236 } 237 238 boolean_t 239 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp, 240 uint64_t blk_birth) 241 { 242 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) || 243 (bp != NULL && BP_IS_HOLE(bp))) 244 return (B_FALSE); 245 246 ddt_prefetch(dsl_dataset_get_spa(ds), bp); 247 248 return (B_TRUE); 249 } 250 251 /* ARGSUSED */ 252 static void 253 dsl_dataset_evict(dmu_buf_t *db, void *dsv) 254 { 255 dsl_dataset_t *ds = dsv; 256 257 ASSERT(ds->ds_owner == NULL); 258 259 unique_remove(ds->ds_fsid_guid); 260 261 if (ds->ds_objset != NULL) 262 dmu_objset_evict(ds->ds_objset); 263 264 if (ds->ds_prev) { 265 dsl_dataset_rele(ds->ds_prev, ds); 266 ds->ds_prev = NULL; 267 } 268 269 bplist_destroy(&ds->ds_pending_deadlist); 270 if (ds->ds_phys->ds_deadlist_obj != 0) 271 dsl_deadlist_close(&ds->ds_deadlist); 272 if (ds->ds_dir) 273 dsl_dir_rele(ds->ds_dir, ds); 274 275 ASSERT(!list_link_active(&ds->ds_synced_link)); 276 277 mutex_destroy(&ds->ds_lock); 278 mutex_destroy(&ds->ds_opening_lock); 279 mutex_destroy(&ds->ds_sendstream_lock); 280 refcount_destroy(&ds->ds_longholds); 281 282 kmem_free(ds, sizeof (dsl_dataset_t)); 283 } 284 285 int 286 dsl_dataset_get_snapname(dsl_dataset_t *ds) 287 { 288 dsl_dataset_phys_t *headphys; 289 int err; 290 dmu_buf_t *headdbuf; 291 dsl_pool_t *dp = ds->ds_dir->dd_pool; 292 objset_t *mos = dp->dp_meta_objset; 293 294 if (ds->ds_snapname[0]) 295 return (0); 296 if (ds->ds_phys->ds_next_snap_obj == 0) 297 return (0); 298 299 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, 300 FTAG, &headdbuf); 301 if (err != 0) 302 return (err); 303 headphys = headdbuf->db_data; 304 err = zap_value_search(dp->dp_meta_objset, 305 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 306 dmu_buf_rele(headdbuf, FTAG); 307 return (err); 308 } 309 310 int 311 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 312 { 313 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 314 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 315 matchtype_t mt; 316 int err; 317 318 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 319 mt = MT_FIRST; 320 else 321 mt = MT_EXACT; 322 323 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 324 value, mt, NULL, 0, NULL); 325 if (err == ENOTSUP && mt == MT_FIRST) 326 err = zap_lookup(mos, snapobj, name, 8, 1, value); 327 return (err); 328 } 329 330 int 331 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 332 boolean_t adj_cnt) 333 { 334 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 335 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 336 matchtype_t mt; 337 int err; 338 339 dsl_dir_snap_cmtime_update(ds->ds_dir); 340 341 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 342 mt = MT_FIRST; 343 else 344 mt = MT_EXACT; 345 346 err = zap_remove_norm(mos, snapobj, name, mt, tx); 347 if (err == ENOTSUP && mt == MT_FIRST) 348 err = zap_remove(mos, snapobj, name, tx); 349 350 if (err == 0 && adj_cnt) 351 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 352 DD_FIELD_SNAPSHOT_COUNT, tx); 353 354 return (err); 355 } 356 357 int 358 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 359 dsl_dataset_t **dsp) 360 { 361 objset_t *mos = dp->dp_meta_objset; 362 dmu_buf_t *dbuf; 363 dsl_dataset_t *ds; 364 int err; 365 dmu_object_info_t doi; 366 367 ASSERT(dsl_pool_config_held(dp)); 368 369 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 370 if (err != 0) 371 return (err); 372 373 /* Make sure dsobj has the correct object type. */ 374 dmu_object_info_from_db(dbuf, &doi); 375 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 376 dmu_buf_rele(dbuf, tag); 377 return (SET_ERROR(EINVAL)); 378 } 379 380 ds = dmu_buf_get_user(dbuf); 381 if (ds == NULL) { 382 dsl_dataset_t *winner = NULL; 383 384 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 385 ds->ds_dbuf = dbuf; 386 ds->ds_object = dsobj; 387 ds->ds_phys = dbuf->db_data; 388 389 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 390 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 391 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 392 refcount_create(&ds->ds_longholds); 393 394 bplist_create(&ds->ds_pending_deadlist); 395 dsl_deadlist_open(&ds->ds_deadlist, 396 mos, ds->ds_phys->ds_deadlist_obj); 397 398 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 399 offsetof(dmu_sendarg_t, dsa_link)); 400 401 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 402 err = zap_contains(mos, dsobj, DS_FIELD_LARGE_BLOCKS); 403 if (err == 0) 404 ds->ds_large_blocks = B_TRUE; 405 else 406 ASSERT3U(err, ==, ENOENT); 407 } 408 409 if (err == 0) { 410 err = dsl_dir_hold_obj(dp, 411 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); 412 } 413 if (err != 0) { 414 mutex_destroy(&ds->ds_lock); 415 mutex_destroy(&ds->ds_opening_lock); 416 mutex_destroy(&ds->ds_sendstream_lock); 417 refcount_destroy(&ds->ds_longholds); 418 bplist_destroy(&ds->ds_pending_deadlist); 419 dsl_deadlist_close(&ds->ds_deadlist); 420 kmem_free(ds, sizeof (dsl_dataset_t)); 421 dmu_buf_rele(dbuf, tag); 422 return (err); 423 } 424 425 if (!dsl_dataset_is_snapshot(ds)) { 426 ds->ds_snapname[0] = '\0'; 427 if (ds->ds_phys->ds_prev_snap_obj != 0) { 428 err = dsl_dataset_hold_obj(dp, 429 ds->ds_phys->ds_prev_snap_obj, 430 ds, &ds->ds_prev); 431 } 432 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 433 int zaperr = zap_lookup(mos, ds->ds_object, 434 DS_FIELD_BOOKMARK_NAMES, 435 sizeof (ds->ds_bookmarks), 1, 436 &ds->ds_bookmarks); 437 if (zaperr != ENOENT) 438 VERIFY0(zaperr); 439 } 440 } else { 441 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 442 err = dsl_dataset_get_snapname(ds); 443 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) { 444 err = zap_count( 445 ds->ds_dir->dd_pool->dp_meta_objset, 446 ds->ds_phys->ds_userrefs_obj, 447 &ds->ds_userrefs); 448 } 449 } 450 451 if (err == 0 && !dsl_dataset_is_snapshot(ds)) { 452 err = dsl_prop_get_int_ds(ds, 453 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 454 &ds->ds_reserved); 455 if (err == 0) { 456 err = dsl_prop_get_int_ds(ds, 457 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 458 &ds->ds_quota); 459 } 460 } else { 461 ds->ds_reserved = ds->ds_quota = 0; 462 } 463 464 if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds, 465 &ds->ds_phys, dsl_dataset_evict)) != NULL) { 466 bplist_destroy(&ds->ds_pending_deadlist); 467 dsl_deadlist_close(&ds->ds_deadlist); 468 if (ds->ds_prev) 469 dsl_dataset_rele(ds->ds_prev, ds); 470 dsl_dir_rele(ds->ds_dir, ds); 471 mutex_destroy(&ds->ds_lock); 472 mutex_destroy(&ds->ds_opening_lock); 473 mutex_destroy(&ds->ds_sendstream_lock); 474 refcount_destroy(&ds->ds_longholds); 475 kmem_free(ds, sizeof (dsl_dataset_t)); 476 if (err != 0) { 477 dmu_buf_rele(dbuf, tag); 478 return (err); 479 } 480 ds = winner; 481 } else { 482 ds->ds_fsid_guid = 483 unique_insert(ds->ds_phys->ds_fsid_guid); 484 } 485 } 486 ASSERT3P(ds->ds_dbuf, ==, dbuf); 487 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); 488 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 || 489 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 490 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 491 *dsp = ds; 492 return (0); 493 } 494 495 int 496 dsl_dataset_hold(dsl_pool_t *dp, const char *name, 497 void *tag, dsl_dataset_t **dsp) 498 { 499 dsl_dir_t *dd; 500 const char *snapname; 501 uint64_t obj; 502 int err = 0; 503 504 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 505 if (err != 0) 506 return (err); 507 508 ASSERT(dsl_pool_config_held(dp)); 509 obj = dd->dd_phys->dd_head_dataset_obj; 510 if (obj != 0) 511 err = dsl_dataset_hold_obj(dp, obj, tag, dsp); 512 else 513 err = SET_ERROR(ENOENT); 514 515 /* we may be looking for a snapshot */ 516 if (err == 0 && snapname != NULL) { 517 dsl_dataset_t *ds; 518 519 if (*snapname++ != '@') { 520 dsl_dataset_rele(*dsp, tag); 521 dsl_dir_rele(dd, FTAG); 522 return (SET_ERROR(ENOENT)); 523 } 524 525 dprintf("looking for snapshot '%s'\n", snapname); 526 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); 527 if (err == 0) 528 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 529 dsl_dataset_rele(*dsp, tag); 530 531 if (err == 0) { 532 mutex_enter(&ds->ds_lock); 533 if (ds->ds_snapname[0] == 0) 534 (void) strlcpy(ds->ds_snapname, snapname, 535 sizeof (ds->ds_snapname)); 536 mutex_exit(&ds->ds_lock); 537 *dsp = ds; 538 } 539 } 540 541 dsl_dir_rele(dd, FTAG); 542 return (err); 543 } 544 545 int 546 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 547 void *tag, dsl_dataset_t **dsp) 548 { 549 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 550 if (err != 0) 551 return (err); 552 if (!dsl_dataset_tryown(*dsp, tag)) { 553 dsl_dataset_rele(*dsp, tag); 554 *dsp = NULL; 555 return (SET_ERROR(EBUSY)); 556 } 557 return (0); 558 } 559 560 int 561 dsl_dataset_own(dsl_pool_t *dp, const char *name, 562 void *tag, dsl_dataset_t **dsp) 563 { 564 int err = dsl_dataset_hold(dp, name, tag, dsp); 565 if (err != 0) 566 return (err); 567 if (!dsl_dataset_tryown(*dsp, tag)) { 568 dsl_dataset_rele(*dsp, tag); 569 return (SET_ERROR(EBUSY)); 570 } 571 return (0); 572 } 573 574 /* 575 * See the comment above dsl_pool_hold() for details. In summary, a long 576 * hold is used to prevent destruction of a dataset while the pool hold 577 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 578 * 579 * The dataset and pool must be held when this function is called. After it 580 * is called, the pool hold may be released while the dataset is still held 581 * and accessed. 582 */ 583 void 584 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 585 { 586 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 587 (void) refcount_add(&ds->ds_longholds, tag); 588 } 589 590 void 591 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 592 { 593 (void) refcount_remove(&ds->ds_longholds, tag); 594 } 595 596 /* Return B_TRUE if there are any long holds on this dataset. */ 597 boolean_t 598 dsl_dataset_long_held(dsl_dataset_t *ds) 599 { 600 return (!refcount_is_zero(&ds->ds_longholds)); 601 } 602 603 void 604 dsl_dataset_name(dsl_dataset_t *ds, char *name) 605 { 606 if (ds == NULL) { 607 (void) strcpy(name, "mos"); 608 } else { 609 dsl_dir_name(ds->ds_dir, name); 610 VERIFY0(dsl_dataset_get_snapname(ds)); 611 if (ds->ds_snapname[0]) { 612 (void) strcat(name, "@"); 613 /* 614 * We use a "recursive" mutex so that we 615 * can call dprintf_ds() with ds_lock held. 616 */ 617 if (!MUTEX_HELD(&ds->ds_lock)) { 618 mutex_enter(&ds->ds_lock); 619 (void) strcat(name, ds->ds_snapname); 620 mutex_exit(&ds->ds_lock); 621 } else { 622 (void) strcat(name, ds->ds_snapname); 623 } 624 } 625 } 626 } 627 628 void 629 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 630 { 631 dmu_buf_rele(ds->ds_dbuf, tag); 632 } 633 634 void 635 dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 636 { 637 ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL); 638 639 mutex_enter(&ds->ds_lock); 640 ds->ds_owner = NULL; 641 mutex_exit(&ds->ds_lock); 642 dsl_dataset_long_rele(ds, tag); 643 if (ds->ds_dbuf != NULL) 644 dsl_dataset_rele(ds, tag); 645 else 646 dsl_dataset_evict(NULL, ds); 647 } 648 649 boolean_t 650 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 651 { 652 boolean_t gotit = FALSE; 653 654 mutex_enter(&ds->ds_lock); 655 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 656 ds->ds_owner = tag; 657 dsl_dataset_long_hold(ds, tag); 658 gotit = TRUE; 659 } 660 mutex_exit(&ds->ds_lock); 661 return (gotit); 662 } 663 664 uint64_t 665 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 666 uint64_t flags, dmu_tx_t *tx) 667 { 668 dsl_pool_t *dp = dd->dd_pool; 669 dmu_buf_t *dbuf; 670 dsl_dataset_phys_t *dsphys; 671 uint64_t dsobj; 672 objset_t *mos = dp->dp_meta_objset; 673 674 if (origin == NULL) 675 origin = dp->dp_origin_snap; 676 677 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 678 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0); 679 ASSERT(dmu_tx_is_syncing(tx)); 680 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 681 682 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 683 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 684 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 685 dmu_buf_will_dirty(dbuf, tx); 686 dsphys = dbuf->db_data; 687 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 688 dsphys->ds_dir_obj = dd->dd_object; 689 dsphys->ds_flags = flags; 690 dsphys->ds_fsid_guid = unique_create(); 691 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 692 sizeof (dsphys->ds_guid)); 693 dsphys->ds_snapnames_zapobj = 694 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 695 DMU_OT_NONE, 0, tx); 696 dsphys->ds_creation_time = gethrestime_sec(); 697 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 698 699 if (origin == NULL) { 700 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 701 } else { 702 dsl_dataset_t *ohds; /* head of the origin snapshot */ 703 704 dsphys->ds_prev_snap_obj = origin->ds_object; 705 dsphys->ds_prev_snap_txg = 706 origin->ds_phys->ds_creation_txg; 707 dsphys->ds_referenced_bytes = 708 origin->ds_phys->ds_referenced_bytes; 709 dsphys->ds_compressed_bytes = 710 origin->ds_phys->ds_compressed_bytes; 711 dsphys->ds_uncompressed_bytes = 712 origin->ds_phys->ds_uncompressed_bytes; 713 dsphys->ds_bp = origin->ds_phys->ds_bp; 714 715 /* 716 * Inherit flags that describe the dataset's contents 717 * (INCONSISTENT) or properties (Case Insensitive). 718 */ 719 dsphys->ds_flags |= origin->ds_phys->ds_flags & 720 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 721 722 if (origin->ds_large_blocks) 723 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); 724 725 dmu_buf_will_dirty(origin->ds_dbuf, tx); 726 origin->ds_phys->ds_num_children++; 727 728 VERIFY0(dsl_dataset_hold_obj(dp, 729 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds)); 730 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 731 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 732 dsl_dataset_rele(ohds, FTAG); 733 734 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 735 if (origin->ds_phys->ds_next_clones_obj == 0) { 736 origin->ds_phys->ds_next_clones_obj = 737 zap_create(mos, 738 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 739 } 740 VERIFY0(zap_add_int(mos, 741 origin->ds_phys->ds_next_clones_obj, dsobj, tx)); 742 } 743 744 dmu_buf_will_dirty(dd->dd_dbuf, tx); 745 dd->dd_phys->dd_origin_obj = origin->ds_object; 746 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 747 if (origin->ds_dir->dd_phys->dd_clones == 0) { 748 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 749 origin->ds_dir->dd_phys->dd_clones = 750 zap_create(mos, 751 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 752 } 753 VERIFY0(zap_add_int(mos, 754 origin->ds_dir->dd_phys->dd_clones, dsobj, tx)); 755 } 756 } 757 758 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 759 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 760 761 dmu_buf_rele(dbuf, FTAG); 762 763 dmu_buf_will_dirty(dd->dd_dbuf, tx); 764 dd->dd_phys->dd_head_dataset_obj = dsobj; 765 766 return (dsobj); 767 } 768 769 static void 770 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 771 { 772 objset_t *os; 773 774 VERIFY0(dmu_objset_from_ds(ds, &os)); 775 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 776 dsl_dataset_dirty(ds, tx); 777 } 778 779 uint64_t 780 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 781 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 782 { 783 dsl_pool_t *dp = pdd->dd_pool; 784 uint64_t dsobj, ddobj; 785 dsl_dir_t *dd; 786 787 ASSERT(dmu_tx_is_syncing(tx)); 788 ASSERT(lastname[0] != '@'); 789 790 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 791 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 792 793 dsobj = dsl_dataset_create_sync_dd(dd, origin, 794 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 795 796 dsl_deleg_set_create_perms(dd, tx, cr); 797 798 /* 799 * Since we're creating a new node we know it's a leaf, so we can 800 * initialize the counts if the limit feature is active. 801 */ 802 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 803 uint64_t cnt = 0; 804 objset_t *os = dd->dd_pool->dp_meta_objset; 805 806 dsl_dir_zapify(dd, tx); 807 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 808 sizeof (cnt), 1, &cnt, tx)); 809 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 810 sizeof (cnt), 1, &cnt, tx)); 811 } 812 813 dsl_dir_rele(dd, FTAG); 814 815 /* 816 * If we are creating a clone, make sure we zero out any stale 817 * data from the origin snapshots zil header. 818 */ 819 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 820 dsl_dataset_t *ds; 821 822 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 823 dsl_dataset_zero_zil(ds, tx); 824 dsl_dataset_rele(ds, FTAG); 825 } 826 827 return (dsobj); 828 } 829 830 /* 831 * The unique space in the head dataset can be calculated by subtracting 832 * the space used in the most recent snapshot, that is still being used 833 * in this file system, from the space currently in use. To figure out 834 * the space in the most recent snapshot still in use, we need to take 835 * the total space used in the snapshot and subtract out the space that 836 * has been freed up since the snapshot was taken. 837 */ 838 void 839 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 840 { 841 uint64_t mrs_used; 842 uint64_t dlused, dlcomp, dluncomp; 843 844 ASSERT(!dsl_dataset_is_snapshot(ds)); 845 846 if (ds->ds_phys->ds_prev_snap_obj != 0) 847 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes; 848 else 849 mrs_used = 0; 850 851 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 852 853 ASSERT3U(dlused, <=, mrs_used); 854 ds->ds_phys->ds_unique_bytes = 855 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused); 856 857 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 858 SPA_VERSION_UNIQUE_ACCURATE) 859 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 860 } 861 862 void 863 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 864 dmu_tx_t *tx) 865 { 866 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 867 uint64_t count; 868 int err; 869 870 ASSERT(ds->ds_phys->ds_num_children >= 2); 871 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx); 872 /* 873 * The err should not be ENOENT, but a bug in a previous version 874 * of the code could cause upgrade_clones_cb() to not set 875 * ds_next_snap_obj when it should, leading to a missing entry. 876 * If we knew that the pool was created after 877 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 878 * ENOENT. However, at least we can check that we don't have 879 * too many entries in the next_clones_obj even after failing to 880 * remove this one. 881 */ 882 if (err != ENOENT) 883 VERIFY0(err); 884 ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj, 885 &count)); 886 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2); 887 } 888 889 890 blkptr_t * 891 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 892 { 893 return (&ds->ds_phys->ds_bp); 894 } 895 896 void 897 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 898 { 899 ASSERT(dmu_tx_is_syncing(tx)); 900 /* If it's the meta-objset, set dp_meta_rootbp */ 901 if (ds == NULL) { 902 tx->tx_pool->dp_meta_rootbp = *bp; 903 } else { 904 dmu_buf_will_dirty(ds->ds_dbuf, tx); 905 ds->ds_phys->ds_bp = *bp; 906 } 907 } 908 909 spa_t * 910 dsl_dataset_get_spa(dsl_dataset_t *ds) 911 { 912 return (ds->ds_dir->dd_pool->dp_spa); 913 } 914 915 void 916 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 917 { 918 dsl_pool_t *dp; 919 920 if (ds == NULL) /* this is the meta-objset */ 921 return; 922 923 ASSERT(ds->ds_objset != NULL); 924 925 if (ds->ds_phys->ds_next_snap_obj != 0) 926 panic("dirtying snapshot!"); 927 928 dp = ds->ds_dir->dd_pool; 929 930 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 931 /* up the hold count until we can be written out */ 932 dmu_buf_add_ref(ds->ds_dbuf, ds); 933 } 934 } 935 936 boolean_t 937 dsl_dataset_is_dirty(dsl_dataset_t *ds) 938 { 939 for (int t = 0; t < TXG_SIZE; t++) { 940 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 941 ds, t)) 942 return (B_TRUE); 943 } 944 return (B_FALSE); 945 } 946 947 static int 948 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 949 { 950 uint64_t asize; 951 952 if (!dmu_tx_is_syncing(tx)) 953 return (0); 954 955 /* 956 * If there's an fs-only reservation, any blocks that might become 957 * owned by the snapshot dataset must be accommodated by space 958 * outside of the reservation. 959 */ 960 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 961 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 962 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 963 return (SET_ERROR(ENOSPC)); 964 965 /* 966 * Propagate any reserved space for this snapshot to other 967 * snapshot checks in this sync group. 968 */ 969 if (asize > 0) 970 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 971 972 return (0); 973 } 974 975 typedef struct dsl_dataset_snapshot_arg { 976 nvlist_t *ddsa_snaps; 977 nvlist_t *ddsa_props; 978 nvlist_t *ddsa_errors; 979 cred_t *ddsa_cr; 980 } dsl_dataset_snapshot_arg_t; 981 982 int 983 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 984 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 985 { 986 int error; 987 uint64_t value; 988 989 ds->ds_trysnap_txg = tx->tx_txg; 990 991 if (!dmu_tx_is_syncing(tx)) 992 return (0); 993 994 /* 995 * We don't allow multiple snapshots of the same txg. If there 996 * is already one, try again. 997 */ 998 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) 999 return (SET_ERROR(EAGAIN)); 1000 1001 /* 1002 * Check for conflicting snapshot name. 1003 */ 1004 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1005 if (error == 0) 1006 return (SET_ERROR(EEXIST)); 1007 if (error != ENOENT) 1008 return (error); 1009 1010 /* 1011 * We don't allow taking snapshots of inconsistent datasets, such as 1012 * those into which we are currently receiving. However, if we are 1013 * creating this snapshot as part of a receive, this check will be 1014 * executed atomically with respect to the completion of the receive 1015 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1016 * case we ignore this, knowing it will be fixed up for us shortly in 1017 * dmu_recv_end_sync(). 1018 */ 1019 if (!recv && DS_IS_INCONSISTENT(ds)) 1020 return (SET_ERROR(EBUSY)); 1021 1022 /* 1023 * Skip the check for temporary snapshots or if we have already checked 1024 * the counts in dsl_dataset_snapshot_check. This means we really only 1025 * check the count here when we're receiving a stream. 1026 */ 1027 if (cnt != 0 && cr != NULL) { 1028 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1029 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 1030 if (error != 0) 1031 return (error); 1032 } 1033 1034 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1035 if (error != 0) 1036 return (error); 1037 1038 return (0); 1039 } 1040 1041 static int 1042 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1043 { 1044 dsl_dataset_snapshot_arg_t *ddsa = arg; 1045 dsl_pool_t *dp = dmu_tx_pool(tx); 1046 nvpair_t *pair; 1047 int rv = 0; 1048 1049 /* 1050 * Pre-compute how many total new snapshots will be created for each 1051 * level in the tree and below. This is needed for validating the 1052 * snapshot limit when either taking a recursive snapshot or when 1053 * taking multiple snapshots. 