1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 24 * Copyright (c) 2014, Joyent, Inc. All rights reserved. 25 * Copyright (c) 2014 RackTop Systems. 26 * 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 54 /* 55 * The SPA supports block sizes up to 16MB. However, very large blocks 56 * can have an impact on i/o latency (e.g. tying up a spinning disk for 57 * ~300ms), and also potentially on the memory allocator. Therefore, 58 * we do not allow the recordsize to be set larger than zfs_max_recordsize 59 * (default 1MB). Larger blocks can be created by changing this tunable, 60 * and pools with larger blocks can always be imported and used, regardless 61 * of this setting. 62 */ 63 int zfs_max_recordsize = 1 * 1024 * 1024; 64 65 #define SWITCH64(x, y) \ 66 { \ 67 uint64_t __tmp = (x); \ 68 (x) = (y); \ 69 (y) = __tmp; \ 70 } 71 72 #define DS_REF_MAX (1ULL << 62) 73 74 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds); 75 76 /* 77 * Figure out how much of this delta should be propogated to the dsl_dir 78 * layer. If there's a refreservation, that space has already been 79 * partially accounted for in our ancestors. 80 */ 81 static int64_t 82 parent_delta(dsl_dataset_t *ds, int64_t delta) 83 { 84 dsl_dataset_phys_t *ds_phys; 85 uint64_t old_bytes, new_bytes; 86 87 if (ds->ds_reserved == 0) 88 return (delta); 89 90 ds_phys = dsl_dataset_phys(ds); 91 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved); 92 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 93 94 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 95 return (new_bytes - old_bytes); 96 } 97 98 void 99 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 100 { 101 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 102 int compressed = BP_GET_PSIZE(bp); 103 int uncompressed = BP_GET_UCSIZE(bp); 104 int64_t delta; 105 106 dprintf_bp(bp, "ds=%p", ds); 107 108 ASSERT(dmu_tx_is_syncing(tx)); 109 /* It could have been compressed away to nothing */ 110 if (BP_IS_HOLE(bp)) 111 return; 112 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 113 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 114 if (ds == NULL) { 115 dsl_pool_mos_diduse_space(tx->tx_pool, 116 used, compressed, uncompressed); 117 return; 118 } 119 120 dmu_buf_will_dirty(ds->ds_dbuf, tx); 121 mutex_enter(&ds->ds_lock); 122 delta = parent_delta(ds, used); 123 dsl_dataset_phys(ds)->ds_referenced_bytes += used; 124 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed; 125 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed; 126 dsl_dataset_phys(ds)->ds_unique_bytes += used; 127 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) 128 ds->ds_need_large_blocks = B_TRUE; 129 mutex_exit(&ds->ds_lock); 130 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 131 compressed, uncompressed, tx); 132 dsl_dir_transfer_space(ds->ds_dir, used - delta, 133 DD_USED_REFRSRV, DD_USED_HEAD, tx); 134 } 135 136 int 137 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 138 boolean_t async) 139 { 140 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 141 int compressed = BP_GET_PSIZE(bp); 142 int uncompressed = BP_GET_UCSIZE(bp); 143 144 if (BP_IS_HOLE(bp)) 145 return (0); 146 147 ASSERT(dmu_tx_is_syncing(tx)); 148 ASSERT(bp->blk_birth <= tx->tx_txg); 149 150 if (ds == NULL) { 151 dsl_free(tx->tx_pool, tx->tx_txg, bp); 152 dsl_pool_mos_diduse_space(tx->tx_pool, 153 -used, -compressed, -uncompressed); 154 return (used); 155 } 156 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 157 158 ASSERT(!ds->ds_is_snapshot); 159 dmu_buf_will_dirty(ds->ds_dbuf, tx); 160 161 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 162 int64_t delta; 163 164 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 165 dsl_free(tx->tx_pool, tx->tx_txg, bp); 166 167 mutex_enter(&ds->ds_lock); 168 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used || 169 !DS_UNIQUE_IS_ACCURATE(ds)); 170 delta = parent_delta(ds, -used); 171 dsl_dataset_phys(ds)->ds_unique_bytes -= used; 172 mutex_exit(&ds->ds_lock); 173 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 174 delta, -compressed, -uncompressed, tx); 175 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 176 DD_USED_REFRSRV, DD_USED_HEAD, tx); 177 } else { 178 dprintf_bp(bp, "putting on dead list: %s", ""); 179 if (async) { 180 /* 181 * We are here as part of zio's write done callback, 182 * which means we're a zio interrupt thread. We can't 183 * call dsl_deadlist_insert() now because it may block 184 * waiting for I/O. Instead, put bp on the deferred 185 * queue and let dsl_pool_sync() finish the job. 186 */ 187 bplist_append(&ds->ds_pending_deadlist, bp); 188 } else { 189 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); 190 } 191 ASSERT3U(ds->ds_prev->ds_object, ==, 192 dsl_dataset_phys(ds)->ds_prev_snap_obj); 193 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0); 194 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 195 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 196 ds->ds_object && bp->blk_birth > 197 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) { 198 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 199 mutex_enter(&ds->ds_prev->ds_lock); 200 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used; 201 mutex_exit(&ds->ds_prev->ds_lock); 202 } 203 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 204 dsl_dir_transfer_space(ds->ds_dir, used, 205 DD_USED_HEAD, DD_USED_SNAP, tx); 206 } 207 } 208 mutex_enter(&ds->ds_lock); 209 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used); 210 dsl_dataset_phys(ds)->ds_referenced_bytes -= used; 211 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed); 212 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed; 213 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed); 214 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed; 215 mutex_exit(&ds->ds_lock); 216 217 return (used); 218 } 219 220 uint64_t 221 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 222 { 223 uint64_t trysnap = 0; 224 225 if (ds == NULL) 226 return (0); 227 /* 228 * The snapshot creation could fail, but that would cause an 229 * incorrect FALSE return, which would only result in an 230 * overestimation of the amount of space that an operation would 231 * consume, which is OK. 232 * 233 * There's also a small window where we could miss a pending 234 * snapshot, because we could set the sync task in the quiescing 235 * phase. So this should only be used as a guess. 236 */ 237 if (ds->ds_trysnap_txg > 238 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 239 trysnap = ds->ds_trysnap_txg; 240 return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap)); 241 } 242 243 boolean_t 244 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp, 245 uint64_t blk_birth) 246 { 247 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) || 248 (bp != NULL && BP_IS_HOLE(bp))) 249 return (B_FALSE); 250 251 ddt_prefetch(dsl_dataset_get_spa(ds), bp); 252 253 return (B_TRUE); 254 } 255 256 static void 257 dsl_dataset_evict(void *dbu) 258 { 259 dsl_dataset_t *ds = dbu; 260 261 ASSERT(ds->ds_owner == NULL); 262 263 ds->ds_dbuf = NULL; 264 265 unique_remove(ds->ds_fsid_guid); 266 267 if (ds->ds_objset != NULL) 268 dmu_objset_evict(ds->ds_objset); 269 270 if (ds->ds_prev) { 271 dsl_dataset_rele(ds->ds_prev, ds); 272 ds->ds_prev = NULL; 273 } 274 275 bplist_destroy(&ds->ds_pending_deadlist); 276 if (ds->ds_deadlist.dl_os != NULL) 277 dsl_deadlist_close(&ds->ds_deadlist); 278 if (ds->ds_dir) 279 dsl_dir_async_rele(ds->ds_dir, ds); 280 281 ASSERT(!list_link_active(&ds->ds_synced_link)); 282 283 mutex_destroy(&ds->ds_lock); 284 mutex_destroy(&ds->ds_opening_lock); 285 mutex_destroy(&ds->ds_sendstream_lock); 286 refcount_destroy(&ds->ds_longholds); 287 288 kmem_free(ds, sizeof (dsl_dataset_t)); 289 } 290 291 int 292 dsl_dataset_get_snapname(dsl_dataset_t *ds) 293 { 294 dsl_dataset_phys_t *headphys; 295 int err; 296 dmu_buf_t *headdbuf; 297 dsl_pool_t *dp = ds->ds_dir->dd_pool; 298 objset_t *mos = dp->dp_meta_objset; 299 300 if (ds->ds_snapname[0]) 301 return (0); 302 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) 303 return (0); 304 305 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, 306 FTAG, &headdbuf); 307 if (err != 0) 308 return (err); 309 headphys = headdbuf->db_data; 310 err = zap_value_search(dp->dp_meta_objset, 311 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 312 dmu_buf_rele(headdbuf, FTAG); 313 return (err); 314 } 315 316 int 317 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 318 { 319 