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, 2015 by Delphix. All rights reserved. 24 * Copyright (c) 2014, Joyent, Inc. All rights reserved. 25 * Copyright (c) 2014 RackTop Systems. 26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 27 */ 28 29 #include <sys/dmu_objset.h> 30 #include <sys/dsl_dataset.h> 31 #include <sys/dsl_dir.h> 32 #include <sys/dsl_prop.h> 33 #include <sys/dsl_synctask.h> 34 #include <sys/dmu_traverse.h> 35 #include <sys/dmu_impl.h> 36 #include <sys/dmu_tx.h> 37 #include <sys/arc.h> 38 #include <sys/zio.h> 39 #include <sys/zap.h> 40 #include <sys/zfeature.h> 41 #include <sys/unique.h> 42 #include <sys/zfs_context.h> 43 #include <sys/zfs_ioctl.h> 44 #include <sys/spa.h> 45 #include <sys/zfs_znode.h> 46 #include <sys/zfs_onexit.h> 47 #include <sys/zvol.h> 48 #include <sys/dsl_scan.h> 49 #include <sys/dsl_deadlist.h> 50 #include <sys/dsl_destroy.h> 51 #include <sys/dsl_userhold.h> 52 #include <sys/dsl_bookmark.h> 53 #include <sys/dmu_send.h> 54 #include <sys/zio_checksum.h> 55 #include <sys/zio_compress.h> 56 #include <zfs_fletcher.h> 57 58 /* 59 * The SPA supports block sizes up to 16MB. However, very large blocks 60 * can have an impact on i/o latency (e.g. tying up a spinning disk for 61 * ~300ms), and also potentially on the memory allocator. Therefore, 62 * we do not allow the recordsize to be set larger than zfs_max_recordsize 63 * (default 1MB). Larger blocks can be created by changing this tunable, 64 * and pools with larger blocks can always be imported and used, regardless 65 * of this setting. 66 */ 67 int zfs_max_recordsize = 1 * 1024 * 1024; 68 69 #define SWITCH64(x, y) \ 70 { \ 71 uint64_t __tmp = (x); \ 72 (x) = (y); \ 73 (y) = __tmp; \ 74 } 75 76 #define DS_REF_MAX (1ULL << 62) 77 78 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds); 79 80 /* 81 * Figure out how much of this delta should be propogated to the dsl_dir 82 * layer. If there's a refreservation, that space has already been 83 * partially accounted for in our ancestors. 84 */ 85 static int64_t 86 parent_delta(dsl_dataset_t *ds, int64_t delta) 87 { 88 dsl_dataset_phys_t *ds_phys; 89 uint64_t old_bytes, new_bytes; 90 91 if (ds->ds_reserved == 0) 92 return (delta); 93 94 ds_phys = dsl_dataset_phys(ds); 95 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved); 96 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 97 98 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 99 return (new_bytes - old_bytes); 100 } 101 102 void 103 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 104 { 105 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 106 int compressed = BP_GET_PSIZE(bp); 107 int uncompressed = BP_GET_UCSIZE(bp); 108 int64_t delta; 109 110 dprintf_bp(bp, "ds=%p", ds); 111 112 ASSERT(dmu_tx_is_syncing(tx)); 113 /* It could have been compressed away to nothing */ 114 if (BP_IS_HOLE(bp)) 115 return; 116 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 117 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 118 if (ds == NULL) { 119 dsl_pool_mos_diduse_space(tx->tx_pool, 120 used, compressed, uncompressed); 121 return; 122 } 123 124 dmu_buf_will_dirty(ds->ds_dbuf, tx); 125 mutex_enter(&ds->ds_lock); 126 delta = parent_delta(ds, used); 127 dsl_dataset_phys(ds)->ds_referenced_bytes += used; 128 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed; 129 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed; 130 dsl_dataset_phys(ds)->ds_unique_bytes += used; 131 132 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) { 133 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] = 134 B_TRUE; 135 } 136 137 spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp)); 138 if (f != SPA_FEATURE_NONE) 139 ds->ds_feature_activation_needed[f] = B_TRUE; 140 141 mutex_exit(&ds->ds_lock); 142 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 143 compressed, uncompressed, tx); 144 dsl_dir_transfer_space(ds->ds_dir, used - delta, 145 DD_USED_REFRSRV, DD_USED_HEAD, tx); 146 } 147 148 int 149 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 150 boolean_t async) 151 { 152 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 153 int compressed = BP_GET_PSIZE(bp); 154 int uncompressed = BP_GET_UCSIZE(bp); 155 156 if (BP_IS_HOLE(bp)) 157 return (0); 158 159 ASSERT(dmu_tx_is_syncing(tx)); 160 ASSERT(bp->blk_birth <= tx->tx_txg); 161 162 if (ds == NULL) { 163 dsl_free(tx->tx_pool, tx->tx_txg, bp); 164 dsl_pool_mos_diduse_space(tx->tx_pool, 165 -used, -compressed, -uncompressed); 166 return (used); 167 } 168 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 169 170 ASSERT(!ds->ds_is_snapshot); 171 dmu_buf_will_dirty(ds->ds_dbuf, tx); 172 173 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 174 int64_t delta; 175 176 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 177 dsl_free(tx->tx_pool, tx->tx_txg, bp); 178 179 mutex_enter(&ds->ds_lock); 180 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used || 181 !DS_UNIQUE_IS_ACCURATE(ds)); 182 delta = parent_delta(ds, -used); 183 dsl_dataset_phys(ds)->ds_unique_bytes -= used; 184 mutex_exit(&ds->ds_lock); 185 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 186 delta, -compressed, -uncompressed, tx); 187 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 188 DD_USED_REFRSRV, DD_USED_HEAD, tx); 189 } else { 190 dprintf_bp(bp, "putting on dead list: %s", ""); 191 if (async) { 192 /* 193 * We are here as part of zio's write done callback, 194 * which means we're a zio interrupt thread. We can't 195 * call dsl_deadlist_insert() now because it may block 196 * waiting for I/O. Instead, put bp on the deferred 197 * queue and let dsl_pool_sync() finish the job. 198 */ 199 bplist_append(&ds->ds_pending_deadlist, bp); 200 } else { 201 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); 202 } 203 ASSERT3U(ds->ds_prev->ds_object, ==, 204 dsl_dataset_phys(ds)->ds_prev_snap_obj); 205 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0); 206 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 207 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 208 ds->ds_object && bp->blk_birth > 209 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) { 210 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 211 mutex_enter(&ds->ds_prev->ds_lock); 212 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used; 213 mutex_exit(&ds->ds_prev->ds_lock); 214 } 215 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 216 dsl_dir_transfer_space(ds->ds_dir, used, 217 DD_USED_HEAD, DD_USED_SNAP, tx); 218 } 219 } 220 mutex_enter(&ds->ds_lock); 221 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used); 222 dsl_dataset_phys(ds)->ds_referenced_bytes -= used; 223 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed); 224 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed; 225 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed); 226 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed; 227 mutex_exit(&ds->ds_lock); 228 229 return (used); 230 } 231 232 uint64_t 233 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 234 { 235 uint64_t trysnap = 0; 236 237 if (ds == NULL) 238 return (0); 239 /* 240 * The snapshot creation could fail, but that would cause an 241 * incorrect FALSE return, which would only result in an 242 * overestimation of the amount of space that an operation would 243 * consume, which is OK. 244 * 245 * There's also a small window where we could miss a pending 246 * snapshot, because we could set the sync task in the quiescing 247 * phase. So this should only be used as a guess. 248 */ 249 if (ds->ds_trysnap_txg > 250 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 251 trysnap = ds->ds_trysnap_txg; 252 return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap)); 253 } 254 255 boolean_t 256 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp, 257 uint64_t blk_birth) 258 { 259 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) || 260 (bp != NULL && BP_IS_HOLE(bp))) 261 return (B_FALSE); 262 263 ddt_prefetch(dsl_dataset_get_spa(ds), bp); 264 265 return (B_TRUE); 266 } 267 268 static void 269 dsl_dataset_evict(void *dbu) 270 { 271 dsl_dataset_t *ds = dbu; 272 273 ASSERT(ds->ds_owner == NULL); 274 275 ds->ds_dbuf = NULL; 276 277 unique_remove(ds->ds_fsid_guid); 278 279 if (ds->ds_objset != NULL) 280 dmu_objset_evict(ds->ds_objset); 281 282 if (ds->ds_prev) { 283 dsl_dataset_rele(ds->ds_prev, ds); 284 ds->ds_prev = NULL; 285 } 286 287 bplist_destroy(&ds->ds_pending_deadlist); 288 if (ds->ds_deadlist.dl_os != NULL) 289 dsl_deadlist_close(&ds->ds_deadlist); 290 if (ds->ds_dir) 291 dsl_dir_async_rele(ds->ds_dir, ds); 292 293 ASSERT(!list_link_active(&ds->ds_synced_link)); 294 295 list_destroy(&ds->ds_prop_cbs); 296 mutex_destroy(&ds->ds_lock); 297 mutex_destroy(&ds->ds_opening_lock); 298 mutex_destroy(&ds->ds_sendstream_lock); 299 refcount_destroy(&ds->ds_longholds); 300 301 kmem_free(ds, sizeof (dsl_dataset_t)); 302 } 303 304 int 305 dsl_dataset_get_snapname(dsl_dataset_t *ds) 306 { 307 dsl_dataset_phys_t *headphys; 308 int err; 309 dmu_buf_t *headdbuf; 310 dsl_pool_t *dp = ds->ds_dir->dd_pool; 311 objset_t *mos = dp->dp_meta_objset; 312 313 if (ds->ds_snapname[0]) 314 return (0); 315 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) 316 return (0); 317 318 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, 319 FTAG, &headdbuf); 320 if (err != 0) 321 return (err); 322 headphys = headdbuf->db_data; 323 err = zap_value_search(dp->dp_meta_objset, 324 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 325 dmu_buf_rele(headdbuf, FTAG); 326 return (err); 327 } 328 329 int 330 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 331 { 332 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 333 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 334 matchtype_t mt; 335 int err; 336 337 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 338 mt = MT_FIRST; 339 else 340 mt = MT_EXACT; 341 342 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 343 value, mt, NULL, 0, NULL); 344 if (err == ENOTSUP && mt == MT_FIRST) 345 err = zap_lookup(mos, snapobj, name, 8, 1, value); 346 return (err); 347 } 348 349 int 350 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 351 boolean_t adj_cnt) 352 { 353 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 354 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 355 matchtype_t mt; 356 int err; 357 358 dsl_dir_snap_cmtime_update(ds->ds_dir); 359 360 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 361 mt = MT_FIRST; 362 else 363 mt = MT_EXACT; 364 365 err = zap_remove_norm(mos, snapobj, name, mt, tx); 366 if (err == ENOTSUP && mt == MT_FIRST) 367 err = zap_remove(mos, snapobj, name, tx); 368 369 if (err == 0 && adj_cnt) 370 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 371 DD_FIELD_SNAPSHOT_COUNT, tx); 372 373 return (err); 374 } 375 376 boolean_t 377 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag) 378 { 379 dmu_buf_t *dbuf = ds->ds_dbuf; 380 boolean_t result = B_FALSE; 381 382 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset, 383 ds->ds_object, DMU_BONUS_BLKID, tag)) { 384 385 if (ds == dmu_buf_get_user(dbuf)) 386 result = B_TRUE; 387 else 388 dmu_buf_rele(dbuf, tag); 389 } 390 391 return (result); 392 } 393 394 int 395 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 396 dsl_dataset_t **dsp) 397 { 398 objset_t *mos = dp->dp_meta_objset; 399 dmu_buf_t *dbuf; 400 dsl_dataset_t *ds; 401 int err; 402 dmu_object_info_t doi; 403 404 ASSERT(dsl_pool_config_held(dp)); 405 406 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 407 if (err != 0) 408 return (err); 409 410 /* Make sure dsobj has the correct object type. */ 411 dmu_object_info_from_db(dbuf, &doi); 412 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 413 dmu_buf_rele(dbuf, tag); 414 return (SET_ERROR(EINVAL)); 415 } 416 417 ds = dmu_buf_get_user(dbuf); 418 if (ds == NULL) { 419 dsl_dataset_t *winner = NULL; 420 421 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 422 ds->ds_dbuf = dbuf; 423 ds->ds_object = dsobj; 424 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0; 425 426 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 427 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 428 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 429 refcount_create(&ds->ds_longholds); 430 431 bplist_create(&ds->ds_pending_deadlist); 432 dsl_deadlist_open(&ds->ds_deadlist, 433 mos, dsl_dataset_phys(ds)->ds_deadlist_obj); 434 435 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 436 offsetof(dmu_sendarg_t, dsa_link)); 437 438 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t), 439 offsetof(dsl_prop_cb_record_t, cbr_ds_node)); 440 441 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 442 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 443 if (!