1054 * 1055 * The problem is that the counts are not actually adjusted when 1056 * we are checking, only when we finally sync. For a single snapshot, 1057 * this is easy, the count will increase by 1 at each node up the tree, 1058 * but its more complicated for the recursive/multiple snapshot case. 1059 * 1060 * The dsl_fs_ss_limit_check function does recursively check the count 1061 * at each level up the tree but since it is validating each snapshot 1062 * independently we need to be sure that we are validating the complete 1063 * count for the entire set of snapshots. We do this by rolling up the 1064 * counts for each component of the name into an nvlist and then 1065 * checking each of those cases with the aggregated count. 1066 * 1067 * This approach properly handles not only the recursive snapshot 1068 * case (where we get all of those on the ddsa_snaps list) but also 1069 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1070 * validate the limit on 'a' using a count of 2). 1071 * 1072 * We validate the snapshot names in the third loop and only report 1073 * name errors once. 1074 */ 1075 if (dmu_tx_is_syncing(tx)) { 1076 nvlist_t *cnt_track = NULL; 1077 cnt_track = fnvlist_alloc(); 1078 1079 /* Rollup aggregated counts into the cnt_track list */ 1080 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1081 pair != NULL; 1082 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1083 char *pdelim; 1084 uint64_t val; 1085 char nm[MAXPATHLEN]; 1086 1087 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1088 pdelim = strchr(nm, '@'); 1089 if (pdelim == NULL) 1090 continue; 1091 *pdelim = '\0'; 1092 1093 do { 1094 if (nvlist_lookup_uint64(cnt_track, nm, 1095 &val) == 0) { 1096 /* update existing entry */ 1097 fnvlist_add_uint64(cnt_track, nm, 1098 val + 1); 1099 } else { 1100 /* add to list */ 1101 fnvlist_add_uint64(cnt_track, nm, 1); 1102 } 1103 1104 pdelim = strrchr(nm, '/'); 1105 if (pdelim != NULL) 1106 *pdelim = '\0'; 1107 } while (pdelim != NULL); 1108 } 1109 1110 /* Check aggregated counts at each level */ 1111 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1112 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1113 int error = 0; 1114 char *name; 1115 uint64_t cnt = 0; 1116 dsl_dataset_t *ds; 1117 1118 name = nvpair_name(pair); 1119 cnt = fnvpair_value_uint64(pair); 1120 ASSERT(cnt > 0); 1121 1122 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1123 if (error == 0) { 1124 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1125 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1126 ddsa->ddsa_cr); 1127 dsl_dataset_rele(ds, FTAG); 1128 } 1129 1130 if (error != 0) { 1131 if (ddsa->ddsa_errors != NULL) 1132 fnvlist_add_int32(ddsa->ddsa_errors, 1133 name, error); 1134 rv = error; 1135 /* only report one error for this check */ 1136 break; 1137 } 1138 } 1139 nvlist_free(cnt_track); 1140 } 1141 1142 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1143 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1144 int error = 0; 1145 dsl_dataset_t *ds; 1146 char *name, *atp; 1147 char dsname[MAXNAMELEN]; 1148 1149 name = nvpair_name(pair); 1150 if (strlen(name) >= MAXNAMELEN) 1151 error = SET_ERROR(ENAMETOOLONG); 1152 if (error == 0) { 1153 atp = strchr(name, '@'); 1154 if (atp == NULL) 1155 error = SET_ERROR(EINVAL); 1156 if (error == 0) 1157 (void) strlcpy(dsname, name, atp - name + 1); 1158 } 1159 if (error == 0) 1160 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1161 if (error == 0) { 1162 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1163 error = dsl_dataset_snapshot_check_impl(ds, 1164 atp + 1, tx, B_FALSE, 0, NULL); 1165 dsl_dataset_rele(ds, FTAG); 1166 } 1167 1168 if (error != 0) { 1169 if (ddsa->ddsa_errors != NULL) { 1170 fnvlist_add_int32(ddsa->ddsa_errors, 1171 name, error); 1172 } 1173 rv = error; 1174 } 1175 } 1176 1177 return (rv); 1178 } 1179 1180 void 1181 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1182 dmu_tx_t *tx) 1183 { 1184 static zil_header_t zero_zil; 1185 1186 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1187 dmu_buf_t *dbuf; 1188 dsl_dataset_phys_t *dsphys; 1189 uint64_t dsobj, crtxg; 1190 objset_t *mos = dp->dp_meta_objset; 1191 objset_t *os; 1192 1193 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1194 1195 /* 1196 * If we are on an old pool, the zil must not be active, in which 1197 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1198 */ 1199 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1200 dmu_objset_from_ds(ds, &os) != 0 || 1201 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1202 sizeof (zero_zil)) == 0); 1203 1204 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1205 1206 /* 1207 * The origin's ds_creation_txg has to be < TXG_INITIAL 1208 */ 1209 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1210 crtxg = 1; 1211 else 1212 crtxg = tx->tx_txg; 1213 1214 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1215 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1216 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1217 dmu_buf_will_dirty(dbuf, tx); 1218 dsphys = dbuf->db_data; 1219 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1220 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1221 dsphys->ds_fsid_guid = unique_create(); 1222 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1223 sizeof (dsphys->ds_guid)); 1224 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; 1225 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; 1226 dsphys->ds_next_snap_obj = ds->ds_object; 1227 dsphys->ds_num_children = 1; 1228 dsphys->ds_creation_time = gethrestime_sec(); 1229 dsphys->ds_creation_txg = crtxg; 1230 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; 1231 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes; 1232 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; 1233 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; 1234 dsphys->ds_flags = ds->ds_phys->ds_flags; 1235 dsphys->ds_bp = ds->ds_phys->ds_bp; 1236 dmu_buf_rele(dbuf, FTAG); 1237 1238 if (ds->ds_large_blocks) 1239 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); 1240 1241 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); 1242 if (ds->ds_prev) { 1243 uint64_t next_clones_obj = 1244 ds->ds_prev->ds_phys->ds_next_clones_obj; 1245 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == 1246 ds->ds_object || 1247 ds->ds_prev->ds_phys->ds_num_children > 1); 1248 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { 1249 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1250 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1251 ds->ds_prev->ds_phys->ds_creation_txg); 1252 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; 1253 } else if (next_clones_obj != 0) { 1254 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1255 dsphys->ds_next_snap_obj, tx); 1256 VERIFY0(zap_add_int(mos, 1257 next_clones_obj, dsobj, tx)); 1258 } 1259 } 1260 1261 /* 1262 * If we have a reference-reservation on this dataset, we will 1263 * need to increase the amount of refreservation being charged 1264 * since our unique space is going to zero. 1265 */ 1266 if (ds->ds_reserved) { 1267 int64_t delta; 1268 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1269 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 1270 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1271 delta, 0, 0, tx); 1272 } 1273 1274 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1275 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist, 1276 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx); 1277 dsl_deadlist_close(&ds->ds_deadlist); 1278 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); 1279 dsl_deadlist_add_key(&ds->ds_deadlist, 1280 ds->ds_phys->ds_prev_snap_txg, tx); 1281 1282 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg); 1283 ds->ds_phys->ds_prev_snap_obj = dsobj; 1284 ds->ds_phys->ds_prev_snap_txg = crtxg; 1285 ds->ds_phys->ds_unique_bytes = 0; 1286 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1287 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1288 1289 VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, 1290 snapname, 8, 1, &dsobj, tx)); 1291 1292 if (ds->ds_prev) 1293 dsl_dataset_rele(ds->ds_prev, ds); 1294 VERIFY0(dsl_dataset_hold_obj(dp, 1295 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 1296 1297 dsl_scan_ds_snapshotted(ds, tx); 1298 1299 dsl_dir_snap_cmtime_update(ds->ds_dir); 1300 1301 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1302 } 1303 1304 static void 1305 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1306 { 1307 dsl_dataset_snapshot_arg_t *ddsa = arg; 1308 dsl_pool_t *dp = dmu_tx_pool(tx); 1309 nvpair_t *pair; 1310 1311 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1312 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1313 dsl_dataset_t *ds; 1314 char *name, *atp; 1315 char dsname[MAXNAMELEN]; 1316 1317 name = nvpair_name(pair); 1318 atp = strchr(name, '@'); 1319 (void) strlcpy(dsname, name, atp - name + 1); 1320 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1321 1322 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1323 if (ddsa->ddsa_props != NULL) { 1324 dsl_props_set_sync_impl(ds->ds_prev, 1325 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1326 } 1327 dsl_dataset_rele(ds, FTAG); 1328 } 1329 } 1330 1331 /* 1332 * The snapshots must all be in the same pool. 1333 * All-or-nothing: if there are any failures, nothing will be modified. 1334 */ 1335 int 1336 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1337 { 1338 dsl_dataset_snapshot_arg_t ddsa; 1339 nvpair_t *pair; 1340 boolean_t needsuspend; 1341 int error; 1342 spa_t *spa; 1343 char *firstname; 1344 nvlist_t *suspended = NULL; 1345 1346 pair = nvlist_next_nvpair(snaps, NULL); 1347 if (pair == NULL) 1348 return (0); 1349 firstname = nvpair_name(pair); 1350 1351 error = spa_open(firstname, &spa, FTAG); 1352 if (error != 0) 1353 return (error); 1354 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1355 spa_close(spa, FTAG); 1356 1357 if (needsuspend) { 1358 suspended = fnvlist_alloc(); 1359 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1360 pair = nvlist_next_nvpair(snaps, pair)) { 1361 char fsname[MAXNAMELEN]; 1362 char *snapname = nvpair_name(pair); 1363 char *atp; 1364 void *cookie; 1365 1366 atp = strchr(snapname, '@'); 1367 if (atp == NULL) { 1368 error = SET_ERROR(EINVAL); 1369 break; 1370 } 1371 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1372 1373 error = zil_suspend(fsname, &cookie); 1374 if (error != 0) 1375 break; 1376 fnvlist_add_uint64(suspended, fsname, 1377 (uintptr_t)cookie); 1378 } 1379 } 1380 1381 ddsa.ddsa_snaps = snaps; 1382 ddsa.ddsa_props = props; 1383 ddsa.ddsa_errors = errors; 1384 ddsa.ddsa_cr = CRED(); 1385 1386 if (error == 0) { 1387 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1388 dsl_dataset_snapshot_sync, &ddsa, 1389 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1390 } 1391 1392 if (suspended != NULL) { 1393 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1394 pair = nvlist_next_nvpair(suspended, pair)) { 1395 zil_resume((void *)(uintptr_t) 1396 fnvpair_value_uint64(pair)); 1397 } 1398 fnvlist_free(suspended); 1399 } 1400 1401 return (error); 1402 } 1403 1404 typedef struct dsl_dataset_snapshot_tmp_arg { 1405 const char *ddsta_fsname; 1406 const char *ddsta_snapname; 1407 minor_t ddsta_cleanup_minor; 1408 const char *ddsta_htag; 1409 } dsl_dataset_snapshot_tmp_arg_t; 1410 1411 static int 1412 