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 320 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 321 matchtype_t mt; 322 int err; 323 324 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 325 mt = MT_FIRST; 326 else 327 mt = MT_EXACT; 328 329 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 330 value, mt, NULL, 0, NULL); 331 if (err == ENOTSUP && mt == MT_FIRST) 332 err = zap_lookup(mos, snapobj, name, 8, 1, value); 333 return (err); 334 } 335 336 int 337 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 338 boolean_t adj_cnt) 339 { 340 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 341 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 342 matchtype_t mt; 343 int err; 344 345 dsl_dir_snap_cmtime_update(ds->ds_dir); 346 347 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET) 348 mt = MT_FIRST; 349 else 350 mt = MT_EXACT; 351 352 err = zap_remove_norm(mos, snapobj, name, mt, tx); 353 if (err == ENOTSUP && mt == MT_FIRST) 354 err = zap_remove(mos, snapobj, name, tx); 355 356 if (err == 0 && adj_cnt) 357 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 358 DD_FIELD_SNAPSHOT_COUNT, tx); 359 360 return (err); 361 } 362 363 boolean_t 364 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag) 365 { 366 dmu_buf_t *dbuf = ds->ds_dbuf; 367 boolean_t result = B_FALSE; 368 369 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset, 370 ds->ds_object, DMU_BONUS_BLKID, tag)) { 371 372 if (ds == dmu_buf_get_user(dbuf)) 373 result = B_TRUE; 374 else 375 dmu_buf_rele(dbuf, tag); 376 } 377 378 return (result); 379 } 380 381 int 382 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 383 dsl_dataset_t **dsp) 384 { 385 objset_t *mos = dp->dp_meta_objset; 386 dmu_buf_t *dbuf; 387 dsl_dataset_t *ds; 388 int err; 389 dmu_object_info_t doi; 390 391 ASSERT(dsl_pool_config_held(dp)); 392 393 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 394 if (err != 0) 395 return (err); 396 397 /* Make sure dsobj has the correct object type. */ 398 dmu_object_info_from_db(dbuf, &doi); 399 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 400 dmu_buf_rele(dbuf, tag); 401 return (SET_ERROR(EINVAL)); 402 } 403 404 ds = dmu_buf_get_user(dbuf); 405 if (ds == NULL) { 406 dsl_dataset_t *winner = NULL; 407 408 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 409 ds->ds_dbuf = dbuf; 410 ds->ds_object = dsobj; 411 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0; 412 413 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 414 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 415 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 416 refcount_create(&ds->ds_longholds); 417 418 bplist_create(&ds->ds_pending_deadlist); 419 dsl_deadlist_open(&ds->ds_deadlist, 420 mos, dsl_dataset_phys(ds)->ds_deadlist_obj); 421 422 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 423 offsetof(dmu_sendarg_t, dsa_link)); 424 425 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 426 int zaperr = zap_contains(mos, dsobj, 427 DS_FIELD_LARGE_BLOCKS); 428 if (zaperr != ENOENT) { 429 VERIFY0(zaperr); 430 ds->ds_large_blocks = B_TRUE; 431 } 432 } 433 434 if (err == 0) { 435 err = dsl_dir_hold_obj(dp, 436 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, 437 &ds->ds_dir); 438 } 439 if (err != 0) { 440 mutex_destroy(&ds->ds_lock); 441 mutex_destroy(&ds->ds_opening_lock); 442 mutex_destroy(&ds->ds_sendstream_lock); 443 refcount_destroy(&ds->ds_longholds); 444 bplist_destroy(&ds->ds_pending_deadlist); 445 dsl_deadlist_close(&ds->ds_deadlist); 446 kmem_free(ds, sizeof (dsl_dataset_t)); 447 dmu_buf_rele(dbuf, tag); 448 return (err); 449 } 450 451 if (!ds->ds_is_snapshot) { 452 ds->ds_snapname[0] = '\0'; 453 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 454 err = dsl_dataset_hold_obj(dp, 455 dsl_dataset_phys(ds)->ds_prev_snap_obj, 456 ds, &ds->ds_prev); 457 } 458 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 459 int zaperr = zap_lookup(mos, ds->ds_object, 460 DS_FIELD_BOOKMARK_NAMES, 461 sizeof (ds->ds_bookmarks), 1, 462 &ds->ds_bookmarks); 463 if (zaperr != ENOENT) 464 VERIFY0(zaperr); 465 } 466 } else { 467 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 468 err = dsl_dataset_get_snapname(ds); 469 if (err == 0 && 470 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) { 471 err = zap_count( 472 ds->ds_dir->dd_pool->dp_meta_objset, 473 dsl_dataset_phys(ds)->ds_userrefs_obj, 474 &ds->ds_userrefs); 475 } 476 } 477 478 if (err == 0 && !ds->ds_is_snapshot) { 479 err = dsl_prop_get_int_ds(ds, 480 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 481 &ds->ds_reserved); 482 if (err == 0) { 483 err = dsl_prop_get_int_ds(ds, 484 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 485 &ds->ds_quota); 486 } 487 } else { 488 ds->ds_reserved = ds->ds_quota = 0; 489 } 490 491 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf); 492 if (err == 0) 493 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu); 494 495 if (err != 0 || winner != NULL) { 496 bplist_destroy(&ds->ds_pending_deadlist); 497 dsl_deadlist_close(&ds->ds_deadlist); 498 if (ds->ds_prev) 499 dsl_dataset_rele(ds->ds_prev, ds); 500 dsl_dir_rele(ds->ds_dir, ds); 501 mutex_destroy(&ds->ds_lock); 502 mutex_destroy(&ds->ds_opening_lock); 503 mutex_destroy(&ds->ds_sendstream_lock); 504 refcount_destroy(&ds->ds_longholds); 505 kmem_free(ds, sizeof (dsl_dataset_t)); 506 if (err != 0) { 507 dmu_buf_rele(dbuf, tag); 508 return (err); 509 } 510 ds = winner; 511 } else { 512 ds->ds_fsid_guid = 513 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid); 514 } 515 } 516 ASSERT3P(ds->ds_dbuf, ==, dbuf); 517 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data); 518 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 || 519 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 520 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 521 *dsp = ds; 522 return (0); 523 } 524 525 int 526 dsl_dataset_hold(dsl_pool_t *dp, const char *name, 527 void *tag, dsl_dataset_t **dsp) 528 { 529 dsl_dir_t *dd; 530 const char *snapname; 531 uint64_t obj; 532 int err = 0; 533 dsl_dataset_t *ds; 534 535 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 536 if (err != 0) 537 return (err); 538 539 ASSERT(dsl_pool_config_held(dp)); 540 obj = dsl_dir_phys(dd)->dd_head_dataset_obj; 541 if (obj != 0) 542 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 543 else 544 err = SET_ERROR(ENOENT); 545 546 /* we may be looking for a snapshot */ 547 if (err == 0 && snapname != NULL) { 548 dsl_dataset_t *snap_ds; 549 550 if (*snapname++ != '@') { 551 dsl_dataset_rele(ds, tag); 552 dsl_dir_rele(dd, FTAG); 553 return (SET_ERROR(ENOENT)); 554 } 555 556 dprintf("looking for snapshot '%s'\n", snapname); 557 err = dsl_dataset_snap_lookup(ds, snapname, &obj); 558 if (err == 0) 559 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds); 560 dsl_dataset_rele(ds, tag); 561 562 if (err == 0) { 563 mutex_enter(&snap_ds->ds_lock); 564 if (snap_ds->ds_snapname[0] == 0) 565 (void) strlcpy(snap_ds->ds_snapname, snapname, 566 sizeof (snap_ds->ds_snapname)); 567 mutex_exit(&snap_ds->ds_lock); 568 ds = snap_ds; 569 } 570 } 571 if (err == 0) 572 *dsp = ds; 573 dsl_dir_rele(dd, FTAG); 574 return (err); 575 } 576 577 int 578 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 579 void *tag, dsl_dataset_t **dsp) 580 { 581 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 582 if (err != 0) 583 return (err); 584 if (!dsl_dataset_tryown(*dsp, tag)) { 585 dsl_dataset_rele(*dsp, tag); 586 *dsp = NULL; 587 return (SET_ERROR(EBUSY)); 588 } 589 return (0); 590 } 591 592 int 593 dsl_dataset_own(dsl_pool_t *dp, const char *name, 594 void *tag, dsl_dataset_t **dsp) 595 { 596 int err = dsl_dataset_hold(dp, name, tag, dsp); 597 if (err != 0) 598 return (err); 599 if (!dsl_dataset_tryown(*dsp, tag)) { 600 dsl_dataset_rele(*dsp, tag); 601 return (SET_ERROR(EBUSY)); 602 } 603 return (0); 604 } 605 606 /* 607 * See the comment above dsl_pool_hold() for details. In summary, a long 608 * hold is used to prevent destruction of a dataset while the pool hold 609 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 610 * 611 * The dataset and pool must be held when this function is called. After it 612 * is called, the pool hold may be released while the dataset is still held 613 * and accessed. 