(spa_feature_table[f].fi_flags & 444 ZFEATURE_FLAG_PER_DATASET)) 445 continue; 446 err = zap_contains(mos, dsobj, 447 spa_feature_table[f].fi_guid); 448 if (err == 0) { 449 ds->ds_feature_inuse[f] = B_TRUE; 450 } else { 451 ASSERT3U(err, ==, ENOENT); 452 err = 0; 453 } 454 } 455 } 456 457 err = dsl_dir_hold_obj(dp, 458 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir); 459 if (err != 0) { 460 mutex_destroy(&ds->ds_lock); 461 mutex_destroy(&ds->ds_opening_lock); 462 mutex_destroy(&ds->ds_sendstream_lock); 463 refcount_destroy(&ds->ds_longholds); 464 bplist_destroy(&ds->ds_pending_deadlist); 465 dsl_deadlist_close(&ds->ds_deadlist); 466 kmem_free(ds, sizeof (dsl_dataset_t)); 467 dmu_buf_rele(dbuf, tag); 468 return (err); 469 } 470 471 if (!ds->ds_is_snapshot) { 472 ds->ds_snapname[0] = '\0'; 473 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 474 err = dsl_dataset_hold_obj(dp, 475 dsl_dataset_phys(ds)->ds_prev_snap_obj, 476 ds, &ds->ds_prev); 477 } 478 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 479 int zaperr = zap_lookup(mos, ds->ds_object, 480 DS_FIELD_BOOKMARK_NAMES, 481 sizeof (ds->ds_bookmarks), 1, 482 &ds->ds_bookmarks); 483 if (zaperr != ENOENT) 484 VERIFY0(zaperr); 485 } 486 } else { 487 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 488 err = dsl_dataset_get_snapname(ds); 489 if (err == 0 && 490 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { 491 err = zap_count( 492 ds->ds_dir->dd_pool->dp_meta_objset, 493 dsl_dataset_phys(ds)->ds_userrefs_obj, 494 &ds->ds_userrefs); 495 } 496 } 497 498 if (err == 0 && !ds->ds_is_snapshot) { 499 err = dsl_prop_get_int_ds(ds, 500 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 501 &ds->ds_reserved); 502 if (err == 0) { 503 err = dsl_prop_get_int_ds(ds, 504 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 505 &ds->ds_quota); 506 } 507 } else { 508 ds->ds_reserved = ds->ds_quota = 0; 509 } 510 511 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf); 512 if (err == 0) 513 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu); 514 515 if (err != 0 || winner != NULL) { 516 bplist_destroy(&ds->ds_pending_deadlist); 517 dsl_deadlist_close(&ds->ds_deadlist); 518 if (ds->ds_prev) 519 dsl_dataset_rele(ds->ds_prev, ds); 520 dsl_dir_rele(ds->ds_dir, ds); 521 mutex_destroy(&ds->ds_lock); 522 mutex_destroy(&ds->ds_opening_lock); 523 mutex_destroy(&ds->ds_sendstream_lock); 524 refcount_destroy(&ds->ds_longholds); 525 kmem_free(ds, sizeof (dsl_dataset_t)); 526 if (err != 0) { 527 dmu_buf_rele(dbuf, tag); 528 return (err); 529 } 530 ds = winner; 531 } else { 532 ds->ds_fsid_guid = 533 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); 534 } 535 } 536 ASSERT3P(ds->ds_dbuf, ==, dbuf); 537 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); 538 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || 539 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 540 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 541 *dsp = ds; 542 return (0); 543 } 544 545 int 546 dsl_dataset_hold(dsl_pool_t *dp, const char *name, 547 void *tag, dsl_dataset_t **dsp) 548 { 549 dsl_dir_t *dd; 550 const char *snapname; 551 uint64_t obj; 552 int err = 0; 553 dsl_dataset_t *ds; 554 555 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 556 if (err != 0) 557 return (err); 558 559 ASSERT(dsl_pool_config_held(dp)); 560 obj = dsl_dir_phys(dd)->dd_head_dataset_obj; 561 if (obj != 0) 562 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 563 else 564 err = SET_ERROR(ENOENT); 565 566 /* we may be looking for a snapshot */ 567 if (err == 0 && snapname != NULL) { 568 dsl_dataset_t *snap_ds; 569 570 if (*snapname++ != '@') { 571 dsl_dataset_rele(ds, tag); 572 dsl_dir_rele(dd, FTAG); 573 return (SET_ERROR(ENOENT)); 574 } 575 576 dprintf("looking for snapshot '%s'\n", snapname); 577 err = dsl_dataset_snap_lookup(ds, snapname, &obj); 578 if (err == 0) 579 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds); 580 dsl_dataset_rele(ds, tag); 581 582 if (err == 0) { 583 mutex_enter(&snap_ds->ds_lock); 584 if (snap_ds->ds_snapname[0] == 0) 585 (void) strlcpy(snap_ds->ds_snapname, snapname, 586 sizeof (snap_ds->ds_snapname)); 587 mutex_exit(&snap_ds->ds_lock); 588 ds = snap_ds; 589 } 590 } 591 if (err == 0) 592 *dsp = ds; 593 dsl_dir_rele(dd, FTAG); 594 return (err); 595 } 596 597 int 598 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 599 void *tag, dsl_dataset_t **dsp) 600 { 601 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 602 if (err != 0) 603 return (err); 604 if (!dsl_dataset_tryown(*dsp, tag)) { 605 dsl_dataset_rele(*dsp, tag); 606 *dsp = NULL; 607 return (SET_ERROR(EBUSY)); 608 } 609 return (0); 610 } 611 612 int 613 dsl_dataset_own(dsl_pool_t *dp, const char *name, 614 void *tag, dsl_dataset_t **dsp) 615 { 616 int err = dsl_dataset_hold(dp, name, tag, dsp); 617 if (err != 0) 618 return (err); 619 if (!dsl_dataset_tryown(*dsp, tag)) { 620 dsl_dataset_rele(*dsp, tag); 621 return (SET_ERROR(EBUSY)); 622 } 623 return (0); 624 } 625 626 /* 627 * See the comment above dsl_pool_hold() for details. In summary, a long 628 * hold is used to prevent destruction of a dataset while the pool hold 629 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 630 * 631 * The dataset and pool must be held when this function is called. After it 632 * is called, the pool hold may be released while the dataset is still held 633 * and accessed. 634 */ 635 void 636 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 637 { 638 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 639 (void) refcount_add(&ds->ds_longholds, tag); 640 } 641 642 void 643 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 644 { 645 (void) refcount_remove(&ds->ds_longholds, tag); 646 } 647 648 /* Return B_TRUE if there are any long holds on this dataset. */ 649 boolean_t 650 dsl_dataset_long_held(dsl_dataset_t *ds) 651 { 652 return (!refcount_is_zero(&ds->ds_longholds)); 653 } 654 655 void 656 dsl_dataset_name(dsl_dataset_t *ds, char *name) 657 { 658 if (ds == NULL) { 659 (void) strcpy(name, "mos"); 660 } else { 661 dsl_dir_name(ds->ds_dir, name); 662 VERIFY0(dsl_dataset_get_snapname(ds)); 663 if (ds->ds_snapname[0]) { 664 (void) strcat(name, "@"); 665 /* 666 * We use a "recursive" mutex so that we 667 * can call dprintf_ds() with ds_lock held. 668 */ 669 if (!MUTEX_HELD(&ds->ds_lock)) { 670 mutex_enter(&ds->ds_lock); 671 (void) strcat(name, ds->ds_snapname); 672 mutex_exit(&ds->ds_lock); 673 } else { 674 (void) strcat(name, ds->ds_snapname); 675 } 676 } 677 } 678 } 679 680 void 681 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 682 { 683 dmu_buf_rele(ds->ds_dbuf, tag); 684 } 685 686 void 687 dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 688 { 689 ASSERT3P(ds->ds_owner, ==, tag); 690 ASSERT(ds->ds_dbuf != NULL); 691 692 mutex_enter(&ds->ds_lock); 693 ds->ds_owner = NULL; 694 mutex_exit(&ds->ds_lock); 695 dsl_dataset_long_rele(ds, tag); 696 dsl_dataset_rele(ds, tag); 697 } 698 699 boolean_t 700 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 701 { 702 boolean_t gotit = FALSE; 703 704 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 705 mutex_enter(&ds->ds_lock); 706 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 707 ds->ds_owner = tag; 708 dsl_dataset_long_hold(ds, tag); 709 gotit = TRUE; 710 } 711 mutex_exit(&ds->ds_lock); 712 return (gotit); 713 } 714 715 boolean_t 716 dsl_dataset_has_owner(dsl_dataset_t *ds) 717 { 718 boolean_t rv; 719 mutex_enter(&ds->ds_lock); 720 rv = (ds->ds_owner != NULL); 721 mutex_exit(&ds->ds_lock); 722 return (rv); 723 } 724 725 static void 726 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx) 727 { 728 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 729 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 730 uint64_t zero = 0; 731 732 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 733 734 spa_feature_incr(spa, f, tx); 735 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 736 737 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid, 738 sizeof (zero), 1, &zero, tx)); 739 } 740 741 void 742 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx) 743 { 744 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 745 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 746 747 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET); 748 749 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx)); 750 spa_feature_decr(spa, f, tx); 751 } 752 753 uint64_t 754 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 755 uint64_t flags, dmu_tx_t *tx) 756 { 757 dsl_pool_t *dp = dd->dd_pool; 758 dmu_buf_t *dbuf; 759 dsl_dataset_phys_t *dsphys; 760 uint64_t dsobj; 761 objset_t *mos = dp->dp_meta_objset; 762 763 if (origin == NULL) 764 origin = dp->dp_origin_snap; 765 766 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 767 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); 768 ASSERT(dmu_tx_is_syncing(tx)); 769 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 770 771 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 772 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 773 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 774 dmu_buf_will_dirty(dbuf, tx); 775 dsphys = dbuf->db_data; 776 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 777 dsphys->ds_dir_obj = dd->dd_object; 778 dsphys->ds_flags = flags; 779 dsphys->ds_fsid_guid = unique_create(); 780 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 781 sizeof (dsphys->ds_guid)); 782 dsphys->ds_snapnames_zapobj = 783 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 784 DMU_OT_NONE, 0, tx); 785 dsphys->ds_creation_time = gethrestime_sec(); 786 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 787 788 if (origin == NULL) { 789 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 790 } else { 791 dsl_dataset_t *ohds; /* head of the origin snapshot */ 792 793 dsphys->ds_prev_snap_obj = origin->ds_object; 794 dsphys->ds_prev_snap_txg = 795 dsl_dataset_phys(origin)->ds_creation_txg; 796 dsphys->ds_referenced_bytes = 797 dsl_dataset_phys(origin)->ds_referenced_bytes; 798 dsphys->ds_compressed_bytes = 799 dsl_dataset_phys(origin)->ds_compressed_bytes; 800 dsphys->ds_uncompressed_bytes = 801 dsl_dataset_phys(origin)->ds_uncompressed_bytes; 802 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; 803 804 /* 805 * Inherit flags that describe the dataset's contents 806 * (INCONSISTENT) or properties (Case Insensitive). 807 */ 808 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & 809 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 810 811 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 812 if (origin->ds_feature_inuse[f]) 813 dsl_dataset_activate_feature(dsobj, f, tx); 814 } 815 816 dmu_buf_will_dirty(origin->ds_dbuf, tx); 817 dsl_dataset_phys(origin)->ds_num_children++; 818 819 VERIFY0(dsl_dataset_hold_obj(dp, 820 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, 821 FTAG, &ohds)); 822 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 823 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 824 dsl_dataset_rele(ohds, FTAG); 825 826 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 827 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { 828 dsl_dataset_phys(origin)->ds_next_clones_obj = 829 zap_create(mos, 830 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 831 } 832 VERIFY0(zap_add_int(mos, 833 dsl_dataset_phys(origin)->ds_next_clones_obj, 834 dsobj, tx)); 835 } 836 837 dmu_buf_will_dirty(dd->dd_dbuf, tx); 838 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; 839 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 840 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 841 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 842 dsl_dir_phys(origin->ds_dir)->dd_clones = 843 zap_create(mos, 844 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 845 } 846 VERIFY0(zap_add_int(mos, 847 dsl_dir_phys(origin->ds_dir)->dd_clones, 848 dsobj, tx)); 849 } 850 } 851 852 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 853 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 854 855 dmu_buf_rele(dbuf, FTAG); 856 857 dmu_buf_will_dirty(dd->dd_dbuf, tx); 858 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; 859 860 return (dsobj); 861 } 862 863 static void 864 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 865 { 866 objset_t *os; 867 868 VERIFY0(dmu_objset_from_ds(ds, &os)); 869 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 870 dsl_dataset_dirty(ds, tx); 871 } 872 873 uint64_t 874 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 875 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 876 { 877 dsl_pool_t *dp = pdd->dd_pool; 878 uint64_t dsobj, ddobj; 879 dsl_dir_t *dd; 880 881 ASSERT(dmu_tx_is_syncing(tx)); 882 ASSERT(lastname[0] != '@'); 883 884 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 885 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 886 887 dsobj = dsl_dataset_create_sync_dd(dd, origin, 888 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 889 890 dsl_deleg_set_create_perms(dd, tx, cr); 891 892 /* 893 * Since we're creating a new node we know it's a leaf, so we can 894 * initialize the counts if the limit feature is active. 895 */ 896 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 897 uint64_t cnt = 0; 898 objset_t *os = dd->dd_pool->dp_meta_objset; 899 900 dsl_dir_zapify(dd, tx); 901 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 902 sizeof (cnt), 1, &cnt, tx)); 903 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 904 sizeof (cnt), 1, &cnt, tx)); 905 } 906 907 dsl_dir_rele(dd, FTAG); 908 909 /* 910 * If we are creating a clone, make sure we zero out any stale 911 * data from the origin snapshots zil header. 912 */ 913 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 914 dsl_dataset_t *ds; 915 916 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 917 dsl_dataset_zero_zil(ds, tx); 918 dsl_dataset_rele(ds, FTAG); 919 } 920 921 return (dsobj); 922 } 923 924 /* 925 * The unique space in the head dataset can be calculated by subtracting 926 * the space used in the most recent snapshot, that is still being used 927 * in this file system, from the space currently in use. To figure out 928 * the space in the most recent snapshot still in use, we need to take 929 * the total space used in the snapshot and subtract out the space that 930 * has been freed up since the snapshot was taken. 