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