614 */ 615 void 616 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 617 { 618 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 619 (void) refcount_add(&ds->ds_longholds, tag); 620 } 621 622 void 623 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 624 { 625 (void) refcount_remove(&ds->ds_longholds, tag); 626 } 627 628 /* Return B_TRUE if there are any long holds on this dataset. */ 629 boolean_t 630 dsl_dataset_long_held(dsl_dataset_t *ds) 631 { 632 return (!refcount_is_zero(&ds->ds_longholds)); 633 } 634 635 void 636 dsl_dataset_name(dsl_dataset_t *ds, char *name) 637 { 638 if (ds == NULL) { 639 (void) strcpy(name, "mos"); 640 } else { 641 dsl_dir_name(ds->ds_dir, name); 642 VERIFY0(dsl_dataset_get_snapname(ds)); 643 if (ds->ds_snapname[0]) { 644 (void) strcat(name, "@"); 645 /* 646 * We use a "recursive" mutex so that we 647 * can call dprintf_ds() with ds_lock held. 648 */ 649 if (!MUTEX_HELD(&ds->ds_lock)) { 650 mutex_enter(&ds->ds_lock); 651 (void) strcat(name, ds->ds_snapname); 652 mutex_exit(&ds->ds_lock); 653 } else { 654 (void) strcat(name, ds->ds_snapname); 655 } 656 } 657 } 658 } 659 660 void 661 dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 662 { 663 dmu_buf_rele(ds->ds_dbuf, tag); 664 } 665 666 void 667 dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 668 { 669 ASSERT3P(ds->ds_owner, ==, tag); 670 ASSERT(ds->ds_dbuf != NULL); 671 672 mutex_enter(&ds->ds_lock); 673 ds->ds_owner = NULL; 674 mutex_exit(&ds->ds_lock); 675 dsl_dataset_long_rele(ds, tag); 676 dsl_dataset_rele(ds, tag); 677 } 678 679 boolean_t 680 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 681 { 682 boolean_t gotit = FALSE; 683 684 mutex_enter(&ds->ds_lock); 685 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 686 ds->ds_owner = tag; 687 dsl_dataset_long_hold(ds, tag); 688 gotit = TRUE; 689 } 690 mutex_exit(&ds->ds_lock); 691 return (gotit); 692 } 693 694 uint64_t 695 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 696 uint64_t flags, dmu_tx_t *tx) 697 { 698 dsl_pool_t *dp = dd->dd_pool; 699 dmu_buf_t *dbuf; 700 dsl_dataset_phys_t *dsphys; 701 uint64_t dsobj; 702 objset_t *mos = dp->dp_meta_objset; 703 704 if (origin == NULL) 705 origin = dp->dp_origin_snap; 706 707 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 708 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0); 709 ASSERT(dmu_tx_is_syncing(tx)); 710 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0); 711 712 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 713 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 714 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 715 dmu_buf_will_dirty(dbuf, tx); 716 dsphys = dbuf->db_data; 717 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 718 dsphys->ds_dir_obj = dd->dd_object; 719 dsphys->ds_flags = flags; 720 dsphys->ds_fsid_guid = unique_create(); 721 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 722 sizeof (dsphys->ds_guid)); 723 dsphys->ds_snapnames_zapobj = 724 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 725 DMU_OT_NONE, 0, tx); 726 dsphys->ds_creation_time = gethrestime_sec(); 727 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 728 729 if (origin == NULL) { 730 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 731 } else { 732 dsl_dataset_t *ohds; /* head of the origin snapshot */ 733 734 dsphys->ds_prev_snap_obj = origin->ds_object; 735 dsphys->ds_prev_snap_txg = 736 dsl_dataset_phys(origin)->ds_creation_txg; 737 dsphys->ds_referenced_bytes = 738 dsl_dataset_phys(origin)->ds_referenced_bytes; 739 dsphys->ds_compressed_bytes = 740 dsl_dataset_phys(origin)->ds_compressed_bytes; 741 dsphys->ds_uncompressed_bytes = 742 dsl_dataset_phys(origin)->ds_uncompressed_bytes; 743 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp; 744 745 /* 746 * Inherit flags that describe the dataset's contents 747 * (INCONSISTENT) or properties (Case Insensitive). 748 */ 749 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags & 750 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 751 752 if (origin->ds_large_blocks) 753 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); 754 755 dmu_buf_will_dirty(origin->ds_dbuf, tx); 756 dsl_dataset_phys(origin)->ds_num_children++; 757 758 VERIFY0(dsl_dataset_hold_obj(dp, 759 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj, 760 FTAG, &ohds)); 761 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 762 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 763 dsl_dataset_rele(ohds, FTAG); 764 765 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 766 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) { 767 dsl_dataset_phys(origin)->ds_next_clones_obj = 768 zap_create(mos, 769 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 770 } 771 VERIFY0(zap_add_int(mos, 772 dsl_dataset_phys(origin)->ds_next_clones_obj, 773 dsobj, tx)); 774 } 775 776 dmu_buf_will_dirty(dd->dd_dbuf, tx); 777 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object; 778 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 779 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 780 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 781 dsl_dir_phys(origin->ds_dir)->dd_clones = 782 zap_create(mos, 783 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 784 } 785 VERIFY0(zap_add_int(mos, 786 dsl_dir_phys(origin->ds_dir)->dd_clones, 787 dsobj, tx)); 788 } 789 } 790 791 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 792 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 793 794 dmu_buf_rele(dbuf, FTAG); 795 796 dmu_buf_will_dirty(dd->dd_dbuf, tx); 797 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj; 798 799 return (dsobj); 800 } 801 802 static void 803 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 804 { 805 objset_t *os; 806 807 VERIFY0(dmu_objset_from_ds(ds, &os)); 808 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 809 dsl_dataset_dirty(ds, tx); 810 } 811 812 uint64_t 813 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 814 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 815 { 816 dsl_pool_t *dp = pdd->dd_pool; 817 uint64_t dsobj, ddobj; 818 dsl_dir_t *dd; 819 820 ASSERT(dmu_tx_is_syncing(tx)); 821 ASSERT(lastname[0] != '@'); 822 823 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 824 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 825 826 dsobj = dsl_dataset_create_sync_dd(dd, origin, 827 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 828 829 dsl_deleg_set_create_perms(dd, tx, cr); 830 831 /* 832 * Since we're creating a new node we know it's a leaf, so we can 833 * initialize the counts if the limit feature is active. 834 */ 835 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 836 uint64_t cnt = 0; 837 objset_t *os = dd->dd_pool->dp_meta_objset; 838 839 dsl_dir_zapify(dd, tx); 840 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 841 sizeof (cnt), 1, &cnt, tx)); 842 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 843 sizeof (cnt), 1, &cnt, tx)); 844 } 845 846 dsl_dir_rele(dd, FTAG); 847 848 /* 849 * If we are creating a clone, make sure we zero out any stale 850 * data from the origin snapshots zil header. 851 */ 852 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 853 dsl_dataset_t *ds; 854 855 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 856 dsl_dataset_zero_zil(ds, tx); 857 dsl_dataset_rele(ds, FTAG); 858 } 859 860 return (dsobj); 861 } 862 863 /* 864 * The unique space in the head dataset can be calculated by subtracting 865 * the space used in the most recent snapshot, that is still being used 866 * in this file system, from the space currently in use. To figure out 867 * the space in the most recent snapshot still in use, we need to take 868 * the total space used in the snapshot and subtract out the space that 869 * has been freed up since the snapshot was taken. 870 */ 871 void 872 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 873 { 874 uint64_t mrs_used; 875 uint64_t dlused, dlcomp, dluncomp; 876 877 ASSERT(!ds->ds_is_snapshot); 878 879 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) 880 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes; 881 else 882 mrs_used = 0; 883 884 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 885 886 ASSERT3U(dlused, <=, mrs_used); 887 dsl_dataset_phys(ds)->ds_unique_bytes = 888 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused); 889 890 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 891 SPA_VERSION_UNIQUE_ACCURATE) 892 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 893 } 894 895 void 896 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 897 dmu_tx_t *tx) 898 { 899 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 900 uint64_t count; 901 int err; 902 903 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2); 904 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 905 obj, tx); 906 /* 907 * The err should not be ENOENT, but a bug in a previous version 908 * of the code could cause upgrade_clones_cb() to not set 909 * ds_next_snap_obj when it should, leading to a missing entry. 