931 */ 932 void 933 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 934 { 935 uint64_t mrs_used; 936 uint64_t dlused, dlcomp, dluncomp; 937 938 ASSERT(!ds->ds_is_snapshot); 939 940 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) 941 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; 942 else 943 mrs_used = 0; 944 945 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 946 947 ASSERT3U(dlused, <=, mrs_used); 948 dsl_dataset_phys(ds)->ds_unique_bytes = 949 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); 950 951 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 952 SPA_VERSION_UNIQUE_ACCURATE) 953 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 954 } 955 956 void 957 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 958 dmu_tx_t *tx) 959 { 960 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 961 uint64_t count; 962 int err; 963 964 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); 965 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 966 obj, tx); 967 /* 968 * The err should not be ENOENT, but a bug in a previous version 969 * of the code could cause upgrade_clones_cb() to not set 970 * ds_next_snap_obj when it should, leading to a missing entry. 971 * If we knew that the pool was created after 972 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 973 * ENOENT. However, at least we can check that we don't have 974 * too many entries in the next_clones_obj even after failing to 975 * remove this one. 976 */ 977 if (err != ENOENT) 978 VERIFY0(err); 979 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 980 &count)); 981 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); 982 } 983 984 985 blkptr_t * 986 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 987 { 988 return (&dsl_dataset_phys(ds)->ds_bp); 989 } 990 991 void 992 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 993 { 994 ASSERT(dmu_tx_is_syncing(tx)); 995 /* If it's the meta-objset, set dp_meta_rootbp */ 996 if (ds == NULL) { 997 tx->tx_pool->dp_meta_rootbp = *bp; 998 } else { 999 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1000 dsl_dataset_phys(ds)->ds_bp = *bp; 1001 } 1002 } 1003 1004 spa_t * 1005 dsl_dataset_get_spa(dsl_dataset_t *ds) 1006 { 1007 return (ds->ds_dir->dd_pool->dp_spa); 1008 } 1009 1010 void 1011 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 1012 { 1013 dsl_pool_t *dp; 1014 1015 if (ds == NULL) /* this is the meta-objset */ 1016 return; 1017 1018 ASSERT(ds->ds_objset != NULL); 1019 1020 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) 1021 panic("dirtying snapshot!"); 1022 1023 dp = ds->ds_dir->dd_pool; 1024 1025 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1026 /* up the hold count until we can be written out */ 1027 dmu_buf_add_ref(ds->ds_dbuf, ds); 1028 } 1029 } 1030 1031 boolean_t 1032 dsl_dataset_is_dirty(dsl_dataset_t *ds) 1033 { 1034 for (int t = 0; t < TXG_SIZE; t++) { 1035 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 1036 ds, t)) 1037 return (B_TRUE); 1038 } 1039 return (B_FALSE); 1040 } 1041 1042 static int 1043 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 1044 { 1045 uint64_t asize; 1046 1047 if (!dmu_tx_is_syncing(tx)) 1048 return (0); 1049 1050 /* 1051 * If there's an fs-only reservation, any blocks that might become 1052 * owned by the snapshot dataset must be accommodated by space 1053 * outside of the reservation. 1054 */ 1055 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 1056 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); 1057 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 1058 return (SET_ERROR(ENOSPC)); 1059 1060 /* 1061 * Propagate any reserved space for this snapshot to other 1062 * snapshot checks in this sync group. 1063 */ 1064 if (asize > 0) 1065 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1066 1067 return (0); 1068 } 1069 1070 typedef struct dsl_dataset_snapshot_arg { 1071 nvlist_t *ddsa_snaps; 1072 nvlist_t *ddsa_props; 1073 nvlist_t *ddsa_errors; 1074 cred_t *ddsa_cr; 1075 } dsl_dataset_snapshot_arg_t; 1076 1077 int 1078 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1079 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 1080 { 1081 int error; 1082 uint64_t value; 1083 1084 ds->ds_trysnap_txg = tx->tx_txg; 1085 1086 if (!dmu_tx_is_syncing(tx)) 1087 return (0); 1088 1089 /* 1090 * We don't allow multiple snapshots of the same txg. If there 1091 * is already one, try again. 1092 */ 1093 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) 1094 return (SET_ERROR(EAGAIN)); 1095 1096 /* 1097 * Check for conflicting snapshot name. 1098 */ 1099 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1100 if (error == 0) 1101 return (SET_ERROR(EEXIST)); 1102 if (error != ENOENT) 1103 return (error); 1104 1105 /* 1106 * We don't allow taking snapshots of inconsistent datasets, such as 1107 * those into which we are currently receiving. However, if we are 1108 * creating this snapshot as part of a receive, this check will be 1109 * executed atomically with respect to the completion of the receive 1110 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1111 * case we ignore this, knowing it will be fixed up for us shortly in 1112 * dmu_recv_end_sync(). 1113 */ 1114 if (!recv && DS_IS_INCONSISTENT(ds)) 1115 return (SET_ERROR(EBUSY)); 1116 1117 /* 1118 * Skip the check for temporary snapshots or if we have already checked 1119 * the counts in dsl_dataset_snapshot_check. This means we really only 1120 * check the count here when we're receiving a stream. 1121 */ 1122 if (cnt != 0 && cr != NULL) { 1123 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1124 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 1125 if (error != 0) 1126 return (error); 1127 } 1128 1129 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1130 if (error != 0) 1131 return (error); 1132 1133 return (0); 1134 } 1135 1136 static int 1137 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1138 { 1139 dsl_dataset_snapshot_arg_t *ddsa = arg; 1140 dsl_pool_t *dp = dmu_tx_pool(tx); 1141 nvpair_t *pair; 1142 int rv = 0; 1143 1144 /* 1145 * Pre-compute how many total new snapshots will be created for each 1146 * level in the tree and below. This is needed for validating the 1147 * snapshot limit when either taking a recursive snapshot or when 1148 * taking multiple snapshots. 1149 * 1150 * The problem is that the counts are not actually adjusted when 1151 * we are checking, only when we finally sync. For a single snapshot, 1152 * this is easy, the count will increase by 1 at each node up the tree, 1153 * but its more complicated for the recursive/multiple snapshot case. 1154 * 1155 * The dsl_fs_ss_limit_check function does recursively check the count 1156 * at each level up the tree but since it is validating each snapshot 1157 * independently we need to be sure that we are validating the complete 1158 * count for the entire set of snapshots. We do this by rolling up the 1159 * counts for each component of the name into an nvlist and then 1160 * checking each of those cases with the aggregated count. 1161 * 1162 * This approach properly handles not only the recursive snapshot 1163 * case (where we get all of those on the ddsa_snaps list) but also 1164 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1165 * validate the limit on 'a' using a count of 2). 1166 * 1167 * We validate the snapshot names in the third loop and only report 1168 * name errors once. 1169 */ 1170 if (dmu_tx_is_syncing(tx)) { 1171 nvlist_t *cnt_track = NULL; 1172 cnt_track = fnvlist_alloc(); 1173 1174 /* Rollup aggregated counts into the cnt_track list */ 1175 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1176 pair != NULL; 1177 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1178 char *pdelim; 1179 uint64_t val; 1180 char nm[MAXPATHLEN]; 1181 1182 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1183 pdelim = strchr(nm, '@'); 1184 if (pdelim == NULL) 1185 continue; 1186 *pdelim = '\0'; 1187 1188 do { 1189 if (nvlist_lookup_uint64(cnt_track, nm, 1190 &val) == 0) { 1191 /* update existing entry */ 1192 fnvlist_add_uint64(cnt_track, nm, 1193 val + 1); 1194 } else { 1195 /* add to list */ 1196 fnvlist_add_uint64(cnt_track, nm, 1); 1197 } 1198 1199 pdelim = strrchr(nm, '/'); 1200 if (pdelim != NULL) 1201 *pdelim = '\0'; 1202 } while (pdelim != NULL); 1203 } 1204 1205 /* Check aggregated counts at each level */ 1206 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1207 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1208 int error = 0; 1209 char *name; 1210 uint64_t cnt = 0; 1211 dsl_dataset_t *ds; 1212 1213 name = nvpair_name(pair); 1214 cnt = fnvpair_value_uint64(pair); 1215 ASSERT(cnt > 0); 1216 1217 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1218 if (error == 0) { 1219 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1220 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1221 ddsa->ddsa_cr); 1222 dsl_dataset_rele(ds, FTAG); 1223 } 1224 1225 if (error != 0) { 1226 if (ddsa->ddsa_errors != NULL) 1227 fnvlist_add_int32(ddsa->ddsa_errors, 1228 name, error); 1229 rv = error; 1230 /* only report one error for this check */ 1231 break; 1232 } 1233 } 1234 nvlist_free(cnt_track); 1235 } 1236 1237 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1238 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1239 int error = 0; 1240 dsl_dataset_t *ds; 1241 char *name, *atp; 1242 char dsname[MAXNAMELEN]; 1243 1244 name = nvpair_name(pair); 1245 if (strlen(name) >= MAXNAMELEN) 1246 error = SET_ERROR(ENAMETOOLONG); 1247 if (error == 0) { 1248 atp = strchr(name, '@'); 1249 if (atp == NULL) 1250 error = SET_ERROR(EINVAL); 1251 if (error == 0) 1252 (void) strlcpy(dsname, name, atp - name + 1); 1253 } 1254 if (error == 0) 1255 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1256 if (error == 0) { 1257 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1258 error = dsl_dataset_snapshot_check_impl(ds, 1259 atp + 1, tx, B_FALSE, 0, NULL); 1260 dsl_dataset_rele(ds, FTAG); 1261 } 1262 1263 if (error != 0) { 1264 if (ddsa->ddsa_errors != NULL) { 1265 fnvlist_add_int32(ddsa->ddsa_errors, 1266 name, error); 1267 } 1268 rv = error; 1269 } 1270 } 1271 1272 return (rv); 1273 } 1274 1275 void 1276 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1277 dmu_tx_t *tx) 1278 { 1279 static zil_header_t zero_zil; 1280 1281 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1282 dmu_buf_t *dbuf; 1283 dsl_dataset_phys_t *dsphys; 1284 uint64_t dsobj, crtxg; 1285 objset_t *mos = dp->dp_meta_objset; 1286 objset_t *os; 1287 1288 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1289 1290 /* 1291 * If we are on an old pool, the zil must not be active, in which 1292 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1293 */ 1294 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1295 dmu_objset_from_ds(ds, &os) != 0 || 1296 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1297 sizeof (zero_zil)) == 0); 1298 1299 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1300 1301 /* 1302 * The origin's ds_creation_txg has to be < TXG_INITIAL 1303 */ 1304 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1305 crtxg = 1; 1306 else 1307 crtxg = tx->tx_txg; 1308 1309 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1310 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1311 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1312 dmu_buf_will_dirty(dbuf, tx); 1313 dsphys = dbuf->db_data; 1314 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1315 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1316 dsphys->ds_fsid_guid = unique_create(); 1317 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1318 sizeof (dsphys->ds_guid)); 1319 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1320 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 1321 dsphys->ds_next_snap_obj = ds->ds_object; 1322 dsphys->ds_num_children = 1; 1323 dsphys->ds_creation_time = gethrestime_sec(); 1324 dsphys->ds_creation_txg = crtxg; 1325 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; 1326 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; 1327 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; 1328 dsphys->ds_uncompressed_bytes = 1329 dsl_dataset_phys(ds)->ds_uncompressed_bytes; 1330 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; 1331 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; 1332 dmu_buf_rele(dbuf, FTAG); 1333 1334 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1335 if (ds->ds_feature_inuse[f]) 1336 dsl_dataset_activate_feature(dsobj, f, tx); 1337 } 1338 1339 ASSERT3U(ds->ds_prev != 0, ==, 1340 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); 1341 if (ds->ds_prev) { 1342 uint64_t next_clones_obj = 1343 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; 1344 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1345 ds->ds_object || 1346 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); 1347 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1348 ds->ds_object) { 1349 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1350 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, 1351 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); 1352 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; 1353 } else if (next_clones_obj != 0) { 1354 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1355 dsphys->ds_next_snap_obj, tx); 1356 VERIFY0(zap_add_int(mos, 1357 next_clones_obj, dsobj, tx)); 1358 } 1359 } 1360 1361 /* 1362 * If we have a reference-reservation on this dataset, we will 1363 * need to increase the amount of refreservation being charged 1364 * since our unique space is going to zero. 1365 */ 1366 if (ds->ds_reserved) { 1367 int64_t delta; 1368 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1369 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, 1370 ds->ds_reserved); 1371 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1372 delta, 0, 0, tx); 1373 } 1374 1375 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1376 dsl_dataset_phys(ds)->ds_deadlist_obj = 1377 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, 1378 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 1379 dsl_deadlist_close(&ds->ds_deadlist); 1380 dsl_deadlist_open(&ds->ds_deadlist, mos, 1381 dsl_dataset_phys(ds)->ds_deadlist_obj); 1382 dsl_deadlist_add_key(&ds->ds_deadlist, 1383 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); 1384 1385 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); 1386 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; 1387 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; 1388 dsl_dataset_phys(ds)->ds_unique_bytes = 0; 1389 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1390 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1391 1392 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1393 snapname, 8, 1, &dsobj, tx)); 1394 1395 if (ds->ds_prev) 1396 dsl_dataset_rele(ds->ds_prev, ds); 1397 VERIFY0(dsl_dataset_hold_obj(dp, 1398 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); 1399 1400 dsl_scan_ds_snapshotted(ds, tx); 1401 1402 dsl_dir_snap_cmtime_update(ds->ds_dir); 1403 1404 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1405 } 1406 1407 static void 1408 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1409 { 1410 dsl_dataset_snapshot_arg_t *ddsa = arg; 1411 dsl_pool_t *dp = dmu_tx_pool(tx); 1412 nvpair_t *pair; 1413 1414 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1415 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1416 dsl_dataset_t *ds; 1417 char *name, *atp; 1418 char dsname[MAXNAMELEN]; 1419 1420 name = nvpair_name(pair); 1421 atp = strchr(name, '@'); 1422 (void) strlcpy(dsname, name, atp - name + 1); 1423 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1424 1425 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1426 if (ddsa->ddsa_props != NULL) { 1427 dsl_props_set_sync_impl(ds->ds_prev, 1428 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1429 } 1430 dsl_dataset_rele(ds, FTAG); 1431 } 1432 } 1433 1434 /* 1435 * The snapshots must all be in the same pool. 1436 * All-or-nothing: if there are any failures, nothing will be modified. 