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1413 { 1414 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1415 dsl_pool_t *dp = dmu_tx_pool(tx); 1416 dsl_dataset_t *ds; 1417 int error; 1418 1419 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1420 if (error != 0) 1421 return (error); 1422 1423 /* NULL cred means no limit check for tmp snapshot */ 1424 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1425 tx, B_FALSE, 0, NULL); 1426 if (error != 0) { 1427 dsl_dataset_rele(ds, FTAG); 1428 return (error); 1429 } 1430 1431 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1432 dsl_dataset_rele(ds, FTAG); 1433 return (SET_ERROR(ENOTSUP)); 1434 } 1435 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1436 B_TRUE, tx); 1437 if (error != 0) { 1438 dsl_dataset_rele(ds, FTAG); 1439 return (error); 1440 } 1441 1442 dsl_dataset_rele(ds, FTAG); 1443 return (0); 1444 } 1445 1446 static void 1447 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1448 { 1449 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1450 dsl_pool_t *dp = dmu_tx_pool(tx); 1451 dsl_dataset_t *ds; 1452 1453 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1454 1455 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1456 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1457 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1458 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1459 1460 dsl_dataset_rele(ds, FTAG); 1461 } 1462 1463 int 1464 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1465 minor_t cleanup_minor, const char *htag) 1466 { 1467 dsl_dataset_snapshot_tmp_arg_t ddsta; 1468 int error; 1469 spa_t *spa; 1470 boolean_t needsuspend; 1471 void *cookie; 1472 1473 ddsta.ddsta_fsname = fsname; 1474 ddsta.ddsta_snapname = snapname; 1475 ddsta.ddsta_cleanup_minor = cleanup_minor; 1476 ddsta.ddsta_htag = htag; 1477 1478 error = spa_open(fsname, &spa, FTAG); 1479 if (error != 0) 1480 return (error); 1481 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1482 spa_close(spa, FTAG); 1483 1484 if (needsuspend) { 1485 error = zil_suspend(fsname, &cookie); 1486 if (error != 0) 1487 return (error); 1488 } 1489 1490 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1491 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 1492 1493 if (needsuspend) 1494 zil_resume(cookie); 1495 return (error); 1496 } 1497 1498 1499 void 1500 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1501 { 1502 ASSERT(dmu_tx_is_syncing(tx)); 1503 ASSERT(ds->ds_objset != NULL); 1504 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); 1505 1506 /* 1507 * in case we had to change ds_fsid_guid when we opened it, 1508 * sync it out now. 1509 */ 1510 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1511 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; 1512 1513 dmu_objset_sync(ds->ds_objset, zio, tx); 1514 1515 if (ds->ds_need_large_blocks && !ds->ds_large_blocks) { 1516 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx); 1517 ds->ds_large_blocks = B_TRUE; 1518 } 1519 } 1520 1521 static void 1522 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1523 { 1524 uint64_t count = 0; 1525 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1526 zap_cursor_t zc; 1527 zap_attribute_t za; 1528 nvlist_t *propval = fnvlist_alloc(); 1529 nvlist_t *val = fnvlist_alloc(); 1530 1531 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1532 1533 /* 1534 * There may be missing entries in ds_next_clones_obj 1535 * due to a bug in a previous version of the code. 1536 * Only trust it if it has the right number of entries. 1537 */ 1538 if (ds->ds_phys->ds_next_clones_obj != 0) { 1539 VERIFY0(zap_count(mos, ds->ds_phys->ds_next_clones_obj, 1540 &count)); 1541 } 1542 if (count != ds->ds_phys->ds_num_children - 1) 1543 goto fail; 1544 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj); 1545 zap_cursor_retrieve(&zc, &za) == 0; 1546 zap_cursor_advance(&zc)) { 1547 dsl_dataset_t *clone; 1548 char buf[ZFS_MAXNAMELEN]; 1549 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1550 za.za_first_integer, FTAG, &clone)); 1551 dsl_dir_name(clone->ds_dir, buf); 1552 fnvlist_add_boolean(val, buf); 1553 dsl_dataset_rele(clone, FTAG); 1554 } 1555 zap_cursor_fini(&zc); 1556 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1557 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); 1558 fail: 1559 nvlist_free(val); 1560 nvlist_free(propval); 1561 } 1562 1563 void 1564 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1565 { 1566 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1567 uint64_t refd, avail, uobjs, aobjs, ratio; 1568 1569 ASSERT(dsl_pool_config_held(dp)); 1570 1571 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 : 1572 (ds->ds_phys->ds_uncompressed_bytes * 100 / 1573 ds->ds_phys->ds_compressed_bytes); 1574 1575 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); 1576 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 1577 ds->ds_phys->ds_uncompressed_bytes); 1578 1579 if (dsl_dataset_is_snapshot(ds)) { 1580 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); 1581 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 1582 ds->ds_phys->ds_unique_bytes); 1583 get_clones_stat(ds, nv); 1584 } else { 1585 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 1586 char buf[MAXNAMELEN]; 1587 dsl_dataset_name(ds->ds_prev, buf); 1588 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf); 1589 } 1590 1591 dsl_dir_stats(ds->ds_dir, nv); 1592 } 1593 1594 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 1595 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 1596 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 1597 1598 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 1599 ds->ds_phys->ds_creation_time); 1600 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 1601 ds->ds_phys->ds_creation_txg); 1602 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 1603 ds->ds_quota); 1604 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 1605 ds->ds_reserved); 1606 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 1607 ds->ds_phys->ds_guid); 1608 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 1609 ds->ds_phys->ds_unique_bytes); 1610 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 1611 ds->ds_object); 1612 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 1613 ds->ds_userrefs); 1614 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 1615 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 1616 1617 if (ds->ds_phys->ds_prev_snap_obj != 0) { 1618 uint64_t written, comp, uncomp; 1619 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1620 dsl_dataset_t *prev; 1621 1622 int err = dsl_dataset_hold_obj(dp, 1623 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev); 1624 if (err == 0) { 1625 err = dsl_dataset_space_written(prev, ds, &written, 1626 &comp, &uncomp); 1627 dsl_dataset_rele(prev, FTAG); 1628 if (err == 0) { 1629 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 1630 written); 1631 } 1632 } 1633 } 1634 } 1635 1636 void 1637 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 1638 { 1639 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1640 ASSERT(dsl_pool_config_held(dp)); 1641 1642 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; 1643 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; 1644 stat->dds_guid = ds->ds_phys->ds_guid; 1645 stat->dds_origin[0] = '\0'; 1646 if (dsl_dataset_is_snapshot(ds)) { 1647 stat->dds_is_snapshot = B_TRUE; 1648 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; 1649 } else { 1650 stat->dds_is_snapshot = B_FALSE; 1651 stat->dds_num_clones = 0; 1652 1653 if (dsl_dir_is_clone(ds->ds_dir)) { 1654 dsl_dataset_t *ods; 1655 1656 VERIFY0(dsl_dataset_hold_obj(dp, 1657 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods)); 1658 dsl_dataset_name(ods, stat->dds_origin); 1659 dsl_dataset_rele(ods, FTAG); 1660 } 1661 } 1662 } 1663 1664 uint64_t 1665 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 1666 { 1667 return (ds->ds_fsid_guid); 1668 } 1669 1670 void 1671 dsl_dataset_space(dsl_dataset_t *ds, 1672 uint64_t *refdbytesp, uint64_t *availbytesp, 1673 uint64_t *usedobjsp, uint64_t *availobjsp) 1674 { 1675 *refdbytesp = ds->ds_phys->ds_referenced_bytes; 1676 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 1677 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) 1678 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes; 1679 if (ds->ds_quota != 0) { 1680 /* 1681 * Adjust available bytes according to refquota 1682 */ 1683 if (*refdbytesp < ds->ds_quota) 1684 *availbytesp = MIN(*availbytesp, 1685 ds->ds_quota - *refdbytesp); 1686 else 1687 *availbytesp = 0; 1688 } 1689 *usedobjsp = BP_GET_FILL(&ds->ds_phys->ds_bp); 1690 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 1691 } 1692 1693 boolean_t 1694 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 1695 { 1696 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1697 1698 ASSERT(dsl_pool_config_held(dp)); 1699 if (snap == NULL) 1700 return (B_FALSE); 1701 if (ds->ds_phys->ds_bp.blk_birth > 1702 snap->ds_phys->ds_creation_txg) { 1703 objset_t *os, *os_snap; 1704 /* 1705 * It may be that only the ZIL differs, because it was 1706 * reset in the head. Don't count that as being 1707 * modified. 1708 */ 1709 if (dmu_objset_from_ds(ds, &os) != 0) 1710 return (B_TRUE); 1711 if (dmu_objset_from_ds(snap, &os_snap) != 0) 1712 return (B_TRUE); 1713 return (bcmp(&os->os_phys->os_meta_dnode, 1714 &os_snap->os_phys->os_meta_dnode, 1715 sizeof (os->os_phys->os_meta_dnode)) != 0); 1716 } 1717 return (B_FALSE); 1718 } 1719 1720 typedef struct dsl_dataset_rename_snapshot_arg { 1721 const char *ddrsa_fsname; 1722 const char *ddrsa_oldsnapname; 1723 const char *ddrsa_newsnapname; 1724 boolean_t ddrsa_recursive; 1725 dmu_tx_t *ddrsa_tx; 1726 } dsl_dataset_rename_snapshot_arg_t; 1727 1728 /* ARGSUSED */ 1729 static int 1730 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 1731 dsl_dataset_t *hds, void *arg) 1732 { 1733 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1734 int error; 1735 uint64_t val; 1736 1737 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1738 if (error != 0) { 1739 /* ignore nonexistent snapshots */ 1740 return (error == ENOENT ? 