910 * If we knew that the pool was created after 911 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 912 * ENOENT. However, at least we can check that we don't have 913 * too many entries in the next_clones_obj even after failing to 914 * remove this one. 915 */ 916 if (err != ENOENT) 917 VERIFY0(err); 918 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 919 &count)); 920 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2); 921 } 922 923 924 blkptr_t * 925 dsl_dataset_get_blkptr(dsl_dataset_t *ds) 926 { 927 return (&dsl_dataset_phys(ds)->ds_bp); 928 } 929 930 void 931 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 932 { 933 ASSERT(dmu_tx_is_syncing(tx)); 934 /* If it's the meta-objset, set dp_meta_rootbp */ 935 if (ds == NULL) { 936 tx->tx_pool->dp_meta_rootbp = *bp; 937 } else { 938 dmu_buf_will_dirty(ds->ds_dbuf, tx); 939 dsl_dataset_phys(ds)->ds_bp = *bp; 940 } 941 } 942 943 spa_t * 944 dsl_dataset_get_spa(dsl_dataset_t *ds) 945 { 946 return (ds->ds_dir->dd_pool->dp_spa); 947 } 948 949 void 950 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 951 { 952 dsl_pool_t *dp; 953 954 if (ds == NULL) /* this is the meta-objset */ 955 return; 956 957 ASSERT(ds->ds_objset != NULL); 958 959 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) 960 panic("dirtying snapshot!"); 961 962 dp = ds->ds_dir->dd_pool; 963 964 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 965 /* up the hold count until we can be written out */ 966 dmu_buf_add_ref(ds->ds_dbuf, ds); 967 } 968 } 969 970 boolean_t 971 dsl_dataset_is_dirty(dsl_dataset_t *ds) 972 { 973 for (int t = 0; t < TXG_SIZE; t++) { 974 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 975 ds, t)) 976 return (B_TRUE); 977 } 978 return (B_FALSE); 979 } 980 981 static int 982 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 983 { 984 uint64_t asize; 985 986 if (!dmu_tx_is_syncing(tx)) 987 return (0); 988 989 /* 990 * If there's an fs-only reservation, any blocks that might become 991 * owned by the snapshot dataset must be accommodated by space 992 * outside of the reservation. 993 */ 994 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 995 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved); 996 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 997 return (SET_ERROR(ENOSPC)); 998 999 /* 1000 * Propagate any reserved space for this snapshot to other 1001 * snapshot checks in this sync group. 1002 */ 1003 if (asize > 0) 1004 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 1005 1006 return (0); 1007 } 1008 1009 typedef struct dsl_dataset_snapshot_arg { 1010 nvlist_t *ddsa_snaps; 1011 nvlist_t *ddsa_props; 1012 nvlist_t *ddsa_errors; 1013 cred_t *ddsa_cr; 1014 } dsl_dataset_snapshot_arg_t; 1015 1016 int 1017 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1018 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 1019 { 1020 int error; 1021 uint64_t value; 1022 1023 ds->ds_trysnap_txg = tx->tx_txg; 1024 1025 if (!dmu_tx_is_syncing(tx)) 1026 return (0); 1027 1028 /* 1029 * We don't allow multiple snapshots of the same txg. If there 1030 * is already one, try again. 1031 */ 1032 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) 1033 return (SET_ERROR(EAGAIN)); 1034 1035 /* 1036 * Check for conflicting snapshot name. 1037 */ 1038 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1039 if (error == 0) 1040 return (SET_ERROR(EEXIST)); 1041 if (error != ENOENT) 1042 return (error); 1043 1044 /* 1045 * We don't allow taking snapshots of inconsistent datasets, such as 1046 * those into which we are currently receiving. However, if we are 1047 * creating this snapshot as part of a receive, this check will be 1048 * executed atomically with respect to the completion of the receive 1049 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1050 * case we ignore this, knowing it will be fixed up for us shortly in 1051 * dmu_recv_end_sync(). 1052 */ 1053 if (!recv && DS_IS_INCONSISTENT(ds)) 1054 return (SET_ERROR(EBUSY)); 1055 1056 /* 1057 * Skip the check for temporary snapshots or if we have already checked 1058 * the counts in dsl_dataset_snapshot_check. This means we really only 1059 * check the count here when we're receiving a stream. 1060 */ 1061 if (cnt != 0 && cr != NULL) { 1062 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1063 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 1064 if (error != 0) 1065 return (error); 1066 } 1067 1068 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1069 if (error != 0) 1070 return (error); 1071 1072 return (0); 1073 } 1074 1075 static int 1076 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1077 { 1078 dsl_dataset_snapshot_arg_t *ddsa = arg; 1079 dsl_pool_t *dp = dmu_tx_pool(tx); 1080 nvpair_t *pair; 1081 int rv = 0; 1082 1083 /* 1084 * Pre-compute how many total new snapshots will be created for each 1085 * level in the tree and below. This is needed for validating the 1086 * snapshot limit when either taking a recursive snapshot or when 1087 * taking multiple snapshots. 1088 * 1089 * The problem is that the counts are not actually adjusted when 1090 * we are checking, only when we finally sync. For a single snapshot, 1091 * this is easy, the count will increase by 1 at each node up the tree, 1092 * but its more complicated for the recursive/multiple snapshot case. 1093 * 1094 * The dsl_fs_ss_limit_check function does recursively check the count 1095 * at each level up the tree but since it is validating each snapshot 1096 * independently we need to be sure that we are validating the complete 1097 * count for the entire set of snapshots. We do this by rolling up the 1098 * counts for each component of the name into an nvlist and then 1099 * checking each of those cases with the aggregated count. 1100 * 1101 * This approach properly handles not only the recursive snapshot 1102 * case (where we get all of those on the ddsa_snaps list) but also 1103 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1104 * validate the limit on 'a' using a count of 2). 1105 * 1106 * We validate the snapshot names in the third loop and only report 1107 * name errors once. 1108 */ 1109 if (dmu_tx_is_syncing(tx)) { 1110 nvlist_t *cnt_track = NULL; 1111 cnt_track = fnvlist_alloc(); 1112 1113 /* Rollup aggregated counts into the cnt_track list */ 1114 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1115 pair != NULL; 1116 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1117 char *pdelim; 1118 uint64_t val; 1119 char nm[MAXPATHLEN]; 1120 1121 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1122 pdelim = strchr(nm, '@'); 1123 if (pdelim == NULL) 1124 continue; 1125 *pdelim = '\0'; 1126 1127 do { 1128 if (nvlist_lookup_uint64(cnt_track, nm, 1129 &val) == 0) { 1130 /* update existing entry */ 1131 fnvlist_add_uint64(cnt_track, nm, 1132 val + 1); 1133 } else { 1134 /* add to list */ 1135 fnvlist_add_uint64(cnt_track, nm, 1); 1136 } 1137 1138 pdelim = strrchr(nm, '/'); 1139 if (pdelim != NULL) 1140 *pdelim = '\0'; 1141 } while (pdelim != NULL); 1142 } 1143 1144 /* Check aggregated counts at each level */ 1145 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1146 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1147 int error = 0; 1148 char *name; 1149 uint64_t cnt = 0; 1150 dsl_dataset_t *ds; 1151 1152 name = nvpair_name(pair); 1153 cnt = fnvpair_value_uint64(pair); 1154 ASSERT(cnt > 0); 1155 1156 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1157 if (error == 0) { 1158 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1159 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1160 ddsa->ddsa_cr); 1161 dsl_dataset_rele(ds, FTAG); 1162 } 1163 1164 if (error != 0) { 1165 if (ddsa->ddsa_errors != NULL) 1166 fnvlist_add_int32(ddsa->ddsa_errors, 1167 name, error); 1168 rv = error; 1169 /* only report one error for this check */ 1170 break; 1171 } 1172 } 1173 nvlist_free(cnt_track); 1174 } 1175 1176 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1177 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1178 int error = 0; 1179 dsl_dataset_t *ds; 1180 char *name, *atp; 1181 char dsname[MAXNAMELEN]; 1182 1183 name = nvpair_name(pair); 1184 if (strlen(name) >= MAXNAMELEN) 1185 error = SET_ERROR(ENAMETOOLONG); 1186 if (error == 0) { 1187 atp = strchr(name, '@'); 1188 if (atp == NULL) 1189 error = SET_ERROR(EINVAL); 1190 if (error == 0) 1191 (void) strlcpy(dsname, name, atp - name + 1); 1192 } 1193 if (error == 0) 1194 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1195 if (error == 0) { 1196 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1197 error = dsl_dataset_snapshot_check_impl(ds, 1198 atp + 1, tx, B_FALSE, 0, NULL); 1199 dsl_dataset_rele(ds, FTAG); 1200 } 1201 1202 if (error != 0) { 1203 if (ddsa->ddsa_errors != NULL) { 1204 fnvlist_add_int32(ddsa->ddsa_errors, 1205 name, error); 1206 } 1207 rv = error; 1208 } 1209 } 1210 1211 return (rv); 1212 } 1213 1214 void 1215 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1216 dmu_tx_t *tx) 1217 { 1218 static zil_header_t zero_zil; 1219 1220 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1221 dmu_buf_t *dbuf; 1222 dsl_dataset_phys_t *dsphys; 1223 uint64_t dsobj, crtxg; 1224 objset_t *mos = dp->dp_meta_objset; 1225 objset_t *os; 1226 1227 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1228 1229 /* 1230 * If we are on an old pool, the zil must not be active, in which 1231 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1232 */ 1233 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1234 dmu_objset_from_ds(ds, &os) != 0 || 1235 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1236 sizeof (zero_zil)) == 0); 1237 1238 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1239 1240 /* 1241 * The origin's ds_creation_txg has to be < TXG_INITIAL 1242 */ 1243 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1244 crtxg = 1; 1245 else 1246 crtxg = tx->tx_txg; 1247 1248 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1249 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1250 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1251 dmu_buf_will_dirty(dbuf, tx); 1252 dsphys = dbuf->db_data; 1253 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1254 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1255 dsphys->ds_fsid_guid = unique_create(); 1256 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1257 sizeof (dsphys->ds_guid)); 1258 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 1259 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg; 1260 dsphys->ds_next_snap_obj = ds->ds_object; 1261 dsphys->ds_num_children = 1; 1262 dsphys->ds_creation_time = gethrestime_sec(); 1263 dsphys->ds_creation_txg = crtxg; 1264 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj; 1265 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes; 1266 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes; 1267 dsphys->ds_uncompressed_bytes = 1268 dsl_dataset_phys(ds)->ds_uncompressed_bytes; 1269 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags; 1270 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp; 1271 dmu_buf_rele(dbuf, FTAG); 1272 1273 if (ds->ds_large_blocks) 1274 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx); 1275 1276 ASSERT3U(ds->ds_prev != 0, ==, 1277 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); 1278 if (ds->ds_prev) { 1279 uint64_t next_clones_obj = 1280 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj; 1281 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1282 ds->ds_object || 1283 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1); 1284 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == 1285 ds->ds_object) { 1286 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1287 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==, 1288 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg); 1289 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj; 1290 } else if (next_clones_obj != 0) { 1291 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1292 dsphys->ds_next_snap_obj, tx); 1293 VERIFY0(zap_add_int(mos, 1294 next_clones_obj, dsobj, tx)); 1295 } 1296 } 1297 1298 /* 1299 * If we have a reference-reservation on this dataset, we will 1300 * need to increase the amount of refreservation being charged 1301 * since our unique space is going to zero. 1302 */ 1303 if (ds->ds_reserved) { 1304 int64_t delta; 1305 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1306 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, 1307 ds->ds_reserved); 1308 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1309 delta, 0, 0, tx); 1310 } 1311 1312 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1313 dsl_dataset_phys(ds)->ds_deadlist_obj = 1314 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX, 1315 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx); 1316 dsl_deadlist_close(&ds->ds_deadlist); 1317 dsl_deadlist_open(&ds->ds_deadlist, mos, 1318 dsl_dataset_phys(ds)->ds_deadlist_obj); 1319 dsl_deadlist_add_key(&ds->ds_deadlist, 1320 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx); 1321 1322 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg); 1323 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj; 1324 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg; 1325 dsl_dataset_phys(ds)->ds_unique_bytes = 0; 1326 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1327 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1328 1329 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj, 1330 snapname, 8, 1, &dsobj, tx)); 1331 1332 if (ds->ds_prev) 1333 dsl_dataset_rele(ds->ds_prev, ds); 1334 VERIFY0(dsl_dataset_hold_obj(dp, 1335 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev)); 1336 1337 dsl_scan_ds_snapshotted(ds, tx); 1338 1339 dsl_dir_snap_cmtime_update(ds->ds_dir); 1340 1341 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1342 } 1343 1344 static void 1345 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1346 { 1347 dsl_dataset_snapshot_arg_t *ddsa = arg; 1348 dsl_pool_t *dp = dmu_tx_pool(tx); 1349 nvpair_t *pair; 1350 1351 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1352 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1353 dsl_dataset_t *ds; 1354 char *name, *atp; 1355 char dsname[MAXNAMELEN]; 1356 1357 name = nvpair_name(pair); 1358 atp = strchr(name, '@'); 1359 (void) strlcpy(dsname, name, atp - name + 1); 1360 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1361 1362 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1363 if (ddsa->ddsa_props != NULL) { 1364 dsl_props_set_sync_impl(ds->ds_prev, 1365 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1366 } 1367 dsl_dataset_rele(ds, FTAG); 1368 } 1369 } 1370 1371 /* 1372 * The snapshots must all be in the same pool. 1373 * All-or-nothing: if there are any failures, nothing will be modified. 1374 */ 1375 int 1376 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1377 { 1378 dsl_dataset_snapshot_arg_t ddsa; 1379 nvpair_t *pair; 1380 boolean_t needsuspend; 1381 int error; 1382 spa_t *spa; 1383 char *firstname; 1384 nvlist_t *suspended = NULL; 1385 1386 pair = nvlist_next_nvpair(snaps, NULL); 1387 if (pair == NULL) 1388 return (0); 1389 firstname = nvpair_name(pair); 1390 1391 error = spa_open(firstname, &spa, FTAG); 1392 if (error != 0) 1393 return (error); 1394 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1395 spa_close(spa, FTAG); 1396 1397 if (needsuspend) { 1398 suspended = fnvlist_alloc(); 1399 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1400 pair = nvlist_next_nvpair(snaps, pair)) { 1401 char fsname[MAXNAMELEN]; 1402 char *snapname = nvpair_name(pair); 1403 char *atp; 1404 void *cookie; 1405 1406 atp = strchr(snapname, '@'); 1407 if (atp == NULL) { 1408 error = SET_ERROR(EINVAL); 1409 break; 1410 } 1411 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1412 1413 error = zil_suspend(fsname, &cookie); 1414 if (error != 0) 1415 break; 1416 fnvlist_add_uint64(suspended, fsname, 1417 (uintptr_t)cookie); 1418 } 1419 } 1420 1421 ddsa.ddsa_snaps = snaps; 1422 ddsa.ddsa_props = props; 1423 ddsa.ddsa_errors = errors; 1424 ddsa.ddsa_cr = CRED(); 1425 1426 if (error == 0) { 1427 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1428 dsl_dataset_snapshot_sync, &ddsa, 1429 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1430 } 1431 1432 if (suspended != NULL) { 1433 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1434 pair = nvlist_next_nvpair(suspended, pair)) { 1435 zil_resume((void *)(uintptr_t) 1436 fnvpair_value_uint64(pair)); 1437 } 1438 fnvlist_free(suspended); 1439 } 1440 1441 return (error); 1442 } 1443 1444 typedef struct dsl_dataset_snapshot_tmp_arg { 1445 const char *ddsta_fsname; 1446 const char *ddsta_snapname; 1447 minor_t ddsta_cleanup_minor; 1448 const char *ddsta_htag; 1449 } dsl_dataset_snapshot_tmp_arg_t; 1450 1451 static int 1452 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1453 { 1454 