1437 */ 1438 int 1439 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1440 { 1441 dsl_dataset_snapshot_arg_t ddsa; 1442 nvpair_t *pair; 1443 boolean_t needsuspend; 1444 int error; 1445 spa_t *spa; 1446 char *firstname; 1447 nvlist_t *suspended = NULL; 1448 1449 pair = nvlist_next_nvpair(snaps, NULL); 1450 if (pair == NULL) 1451 return (0); 1452 firstname = nvpair_name(pair); 1453 1454 error = spa_open(firstname, &spa, FTAG); 1455 if (error != 0) 1456 return (error); 1457 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1458 spa_close(spa, FTAG); 1459 1460 if (needsuspend) { 1461 suspended = fnvlist_alloc(); 1462 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1463 pair = nvlist_next_nvpair(snaps, pair)) { 1464 char fsname[MAXNAMELEN]; 1465 char *snapname = nvpair_name(pair); 1466 char *atp; 1467 void *cookie; 1468 1469 atp = strchr(snapname, '@'); 1470 if (atp == NULL) { 1471 error = SET_ERROR(EINVAL); 1472 break; 1473 } 1474 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1475 1476 error = zil_suspend(fsname, &cookie); 1477 if (error != 0) 1478 break; 1479 fnvlist_add_uint64(suspended, fsname, 1480 (uintptr_t)cookie); 1481 } 1482 } 1483 1484 ddsa.ddsa_snaps = snaps; 1485 ddsa.ddsa_props = props; 1486 ddsa.ddsa_errors = errors; 1487 ddsa.ddsa_cr = CRED(); 1488 1489 if (error == 0) { 1490 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1491 dsl_dataset_snapshot_sync, &ddsa, 1492 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1493 } 1494 1495 if (suspended != NULL) { 1496 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1497 pair = nvlist_next_nvpair(suspended, pair)) { 1498 zil_resume((void *)(uintptr_t) 1499 fnvpair_value_uint64(pair)); 1500 } 1501 fnvlist_free(suspended); 1502 } 1503 1504 return (error); 1505 } 1506 1507 typedef struct dsl_dataset_snapshot_tmp_arg { 1508 const char *ddsta_fsname; 1509 const char *ddsta_snapname; 1510 minor_t ddsta_cleanup_minor; 1511 const char *ddsta_htag; 1512 } dsl_dataset_snapshot_tmp_arg_t; 1513 1514 static int 1515 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1516 { 1517 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1518 dsl_pool_t *dp = dmu_tx_pool(tx); 1519 dsl_dataset_t *ds; 1520 int error; 1521 1522 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1523 if (error != 0) 1524 return (error); 1525 1526 /* NULL cred means no limit check for tmp snapshot */ 1527 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1528 tx, B_FALSE, 0, NULL); 1529 if (error != 0) { 1530 dsl_dataset_rele(ds, FTAG); 1531 return (error); 1532 } 1533 1534 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1535 dsl_dataset_rele(ds, FTAG); 1536 return (SET_ERROR(ENOTSUP)); 1537 } 1538 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1539 B_TRUE, tx); 1540 if (error != 0) { 1541 dsl_dataset_rele(ds, FTAG); 1542 return (error); 1543 } 1544 1545 dsl_dataset_rele(ds, FTAG); 1546 return (0); 1547 } 1548 1549 static void 1550 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1551 { 1552 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1553 dsl_pool_t *dp = dmu_tx_pool(tx); 1554 dsl_dataset_t *ds; 1555 1556 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1557 1558 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1559 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1560 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1561 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1562 1563 dsl_dataset_rele(ds, FTAG); 1564 } 1565 1566 int 1567 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1568 minor_t cleanup_minor, const char *htag) 1569 { 1570 dsl_dataset_snapshot_tmp_arg_t ddsta; 1571 int error; 1572 spa_t *spa; 1573 boolean_t needsuspend; 1574 void *cookie; 1575 1576 ddsta.ddsta_fsname = fsname; 1577 ddsta.ddsta_snapname = snapname; 1578 ddsta.ddsta_cleanup_minor = cleanup_minor; 1579 ddsta.ddsta_htag = htag; 1580 1581 error = spa_open(fsname, &spa, FTAG); 1582 if (error != 0) 1583 return (error); 1584 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1585 spa_close(spa, FTAG); 1586 1587 if (needsuspend) { 1588 error = zil_suspend(fsname, &cookie); 1589 if (error != 0) 1590 return (error); 1591 } 1592 1593 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1594 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 1595 1596 if (needsuspend) 1597 zil_resume(cookie); 1598 return (error); 1599 } 1600 1601 1602 void 1603 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1604 { 1605 ASSERT(dmu_tx_is_syncing(tx)); 1606 ASSERT(ds->ds_objset != NULL); 1607 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); 1608 1609 /* 1610 * in case we had to change ds_fsid_guid when we opened it, 1611 * sync it out now. 1612 */ 1613 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1614 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; 1615 1616 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) { 1617 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1618 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1, 1619 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx)); 1620 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1621 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1, 1622 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx)); 1623 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset, 1624 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1, 1625 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx)); 1626 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0; 1627 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0; 1628 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0; 1629 } 1630 1631 dmu_objset_sync(ds->ds_objset, zio, tx); 1632 1633 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 1634 if (ds->ds_feature_activation_needed[f]) { 1635 if (ds->ds_feature_inuse[f]) 1636 continue; 1637 dsl_dataset_activate_feature(ds->ds_object, f, tx); 1638 ds->ds_feature_inuse[f] = B_TRUE; 1639 } 1640 } 1641 } 1642 1643 static void 1644 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1645 { 1646 uint64_t count = 0; 1647 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1648 zap_cursor_t zc; 1649 zap_attribute_t za; 1650 nvlist_t *propval = fnvlist_alloc(); 1651 nvlist_t *val = fnvlist_alloc(); 1652 1653 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1654 1655 /* 1656 * There may be missing entries in ds_next_clones_obj 1657 * due to a bug in a previous version of the code. 1658 * Only trust it if it has the right number of entries. 1659 */ 1660 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { 1661 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1662 &count)); 1663 } 1664 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) 1665 goto fail; 1666 for (zap_cursor_init(&zc, mos, 1667 dsl_dataset_phys(ds)->ds_next_clones_obj); 1668 zap_cursor_retrieve(&zc, &za) == 0; 1669 zap_cursor_advance(&zc)) { 1670 dsl_dataset_t *clone; 1671 char buf[ZFS_MAXNAMELEN]; 1672 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1673 za.za_first_integer, FTAG, &clone)); 1674 dsl_dir_name(clone->ds_dir, buf); 1675 fnvlist_add_boolean(val, buf); 1676 dsl_dataset_rele(clone, FTAG); 1677 } 1678 zap_cursor_fini(&zc); 1679 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1680 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); 1681 fail: 1682 nvlist_free(val); 1683 nvlist_free(propval); 1684 } 1685 1686 static void 1687 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv) 1688 { 1689 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1690 1691 if (dsl_dataset_has_resume_receive_state(ds)) { 1692 char *str; 1693 void *packed; 1694 uint8_t *compressed; 1695 uint64_t val; 1696 nvlist_t *token_nv = fnvlist_alloc(); 1697 size_t packed_size, compressed_size; 1698 1699 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1700 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) { 1701 fnvlist_add_uint64(token_nv, "fromguid", val); 1702 } 1703 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1704 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) { 1705 fnvlist_add_uint64(token_nv, "object", val); 1706 } 1707 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1708 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) { 1709 fnvlist_add_uint64(token_nv, "offset", val); 1710 } 1711 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1712 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) { 1713 fnvlist_add_uint64(token_nv, "bytes", val); 1714 } 1715 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1716 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) { 1717 fnvlist_add_uint64(token_nv, "toguid", val); 1718 } 1719 char buf[256]; 1720 if (zap_lookup(dp->dp_meta_objset, ds->ds_object, 1721 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) { 1722 fnvlist_add_string(token_nv, "toname", buf); 1723 } 1724 if (zap_contains(dp->dp_meta_objset, ds->ds_object, 1725 DS_FIELD_RESUME_EMBEDOK) == 0) { 1726 fnvlist_add_boolean(token_nv, "embedok"); 1727 } 1728 packed = fnvlist_pack(token_nv, &packed_size); 1729 fnvlist_free(token_nv); 1730 compressed = kmem_alloc(packed_size, KM_SLEEP); 1731 1732 compressed_size = gzip_compress(packed, compressed, 1733 packed_size, packed_size, 6); 1734 1735 zio_cksum_t cksum; 1736 fletcher_4_native(compressed, compressed_size, NULL, &cksum); 1737 1738 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP); 1739 for (int i = 0; i < compressed_size; i++) { 1740 (void) sprintf(str + i * 2, "%02x", compressed[i]); 1741 } 1742 str[compressed_size * 2] = '\0'; 1743 char *propval = kmem_asprintf("%u-%llx-%llx-%s", 1744 ZFS_SEND_RESUME_TOKEN_VERSION, 1745 (longlong_t)cksum.zc_word[0], 1746 (longlong_t)packed_size, str); 1747 dsl_prop_nvlist_add_string(nv, 1748 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval); 1749 kmem_free(packed, packed_size); 1750 kmem_free(str, compressed_size * 2 + 1); 1751 kmem_free(compressed, packed_size); 1752 strfree(propval); 1753 } 1754 } 1755 1756 void 1757 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1758 { 1759 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1760 uint64_t refd, avail, uobjs, aobjs, ratio; 1761 1762 ASSERT(dsl_pool_config_held(dp)); 1763 1764 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : 1765 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / 1766 dsl_dataset_phys(ds)->ds_compressed_bytes); 1767 1768 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); 1769 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 1770 dsl_dataset_phys(ds)->ds_uncompressed_bytes); 1771 1772 if (ds->ds_is_snapshot) { 1773 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); 1774 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 1775 dsl_dataset_phys(ds)->ds_unique_bytes); 1776 get_clones_stat(ds, nv); 1777 } else { 1778 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 1779 char buf[MAXNAMELEN]; 1780 dsl_dataset_name(ds->ds_prev, buf); 1781 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf); 1782 } 1783 1784 dsl_dir_stats(ds->ds_dir, nv); 1785 } 1786 1787 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 1788 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 1789 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 1790 1791 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 1792 dsl_dataset_phys(ds)->ds_creation_time); 1793 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 1794 dsl_dataset_phys(ds)->ds_creation_txg); 1795 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 1796 ds->ds_quota); 1797 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 1798 ds->ds_reserved); 1799 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 1800 dsl_dataset_phys(ds)->ds_guid); 1801 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_FSID_GUID, 1802 dsl_dataset_phys(ds)->ds_fsid_guid); 1803 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 1804 