0 : error); 1741 } 1742 1743 /* new name should not exist */ 1744 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 1745 if (error == 0) 1746 error = SET_ERROR(EEXIST); 1747 else if (error == ENOENT) 1748 error = 0; 1749 1750 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 1751 if (dsl_dir_namelen(hds->ds_dir) + 1 + 1752 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN) 1753 error = SET_ERROR(ENAMETOOLONG); 1754 1755 return (error); 1756 } 1757 1758 static int 1759 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 1760 { 1761 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1762 dsl_pool_t *dp = dmu_tx_pool(tx); 1763 dsl_dataset_t *hds; 1764 int error; 1765 1766 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 1767 if (error != 0) 1768 return (error); 1769 1770 if (ddrsa->ddrsa_recursive) { 1771 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1772 dsl_dataset_rename_snapshot_check_impl, ddrsa, 1773 DS_FIND_CHILDREN); 1774 } else { 1775 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 1776 } 1777 dsl_dataset_rele(hds, FTAG); 1778 return (error); 1779 } 1780 1781 static int 1782 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 1783 dsl_dataset_t *hds, void *arg) 1784 { 1785 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1786 dsl_dataset_t *ds; 1787 uint64_t val; 1788 dmu_tx_t *tx = ddrsa->ddrsa_tx; 1789 int error; 1790 1791 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1792 ASSERT(error == 0 || error == ENOENT); 1793 if (error == ENOENT) { 1794 /* ignore nonexistent snapshots */ 1795 return (0); 1796 } 1797 1798 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 1799 1800 /* log before we change the name */ 1801 spa_history_log_internal_ds(ds, "rename", tx, 1802 "-> @%s", ddrsa->ddrsa_newsnapname); 1803 1804 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 1805 B_FALSE)); 1806 mutex_enter(&ds->ds_lock); 1807 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 1808 mutex_exit(&ds->ds_lock); 1809 VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj, 1810 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 1811 1812 dsl_dataset_rele(ds, FTAG); 1813 return (0); 1814 } 1815 1816 static void 1817 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 1818 { 1819 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1820 dsl_pool_t *dp = dmu_tx_pool(tx); 1821 dsl_dataset_t *hds; 1822 1823 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 1824 ddrsa->ddrsa_tx = tx; 1825 if (ddrsa->ddrsa_recursive) { 1826 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1827 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 1828 DS_FIND_CHILDREN)); 1829 } else { 1830 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 1831 } 1832 dsl_dataset_rele(hds, FTAG); 1833 } 1834 1835 int 1836 dsl_dataset_rename_snapshot(const char *fsname, 1837 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 1838 { 1839 dsl_dataset_rename_snapshot_arg_t ddrsa; 1840 1841 ddrsa.ddrsa_fsname = fsname; 1842 ddrsa.ddrsa_oldsnapname = oldsnapname; 1843 ddrsa.ddrsa_newsnapname = newsnapname; 1844 ddrsa.ddrsa_recursive = recursive; 1845 1846 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 1847 dsl_dataset_rename_snapshot_sync, &ddrsa, 1848 1, ZFS_SPACE_CHECK_RESERVED)); 1849 } 1850 1851 /* 1852 * If we're doing an ownership handoff, we need to make sure that there is 1853 * only one long hold on the dataset. We're not allowed to change anything here 1854 * so we don't permanently release the long hold or regular hold here. We want 1855 * to do this only when syncing to avoid the dataset unexpectedly going away 1856 * when we release the long hold. 1857 */ 1858 static int 1859 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 1860 { 1861 boolean_t held; 1862 1863 if (!dmu_tx_is_syncing(tx)) 1864 return (0); 1865 1866 if (owner != NULL) { 1867 VERIFY3P(ds->ds_owner, ==, owner); 1868 dsl_dataset_long_rele(ds, owner); 1869 } 1870 1871 held = dsl_dataset_long_held(ds); 1872 1873 if (owner != NULL) 1874 dsl_dataset_long_hold(ds, owner); 1875 1876 if (held) 1877 return (SET_ERROR(EBUSY)); 1878 1879 return (0); 1880 } 1881 1882 typedef struct dsl_dataset_rollback_arg { 1883 const char *ddra_fsname; 1884 void *ddra_owner; 1885 nvlist_t *ddra_result; 1886 } dsl_dataset_rollback_arg_t; 1887 1888 static int 1889 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 1890 { 1891 dsl_dataset_rollback_arg_t *ddra = arg; 1892 dsl_pool_t *dp = dmu_tx_pool(tx); 1893 dsl_dataset_t *ds; 1894 int64_t unused_refres_delta; 1895 int error; 1896 1897 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 1898 if (error != 0) 1899 return (error); 1900 1901 /* must not be a snapshot */ 1902 if (dsl_dataset_is_snapshot(ds)) { 1903 dsl_dataset_rele(ds, FTAG); 1904 return (SET_ERROR(EINVAL)); 1905 } 1906 1907 /* must have a most recent snapshot */ 1908 if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) { 1909 dsl_dataset_rele(ds, FTAG); 1910 return (SET_ERROR(EINVAL)); 1911 } 1912 1913 /* must not have any bookmarks after the most recent snapshot */ 1914 nvlist_t *proprequest = fnvlist_alloc(); 1915 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 1916 nvlist_t *bookmarks = fnvlist_alloc(); 1917 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 1918 fnvlist_free(proprequest); 1919 if (error != 0) 1920 return (error); 1921 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 1922 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 1923 nvlist_t *valuenv = 1924 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 1925 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 1926 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 1927 if (createtxg > ds->ds_phys->ds_prev_snap_txg) { 1928 fnvlist_free(bookmarks); 1929 dsl_dataset_rele(ds, FTAG); 1930 return (SET_ERROR(EEXIST)); 1931 } 1932 } 1933 fnvlist_free(bookmarks); 1934 1935 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 1936 if (error != 0) { 1937 dsl_dataset_rele(ds, FTAG); 1938 return (error); 1939 } 1940 1941 /* 1942 * Check if the snap we are rolling back to uses more than 1943 * the refquota. 1944 */ 1945 if (ds->ds_quota != 0 && 1946 ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) { 1947 dsl_dataset_rele(ds, FTAG); 1948 return (SET_ERROR(EDQUOT)); 1949 } 1950 1951 /* 1952 * When we do the clone swap, we will temporarily use more space 1953 * due to the refreservation (the head will no longer have any 1954 * unique space, so the entire amount of the refreservation will need 1955 * to be free). We will immediately destroy the clone, freeing 1956 * this space, but the freeing happens over many txg's. 1957 */ 1958 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 1959 ds->ds_phys->ds_unique_bytes); 1960 1961 if (unused_refres_delta > 0 && 1962 unused_refres_delta > 1963 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 1964 dsl_dataset_rele(ds, FTAG); 1965 return (SET_ERROR(ENOSPC)); 1966 } 1967 1968 dsl_dataset_rele(ds, FTAG); 1969 return (0); 1970 } 1971 1972 static void 1973 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 1974 { 1975 dsl_dataset_rollback_arg_t *ddra = arg; 1976 dsl_pool_t *dp = dmu_tx_pool(tx); 1977 dsl_dataset_t *ds, *clone; 1978 uint64_t cloneobj; 1979 char namebuf[ZFS_MAXNAMELEN]; 1980 1981 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 1982 1983 dsl_dataset_name(ds->ds_prev, namebuf); 1984 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 1985 1986 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 1987 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 1988 1989 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 1990 1991 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 1992 dsl_dataset_zero_zil(ds, tx); 1993 1994 dsl_destroy_head_sync_impl(clone, tx); 1995 1996 dsl_dataset_rele(clone, FTAG); 1997 dsl_dataset_rele(ds, FTAG); 1998 } 1999 2000 /* 2001 * Rolls back the given filesystem or volume to the most recent snapshot. 2002 * The name of the most recent snapshot will be returned under key "target" 2003 * in the result nvlist. 2004 * 2005 * If owner != NULL: 2006 * - The existing dataset MUST be owned by the specified owner at entry 2007 * - Upon return, dataset will still be held by the same owner, whether we 2008 * succeed or not. 2009 * 2010 * This mode is required any time the existing filesystem is mounted. See 2011 * notes above zfs_suspend_fs() for further details. 2012 */ 2013 int 2014 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result) 2015 { 2016 dsl_dataset_rollback_arg_t ddra; 2017 2018 ddra.ddra_fsname = fsname; 2019 ddra.ddra_owner = owner; 2020 ddra.ddra_result = result; 2021 2022 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 2023 dsl_dataset_rollback_sync, &ddra, 2024 1, ZFS_SPACE_CHECK_RESERVED)); 2025 } 2026 2027 struct promotenode { 2028 list_node_t link; 2029 dsl_dataset_t *ds; 2030 }; 2031 2032 typedef struct dsl_dataset_promote_arg { 2033 const char *ddpa_clonename; 2034 dsl_dataset_t *ddpa_clone; 2035 list_t shared_snaps, origin_snaps, clone_snaps; 2036 dsl_dataset_t *origin_origin; /* origin of the origin */ 2037 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2038 char *err_ds; 2039 cred_t *cr; 2040 } dsl_dataset_promote_arg_t; 2041 2042 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2043 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2044 void *tag); 2045 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2046 2047 static int 2048 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2049 { 2050 dsl_dataset_promote_arg_t *ddpa = arg; 2051 dsl_pool_t *dp = dmu_tx_pool(tx); 2052 dsl_dataset_t *hds; 2053 struct promotenode *snap; 2054 dsl_dataset_t *origin_ds; 2055 int err; 2056 uint64_t unused; 2057 uint64_t ss_mv_cnt; 2058 2059 err = promote_hold(ddpa, dp, FTAG); 2060 if (err != 0) 2061 return (err); 2062 2063 hds = ddpa->ddpa_clone; 2064 2065 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) { 2066 promote_rele(ddpa, FTAG); 2067 return (SET_ERROR(EXDEV)); 2068 } 2069 2070 /* 2071 * Compute and check the amount of space to transfer. Since this is 2072 * so expensive, don't do the preliminary check. 2073 */ 2074 if (!dmu_tx_is_syncing(tx)) { 2075 promote_rele(ddpa, FTAG); 2076 return (0); 2077 } 2078 2079 snap = list_head(&ddpa->shared_snaps); 2080 origin_ds = snap->ds; 2081 2082 /* compute origin's new unique space */ 2083 snap = list_tail(&ddpa->clone_snaps); 2084 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2085 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2086 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2087 &ddpa->unique, &unused, &unused); 2088 2089 /* 2090 * Walk the snapshots that we are moving 2091 * 2092 * Compute space to transfer. Consider the incremental changes 2093 * to used by each snapshot: 2094 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2095 * So each snapshot gave birth to: 2096 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2097 * So a sequence would look like: 2098 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2099 * Which simplifies to: 2100 * uN + kN + kN-1 + ... + k1 + k0 2101 * Note however, if we stop before we reach the ORIGIN we get: 2102 * uN + kN + kN-1 + ... + kM - uM-1 2103 */ 2104 ss_mv_cnt = 0; 2105 ddpa->used = origin_ds->ds_phys->ds_referenced_bytes; 2106 ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes; 2107 ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; 2108 for (snap = list_head(&ddpa->shared_snaps); snap; 2109 snap = list_next(&ddpa->shared_snaps, snap)) { 2110 uint64_t val, dlused, dlcomp, dluncomp; 2111 dsl_dataset_t *ds = snap->ds; 2112 2113 ss_mv_cnt++; 2114 2115 /* 2116 * If there are long holds, we won't be able to evict 2117 * the objset. 2118 */ 2119 if (dsl_dataset_long_held(ds)) { 2120 err = SET_ERROR(EBUSY); 2121 goto out; 2122 } 2123 2124 /* Check that the snapshot name does not conflict */ 2125 VERIFY0(dsl_dataset_get_snapname(ds)); 2126 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2127 if (err == 0) { 2128 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname); 2129 err = SET_ERROR(EEXIST); 2130 goto out; 2131 } 2132 if (err != ENOENT) 2133 goto out; 2134 2135 /* The very first snapshot does not have a deadlist */ 2136 if (ds->ds_phys->ds_prev_snap_obj == 0) 2137 continue; 2138 2139 dsl_deadlist_space(&ds->ds_deadlist, 2140 &dlused, &dlcomp, &dluncomp); 2141 ddpa->used += dlused; 2142 ddpa->comp += dlcomp; 2143 ddpa->uncomp += dluncomp; 2144 } 2145 2146 /* 2147 * If we are a clone of a clone then we never reached ORIGIN, 2148 * so we need to subtract out the clone origin's used space. 2149 */ 2150 if (ddpa->origin_origin) { 2151 ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes; 2152 ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes; 2153 ddpa->uncomp -= 2154 ddpa->origin_origin->ds_phys->ds_uncompressed_bytes; 2155 } 2156 2157 /* Check that there is enough space and limit headroom here */ 2158 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2159 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2160 if (err != 0) 2161 goto out; 2162 2163 /* 2164 * Compute the amounts of space that will be used by snapshots 2165 * after the promotion (for both origin and clone). For each, 2166 * it is the amount of space that will be on all of their 2167 * deadlists (that was not born before their new origin). 