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1455 dsl_pool_t *dp = dmu_tx_pool(tx); 1456 dsl_dataset_t *ds; 1457 int error; 1458 1459 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1460 if (error != 0) 1461 return (error); 1462 1463 /* NULL cred means no limit check for tmp snapshot */ 1464 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1465 tx, B_FALSE, 0, NULL); 1466 if (error != 0) { 1467 dsl_dataset_rele(ds, FTAG); 1468 return (error); 1469 } 1470 1471 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1472 dsl_dataset_rele(ds, FTAG); 1473 return (SET_ERROR(ENOTSUP)); 1474 } 1475 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1476 B_TRUE, tx); 1477 if (error != 0) { 1478 dsl_dataset_rele(ds, FTAG); 1479 return (error); 1480 } 1481 1482 dsl_dataset_rele(ds, FTAG); 1483 return (0); 1484 } 1485 1486 static void 1487 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1488 { 1489 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1490 dsl_pool_t *dp = dmu_tx_pool(tx); 1491 dsl_dataset_t *ds; 1492 1493 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1494 1495 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1496 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1497 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1498 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1499 1500 dsl_dataset_rele(ds, FTAG); 1501 } 1502 1503 int 1504 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1505 minor_t cleanup_minor, const char *htag) 1506 { 1507 dsl_dataset_snapshot_tmp_arg_t ddsta; 1508 int error; 1509 spa_t *spa; 1510 boolean_t needsuspend; 1511 void *cookie; 1512 1513 ddsta.ddsta_fsname = fsname; 1514 ddsta.ddsta_snapname = snapname; 1515 ddsta.ddsta_cleanup_minor = cleanup_minor; 1516 ddsta.ddsta_htag = htag; 1517 1518 error = spa_open(fsname, &spa, FTAG); 1519 if (error != 0) 1520 return (error); 1521 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1522 spa_close(spa, FTAG); 1523 1524 if (needsuspend) { 1525 error = zil_suspend(fsname, &cookie); 1526 if (error != 0) 1527 return (error); 1528 } 1529 1530 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1531 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 1532 1533 if (needsuspend) 1534 zil_resume(cookie); 1535 return (error); 1536 } 1537 1538 1539 void 1540 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1541 { 1542 ASSERT(dmu_tx_is_syncing(tx)); 1543 ASSERT(ds->ds_objset != NULL); 1544 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0); 1545 1546 /* 1547 * in case we had to change ds_fsid_guid when we opened it, 1548 * sync it out now. 1549 */ 1550 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1551 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid; 1552 1553 dmu_objset_sync(ds->ds_objset, zio, tx); 1554 1555 if (ds->ds_need_large_blocks && !ds->ds_large_blocks) { 1556 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx); 1557 ds->ds_large_blocks = B_TRUE; 1558 } 1559 } 1560 1561 static void 1562 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1563 { 1564 uint64_t count = 0; 1565 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1566 zap_cursor_t zc; 1567 zap_attribute_t za; 1568 nvlist_t *propval = fnvlist_alloc(); 1569 nvlist_t *val = fnvlist_alloc(); 1570 1571 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1572 1573 /* 1574 * There may be missing entries in ds_next_clones_obj 1575 * due to a bug in a previous version of the code. 1576 * Only trust it if it has the right number of entries. 1577 */ 1578 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { 1579 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj, 1580 &count)); 1581 } 1582 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) 1583 goto fail; 1584 for (zap_cursor_init(&zc, mos, 1585 dsl_dataset_phys(ds)->ds_next_clones_obj); 1586 zap_cursor_retrieve(&zc, &za) == 0; 1587 zap_cursor_advance(&zc)) { 1588 dsl_dataset_t *clone; 1589 char buf[ZFS_MAXNAMELEN]; 1590 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1591 za.za_first_integer, FTAG, &clone)); 1592 dsl_dir_name(clone->ds_dir, buf); 1593 fnvlist_add_boolean(val, buf); 1594 dsl_dataset_rele(clone, FTAG); 1595 } 1596 zap_cursor_fini(&zc); 1597 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1598 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); 1599 fail: 1600 nvlist_free(val); 1601 nvlist_free(propval); 1602 } 1603 1604 void 1605 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1606 { 1607 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1608 uint64_t refd, avail, uobjs, aobjs, ratio; 1609 1610 ASSERT(dsl_pool_config_held(dp)); 1611 1612 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 : 1613 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 / 1614 dsl_dataset_phys(ds)->ds_compressed_bytes); 1615 1616 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); 1617 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 1618 dsl_dataset_phys(ds)->ds_uncompressed_bytes); 1619 1620 if (ds->ds_is_snapshot) { 1621 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); 1622 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 1623 dsl_dataset_phys(ds)->ds_unique_bytes); 1624 get_clones_stat(ds, nv); 1625 } else { 1626 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 1627 char buf[MAXNAMELEN]; 1628 dsl_dataset_name(ds->ds_prev, buf); 1629 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf); 1630 } 1631 1632 dsl_dir_stats(ds->ds_dir, nv); 1633 } 1634 1635 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 1636 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 1637 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 1638 1639 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 1640 dsl_dataset_phys(ds)->ds_creation_time); 1641 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 1642 dsl_dataset_phys(ds)->ds_creation_txg); 1643 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 1644 ds->ds_quota); 1645 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 1646 ds->ds_reserved); 1647 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 1648 dsl_dataset_phys(ds)->ds_guid); 1649 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 1650 dsl_dataset_phys(ds)->ds_unique_bytes); 1651 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 1652 ds->ds_object); 1653 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 1654 ds->ds_userrefs); 1655 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 1656 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 1657 1658 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 1659 uint64_t written, comp, uncomp; 1660 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1661 dsl_dataset_t *prev; 1662 1663 int err = dsl_dataset_hold_obj(dp, 1664 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 1665 if (err == 0) { 1666 err = dsl_dataset_space_written(prev, ds, &written, 1667 &comp, &uncomp); 1668 dsl_dataset_rele(prev, FTAG); 1669 if (err == 0) { 1670 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 1671 written); 1672 } 1673 } 1674 } 1675 } 1676 1677 void 1678 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 1679 { 1680 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1681 ASSERT(dsl_pool_config_held(dp)); 1682 1683 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg; 1684 stat->dds_inconsistent = 1685 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT; 1686 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid; 1687 stat->dds_origin[0] = '\0'; 1688 if (ds->ds_is_snapshot) { 1689 stat->dds_is_snapshot = B_TRUE; 1690 stat->dds_num_clones = 1691 dsl_dataset_phys(ds)->ds_num_children - 1; 1692 } else { 1693 stat->dds_is_snapshot = B_FALSE; 1694 stat->dds_num_clones = 0; 1695 1696 if (dsl_dir_is_clone(ds->ds_dir)) { 1697 dsl_dataset_t *ods; 1698 1699 VERIFY0(dsl_dataset_hold_obj(dp, 1700 dsl_dir_phys(ds->ds_dir)->dd_origin_obj, 1701 FTAG, &ods)); 1702 dsl_dataset_name(ods, stat->dds_origin); 1703 dsl_dataset_rele(ods, FTAG); 1704 } 1705 } 1706 } 1707 1708 uint64_t 1709 dsl_dataset_fsid_guid(dsl_dataset_t *ds) 1710 { 1711 return (ds->ds_fsid_guid); 1712 } 1713 1714 void 1715 dsl_dataset_space(dsl_dataset_t *ds, 1716 uint64_t *refdbytesp, uint64_t *availbytesp, 1717 uint64_t *usedobjsp, uint64_t *availobjsp) 1718 { 1719 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes; 1720 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 1721 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) 1722 *availbytesp += 1723 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes; 1724 if (ds->ds_quota != 0) { 1725 /* 1726 * Adjust available bytes according to refquota 1727 */ 1728 if (*refdbytesp < ds->ds_quota) 1729 *availbytesp = MIN(*availbytesp, 1730 ds->ds_quota - *refdbytesp); 1731 else 1732 *availbytesp = 0; 1733 } 1734 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp); 1735 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 1736 } 1737 1738 boolean_t 1739 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 1740 { 1741 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1742 1743 ASSERT(dsl_pool_config_held(dp)); 1744 if (snap == NULL) 1745 return (B_FALSE); 1746 if (dsl_dataset_phys(ds)->ds_bp.blk_birth > 1747 dsl_dataset_phys(snap)->ds_creation_txg) { 1748 objset_t *os, *os_snap; 1749 /* 1750 * It may be that only the ZIL differs, because it was 1751 * reset in the head. Don't count that as being 1752 * modified. 