dsl_dataset_phys(ds)->ds_unique_bytes); 1805 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 1806 ds->ds_object); 1807 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 1808 ds->ds_userrefs); 1809 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 1810 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 1811 1812 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 1813 uint64_t written, comp, uncomp; 1814 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1815 dsl_dataset_t *prev; 1816 1817 int err = dsl_dataset_hold_obj(dp, 1818 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 1819 if (err == 0) { 1820 err = dsl_dataset_space_written(prev, ds, &written, 1821 &comp, &uncomp); 1822 dsl_dataset_rele(prev, FTAG); 1823 if (err == 0) { 1824 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 1825 written); 1826 } 1827 } 1828 } 1829 1830 if (!dsl_dataset_is_snapshot(ds)) { 1831 /* 1832 * A failed "newfs" (e.g. full) resumable receive leaves 1833 * the stats set on this dataset. Check here for the prop. 1834 */ 1835 get_receive_resume_stats(ds, nv); 1836 1837 /* 1838 * A failed incremental resumable receive leaves the 1839 * stats set on our child named "%recv". Check the child 1840 * for the prop. 1841 */ 1842 char recvname[ZFS_MAXNAMELEN]; 1843 dsl_dataset_t *recv_ds; 1844 dsl_dataset_name(ds, recvname); 1845 (void) strcat(recvname, "/"); 1846 (void) strcat(recvname, recv_clone_name); 1847 if (dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) { 1848 get_receive_resume_stats(recv_ds, nv); 1849 dsl_dataset_rele(recv_ds, FTAG); 1850 } 1851 } 1852 } 1853 1854 void 1855 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 1856 { 1857 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1858 ASSERT(dsl_pool_config_held(dp)); 1859 1860 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg; 1861 stat->dds_inconsistent = 1862 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT; 1863 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid; 1864 stat->dds_origin[0] = '\0'; 1865 if (ds->ds_is_snapshot) { 1866 stat->dds_is_snapshot = B_TRUE; 1867 stat->dds_num_clones = 1868 dsl_dataset_phys(ds)->ds_num_children - 1; 1869 } else { 1870 stat->dds_is_snapshot = B_FALSE; 1871 stat->dds_num_clones = 0; 1872 1873 if (dsl_dir_is_clone(ds->ds_dir)) { 1874 dsl_dataset_t *ods; 1875 1876 VERIFY0(dsl_dataset_hold_obj(dp, 1877 dsl_dir_phys(ds->ds_dir)->dd_origin_obj, 1878 FTAG, &ods)); 1879 dsl_dataset_name(ods, stat->dds_origin); 1880 dsl_dataset_rele(ods, FTAG); 1881 } 1882 } 1883 } 1884 1885 uint64_t 1886 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 1887 { 1888 return (ds->ds_fsid_guid); 1889 } 1890 1891 void 1892 dsl_dataset_space(dsl_dataset_t *ds, 1893 uint64_t *refdbytesp, uint64_t *availbytesp, 1894 uint64_t *usedobjsp, uint64_t *availobjsp) 1895 { 1896 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; 1897 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 1898 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) 1899 *availbytesp += 1900 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 1901 if (ds->ds_quota != 0) { 1902 /* 1903 * Adjust available bytes according to refquota 1904 */ 1905 if (*refdbytesp < ds->ds_quota) 1906 *availbytesp = MIN(*availbytesp, 1907 ds->ds_quota - *refdbytesp); 1908 else 1909 *availbytesp = 0; 1910 } 1911 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); 1912 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 1913 } 1914 1915 boolean_t 1916 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 1917 { 1918 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1919 1920 ASSERT(dsl_pool_config_held(dp)); 1921 if (snap == NULL) 1922 return (B_FALSE); 1923 if (dsl_dataset_phys(ds)->ds_bp.blk_birth > 1924 dsl_dataset_phys(snap)->ds_creation_txg) { 1925 objset_t *os, *os_snap; 1926 /* 1927 * It may be that only the ZIL differs, because it was 1928 * reset in the head. Don't count that as being 1929 * modified. 1930 */ 1931 if (dmu_objset_from_ds(ds, &os) != 0) 1932 return (B_TRUE); 1933 if (dmu_objset_from_ds(snap, &os_snap) != 0) 1934 return (B_TRUE); 1935 return (bcmp(&os->os_phys->os_meta_dnode, 1936 &os_snap->os_phys->os_meta_dnode, 1937 sizeof (os->os_phys->os_meta_dnode)) != 0); 1938 } 1939 return (B_FALSE); 1940 } 1941 1942 typedef struct dsl_dataset_rename_snapshot_arg { 1943 const char *ddrsa_fsname; 1944 const char *ddrsa_oldsnapname; 1945 const char *ddrsa_newsnapname; 1946 boolean_t ddrsa_recursive; 1947 dmu_tx_t *ddrsa_tx; 1948 } dsl_dataset_rename_snapshot_arg_t; 1949 1950 /* ARGSUSED */ 1951 static int 1952 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 1953 dsl_dataset_t *hds, void *arg) 1954 { 1955 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1956 int error; 1957 uint64_t val; 1958 1959 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1960 if (error != 0) { 1961 /* ignore nonexistent snapshots */ 1962 return (error == ENOENT ? 0 : error); 1963 } 1964 1965 /* new name should not exist */ 1966 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 1967 if (error == 0) 1968 error = SET_ERROR(EEXIST); 1969 else if (error == ENOENT) 1970 error = 0; 1971 1972 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 1973 if (dsl_dir_namelen(hds->ds_dir) + 1 + 1974 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN) 1975 error = SET_ERROR(ENAMETOOLONG); 1976 1977 return (error); 1978 } 1979 1980 static int 1981 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 1982 { 1983 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1984 dsl_pool_t *dp = dmu_tx_pool(tx); 1985 dsl_dataset_t *hds; 1986 int error; 1987 1988 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 1989 if (error != 0) 1990 return (error); 1991 1992 if (ddrsa->ddrsa_recursive) { 1993 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1994 dsl_dataset_rename_snapshot_check_impl, ddrsa, 1995 DS_FIND_CHILDREN); 1996 } else { 1997 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 1998 } 1999 dsl_dataset_rele(hds, FTAG); 2000 return (error); 2001 } 2002 2003 static int 2004 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 2005 dsl_dataset_t *hds, void *arg) 2006 { 2007 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2008 dsl_dataset_t *ds; 2009 uint64_t val; 2010 dmu_tx_t *tx = ddrsa->ddrsa_tx; 2011 int error; 2012 2013 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 2014 ASSERT(error == 0 || error == ENOENT); 2015 if (error == ENOENT) { 2016 /* ignore nonexistent snapshots */ 2017 return (0); 2018 } 2019 2020 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 2021 2022 /* log before we change the name */ 2023 spa_history_log_internal_ds(ds, "rename", tx, 2024 "-> @%s", ddrsa->ddrsa_newsnapname); 2025 2026 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 2027 B_FALSE)); 2028 mutex_enter(&ds->ds_lock); 2029 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 2030 mutex_exit(&ds->ds_lock); 2031 VERIFY0(zap_add(dp->dp_meta_objset, 2032 dsl_dataset_phys(hds)->ds_snapnames_zapobj, 2033 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 2034 2035 dsl_dataset_rele(ds, FTAG); 2036 return (0); 2037 } 2038 2039 static void 2040 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 2041 { 2042 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 2043 dsl_pool_t *dp = dmu_tx_pool(tx); 2044 dsl_dataset_t *hds; 2045 2046 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 2047 ddrsa->ddrsa_tx = tx; 2048 if (ddrsa->ddrsa_recursive) { 2049 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 2050 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 2051 DS_FIND_CHILDREN)); 2052 } else { 2053 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 2054 } 2055 dsl_dataset_rele(hds, FTAG); 2056 } 2057 2058 int 2059 dsl_dataset_rename_snapshot(const char *fsname, 2060 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 2061 { 2062 dsl_dataset_rename_snapshot_arg_t ddrsa; 2063 2064 ddrsa.ddrsa_fsname = fsname; 2065 ddrsa.ddrsa_oldsnapname = oldsnapname; 2066 ddrsa.ddrsa_newsnapname = newsnapname; 2067 ddrsa.ddrsa_recursive = recursive; 2068 2069 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 2070 dsl_dataset_rename_snapshot_sync, &ddrsa, 2071 1, ZFS_SPACE_CHECK_RESERVED)); 2072 } 2073 2074 /* 2075 * If we're doing an ownership handoff, we need to make sure that there is 2076 * only one long hold on the dataset. We're not allowed to change anything here 2077 * so we don't permanently release the long hold or regular hold here. We want 2078 * to do this only when syncing to avoid the dataset unexpectedly going away 2079 * when we release the long hold. 2080 */ 2081 static int 2082 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 2083 { 2084 boolean_t held; 2085 2086 if (!dmu_tx_is_syncing(tx)) 2087 return (0); 2088 2089 if (owner != NULL) { 2090 VERIFY3P(ds->ds_owner, ==, owner); 2091 dsl_dataset_long_rele(ds, owner); 2092 } 2093 2094 held = dsl_dataset_long_held(ds); 2095 2096 if (owner != NULL) 2097 dsl_dataset_long_hold(ds, owner); 2098 2099 if (held) 2100 return (SET_ERROR(EBUSY)); 2101 2102 return (0); 2103 } 2104 2105 typedef struct dsl_dataset_rollback_arg { 2106 const char *ddra_fsname; 2107 void *ddra_owner; 2108 nvlist_t *ddra_result; 2109 } dsl_dataset_rollback_arg_t; 2110 2111 static int 2112 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 2113 { 2114 dsl_dataset_rollback_arg_t *ddra = arg; 2115 dsl_pool_t *dp = dmu_tx_pool(tx); 2116 dsl_dataset_t *ds; 2117 int64_t unused_refres_delta; 2118 int error; 2119 2120 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 2121 if (error != 0) 2122 return (error); 2123 2124 /* must not be a snapshot */ 2125 if (ds->ds_is_snapshot) { 2126 dsl_dataset_rele(ds, FTAG); 2127 return (SET_ERROR(EINVAL)); 2128 } 2129 2130 /* must have a most recent snapshot */ 2131 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { 2132 dsl_dataset_rele(ds, FTAG); 2133 return (SET_ERROR(EINVAL)); 2134 } 2135 2136 /* must not have any bookmarks after the most recent snapshot */ 2137 nvlist_t *proprequest = fnvlist_alloc(); 2138 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2139 nvlist_t *bookmarks = fnvlist_alloc(); 2140 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 2141 fnvlist_free(proprequest); 2142 if (error != 0) 2143 return (error); 2144 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 2145 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 2146 nvlist_t *valuenv = 2147 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 2148 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 2149 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 2150 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 2151 fnvlist_free(bookmarks); 2152 dsl_dataset_rele(ds, FTAG); 2153 return (SET_ERROR(EEXIST)); 2154 } 2155 } 2156 fnvlist_free(bookmarks); 2157 2158 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 2159 if (error != 0) { 2160 dsl_dataset_rele(ds, FTAG); 2161 return (error); 2162 } 2163 2164 /* 2165 * Check if the snap we are rolling back to uses more than 2166 * the refquota. 2167 */ 2168 if (ds->ds_quota != 0 && 2169 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { 2170 dsl_dataset_rele(ds, FTAG); 2171 return (SET_ERROR(EDQUOT)); 2172 } 2173 2174 /* 2175 * When we do the clone swap, we will temporarily use more space 2176 * due to the refreservation (the head will no longer have any 2177 * unique space, so the entire amount of the refreservation will need 2178 * to be free). We will immediately destroy the clone, freeing 2179 * this space, but the freeing happens over many txg's. 2180 */ 2181 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 2182 dsl_dataset_phys(ds)->ds_unique_bytes); 2183 2184 if (unused_refres_delta > 0 && 2185 unused_refres_delta > 2186 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 2187 dsl_dataset_rele(ds, FTAG); 2188 return (SET_ERROR(ENOSPC)); 2189 } 2190 2191 dsl_dataset_rele(ds, FTAG); 2192 return (0); 2193 } 2194 2195 static void 2196 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 2197 { 2198 dsl_dataset_rollback_arg_t *ddra = arg; 2199 dsl_pool_t *dp = dmu_tx_pool(tx); 2200 dsl_dataset_t *ds, *clone; 2201 uint64_t cloneobj; 2202 char namebuf[ZFS_MAXNAMELEN]; 2203 2204 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 2205 2206 dsl_dataset_name(ds->ds_prev, namebuf); 2207 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 2208 2209 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 2210 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 2211 2212 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 2213 2214 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 2215 dsl_dataset_zero_zil(ds, tx); 2216 2217 dsl_destroy_head_sync_impl(clone, tx); 2218 2219 dsl_dataset_rele(clone, FTAG); 2220 dsl_dataset_rele(ds, FTAG); 2221 } 2222 2223 /* 2224 * Rolls back the given filesystem or volume to the most recent snapshot. 2225 * The name of the most recent snapshot will be returned under key "target" 2226 * in the result nvlist. 2227 * 2228 * If owner != NULL: 2229 * - The existing dataset MUST be owned by the specified owner at entry 2230 * - Upon return, dataset will still be held by the same owner, whether we 2231 * succeed or not. 2232 * 2233 * This mode is required any time the existing filesystem is mounted. See 2234 * notes above zfs_suspend_fs() for further details. 