2168 */ 2169 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2170 uint64_t space; 2171 2172 /* 2173 * Note, typically this will not be a clone of a clone, 2174 * so dd_origin_txg will be < TXG_INITIAL, so 2175 * these snaplist_space() -> dsl_deadlist_space_range() 2176 * calls will be fast because they do not have to 2177 * iterate over all bps. 2178 */ 2179 snap = list_head(&ddpa->origin_snaps); 2180 err = snaplist_space(&ddpa->shared_snaps, 2181 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2182 if (err != 0) 2183 goto out; 2184 2185 err = snaplist_space(&ddpa->clone_snaps, 2186 snap->ds->ds_dir->dd_origin_txg, &space); 2187 if (err != 0) 2188 goto out; 2189 ddpa->cloneusedsnap += space; 2190 } 2191 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2192 err = snaplist_space(&ddpa->origin_snaps, 2193 origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap); 2194 if (err != 0) 2195 goto out; 2196 } 2197 2198 out: 2199 promote_rele(ddpa, FTAG); 2200 return (err); 2201 } 2202 2203 static void 2204 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2205 { 2206 dsl_dataset_promote_arg_t *ddpa = arg; 2207 dsl_pool_t *dp = dmu_tx_pool(tx); 2208 dsl_dataset_t *hds; 2209 struct promotenode *snap; 2210 dsl_dataset_t *origin_ds; 2211 dsl_dataset_t *origin_head; 2212 dsl_dir_t *dd; 2213 dsl_dir_t *odd = NULL; 2214 uint64_t oldnext_obj; 2215 int64_t delta; 2216 2217 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2218 hds = ddpa->ddpa_clone; 2219 2220 ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE); 2221 2222 snap = list_head(&ddpa->shared_snaps); 2223 origin_ds = snap->ds; 2224 dd = hds->ds_dir; 2225 2226 snap = list_head(&ddpa->origin_snaps); 2227 origin_head = snap->ds; 2228 2229 /* 2230 * We need to explicitly open odd, since origin_ds's dd will be 2231 * changing. 2232 */ 2233 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2234 NULL, FTAG, &odd)); 2235 2236 /* change origin's next snap */ 2237 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2238 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj; 2239 snap = list_tail(&ddpa->clone_snaps); 2240 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2241 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object; 2242 2243 /* change the origin's next clone */ 2244 if (origin_ds->ds_phys->ds_next_clones_obj) { 2245 dsl_dataset_remove_from_next_clones(origin_ds, 2246 snap->ds->ds_object, tx); 2247 VERIFY0(zap_add_int(dp->dp_meta_objset, 2248 origin_ds->ds_phys->ds_next_clones_obj, 2249 oldnext_obj, tx)); 2250 } 2251 2252 /* change origin */ 2253 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2254 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object); 2255 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj; 2256 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2257 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2258 odd->dd_phys->dd_origin_obj = origin_ds->ds_object; 2259 origin_head->ds_dir->dd_origin_txg = 2260 origin_ds->ds_phys->ds_creation_txg; 2261 2262 /* change dd_clone entries */ 2263 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2264 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2265 odd->dd_phys->dd_clones, hds->ds_object, tx)); 2266 VERIFY0(zap_add_int(dp->dp_meta_objset, 2267 ddpa->origin_origin->ds_dir->dd_phys->dd_clones, 2268 hds->ds_object, tx)); 2269 2270 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2271 ddpa->origin_origin->ds_dir->dd_phys->dd_clones, 2272 origin_head->ds_object, tx)); 2273 if (dd->dd_phys->dd_clones == 0) { 2274 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset, 2275 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 2276 } 2277 VERIFY0(zap_add_int(dp->dp_meta_objset, 2278 dd->dd_phys->dd_clones, origin_head->ds_object, tx)); 2279 } 2280 2281 /* move snapshots to this dir */ 2282 for (snap = list_head(&ddpa->shared_snaps); snap; 2283 snap = list_next(&ddpa->shared_snaps, snap)) { 2284 dsl_dataset_t *ds = snap->ds; 2285 2286 /* 2287 * Property callbacks are registered to a particular 2288 * dsl_dir. Since ours is changing, evict the objset 2289 * so that they will be unregistered from the old dsl_dir. 2290 */ 2291 if (ds->ds_objset) { 2292 dmu_objset_evict(ds->ds_objset); 2293 ds->ds_objset = NULL; 2294 } 2295 2296 /* move snap name entry */ 2297 VERIFY0(dsl_dataset_get_snapname(ds)); 2298 VERIFY0(dsl_dataset_snap_remove(origin_head, 2299 ds->ds_snapname, tx, B_TRUE)); 2300 VERIFY0(zap_add(dp->dp_meta_objset, 2301 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, 2302 8, 1, &ds->ds_object, tx)); 2303 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 2304 DD_FIELD_SNAPSHOT_COUNT, tx); 2305 2306 /* change containing dsl_dir */ 2307 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2308 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object); 2309 ds->ds_phys->ds_dir_obj = dd->dd_object; 2310 ASSERT3P(ds->ds_dir, ==, odd); 2311 dsl_dir_rele(ds->ds_dir, ds); 2312 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 2313 NULL, ds, &ds->ds_dir)); 2314 2315 /* move any clone references */ 2316 if (ds->ds_phys->ds_next_clones_obj && 2317 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2318 zap_cursor_t zc; 2319 zap_attribute_t za; 2320 2321 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2322 ds->ds_phys->ds_next_clones_obj); 2323 zap_cursor_retrieve(&zc, &za) == 0; 2324 zap_cursor_advance(&zc)) { 2325 dsl_dataset_t *cnds; 2326 uint64_t o; 2327 2328 if (za.za_first_integer == oldnext_obj) { 2329 /* 2330 * We've already moved the 2331 * origin's reference. 2332 */ 2333 continue; 2334 } 2335 2336 VERIFY0(dsl_dataset_hold_obj(dp, 2337 za.za_first_integer, FTAG, &cnds)); 2338 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj; 2339 2340 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2341 odd->dd_phys->dd_clones, o, tx)); 2342 VERIFY0(zap_add_int(dp->dp_meta_objset, 2343 dd->dd_phys->dd_clones, o, tx)); 2344 dsl_dataset_rele(cnds, FTAG); 2345 } 2346 zap_cursor_fini(&zc); 2347 } 2348 2349 ASSERT(!dsl_prop_hascb(ds)); 2350 } 2351 2352 /* 2353 * Change space accounting. 2354 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2355 * both be valid, or both be 0 (resulting in delta == 0). This 2356 * is true for each of {clone,origin} independently. 2357 */ 2358 2359 delta = ddpa->cloneusedsnap - 2360 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2361 ASSERT3S(delta, >=, 0); 2362 ASSERT3U(ddpa->used, >=, delta); 2363 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2364 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2365 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 2366 2367 delta = ddpa->originusedsnap - 2368 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2369 ASSERT3S(delta, <=, 0); 2370 ASSERT3U(ddpa->used, >=, -delta); 2371 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2372 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2373 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 2374 2375 origin_ds->ds_phys->ds_unique_bytes = ddpa->unique; 2376 2377 /* log history record */ 2378 spa_history_log_internal_ds(hds, "promote", tx, ""); 2379 2380 dsl_dir_rele(odd, FTAG); 2381 promote_rele(ddpa, FTAG); 2382 } 2383 2384 /* 2385 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2386 * (exclusive) and last_obj (inclusive). The list will be in reverse 2387 * order (last_obj will be the list_head()). If first_obj == 0, do all 2388 * snapshots back to this dataset's origin. 2389 */ 2390 static int 2391 snaplist_make(dsl_pool_t *dp, 2392 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 2393 { 2394 uint64_t obj = last_obj; 2395 2396 list_create(l, sizeof (struct promotenode), 2397 offsetof(struct promotenode, link)); 2398 2399 while (obj != first_obj) { 2400 dsl_dataset_t *ds; 2401 struct promotenode *snap; 2402 int err; 2403 2404 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 2405 ASSERT(err != ENOENT); 2406 if (err != 0) 2407 return (err); 2408 2409 if (first_obj == 0) 2410 first_obj = ds->ds_dir->dd_phys->dd_origin_obj; 2411 2412 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 2413 snap->ds = ds; 2414 list_insert_tail(l, snap); 2415 obj = ds->ds_phys->ds_prev_snap_obj; 2416 } 2417 2418 return (0); 2419 } 2420 2421 static int 2422 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2423 { 2424 struct promotenode *snap; 2425 2426 *spacep = 0; 2427 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2428 uint64_t used, comp, uncomp; 2429 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2430 mintxg, UINT64_MAX, &used, &comp, &uncomp); 2431 *spacep += used; 2432 } 2433 return (0); 2434 } 2435 2436 static void 2437 snaplist_destroy(list_t *l, void *tag) 2438 { 2439 struct promotenode *snap; 2440 2441 if (l == NULL || !list_link_active(&l->list_head)) 2442 return; 2443 2444 while ((snap = list_tail(l)) != NULL) { 2445 list_remove(l, snap); 2446 dsl_dataset_rele(snap->ds, tag); 2447 kmem_free(snap, sizeof (*snap)); 2448 } 2449 list_destroy(l); 2450 } 2451 2452 static int 2453 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 2454 { 2455 int error; 2456 dsl_dir_t *dd; 2457 struct promotenode *snap; 2458 2459 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 2460 &ddpa->ddpa_clone); 2461 if (error != 0) 2462 return (error); 2463 dd = ddpa->ddpa_clone->ds_dir; 2464 2465 if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) || 2466 !dsl_dir_is_clone(dd)) { 2467 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2468 return (SET_ERROR(EINVAL)); 2469 } 2470 2471 error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj, 2472 &ddpa->shared_snaps, tag); 2473 if (error != 0) 2474 goto out; 2475 2476 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 2477 &ddpa->clone_snaps, tag); 2478 if (error != 0) 2479 goto out; 2480 2481 snap = list_head(&ddpa->shared_snaps); 2482 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); 2483 error = snaplist_make(dp, dd->dd_phys->dd_origin_obj, 2484 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, 2485 &ddpa->origin_snaps, tag); 2486 if (error != 0) 2487 goto out; 2488 2489 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) { 2490 error = dsl_dataset_hold_obj(dp, 2491 snap->ds->ds_dir->dd_phys->dd_origin_obj, 2492 tag, &ddpa->origin_origin); 2493 if (error != 0) 2494 goto out; 2495 } 2496 out: 2497 if (error != 0) 2498 promote_rele(ddpa, tag); 2499 return (error); 2500 } 2501 2502 static void 2503 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 2504 { 2505 snaplist_destroy(&ddpa->shared_snaps, tag); 2506 snaplist_destroy(&ddpa->clone_snaps, tag); 2507 snaplist_destroy(&ddpa->origin_snaps, tag); 2508 if (ddpa->origin_origin != NULL) 2509 dsl_dataset_rele(ddpa->origin_origin, tag); 2510 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2511 } 2512 2513 /* 2514 * Promote a clone. 2515 * 2516 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 2517 * in with the name. (It must be at least MAXNAMELEN bytes long.) 2518 */ 2519 int 2520 dsl_dataset_promote(const char *name, char *conflsnap) 2521 { 2522 dsl_dataset_promote_arg_t ddpa = { 0 }; 2523 uint64_t numsnaps; 2524 int error; 2525 objset_t *os; 2526 2527 /* 2528 * We will modify space proportional to the number of 2529 * snapshots. Compute numsnaps. 