1753 */ 1754 if (dmu_objset_from_ds(ds, &os) != 0) 1755 return (B_TRUE); 1756 if (dmu_objset_from_ds(snap, &os_snap) != 0) 1757 return (B_TRUE); 1758 return (bcmp(&os->os_phys->os_meta_dnode, 1759 &os_snap->os_phys->os_meta_dnode, 1760 sizeof (os->os_phys->os_meta_dnode)) != 0); 1761 } 1762 return (B_FALSE); 1763 } 1764 1765 typedef struct dsl_dataset_rename_snapshot_arg { 1766 const char *ddrsa_fsname; 1767 const char *ddrsa_oldsnapname; 1768 const char *ddrsa_newsnapname; 1769 boolean_t ddrsa_recursive; 1770 dmu_tx_t *ddrsa_tx; 1771 } dsl_dataset_rename_snapshot_arg_t; 1772 1773 /* ARGSUSED */ 1774 static int 1775 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 1776 dsl_dataset_t *hds, void *arg) 1777 { 1778 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1779 int error; 1780 uint64_t val; 1781 1782 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1783 if (error != 0) { 1784 /* ignore nonexistent snapshots */ 1785 return (error == ENOENT ? 0 : error); 1786 } 1787 1788 /* new name should not exist */ 1789 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 1790 if (error == 0) 1791 error = SET_ERROR(EEXIST); 1792 else if (error == ENOENT) 1793 error = 0; 1794 1795 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 1796 if (dsl_dir_namelen(hds->ds_dir) + 1 + 1797 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN) 1798 error = SET_ERROR(ENAMETOOLONG); 1799 1800 return (error); 1801 } 1802 1803 static int 1804 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 1805 { 1806 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1807 dsl_pool_t *dp = dmu_tx_pool(tx); 1808 dsl_dataset_t *hds; 1809 int error; 1810 1811 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 1812 if (error != 0) 1813 return (error); 1814 1815 if (ddrsa->ddrsa_recursive) { 1816 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1817 dsl_dataset_rename_snapshot_check_impl, ddrsa, 1818 DS_FIND_CHILDREN); 1819 } else { 1820 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 1821 } 1822 dsl_dataset_rele(hds, FTAG); 1823 return (error); 1824 } 1825 1826 static int 1827 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 1828 dsl_dataset_t *hds, void *arg) 1829 { 1830 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1831 dsl_dataset_t *ds; 1832 uint64_t val; 1833 dmu_tx_t *tx = ddrsa->ddrsa_tx; 1834 int error; 1835 1836 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1837 ASSERT(error == 0 || error == ENOENT); 1838 if (error == ENOENT) { 1839 /* ignore nonexistent snapshots */ 1840 return (0); 1841 } 1842 1843 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 1844 1845 /* log before we change the name */ 1846 spa_history_log_internal_ds(ds, "rename", tx, 1847 "-> @%s", ddrsa->ddrsa_newsnapname); 1848 1849 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 1850 B_FALSE)); 1851 mutex_enter(&ds->ds_lock); 1852 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 1853 mutex_exit(&ds->ds_lock); 1854 VERIFY0(zap_add(dp->dp_meta_objset, 1855 dsl_dataset_phys(hds)->ds_snapnames_zapobj, 1856 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 1857 1858 dsl_dataset_rele(ds, FTAG); 1859 return (0); 1860 } 1861 1862 static void 1863 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 1864 { 1865 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1866 dsl_pool_t *dp = dmu_tx_pool(tx); 1867 dsl_dataset_t *hds; 1868 1869 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 1870 ddrsa->ddrsa_tx = tx; 1871 if (ddrsa->ddrsa_recursive) { 1872 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1873 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 1874 DS_FIND_CHILDREN)); 1875 } else { 1876 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 1877 } 1878 dsl_dataset_rele(hds, FTAG); 1879 } 1880 1881 int 1882 dsl_dataset_rename_snapshot(const char *fsname, 1883 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 1884 { 1885 dsl_dataset_rename_snapshot_arg_t ddrsa; 1886 1887 ddrsa.ddrsa_fsname = fsname; 1888 ddrsa.ddrsa_oldsnapname = oldsnapname; 1889 ddrsa.ddrsa_newsnapname = newsnapname; 1890 ddrsa.ddrsa_recursive = recursive; 1891 1892 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 1893 dsl_dataset_rename_snapshot_sync, &ddrsa, 1894 1, ZFS_SPACE_CHECK_RESERVED)); 1895 } 1896 1897 /* 1898 * If we're doing an ownership handoff, we need to make sure that there is 1899 * only one long hold on the dataset. We're not allowed to change anything here 1900 * so we don't permanently release the long hold or regular hold here. We want 1901 * to do this only when syncing to avoid the dataset unexpectedly going away 1902 * when we release the long hold. 1903 */ 1904 static int 1905 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 1906 { 1907 boolean_t held; 1908 1909 if (!dmu_tx_is_syncing(tx)) 1910 return (0); 1911 1912 if (owner != NULL) { 1913 VERIFY3P(ds->ds_owner, ==, owner); 1914 dsl_dataset_long_rele(ds, owner); 1915 } 1916 1917 held = dsl_dataset_long_held(ds); 1918 1919 if (owner != NULL) 1920 dsl_dataset_long_hold(ds, owner); 1921 1922 if (held) 1923 return (SET_ERROR(EBUSY)); 1924 1925 return (0); 1926 } 1927 1928 typedef struct dsl_dataset_rollback_arg { 1929 const char *ddra_fsname; 1930 void *ddra_owner; 1931 nvlist_t *ddra_result; 1932 } dsl_dataset_rollback_arg_t; 1933 1934 static int 1935 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 1936 { 1937 dsl_dataset_rollback_arg_t *ddra = arg; 1938 dsl_pool_t *dp = dmu_tx_pool(tx); 1939 dsl_dataset_t *ds; 1940 int64_t unused_refres_delta; 1941 int error; 1942 1943 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 1944 if (error != 0) 1945 return (error); 1946 1947 /* must not be a snapshot */ 1948 if (ds->ds_is_snapshot) { 1949 dsl_dataset_rele(ds, FTAG); 1950 return (SET_ERROR(EINVAL)); 1951 } 1952 1953 /* must have a most recent snapshot */ 1954 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) { 1955 dsl_dataset_rele(ds, FTAG); 1956 return (SET_ERROR(EINVAL)); 1957 } 1958 1959 /* must not have any bookmarks after the most recent snapshot */ 1960 nvlist_t *proprequest = fnvlist_alloc(); 1961 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 1962 nvlist_t *bookmarks = fnvlist_alloc(); 1963 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 1964 fnvlist_free(proprequest); 1965 if (error != 0) 1966 return (error); 1967 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 1968 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 1969 nvlist_t *valuenv = 1970 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 1971 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 1972 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 1973 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) { 1974 fnvlist_free(bookmarks); 1975 dsl_dataset_rele(ds, FTAG); 1976 return (SET_ERROR(EEXIST)); 1977 } 1978 } 1979 fnvlist_free(bookmarks); 1980 1981 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 1982 if (error != 0) { 1983 dsl_dataset_rele(ds, FTAG); 1984 return (error); 1985 } 1986 1987 /* 1988 * Check if the snap we are rolling back to uses more than 1989 * the refquota. 