2235 */ 2236 int 2237 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result) 2238 { 2239 dsl_dataset_rollback_arg_t ddra; 2240 2241 ddra.ddra_fsname = fsname; 2242 ddra.ddra_owner = owner; 2243 ddra.ddra_result = result; 2244 2245 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 2246 dsl_dataset_rollback_sync, &ddra, 2247 1, ZFS_SPACE_CHECK_RESERVED)); 2248 } 2249 2250 struct promotenode { 2251 list_node_t link; 2252 dsl_dataset_t *ds; 2253 }; 2254 2255 typedef struct dsl_dataset_promote_arg { 2256 const char *ddpa_clonename; 2257 dsl_dataset_t *ddpa_clone; 2258 list_t shared_snaps, origin_snaps, clone_snaps; 2259 dsl_dataset_t *origin_origin; /* origin of the origin */ 2260 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2261 char *err_ds; 2262 cred_t *cr; 2263 } dsl_dataset_promote_arg_t; 2264 2265 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2266 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2267 void *tag); 2268 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2269 2270 static int 2271 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2272 { 2273 dsl_dataset_promote_arg_t *ddpa = arg; 2274 dsl_pool_t *dp = dmu_tx_pool(tx); 2275 dsl_dataset_t *hds; 2276 struct promotenode *snap; 2277 dsl_dataset_t *origin_ds; 2278 int err; 2279 uint64_t unused; 2280 uint64_t ss_mv_cnt; 2281 size_t max_snap_len; 2282 2283 err = promote_hold(ddpa, dp, FTAG); 2284 if (err != 0) 2285 return (err); 2286 2287 hds = ddpa->ddpa_clone; 2288 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; 2289 2290 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { 2291 promote_rele(ddpa, FTAG); 2292 return (SET_ERROR(EXDEV)); 2293 } 2294 2295 /* 2296 * Compute and check the amount of space to transfer. Since this is 2297 * so expensive, don't do the preliminary check. 2298 */ 2299 if (!dmu_tx_is_syncing(tx)) { 2300 promote_rele(ddpa, FTAG); 2301 return (0); 2302 } 2303 2304 snap = list_head(&ddpa->shared_snaps); 2305 origin_ds = snap->ds; 2306 2307 /* compute origin's new unique space */ 2308 snap = list_tail(&ddpa->clone_snaps); 2309 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2310 origin_ds->ds_object); 2311 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2312 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, 2313 &ddpa->unique, &unused, &unused); 2314 2315 /* 2316 * Walk the snapshots that we are moving 2317 * 2318 * Compute space to transfer. Consider the incremental changes 2319 * to used by each snapshot: 2320 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2321 * So each snapshot gave birth to: 2322 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2323 * So a sequence would look like: 2324 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2325 * Which simplifies to: 2326 * uN + kN + kN-1 + ... + k1 + k0 2327 * Note however, if we stop before we reach the ORIGIN we get: 2328 * uN + kN + kN-1 + ... + kM - uM-1 2329 */ 2330 ss_mv_cnt = 0; 2331 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; 2332 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; 2333 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; 2334 for (snap = list_head(&ddpa->shared_snaps); snap; 2335 snap = list_next(&ddpa->shared_snaps, snap)) { 2336 uint64_t val, dlused, dlcomp, dluncomp; 2337 dsl_dataset_t *ds = snap->ds; 2338 2339 ss_mv_cnt++; 2340 2341 /* 2342 * If there are long holds, we won't be able to evict 2343 * the objset. 2344 */ 2345 if (dsl_dataset_long_held(ds)) { 2346 err = SET_ERROR(EBUSY); 2347 goto out; 2348 } 2349 2350 /* Check that the snapshot name does not conflict */ 2351 VERIFY0(dsl_dataset_get_snapname(ds)); 2352 if (strlen(ds->ds_snapname) >= max_snap_len) { 2353 err = SET_ERROR(ENAMETOOLONG); 2354 goto out; 2355 } 2356 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2357 if (err == 0) { 2358 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname); 2359 err = SET_ERROR(EEXIST); 2360 goto out; 2361 } 2362 if (err != ENOENT) 2363 goto out; 2364 2365 /* The very first snapshot does not have a deadlist */ 2366 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) 2367 continue; 2368 2369 dsl_deadlist_space(&ds->ds_deadlist, 2370 &dlused, &dlcomp, &dluncomp); 2371 ddpa->used += dlused; 2372 ddpa->comp += dlcomp; 2373 ddpa->uncomp += dluncomp; 2374 } 2375 2376 /* 2377 * If we are a clone of a clone then we never reached ORIGIN, 2378 * so we need to subtract out the clone origin's used space. 2379 */ 2380 if (ddpa->origin_origin) { 2381 ddpa->used -= 2382 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; 2383 ddpa->comp -= 2384 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; 2385 ddpa->uncomp -= 2386 dsl_dataset_phys(ddpa->origin_origin)-> 2387 ds_uncompressed_bytes; 2388 } 2389 2390 /* Check that there is enough space and limit headroom here */ 2391 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2392 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2393 if (err != 0) 2394 goto out; 2395 2396 /* 2397 * Compute the amounts of space that will be used by snapshots 2398 * after the promotion (for both origin and clone). For each, 2399 * it is the amount of space that will be on all of their 2400 * deadlists (that was not born before their new origin). 2401 */ 2402 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2403 uint64_t space; 2404 2405 /* 2406 * Note, typically this will not be a clone of a clone, 2407 * so dd_origin_txg will be < TXG_INITIAL, so 2408 * these snaplist_space() -> dsl_deadlist_space_range() 2409 * calls will be fast because they do not have to 2410 * iterate over all bps. 2411 */ 2412 snap = list_head(&ddpa->origin_snaps); 2413 err = snaplist_space(&ddpa->shared_snaps, 2414 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2415 if (err != 0) 2416 goto out; 2417 2418 err = snaplist_space(&ddpa->clone_snaps, 2419 snap->ds->ds_dir->dd_origin_txg, &space); 2420 if (err != 0) 2421 goto out; 2422 ddpa->cloneusedsnap += space; 2423 } 2424 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & 2425 DD_FLAG_USED_BREAKDOWN) { 2426 err = snaplist_space(&ddpa->origin_snaps, 2427 dsl_dataset_phys(origin_ds)->ds_creation_txg, 2428 &ddpa->originusedsnap); 2429 if (err != 0) 2430 goto out; 2431 } 2432 2433 out: 2434 promote_rele(ddpa, FTAG); 2435 return (err); 2436 } 2437 2438 static void 2439 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2440 { 2441 dsl_dataset_promote_arg_t *ddpa = arg; 2442 dsl_pool_t *dp = dmu_tx_pool(tx); 2443 dsl_dataset_t *hds; 2444 struct promotenode *snap; 2445 dsl_dataset_t *origin_ds; 2446 dsl_dataset_t *origin_head; 2447 dsl_dir_t *dd; 2448 dsl_dir_t *odd = NULL; 2449 uint64_t oldnext_obj; 2450 int64_t delta; 2451 2452 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2453 hds = ddpa->ddpa_clone; 2454 2455 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); 2456 2457 snap = list_head(&ddpa->shared_snaps); 2458 origin_ds = snap->ds; 2459 dd = hds->ds_dir; 2460 2461 snap = list_head(&ddpa->origin_snaps); 2462 origin_head = snap->ds; 2463 2464 /* 2465 * We need to explicitly open odd, since origin_ds's dd will be 2466 * changing. 2467 */ 2468 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2469 NULL, FTAG, &odd)); 2470 2471 /* change origin's next snap */ 2472 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2473 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; 2474 snap = list_tail(&ddpa->clone_snaps); 2475 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2476 origin_ds->ds_object); 2477 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; 2478 2479 /* change the origin's next clone */ 2480 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { 2481 dsl_dataset_remove_from_next_clones(origin_ds, 2482 snap->ds->ds_object, tx); 2483 VERIFY0(zap_add_int(dp->dp_meta_objset, 2484 dsl_dataset_phys(origin_ds)->ds_next_clones_obj, 2485 oldnext_obj, tx)); 2486 } 2487 2488 /* change origin */ 2489 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2490 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); 2491 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; 2492 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2493 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2494 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; 2495 origin_head->ds_dir->dd_origin_txg = 2496 dsl_dataset_phys(origin_ds)->ds_creation_txg; 2497 2498 /* change dd_clone entries */ 2499 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2500 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2501 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); 2502 VERIFY0(zap_add_int(dp->dp_meta_objset, 2503 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2504 hds->ds_object, tx)); 2505 2506 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2507 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2508 origin_head->ds_object, tx)); 2509 if (dsl_dir_phys(dd)->dd_clones == 0) { 2510 dsl_dir_phys(dd)->dd_clones = 2511 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, 2512 DMU_OT_NONE, 0, tx); 2513 } 2514 VERIFY0(zap_add_int(dp->dp_meta_objset, 2515 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); 2516 } 2517 2518 /* move snapshots to this dir */ 2519 for (snap = list_head(&ddpa->shared_snaps); snap; 2520 snap = list_next(&ddpa->shared_snaps, snap)) { 2521 dsl_dataset_t *ds = snap->ds; 2522 2523 /* 2524 * Property callbacks are registered to a particular 2525 * dsl_dir. Since ours is changing, evict the objset 2526 * so that they will be unregistered from the old dsl_dir. 2527 */ 2528 if (ds->ds_objset) { 2529 dmu_objset_evict(ds->ds_objset); 2530 ds->ds_objset = NULL; 2531 } 2532 2533 /* move snap name entry */ 2534 VERIFY0(dsl_dataset_get_snapname(ds)); 2535 VERIFY0(dsl_dataset_snap_remove(origin_head, 2536 ds->ds_snapname, tx, B_TRUE)); 2537 VERIFY0(zap_add(dp->dp_meta_objset, 2538 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 2539 8, 1, &ds->ds_object, tx)); 2540 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 2541 DD_FIELD_SNAPSHOT_COUNT, tx); 2542 2543 /* change containing dsl_dir */ 2544 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2545 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); 2546 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; 2547 ASSERT3P(ds->ds_dir, ==, odd); 2548 dsl_dir_rele(ds->ds_dir, ds); 2549 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 2550 NULL, ds, &ds->ds_dir)); 2551 2552 /* move any clone references */ 2553 if (dsl_dataset_phys(ds)->ds_next_clones_obj && 2554 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2555 zap_cursor_t zc; 2556 zap_attribute_t za; 2557 2558 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2559 dsl_dataset_phys(ds)->ds_next_clones_obj); 2560 zap_cursor_retrieve(&zc, &za) == 0; 2561 zap_cursor_advance(&zc)) { 2562 dsl_dataset_t *cnds; 2563 uint64_t o; 2564 2565 if (za.za_first_integer == oldnext_obj) { 2566 /* 2567 * We've already moved the 2568 * origin's reference. 2569 */ 2570 continue; 2571 } 2572 2573 VERIFY0(dsl_dataset_hold_obj(dp, 2574 za.za_first_integer, FTAG, &cnds)); 2575 o = dsl_dir_phys(cnds->ds_dir)-> 2576 dd_head_dataset_obj; 2577 2578 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2579 dsl_dir_phys(odd)->dd_clones, o, tx)); 2580 VERIFY0(zap_add_int(dp->dp_meta_objset, 2581 dsl_dir_phys(dd)->dd_clones, o, tx)); 2582 dsl_dataset_rele(cnds, FTAG); 2583 } 2584 zap_cursor_fini(&zc); 2585 } 2586 2587 ASSERT(!dsl_prop_hascb(ds)); 2588 } 2589 2590 /* 2591 * Change space accounting. 2592 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2593 * both be valid, or both be 0 (resulting in delta == 0). This 2594 * is true for each of {clone,origin} independently. 2595 */ 2596 2597 delta = ddpa->cloneusedsnap - 2598 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; 2599 ASSERT3S(delta, >=, 0); 2600 ASSERT3U(ddpa->used, >=, delta); 2601 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2602 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2603 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 2604 2605 delta = ddpa->originusedsnap - 2606 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; 2607 ASSERT3S(delta, <=, 0); 2608 ASSERT3U(ddpa->used, >=, -delta); 2609 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2610 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2611 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 2612 2613 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; 2614 2615 /* log history record */ 2616 spa_history_log_internal_ds(hds, "promote", tx, ""); 2617 2618 dsl_dir_rele(odd, FTAG); 2619 promote_rele(ddpa, FTAG); 2620 } 2621 2622 /* 2623 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2624 * (exclusive) and last_obj (inclusive). The list will be in reverse 2625 * order (last_obj will be the list_head()). If first_obj == 0, do all 2626 * snapshots back to this dataset's origin. 