2530 */ 2531 error = dmu_objset_hold(name, FTAG, &os); 2532 if (error != 0) 2533 return (error); 2534 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 2535 dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps); 2536 dmu_objset_rele(os, FTAG); 2537 if (error != 0) 2538 return (error); 2539 2540 ddpa.ddpa_clonename = name; 2541 ddpa.err_ds = conflsnap; 2542 ddpa.cr = CRED(); 2543 2544 return (dsl_sync_task(name, dsl_dataset_promote_check, 2545 dsl_dataset_promote_sync, &ddpa, 2546 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED)); 2547 } 2548 2549 int 2550 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 2551 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 2552 { 2553 int64_t unused_refres_delta; 2554 2555 /* they should both be heads */ 2556 if (dsl_dataset_is_snapshot(clone) || 2557 dsl_dataset_is_snapshot(origin_head)) 2558 return (SET_ERROR(EINVAL)); 2559 2560 /* if we are not forcing, the branch point should be just before them */ 2561 if (!force && clone->ds_prev != origin_head->ds_prev) 2562 return (SET_ERROR(EINVAL)); 2563 2564 /* clone should be the clone (unless they are unrelated) */ 2565 if (clone->ds_prev != NULL && 2566 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 2567 origin_head->ds_dir != clone->ds_prev->ds_dir) 2568 return (SET_ERROR(EINVAL)); 2569 2570 /* the clone should be a child of the origin */ 2571 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 2572 return (SET_ERROR(EINVAL)); 2573 2574 /* origin_head shouldn't be modified unless 'force' */ 2575 if (!force && 2576 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 2577 return (SET_ERROR(ETXTBSY)); 2578 2579 /* origin_head should have no long holds (e.g. is not mounted) */ 2580 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 2581 return (SET_ERROR(EBUSY)); 2582 2583 /* check amount of any unconsumed refreservation */ 2584 unused_refres_delta = 2585 (int64_t)MIN(origin_head->ds_reserved, 2586 origin_head->ds_phys->ds_unique_bytes) - 2587 (int64_t)MIN(origin_head->ds_reserved, 2588 clone->ds_phys->ds_unique_bytes); 2589 2590 if (unused_refres_delta > 0 && 2591 unused_refres_delta > 2592 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 2593 return (SET_ERROR(ENOSPC)); 2594 2595 /* clone can't be over the head's refquota */ 2596 if (origin_head->ds_quota != 0 && 2597 clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota) 2598 return (SET_ERROR(EDQUOT)); 2599 2600 return (0); 2601 } 2602 2603 void 2604 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 2605 dsl_dataset_t *origin_head, dmu_tx_t *tx) 2606 { 2607 dsl_pool_t *dp = dmu_tx_pool(tx); 2608 int64_t unused_refres_delta; 2609 2610 ASSERT(clone->ds_reserved == 0); 2611 ASSERT(origin_head->ds_quota == 0 || 2612 clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota); 2613 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 2614 2615 dmu_buf_will_dirty(clone->ds_dbuf, tx); 2616 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 2617 2618 if (clone->ds_objset != NULL) { 2619 dmu_objset_evict(clone->ds_objset); 2620 clone->ds_objset = NULL; 2621 } 2622 2623 if (origin_head->ds_objset != NULL) { 2624 dmu_objset_evict(origin_head->ds_objset); 2625 origin_head->ds_objset = NULL; 2626 } 2627 2628 unused_refres_delta = 2629 (int64_t)MIN(origin_head->ds_reserved, 2630 origin_head->ds_phys->ds_unique_bytes) - 2631 (int64_t)MIN(origin_head->ds_reserved, 2632 clone->ds_phys->ds_unique_bytes); 2633 2634 /* 2635 * Reset origin's unique bytes, if it exists. 2636 */ 2637 if (clone->ds_prev) { 2638 dsl_dataset_t *origin = clone->ds_prev; 2639 uint64_t comp, uncomp; 2640 2641 dmu_buf_will_dirty(origin->ds_dbuf, tx); 2642 dsl_deadlist_space_range(&clone->ds_deadlist, 2643 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2644 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp); 2645 } 2646 2647 /* swap blkptrs */ 2648 { 2649 blkptr_t tmp; 2650 tmp = origin_head->ds_phys->ds_bp; 2651 origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp; 2652 clone->ds_phys->ds_bp = tmp; 2653 } 2654 2655 /* set dd_*_bytes */ 2656 { 2657 int64_t dused, dcomp, duncomp; 2658 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2659 uint64_t odl_used, odl_comp, odl_uncomp; 2660 2661 ASSERT3U(clone->ds_dir->dd_phys-> 2662 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2663 2664 dsl_deadlist_space(&clone->ds_deadlist, 2665 &cdl_used, &cdl_comp, &cdl_uncomp); 2666 dsl_deadlist_space(&origin_head->ds_deadlist, 2667 &odl_used, &odl_comp, &odl_uncomp); 2668 2669 dused = clone->ds_phys->ds_referenced_bytes + cdl_used - 2670 (origin_head->ds_phys->ds_referenced_bytes + odl_used); 2671 dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp - 2672 (origin_head->ds_phys->ds_compressed_bytes + odl_comp); 2673 duncomp = clone->ds_phys->ds_uncompressed_bytes + 2674 cdl_uncomp - 2675 (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp); 2676 2677 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 2678 dused, dcomp, duncomp, tx); 2679 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 2680 -dused, -dcomp, -duncomp, tx); 2681 2682 /* 2683 * The difference in the space used by snapshots is the 2684 * difference in snapshot space due to the head's 2685 * deadlist (since that's the only thing that's 2686 * changing that affects the snapused). 2687 */ 2688 dsl_deadlist_space_range(&clone->ds_deadlist, 2689 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2690 &cdl_used, &cdl_comp, &cdl_uncomp); 2691 dsl_deadlist_space_range(&origin_head->ds_deadlist, 2692 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2693 &odl_used, &odl_comp, &odl_uncomp); 2694 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 2695 DD_USED_HEAD, DD_USED_SNAP, tx); 2696 } 2697 2698 /* swap ds_*_bytes */ 2699 SWITCH64(origin_head->ds_phys->ds_referenced_bytes, 2700 clone->ds_phys->ds_referenced_bytes); 2701 SWITCH64(origin_head->ds_phys->ds_compressed_bytes, 2702 clone->ds_phys->ds_compressed_bytes); 2703 SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes, 2704 clone->ds_phys->ds_uncompressed_bytes); 2705 SWITCH64(origin_head->ds_phys->ds_unique_bytes, 2706 clone->ds_phys->ds_unique_bytes); 2707 2708 /* apply any parent delta for change in unconsumed refreservation */ 2709 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 2710 unused_refres_delta, 0, 0, tx); 2711 2712 /* 2713 * Swap deadlists. 2714 */ 2715 dsl_deadlist_close(&clone->ds_deadlist); 2716 dsl_deadlist_close(&origin_head->ds_deadlist); 2717 SWITCH64(origin_head->ds_phys->ds_deadlist_obj, 2718 clone->ds_phys->ds_deadlist_obj); 2719 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 2720 clone->ds_phys->ds_deadlist_obj); 2721 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 2722 origin_head->ds_phys->ds_deadlist_obj); 2723 2724 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 2725 2726 spa_history_log_internal_ds(clone, "clone swap", tx, 2727 "parent=%s", origin_head->ds_dir->dd_myname); 2728 } 2729 2730 /* 2731 * Given a pool name and a dataset object number in that pool, 2732 * return the name of that dataset. 2733 */ 2734 int 2735 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 2736 { 2737 dsl_pool_t *dp; 2738 dsl_dataset_t *ds; 2739 int error; 2740 2741 error = dsl_pool_hold(pname, FTAG, &dp); 2742 if (error != 0) 2743 return (error); 2744 2745 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 2746 if (error == 0) { 2747 dsl_dataset_name(ds, buf); 2748 dsl_dataset_rele(ds, FTAG); 2749 } 2750 dsl_pool_rele(dp, FTAG); 2751 2752 return (error); 2753 } 2754 2755 int 2756 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 2757 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 2758 { 2759 int error = 0; 2760 2761 ASSERT3S(asize, >, 0); 2762 2763 /* 2764 * *ref_rsrv is the portion of asize that will come from any 2765 * unconsumed refreservation space. 2766 */ 2767 *ref_rsrv = 0; 2768 2769 mutex_enter(&ds->ds_lock); 2770 /* 2771 * Make a space adjustment for reserved bytes. 2772 */ 2773 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) { 2774 ASSERT3U(*used, >=, 2775 ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 2776 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 2777 *ref_rsrv = 2778 asize - MIN(asize, parent_delta(ds, asize + inflight)); 2779 } 2780 2781 if (!check_quota || ds->ds_quota == 0) { 2782 mutex_exit(&ds->ds_lock); 2783 return (0); 2784 } 2785 /* 2786 * If they are requesting more space, and our current estimate 2787 * is over quota, they get to try again unless the actual 2788 * on-disk is over quota and there are no pending changes (which 2789 * may free up space for us). 2790 */ 2791 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) { 2792 if (inflight > 0 || 2793 ds->ds_phys->ds_referenced_bytes < ds->ds_quota) 2794 error = SET_ERROR(ERESTART); 2795 else 2796 error = SET_ERROR(EDQUOT); 2797 } 2798 mutex_exit(&ds->ds_lock); 2799 2800 return (error); 2801 } 2802 2803 typedef struct dsl_dataset_set_qr_arg { 2804 const char *ddsqra_name; 2805 zprop_source_t ddsqra_source; 2806 uint64_t ddsqra_value; 2807 } dsl_dataset_set_qr_arg_t; 2808 2809 2810 /* ARGSUSED */ 2811 static int 2812 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 2813 { 2814 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2815 dsl_pool_t *dp = dmu_tx_pool(tx); 2816 dsl_dataset_t *ds; 2817 int error; 2818 uint64_t newval; 2819 2820 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 2821 return (SET_ERROR(ENOTSUP)); 2822 2823 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 2824 if (error != 0) 2825 return (error); 2826 2827 if (dsl_dataset_is_snapshot(ds)) { 2828 dsl_dataset_rele(ds, FTAG); 2829 return (SET_ERROR(EINVAL)); 2830 } 2831 2832 error = dsl_prop_predict(ds->ds_dir, 2833 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 2834 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 2835 if (error != 0) { 2836 dsl_dataset_rele(ds, FTAG); 2837 return (error); 2838 } 2839 2840 if (newval == 0) { 2841 dsl_dataset_rele(ds, FTAG); 2842 return (0); 2843 } 2844 2845 if (newval < ds->ds_phys->ds_referenced_bytes || 2846 newval < ds->ds_reserved) { 2847 dsl_dataset_rele(ds, FTAG); 2848 return (SET_ERROR(ENOSPC)); 2849 } 2850 2851 dsl_dataset_rele(ds, FTAG); 2852 return (0); 2853 } 2854 2855 static void 2856 dsl_dataset_set_refquota_sync(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 uint64_t newval; 2862 2863 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 2864 2865 dsl_prop_set_sync_impl(ds, 2866 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 2867 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 2868 &ddsqra->ddsqra_value, tx); 2869 2870 VERIFY0(dsl_prop_get_int_ds(ds, 2871 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 2872 2873 if (ds->ds_quota != newval) { 2874 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2875 ds->ds_quota = newval; 2876 } 2877 dsl_dataset_rele(ds, FTAG); 2878 } 2879 2880 int 2881 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 2882 uint64_t refquota) 2883 { 2884 dsl_dataset_set_qr_arg_t ddsqra; 2885 2886 ddsqra.ddsqra_name = dsname; 2887 ddsqra.ddsqra_source = source; 2888 ddsqra.ddsqra_value = refquota; 2889 2890 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 2891 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 2892 } 2893 2894 static int 2895 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 2896 { 2897 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2898 dsl_pool_t *dp = dmu_tx_pool(tx); 2899 dsl_dataset_t *ds; 2900 int error; 2901 uint64_t newval, unique; 2902 2903 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 2904 return (SET_ERROR(ENOTSUP)); 2905 2906 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 2907 if (error != 0) 2908 return (error); 2909 2910 if (dsl_dataset_is_snapshot(ds)) { 2911 dsl_dataset_rele(ds, FTAG); 2912 return (SET_ERROR(EINVAL)); 2913 } 2914 2915 error = dsl_prop_predict(ds->ds_dir, 2916 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 2917 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 2918 if (error != 0) { 2919 dsl_dataset_rele(ds, FTAG); 2920 return (error); 2921 } 2922 2923 /* 2924 * If we are doing the preliminary check in open context, the 2925 * space estimates may be inaccurate. 