2627 */ 2628 static int 2629 snaplist_make(dsl_pool_t *dp, 2630 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 2631 { 2632 uint64_t obj = last_obj; 2633 2634 list_create(l, sizeof (struct promotenode), 2635 offsetof(struct promotenode, link)); 2636 2637 while (obj != first_obj) { 2638 dsl_dataset_t *ds; 2639 struct promotenode *snap; 2640 int err; 2641 2642 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 2643 ASSERT(err != ENOENT); 2644 if (err != 0) 2645 return (err); 2646 2647 if (first_obj == 0) 2648 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; 2649 2650 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 2651 snap->ds = ds; 2652 list_insert_tail(l, snap); 2653 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 2654 } 2655 2656 return (0); 2657 } 2658 2659 static int 2660 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2661 { 2662 struct promotenode *snap; 2663 2664 *spacep = 0; 2665 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2666 uint64_t used, comp, uncomp; 2667 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2668 mintxg, UINT64_MAX, &used, &comp, &uncomp); 2669 *spacep += used; 2670 } 2671 return (0); 2672 } 2673 2674 static void 2675 snaplist_destroy(list_t *l, void *tag) 2676 { 2677 struct promotenode *snap; 2678 2679 if (l == NULL || !list_link_active(&l->list_head)) 2680 return; 2681 2682 while ((snap = list_tail(l)) != NULL) { 2683 list_remove(l, snap); 2684 dsl_dataset_rele(snap->ds, tag); 2685 kmem_free(snap, sizeof (*snap)); 2686 } 2687 list_destroy(l); 2688 } 2689 2690 static int 2691 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 2692 { 2693 int error; 2694 dsl_dir_t *dd; 2695 struct promotenode *snap; 2696 2697 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 2698 &ddpa->ddpa_clone); 2699 if (error != 0) 2700 return (error); 2701 dd = ddpa->ddpa_clone->ds_dir; 2702 2703 if (ddpa->ddpa_clone->ds_is_snapshot || 2704 !dsl_dir_is_clone(dd)) { 2705 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2706 return (SET_ERROR(EINVAL)); 2707 } 2708 2709 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, 2710 &ddpa->shared_snaps, tag); 2711 if (error != 0) 2712 goto out; 2713 2714 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 2715 &ddpa->clone_snaps, tag); 2716 if (error != 0) 2717 goto out; 2718 2719 snap = list_head(&ddpa->shared_snaps); 2720 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); 2721 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, 2722 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, 2723 &ddpa->origin_snaps, tag); 2724 if (error != 0) 2725 goto out; 2726 2727 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { 2728 error = dsl_dataset_hold_obj(dp, 2729 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, 2730 tag, &ddpa->origin_origin); 2731 if (error != 0) 2732 goto out; 2733 } 2734 out: 2735 if (error != 0) 2736 promote_rele(ddpa, tag); 2737 return (error); 2738 } 2739 2740 static void 2741 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 2742 { 2743 snaplist_destroy(&ddpa->shared_snaps, tag); 2744 snaplist_destroy(&ddpa->clone_snaps, tag); 2745 snaplist_destroy(&ddpa->origin_snaps, tag); 2746 if (ddpa->origin_origin != NULL) 2747 dsl_dataset_rele(ddpa->origin_origin, tag); 2748 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2749 } 2750 2751 /* 2752 * Promote a clone. 2753 * 2754 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 2755 * in with the name. (It must be at least MAXNAMELEN bytes long.) 2756 */ 2757 int 2758 dsl_dataset_promote(const char *name, char *conflsnap) 2759 { 2760 dsl_dataset_promote_arg_t ddpa = { 0 }; 2761 uint64_t numsnaps; 2762 int error; 2763 objset_t *os; 2764 2765 /* 2766 * We will modify space proportional to the number of 2767 * snapshots. Compute numsnaps. 2768 */ 2769 error = dmu_objset_hold(name, FTAG, &os); 2770 if (error != 0) 2771 return (error); 2772 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 2773 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, 2774 &numsnaps); 2775 dmu_objset_rele(os, FTAG); 2776 if (error != 0) 2777 return (error); 2778 2779 ddpa.ddpa_clonename = name; 2780 ddpa.err_ds = conflsnap; 2781 ddpa.cr = CRED(); 2782 2783 return (dsl_sync_task(name, dsl_dataset_promote_check, 2784 dsl_dataset_promote_sync, &ddpa, 2785 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED)); 2786 } 2787 2788 int 2789 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 2790 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 2791 { 2792 int64_t unused_refres_delta; 2793 2794 /* they should both be heads */ 2795 if (clone->ds_is_snapshot || 2796 origin_head->ds_is_snapshot) 2797 return (SET_ERROR(EINVAL)); 2798 2799 /* if we are not forcing, the branch point should be just before them */ 2800 if (!force && clone->ds_prev != origin_head->ds_prev) 2801 return (SET_ERROR(EINVAL)); 2802 2803 /* clone should be the clone (unless they are unrelated) */ 2804 if (clone->ds_prev != NULL && 2805 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 2806 origin_head->ds_dir != clone->ds_prev->ds_dir) 2807 return (SET_ERROR(EINVAL)); 2808 2809 /* the clone should be a child of the origin */ 2810 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 2811 return (SET_ERROR(EINVAL)); 2812 2813 /* origin_head shouldn't be modified unless 'force' */ 2814 if (!force && 2815 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 2816 return (SET_ERROR(ETXTBSY)); 2817 2818 /* origin_head should have no long holds (e.g. is not mounted) */ 2819 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 2820 return (SET_ERROR(EBUSY)); 2821 2822 /* check amount of any unconsumed refreservation */ 2823 unused_refres_delta = 2824 (int64_t)MIN(origin_head->ds_reserved, 2825 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 2826 (int64_t)MIN(origin_head->ds_reserved, 2827 dsl_dataset_phys(clone)->ds_unique_bytes); 2828 2829 if (unused_refres_delta > 0 && 2830 unused_refres_delta > 2831 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 2832 return (SET_ERROR(ENOSPC)); 2833 2834 /* clone can't be over the head's refquota */ 2835 if (origin_head->ds_quota != 0 && 2836 dsl_dataset_phys(clone)->ds_referenced_bytes > 2837 origin_head->ds_quota) 2838 return (SET_ERROR(EDQUOT)); 2839 2840 return (0); 2841 } 2842 2843 void 2844 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 2845 dsl_dataset_t *origin_head, dmu_tx_t *tx) 2846 { 2847 dsl_pool_t *dp = dmu_tx_pool(tx); 2848 int64_t unused_refres_delta; 2849 2850 ASSERT(clone->ds_reserved == 0); 2851 ASSERT(origin_head->ds_quota == 0 || 2852 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota); 2853 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 2854 2855 /* 2856 * Swap per-dataset feature flags. 2857 */ 2858 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) { 2859 if (!(spa_feature_table[f].fi_flags & 2860 ZFEATURE_FLAG_PER_DATASET)) { 2861 ASSERT(!clone->ds_feature_inuse[f]); 2862 ASSERT(!origin_head->ds_feature_inuse[f]); 2863 continue; 2864 } 2865 2866 boolean_t clone_inuse = clone->ds_feature_inuse[f]; 2867 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f]; 2868 2869 if (clone_inuse) { 2870 dsl_dataset_deactivate_feature(clone->ds_object, f, tx); 2871 clone->ds_feature_inuse[f] = B_FALSE; 2872 } 2873 if (origin_head_inuse) { 2874 dsl_dataset_deactivate_feature(origin_head->ds_object, 2875 f, tx); 2876 origin_head->ds_feature_inuse[f] = B_FALSE; 2877 } 2878 if (clone_inuse) { 2879 dsl_dataset_activate_feature(origin_head->ds_object, 2880 f, tx); 2881 origin_head->ds_feature_inuse[f] = B_TRUE; 2882 } 2883 if (origin_head_inuse) { 2884 dsl_dataset_activate_feature(clone->ds_object, f, tx); 2885 clone->ds_feature_inuse[f] = B_TRUE; 2886 } 2887 } 2888 2889 dmu_buf_will_dirty(clone->ds_dbuf, tx); 2890 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 2891 2892 if (clone->ds_objset != NULL) { 2893 dmu_objset_evict(clone->ds_objset); 2894 clone->ds_objset = NULL; 2895 } 2896 2897 if (origin_head->ds_objset != NULL) { 2898 dmu_objset_evict(origin_head->ds_objset); 2899 origin_head->ds_objset = NULL; 2900 } 2901 2902 unused_refres_delta = 2903 (int64_t)MIN(origin_head->ds_reserved, 2904 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 2905 (int64_t)MIN(origin_head->ds_reserved, 2906 dsl_dataset_phys(clone)->ds_unique_bytes); 2907 2908 /* 2909 * Reset origin's unique bytes, if it exists. 2910 */ 2911 if (clone->ds_prev) { 2912 dsl_dataset_t *origin = clone->ds_prev; 2913 uint64_t comp, uncomp; 2914 2915 dmu_buf_will_dirty(origin->ds_dbuf, tx); 2916 dsl_deadlist_space_range(&clone->ds_deadlist, 2917 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, 2918 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); 2919 } 2920 2921 /* swap blkptrs */ 2922 { 2923 blkptr_t tmp; 2924 tmp = dsl_dataset_phys(origin_head)->ds_bp; 2925 dsl_dataset_phys(origin_head)->ds_bp = 2926 dsl_dataset_phys(clone)->ds_bp; 2927 dsl_dataset_phys(clone)->ds_bp = tmp; 2928 } 2929 2930 /* set dd_*_bytes */ 2931 { 2932 int64_t dused, dcomp, duncomp; 2933 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2934 uint64_t odl_used, odl_comp, odl_uncomp; 2935 2936 ASSERT3U(dsl_dir_phys(clone->ds_dir)-> 2937 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2938 2939 dsl_deadlist_space(&clone->ds_deadlist, 2940 &cdl_used, &cdl_comp, &cdl_uncomp); 2941 dsl_deadlist_space(&origin_head->ds_deadlist, 2942 &odl_used, &odl_comp, &odl_uncomp); 2943 2944 dused = dsl_dataset_phys(clone)->ds_referenced_bytes + 2945 cdl_used - 2946 (dsl_dataset_phys(origin_head)->ds_referenced_bytes + 2947 odl_used); 2948 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + 2949 cdl_comp - 2950 (dsl_dataset_phys(origin_head)->ds_compressed_bytes + 2951 odl_comp); 2952 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + 2953 cdl_uncomp - 2954 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + 2955 odl_uncomp); 2956 2957 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 2958 dused, dcomp, duncomp, tx); 2959 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 2960 -dused, -dcomp, -duncomp, tx); 2961 2962 /* 2963 * The difference in the space used by snapshots is the 2964 * difference in snapshot space due to the head's 2965 * deadlist (since that's the only thing that's 2966 * changing that affects the snapused). 2967 */ 2968 dsl_deadlist_space_range(&clone->ds_deadlist, 2969 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2970 &cdl_used, &cdl_comp, &cdl_uncomp); 2971 dsl_deadlist_space_range(&origin_head->ds_deadlist, 2972 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2973 &odl_used, &odl_comp, &odl_uncomp); 2974 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 2975 DD_USED_HEAD, DD_USED_SNAP, tx); 2976 } 2977 2978 /* swap ds_*_bytes */ 2979 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, 2980 dsl_dataset_phys(clone)->ds_referenced_bytes); 2981 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, 2982 dsl_dataset_phys(clone)->ds_compressed_bytes); 2983 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, 2984 dsl_dataset_phys(clone)->ds_uncompressed_bytes); 2985 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, 2986 dsl_dataset_phys(clone)->ds_unique_bytes); 2987 2988 /* apply any parent delta for change in unconsumed refreservation */ 2989 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 2990 unused_refres_delta, 0, 0, tx); 2991 2992 /* 2993 * Swap deadlists. 2994 */ 2995 dsl_deadlist_close(&clone->ds_deadlist); 2996 dsl_deadlist_close(&origin_head->ds_deadlist); 2997 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, 2998 dsl_dataset_phys(clone)->ds_deadlist_obj); 2999 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 3000 dsl_dataset_phys(clone)->ds_deadlist_obj); 3001 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 3002 dsl_dataset_phys(origin_head)->ds_deadlist_obj); 3003 3004 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 3005 3006 spa_history_log_internal_ds(clone, "clone swap", tx, 3007 "parent=%s", origin_head->ds_dir->dd_myname); 3008 } 3009 3010 /* 3011 * Given a pool name and a dataset object number in that pool, 3012 * return the name of that dataset. 3013 */ 3014 int 3015 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 3016 { 3017 dsl_pool_t *dp; 3018 dsl_dataset_t *ds; 3019 int error; 3020 3021 error = dsl_pool_hold(pname, FTAG, &dp); 3022 if (error != 0) 3023 return (error); 3024 3025 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 3026 if (error == 0) { 3027 dsl_dataset_name(ds, buf); 3028 dsl_dataset_rele(ds, FTAG); 3029 } 3030 dsl_pool_rele(dp, FTAG); 3031 3032 return (error); 3033 } 3034 3035 int 3036 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 3037 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 3038 { 3039 int error = 0; 3040 3041 ASSERT3S(asize, >, 0); 3042 3043 /* 3044 * *ref_rsrv is the portion of asize that will come from any 3045 * unconsumed refreservation space. 3046 */ 3047 *ref_rsrv = 0; 3048 3049 mutex_enter(&ds->ds_lock); 3050 /* 3051 * Make a space adjustment for reserved bytes. 3052 */ 3053 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 3054 ASSERT3U(*used, >=, 3055 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 3056 *used -= 3057 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 3058 *ref_rsrv = 3059 asize - MIN(asize, parent_delta(ds, asize + inflight)); 3060 } 3061 3062 if (!check_quota || ds->ds_quota == 0) { 3063 mutex_exit(&ds->ds_lock); 3064 return (0); 3065 } 3066 /* 3067 * If they are requesting more space, and our current estimate 3068 * is over quota, they get to try again unless the actual 3069 * on-disk is over quota and there are no pending changes (which 3070 * may free up space for us). 