2926 */ 2927 if (!dmu_tx_is_syncing(tx)) { 2928 dsl_dataset_rele(ds, FTAG); 2929 return (0); 2930 } 2931 2932 mutex_enter(&ds->ds_lock); 2933 if (!DS_UNIQUE_IS_ACCURATE(ds)) 2934 dsl_dataset_recalc_head_uniq(ds); 2935 unique = ds->ds_phys->ds_unique_bytes; 2936 mutex_exit(&ds->ds_lock); 2937 2938 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 2939 uint64_t delta = MAX(unique, newval) - 2940 MAX(unique, ds->ds_reserved); 2941 2942 if (delta > 2943 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 2944 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 2945 dsl_dataset_rele(ds, FTAG); 2946 return (SET_ERROR(ENOSPC)); 2947 } 2948 } 2949 2950 dsl_dataset_rele(ds, FTAG); 2951 return (0); 2952 } 2953 2954 void 2955 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 2956 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 2957 { 2958 uint64_t newval; 2959 uint64_t unique; 2960 int64_t delta; 2961 2962 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 2963 source, sizeof (value), 1, &value, tx); 2964 2965 VERIFY0(dsl_prop_get_int_ds(ds, 2966 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 2967 2968 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2969 mutex_enter(&ds->ds_dir->dd_lock); 2970 mutex_enter(&ds->ds_lock); 2971 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 2972 unique = ds->ds_phys->ds_unique_bytes; 2973 delta = MAX(0, (int64_t)(newval - unique)) - 2974 MAX(0, (int64_t)(ds->ds_reserved - unique)); 2975 ds->ds_reserved = newval; 2976 mutex_exit(&ds->ds_lock); 2977 2978 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 2979 mutex_exit(&ds->ds_dir->dd_lock); 2980 } 2981 2982 static void 2983 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 2984 { 2985 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2986 dsl_pool_t *dp = dmu_tx_pool(tx); 2987 dsl_dataset_t *ds; 2988 2989 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 2990 dsl_dataset_set_refreservation_sync_impl(ds, 2991 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 2992 dsl_dataset_rele(ds, FTAG); 2993 } 2994 2995 int 2996 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 2997 uint64_t refreservation) 2998 { 2999 dsl_dataset_set_qr_arg_t ddsqra; 3000 3001 ddsqra.ddsqra_name = dsname; 3002 ddsqra.ddsqra_source = source; 3003 ddsqra.ddsqra_value = refreservation; 3004 3005 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 3006 dsl_dataset_set_refreservation_sync, &ddsqra, 3007 0, ZFS_SPACE_CHECK_NONE)); 3008 } 3009 3010 /* 3011 * Return (in *usedp) the amount of space written in new that is not 3012 * present in oldsnap. New may be a snapshot or the head. Old must be 3013 * a snapshot before new, in new's filesystem (or its origin). If not then 3014 * fail and return EINVAL. 3015 * 3016 * The written space is calculated by considering two components: First, we 3017 * ignore any freed space, and calculate the written as new's used space 3018 * minus old's used space. Next, we add in the amount of space that was freed 3019 * between the two snapshots, thus reducing new's used space relative to old's. 3020 * Specifically, this is the space that was born before old->ds_creation_txg, 3021 * and freed before new (ie. on new's deadlist or a previous deadlist). 3022 * 3023 * space freed [---------------------] 3024 * snapshots ---O-------O--------O-------O------ 3025 * oldsnap new 3026 */ 3027 int 3028 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 3029 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3030 { 3031 int err = 0; 3032 uint64_t snapobj; 3033 dsl_pool_t *dp = new->ds_dir->dd_pool; 3034 3035 ASSERT(dsl_pool_config_held(dp)); 3036 3037 *usedp = 0; 3038 *usedp += new->ds_phys->ds_referenced_bytes; 3039 *usedp -= oldsnap->ds_phys->ds_referenced_bytes; 3040 3041 *compp = 0; 3042 *compp += new->ds_phys->ds_compressed_bytes; 3043 *compp -= oldsnap->ds_phys->ds_compressed_bytes; 3044 3045 *uncompp = 0; 3046 *uncompp += new->ds_phys->ds_uncompressed_bytes; 3047 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes; 3048 3049 snapobj = new->ds_object; 3050 while (snapobj != oldsnap->ds_object) { 3051 dsl_dataset_t *snap; 3052 uint64_t used, comp, uncomp; 3053 3054 if (snapobj == new->ds_object) { 3055 snap = new; 3056 } else { 3057 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3058 if (err != 0) 3059 break; 3060 } 3061 3062 if (snap->ds_phys->ds_prev_snap_txg == 3063 oldsnap->ds_phys->ds_creation_txg) { 3064 /* 3065 * The blocks in the deadlist can not be born after 3066 * ds_prev_snap_txg, so get the whole deadlist space, 3067 * which is more efficient (especially for old-format 3068 * deadlists). Unfortunately the deadlist code 3069 * doesn't have enough information to make this 3070 * optimization itself. 3071 */ 3072 dsl_deadlist_space(&snap->ds_deadlist, 3073 &used, &comp, &uncomp); 3074 } else { 3075 dsl_deadlist_space_range(&snap->ds_deadlist, 3076 0, oldsnap->ds_phys->ds_creation_txg, 3077 &used, &comp, &uncomp); 3078 } 3079 *usedp += used; 3080 *compp += comp; 3081 *uncompp += uncomp; 3082 3083 /* 3084 * If we get to the beginning of the chain of snapshots 3085 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3086 * was not a snapshot of/before new. 3087 */ 3088 snapobj = snap->ds_phys->ds_prev_snap_obj; 3089 if (snap != new) 3090 dsl_dataset_rele(snap, FTAG); 3091 if (snapobj == 0) { 3092 err = SET_ERROR(EINVAL); 3093 break; 3094 } 3095 3096 } 3097 return (err); 3098 } 3099 3100 /* 3101 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3102 * lastsnap, and all snapshots in between are deleted. 3103 * 3104 * blocks that would be freed [---------------------------] 3105 * snapshots ---O-------O--------O-------O--------O 3106 * firstsnap lastsnap 3107 * 3108 * This is the set of blocks that were born after the snap before firstsnap, 3109 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3110 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3111 * We calculate this by iterating over the relevant deadlists (from the snap 3112 * after lastsnap, backward to the snap after firstsnap), summing up the 3113 * space on the deadlist that was born after the snap before firstsnap. 3114 */ 3115 int 3116 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3117 dsl_dataset_t *lastsnap, 3118 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3119 { 3120 int err = 0; 3121 uint64_t snapobj; 3122 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3123 3124 ASSERT(dsl_dataset_is_snapshot(firstsnap)); 3125 ASSERT(dsl_dataset_is_snapshot(lastsnap)); 3126 3127 /* 3128 * Check that the snapshots are in the same dsl_dir, and firstsnap 3129 * is before lastsnap. 3130 */ 3131 if (firstsnap->ds_dir != lastsnap->ds_dir || 3132 firstsnap->ds_phys->ds_creation_txg > 3133 lastsnap->ds_phys->ds_creation_txg) 3134 return (SET_ERROR(EINVAL)); 3135 3136 *usedp = *compp = *uncompp = 0; 3137 3138 snapobj = lastsnap->ds_phys->ds_next_snap_obj; 3139 while (snapobj != firstsnap->ds_object) { 3140 dsl_dataset_t *ds; 3141 uint64_t used, comp, uncomp; 3142 3143 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3144 if (err != 0) 3145 break; 3146 3147 dsl_deadlist_space_range(&ds->ds_deadlist, 3148 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX, 3149 &used, &comp, &uncomp); 3150 *usedp += used; 3151 *compp += comp; 3152 *uncompp += uncomp; 3153 3154 snapobj = ds->ds_phys->ds_prev_snap_obj; 3155 ASSERT3U(snapobj, !=, 0); 3156 dsl_dataset_rele(ds, FTAG); 3157 } 3158 return (err); 3159 } 3160 3161 static int 3162 dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx) 3163 { 3164 const char *dsname = arg; 3165 dsl_dataset_t *ds; 3166 dsl_pool_t *dp = dmu_tx_pool(tx); 3167 int error = 0; 3168 3169 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) 3170 return (SET_ERROR(ENOTSUP)); 3171 3172 ASSERT(spa_feature_is_enabled(dp->dp_spa, 3173 SPA_FEATURE_EXTENSIBLE_DATASET)); 3174 3175 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 3176 if (error != 0) 3177 return (error); 3178 3179 if (ds->ds_large_blocks) 3180 error = EALREADY; 3181 dsl_dataset_rele(ds, FTAG); 3182 3183 return (error); 3184 } 3185 3186 void 3187 dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx) 3188 { 3189 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 3190 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 3191 uint64_t zero = 0; 3192 3193 spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx); 3194 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 3195 3196 VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS, 3197 sizeof (zero), 1, &zero, tx)); 3198 } 3199 3200 static void 3201 dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx) 3202 { 3203 const char *dsname = arg; 3204 dsl_dataset_t *ds; 3205 3206 VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds)); 3207 3208 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx); 3209 ASSERT(!ds->ds_large_blocks); 3210 ds->ds_large_blocks = B_TRUE; 3211 dsl_dataset_rele(ds, FTAG); 3212 } 3213 3214 int 3215 dsl_dataset_activate_large_blocks(const char *dsname) 3216 { 3217 int error; 3218 3219 error = dsl_sync_task(dsname, 3220 dsl_dataset_activate_large_blocks_check, 3221 dsl_dataset_activate_large_blocks_sync, (void *)dsname, 3222 1, ZFS_SPACE_CHECK_RESERVED); 3223 3224 /* 3225 * EALREADY indicates that this dataset already supports large blocks. 3226 */ 3227 if (error == EALREADY) 3228 error = 0; 3229 return (error); 3230 } 3231 3232 /* 3233 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 3234 * For example, they could both be snapshots of the same filesystem, and 3235 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 3236 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 3237 * filesystem. Or 'earlier' could be the origin's origin. 3238 * 3239 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 3240 */ 3241 boolean_t 3242 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 3243 uint64_t earlier_txg) 3244 { 3245 dsl_pool_t *dp = later->ds_dir->dd_pool; 3246 int error; 3247 boolean_t ret; 3248 3249 ASSERT(dsl_pool_config_held(dp)); 3250 ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0); 3251 3252 if (earlier_txg == 0) 3253 earlier_txg = earlier->ds_phys->ds_creation_txg; 3254 3255 if (dsl_dataset_is_snapshot(later) && 3256 earlier_txg >= later->ds_phys->ds_creation_txg) 3257 return (B_FALSE); 3258 3259 if (later->ds_dir == earlier->ds_dir) 3260 return (B_TRUE); 3261 if (!dsl_dir_is_clone(later->ds_dir)) 3262 return (B_FALSE); 3263 3264 if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object) 3265 return (B_TRUE); 3266 dsl_dataset_t *origin; 3267 error = dsl_dataset_hold_obj(dp, 3268 later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin); 3269 if (error != 0) 3270 return (B_FALSE); 3271 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 3272 dsl_dataset_rele(origin, FTAG); 3273 return (ret); 3274 } 3275 3276 3277 void 3278 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 3279 { 3280 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3281 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 3282 } 3283