3071 */ 3072 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= 3073 ds->ds_quota) { 3074 if (inflight > 0 || 3075 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) 3076 error = SET_ERROR(ERESTART); 3077 else 3078 error = SET_ERROR(EDQUOT); 3079 } 3080 mutex_exit(&ds->ds_lock); 3081 3082 return (error); 3083 } 3084 3085 typedef struct dsl_dataset_set_qr_arg { 3086 const char *ddsqra_name; 3087 zprop_source_t ddsqra_source; 3088 uint64_t ddsqra_value; 3089 } dsl_dataset_set_qr_arg_t; 3090 3091 3092 /* ARGSUSED */ 3093 static int 3094 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 3095 { 3096 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3097 dsl_pool_t *dp = dmu_tx_pool(tx); 3098 dsl_dataset_t *ds; 3099 int error; 3100 uint64_t newval; 3101 3102 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 3103 return (SET_ERROR(ENOTSUP)); 3104 3105 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3106 if (error != 0) 3107 return (error); 3108 3109 if (ds->ds_is_snapshot) { 3110 dsl_dataset_rele(ds, FTAG); 3111 return (SET_ERROR(EINVAL)); 3112 } 3113 3114 error = dsl_prop_predict(ds->ds_dir, 3115 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3116 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3117 if (error != 0) { 3118 dsl_dataset_rele(ds, FTAG); 3119 return (error); 3120 } 3121 3122 if (newval == 0) { 3123 dsl_dataset_rele(ds, FTAG); 3124 return (0); 3125 } 3126 3127 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || 3128 newval < ds->ds_reserved) { 3129 dsl_dataset_rele(ds, FTAG); 3130 return (SET_ERROR(ENOSPC)); 3131 } 3132 3133 dsl_dataset_rele(ds, FTAG); 3134 return (0); 3135 } 3136 3137 static void 3138 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 3139 { 3140 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3141 dsl_pool_t *dp = dmu_tx_pool(tx); 3142 dsl_dataset_t *ds; 3143 uint64_t newval; 3144 3145 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3146 3147 dsl_prop_set_sync_impl(ds, 3148 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 3149 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 3150 &ddsqra->ddsqra_value, tx); 3151 3152 VERIFY0(dsl_prop_get_int_ds(ds, 3153 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 3154 3155 if (ds->ds_quota != newval) { 3156 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3157 ds->ds_quota = newval; 3158 } 3159 dsl_dataset_rele(ds, FTAG); 3160 } 3161 3162 int 3163 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 3164 uint64_t refquota) 3165 { 3166 dsl_dataset_set_qr_arg_t ddsqra; 3167 3168 ddsqra.ddsqra_name = dsname; 3169 ddsqra.ddsqra_source = source; 3170 ddsqra.ddsqra_value = refquota; 3171 3172 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 3173 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 3174 } 3175 3176 static int 3177 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 3178 { 3179 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3180 dsl_pool_t *dp = dmu_tx_pool(tx); 3181 dsl_dataset_t *ds; 3182 int error; 3183 uint64_t newval, unique; 3184 3185 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 3186 return (SET_ERROR(ENOTSUP)); 3187 3188 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 3189 if (error != 0) 3190 return (error); 3191 3192 if (ds->ds_is_snapshot) { 3193 dsl_dataset_rele(ds, FTAG); 3194 return (SET_ERROR(EINVAL)); 3195 } 3196 3197 error = dsl_prop_predict(ds->ds_dir, 3198 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3199 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 3200 if (error != 0) { 3201 dsl_dataset_rele(ds, FTAG); 3202 return (error); 3203 } 3204 3205 /* 3206 * If we are doing the preliminary check in open context, the 3207 * space estimates may be inaccurate. 3208 */ 3209 if (!dmu_tx_is_syncing(tx)) { 3210 dsl_dataset_rele(ds, FTAG); 3211 return (0); 3212 } 3213 3214 mutex_enter(&ds->ds_lock); 3215 if (!DS_UNIQUE_IS_ACCURATE(ds)) 3216 dsl_dataset_recalc_head_uniq(ds); 3217 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3218 mutex_exit(&ds->ds_lock); 3219 3220 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 3221 uint64_t delta = MAX(unique, newval) - 3222 MAX(unique, ds->ds_reserved); 3223 3224 if (delta > 3225 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 3226 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 3227 dsl_dataset_rele(ds, FTAG); 3228 return (SET_ERROR(ENOSPC)); 3229 } 3230 } 3231 3232 dsl_dataset_rele(ds, FTAG); 3233 return (0); 3234 } 3235 3236 void 3237 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 3238 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 3239 { 3240 uint64_t newval; 3241 uint64_t unique; 3242 int64_t delta; 3243 3244 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3245 source, sizeof (value), 1, &value, tx); 3246 3247 VERIFY0(dsl_prop_get_int_ds(ds, 3248 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 3249 3250 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3251 mutex_enter(&ds->ds_dir->dd_lock); 3252 mutex_enter(&ds->ds_lock); 3253 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3254 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3255 delta = MAX(0, (int64_t)(newval - unique)) - 3256 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3257 ds->ds_reserved = newval; 3258 mutex_exit(&ds->ds_lock); 3259 3260 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3261 mutex_exit(&ds->ds_dir->dd_lock); 3262 } 3263 3264 static void 3265 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 3266 { 3267 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3268 dsl_pool_t *dp = dmu_tx_pool(tx); 3269 dsl_dataset_t *ds; 3270 3271 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3272 dsl_dataset_set_refreservation_sync_impl(ds, 3273 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 3274 dsl_dataset_rele(ds, FTAG); 3275 } 3276 3277 int 3278 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 3279 uint64_t refreservation) 3280 { 3281 dsl_dataset_set_qr_arg_t ddsqra; 3282 3283 ddsqra.ddsqra_name = dsname; 3284 ddsqra.ddsqra_source = source; 3285 ddsqra.ddsqra_value = refreservation; 3286 3287 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 3288 dsl_dataset_set_refreservation_sync, &ddsqra, 3289 0, ZFS_SPACE_CHECK_NONE)); 3290 } 3291 3292 /* 3293 * Return (in *usedp) the amount of space written in new that is not 3294 * present in oldsnap. New may be a snapshot or the head. Old must be 3295 * a snapshot before new, in new's filesystem (or its origin). If not then 3296 * fail and return EINVAL. 3297 * 3298 * The written space is calculated by considering two components: First, we 3299 * ignore any freed space, and calculate the written as new's used space 3300 * minus old's used space. Next, we add in the amount of space that was freed 3301 * between the two snapshots, thus reducing new's used space relative to old's. 3302 * Specifically, this is the space that was born before old->ds_creation_txg, 3303 * and freed before new (ie. on new's deadlist or a previous deadlist). 3304 * 3305 * space freed [---------------------] 3306 * snapshots ---O-------O--------O-------O------ 3307 * oldsnap new 3308 */ 3309 int 3310 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 3311 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3312 { 3313 int err = 0; 3314 uint64_t snapobj; 3315 dsl_pool_t *dp = new->ds_dir->dd_pool; 3316 3317 ASSERT(dsl_pool_config_held(dp)); 3318 3319 *usedp = 0; 3320 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; 3321 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes; 3322 3323 *compp = 0; 3324 *compp += dsl_dataset_phys(new)->ds_compressed_bytes; 3325 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes; 3326 3327 *uncompp = 0; 3328 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; 3329 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes; 3330 3331 snapobj = new->ds_object; 3332 while (snapobj != oldsnap->ds_object) { 3333 dsl_dataset_t *snap; 3334 uint64_t used, comp, uncomp; 3335 3336 if (snapobj == new->ds_object) { 3337 snap = new; 3338 } else { 3339 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3340 if (err != 0) 3341 break; 3342 } 3343 3344 if (dsl_dataset_phys(snap)->ds_prev_snap_txg == 3345 dsl_dataset_phys(oldsnap)->ds_creation_txg) { 3346 /* 3347 * The blocks in the deadlist can not be born after 3348 * ds_prev_snap_txg, so get the whole deadlist space, 3349 * which is more efficient (especially for old-format 3350 * deadlists). Unfortunately the deadlist code 3351 * doesn't have enough information to make this 3352 * optimization itself. 3353 */ 3354 dsl_deadlist_space(&snap->ds_deadlist, 3355 &used, &comp, &uncomp); 3356 } else { 3357 dsl_deadlist_space_range(&snap->ds_deadlist, 3358 0, dsl_dataset_phys(oldsnap)->ds_creation_txg, 3359 &used, &comp, &uncomp); 3360 } 3361 *usedp += used; 3362 *compp += comp; 3363 *uncompp += uncomp; 3364 3365 /* 3366 * If we get to the beginning of the chain of snapshots 3367 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3368 * was not a snapshot of/before new. 3369 */ 3370 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 3371 if (snap != new) 3372 dsl_dataset_rele(snap, FTAG); 3373 if (snapobj == 0) { 3374 err = SET_ERROR(EINVAL); 3375 break; 3376 } 3377 3378 } 3379 return (err); 3380 } 3381 3382 /* 3383 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3384 * lastsnap, and all snapshots in between are deleted. 3385 * 3386 * blocks that would be freed [---------------------------] 3387 * snapshots ---O-------O--------O-------O--------O 3388 * firstsnap lastsnap 3389 * 3390 * This is the set of blocks that were born after the snap before firstsnap, 3391 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3392 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3393 * We calculate this by iterating over the relevant deadlists (from the snap 3394 * after lastsnap, backward to the snap after firstsnap), summing up the 3395 * space on the deadlist that was born after the snap before firstsnap. 3396 */ 3397 int 3398 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3399 dsl_dataset_t *lastsnap, 3400 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3401 { 3402 int err = 0; 3403 uint64_t snapobj; 3404 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3405 3406 ASSERT(firstsnap->ds_is_snapshot); 3407 ASSERT(lastsnap->ds_is_snapshot); 3408 3409 /* 3410 * Check that the snapshots are in the same dsl_dir, and firstsnap 3411 * is before lastsnap. 3412 */ 3413 if (firstsnap->ds_dir != lastsnap->ds_dir || 3414 dsl_dataset_phys(firstsnap)->ds_creation_txg > 3415 dsl_dataset_phys(lastsnap)->ds_creation_txg) 3416 return (SET_ERROR(EINVAL)); 3417 3418 *usedp = *compp = *uncompp = 0; 3419 3420 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; 3421 while (snapobj != firstsnap->ds_object) { 3422 dsl_dataset_t *ds; 3423 uint64_t used, comp, uncomp; 3424 3425 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3426 if (err != 0) 3427 break; 3428 3429 dsl_deadlist_space_range(&ds->ds_deadlist, 3430 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, 3431 &used, &comp, &uncomp); 3432 *usedp += used; 3433 *compp += comp; 3434 *uncompp += uncomp; 3435 3436 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3437 ASSERT3U(snapobj, !=, 0); 3438 dsl_dataset_rele(ds, FTAG); 3439 } 3440 return (err); 3441 } 3442 3443 /* 3444 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 3445 * For example, they could both be snapshots of the same filesystem, and 3446 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 3447 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 3448 * filesystem. Or 'earlier' could be the origin's origin. 3449 * 3450 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 3451 */ 3452 boolean_t 3453 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 3454 uint64_t earlier_txg) 3455 { 3456 dsl_pool_t *dp = later->ds_dir->dd_pool; 3457 int error; 3458 boolean_t ret; 3459 3460 ASSERT(dsl_pool_config_held(dp)); 3461 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); 3462 3463 if (earlier_txg == 0) 3464 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; 3465 3466 if (later->ds_is_snapshot && 3467 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) 3468 return (B_FALSE); 3469 3470 if (later->ds_dir == earlier->ds_dir) 3471 return (B_TRUE); 3472 if (!dsl_dir_is_clone(later->ds_dir)) 3473 return (B_FALSE); 3474 3475 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object) 3476 return (B_TRUE); 3477 dsl_dataset_t *origin; 3478 error = dsl_dataset_hold_obj(dp, 3479 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); 3480 if (error != 0) 3481 return (B_FALSE); 3482 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 3483 dsl_dataset_rele(origin, FTAG); 3484 return (ret); 3485 } 3486 3487 void 3488 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 3489 { 3490 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3491 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 3492 } 3493 3494 boolean_t 3495 dsl_dataset_is_zapified(dsl_dataset_t *ds) 3496 { 3497 dmu_object_info_t doi; 3498 3499 dmu_object_info_from_db(ds->ds_dbuf, &doi); 3500 return (doi.doi_type == DMU_OTN_ZAP_METADATA); 3501 } 3502 3503 boolean_t 3504 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds) 3505 { 3506 return (dsl_dataset_is_zapified(ds) && 3507 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset, 3508 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0); 3509 } 3510 3511 typedef struct dsl_dataset_set_fsid_guid_arg { 3512 const char *ddsfg_name; 3513 uint64_t ddsfg_value; 3514 } dsl_dataset_set_fsid_guid_arg_t; 3515 3516 static int 3517 dsl_dataset_set_fsid_guid_check(void *arg, dmu_tx_t *tx) 3518 { 3519 int err = 0; 3520 dsl_dataset_set_fsid_guid_arg_t *ddsfg = arg; 3521 dsl_pool_t *dp = dmu_tx_pool(tx); 3522 dsl_dataset_t *ds; 3523 uint64_t newval; 3524 3525 newval = ddsfg->ddsfg_value; 3526 3527 err = dsl_dataset_hold(dp, ddsfg->ddsfg_name, FTAG, &ds); 3528 if (err != 0) 3529 return (err); 3530 3531 /* Check if the FS is mounted */ 3532 if (ds->ds_owner != NULL) { 3533 dsl_dataset_rele(ds, FTAG); 3534 return (SET_ERROR(EBUSY)); 3535 } 3536 dsl_dataset_rele(ds, FTAG); 3537 3538 if (!unique_valid(newval)) 3539 return (SET_ERROR(EEXIST)); 3540 3541 return (err); 3542 } 3543 3544 static void 3545 dsl_dataset_set_fsid_guid_sync(void *arg, dmu_tx_t *tx) 3546 { 3547 dsl_dataset_set_fsid_guid_arg_t *ddsfg = arg; 3548 dsl_pool_t *dp = dmu_tx_pool(tx); 3549 dsl_dataset_t *ds; 3550 uint64_t newval; 3551 3552 VERIFY0(dsl_dataset_hold(dp, ddsfg->ddsfg_name, FTAG, &ds)); 3553 3554 newval = ddsfg->ddsfg_value; 3555 3556 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3557 mutex_enter(&ds->ds_lock); 3558 ds->ds_fsid_guid = newval; 3559 dsl_dataset_phys(ds)->ds_fsid_guid = newval; 3560 mutex_exit(&ds->ds_lock); 3561 3562 dsl_dataset_rele(ds, FTAG); 3563 } 3564 3565 int 3566 dsl_dataset_set_fsid_guid(const char *ddname, zprop_source_t source, 3567 uint64_t fsid_guid) 3568 { 3569 dsl_dataset_set_fsid_guid_arg_t ddsfg; 3570 3571 ddsfg.ddsfg_name = ddname; 3572 ddsfg.ddsfg_value = fsid_guid; 3573 3574 return (dsl_sync_task(ddname, dsl_dataset_set_fsid_guid_check, 3575 dsl_dataset_set_fsid_guid_sync, &ddsfg, 0, 3576 ZFS_SPACE_CHECK_RESERVED)); 3577 }