1990 */ 1991 if (ds->ds_quota != 0 && 1992 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) { 1993 dsl_dataset_rele(ds, FTAG); 1994 return (SET_ERROR(EDQUOT)); 1995 } 1996 1997 /* 1998 * When we do the clone swap, we will temporarily use more space 1999 * due to the refreservation (the head will no longer have any 2000 * unique space, so the entire amount of the refreservation will need 2001 * to be free). We will immediately destroy the clone, freeing 2002 * this space, but the freeing happens over many txg's. 2003 */ 2004 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 2005 dsl_dataset_phys(ds)->ds_unique_bytes); 2006 2007 if (unused_refres_delta > 0 && 2008 unused_refres_delta > 2009 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 2010 dsl_dataset_rele(ds, FTAG); 2011 return (SET_ERROR(ENOSPC)); 2012 } 2013 2014 dsl_dataset_rele(ds, FTAG); 2015 return (0); 2016 } 2017 2018 static void 2019 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 2020 { 2021 dsl_dataset_rollback_arg_t *ddra = arg; 2022 dsl_pool_t *dp = dmu_tx_pool(tx); 2023 dsl_dataset_t *ds, *clone; 2024 uint64_t cloneobj; 2025 char namebuf[ZFS_MAXNAMELEN]; 2026 2027 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 2028 2029 dsl_dataset_name(ds->ds_prev, namebuf); 2030 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 2031 2032 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 2033 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 2034 2035 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 2036 2037 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 2038 dsl_dataset_zero_zil(ds, tx); 2039 2040 dsl_destroy_head_sync_impl(clone, tx); 2041 2042 dsl_dataset_rele(clone, FTAG); 2043 dsl_dataset_rele(ds, FTAG); 2044 } 2045 2046 /* 2047 * Rolls back the given filesystem or volume to the most recent snapshot. 2048 * The name of the most recent snapshot will be returned under key "target" 2049 * in the result nvlist. 2050 * 2051 * If owner != NULL: 2052 * - The existing dataset MUST be owned by the specified owner at entry 2053 * - Upon return, dataset will still be held by the same owner, whether we 2054 * succeed or not. 2055 * 2056 * This mode is required any time the existing filesystem is mounted. See 2057 * notes above zfs_suspend_fs() for further details. 2058 */ 2059 int 2060 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result) 2061 { 2062 dsl_dataset_rollback_arg_t ddra; 2063 2064 ddra.ddra_fsname = fsname; 2065 ddra.ddra_owner = owner; 2066 ddra.ddra_result = result; 2067 2068 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 2069 dsl_dataset_rollback_sync, &ddra, 2070 1, ZFS_SPACE_CHECK_RESERVED)); 2071 } 2072 2073 struct promotenode { 2074 list_node_t link; 2075 dsl_dataset_t *ds; 2076 }; 2077 2078 typedef struct dsl_dataset_promote_arg { 2079 const char *ddpa_clonename; 2080 dsl_dataset_t *ddpa_clone; 2081 list_t shared_snaps, origin_snaps, clone_snaps; 2082 dsl_dataset_t *origin_origin; /* origin of the origin */ 2083 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2084 char *err_ds; 2085 cred_t *cr; 2086 } dsl_dataset_promote_arg_t; 2087 2088 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2089 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2090 void *tag); 2091 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2092 2093 static int 2094 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2095 { 2096 dsl_dataset_promote_arg_t *ddpa = arg; 2097 dsl_pool_t *dp = dmu_tx_pool(tx); 2098 dsl_dataset_t *hds; 2099 struct promotenode *snap; 2100 dsl_dataset_t *origin_ds; 2101 int err; 2102 uint64_t unused; 2103 uint64_t ss_mv_cnt; 2104 size_t max_snap_len; 2105 2106 err = promote_hold(ddpa, dp, FTAG); 2107 if (err != 0) 2108 return (err); 2109 2110 hds = ddpa->ddpa_clone; 2111 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1; 2112 2113 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) { 2114 promote_rele(ddpa, FTAG); 2115 return (SET_ERROR(EXDEV)); 2116 } 2117 2118 /* 2119 * Compute and check the amount of space to transfer. Since this is 2120 * so expensive, don't do the preliminary check. 2121 */ 2122 if (!dmu_tx_is_syncing(tx)) { 2123 promote_rele(ddpa, FTAG); 2124 return (0); 2125 } 2126 2127 snap = list_head(&ddpa->shared_snaps); 2128 origin_ds = snap->ds; 2129 2130 /* compute origin's new unique space */ 2131 snap = list_tail(&ddpa->clone_snaps); 2132 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2133 origin_ds->ds_object); 2134 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2135 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX, 2136 &ddpa->unique, &unused, &unused); 2137 2138 /* 2139 * Walk the snapshots that we are moving 2140 * 2141 * Compute space to transfer. Consider the incremental changes 2142 * to used by each snapshot: 2143 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2144 * So each snapshot gave birth to: 2145 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2146 * So a sequence would look like: 2147 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2148 * Which simplifies to: 2149 * uN + kN + kN-1 + ... + k1 + k0 2150 * Note however, if we stop before we reach the ORIGIN we get: 2151 * uN + kN + kN-1 + ... + kM - uM-1 2152 */ 2153 ss_mv_cnt = 0; 2154 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes; 2155 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes; 2156 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes; 2157 for (snap = list_head(&ddpa->shared_snaps); snap; 2158 snap = list_next(&ddpa->shared_snaps, snap)) { 2159 uint64_t val, dlused, dlcomp, dluncomp; 2160 dsl_dataset_t *ds = snap->ds; 2161 2162 ss_mv_cnt++; 2163 2164 /* 2165 * If there are long holds, we won't be able to evict 2166 * the objset. 2167 */ 2168 if (dsl_dataset_long_held(ds)) { 2169 err = SET_ERROR(EBUSY); 2170 goto out; 2171 } 2172 2173 /* Check that the snapshot name does not conflict */ 2174 VERIFY0(dsl_dataset_get_snapname(ds)); 2175 if (strlen(ds->ds_snapname) >= max_snap_len) { 2176 err = SET_ERROR(ENAMETOOLONG); 2177 goto out; 2178 } 2179 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2180 if (err == 0) { 2181 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname); 2182 err = SET_ERROR(EEXIST); 2183 goto out; 2184 } 2185 if (err != ENOENT) 2186 goto out; 2187 2188 /* The very first snapshot does not have a deadlist */ 2189 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0) 2190 continue; 2191 2192 dsl_deadlist_space(&ds->ds_deadlist, 2193 &dlused, &dlcomp, &dluncomp); 2194 ddpa->used += dlused; 2195 ddpa->comp += dlcomp; 2196 ddpa->uncomp += dluncomp; 2197 } 2198 2199 /* 2200 * If we are a clone of a clone then we never reached ORIGIN, 2201 * so we need to subtract out the clone origin's used space. 2202 */ 2203 if (ddpa->origin_origin) { 2204 ddpa->used -= 2205 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes; 2206 ddpa->comp -= 2207 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes; 2208 ddpa->uncomp -= 2209 dsl_dataset_phys(ddpa->origin_origin)-> 2210 ds_uncompressed_bytes; 2211 } 2212 2213 /* Check that there is enough space and limit headroom here */ 2214 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2215 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2216 if (err != 0) 2217 goto out; 2218 2219 /* 2220 * Compute the amounts of space that will be used by snapshots 2221 * after the promotion (for both origin and clone). For each, 2222 * it is the amount of space that will be on all of their 2223 * deadlists (that was not born before their new origin). 2224 */ 2225 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2226 uint64_t space; 2227 2228 /* 2229 * Note, typically this will not be a clone of a clone, 2230 * so dd_origin_txg will be < TXG_INITIAL, so 2231 * these snaplist_space() -> dsl_deadlist_space_range() 2232 * calls will be fast because they do not have to 2233 * iterate over all bps. 2234 */ 2235 snap = list_head(&ddpa->origin_snaps); 2236 err = snaplist_space(&ddpa->shared_snaps, 2237 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2238 if (err != 0) 2239 goto out; 2240 2241 err = snaplist_space(&ddpa->clone_snaps, 2242 snap->ds->ds_dir->dd_origin_txg, &space); 2243 if (err != 0) 2244 goto out; 2245 ddpa->cloneusedsnap += space; 2246 } 2247 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags & 2248 DD_FLAG_USED_BREAKDOWN) { 2249 err = snaplist_space(&ddpa->origin_snaps, 2250 dsl_dataset_phys(origin_ds)->ds_creation_txg, 2251 &ddpa->originusedsnap); 2252 if (err != 0) 2253 goto out; 2254 } 2255 2256 out: 2257 promote_rele(ddpa, FTAG); 2258 return (err); 2259 } 2260 2261 static void 2262 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2263 { 2264 dsl_dataset_promote_arg_t *ddpa = arg; 2265 dsl_pool_t *dp = dmu_tx_pool(tx); 2266 dsl_dataset_t *hds; 2267 struct promotenode *snap; 2268 dsl_dataset_t *origin_ds; 2269 dsl_dataset_t *origin_head; 2270 dsl_dir_t *dd; 2271 dsl_dir_t *odd = NULL; 2272 uint64_t oldnext_obj; 2273 int64_t delta; 2274 2275 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2276 hds = ddpa->ddpa_clone; 2277 2278 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE); 2279 2280 snap = list_head(&ddpa->shared_snaps); 2281 origin_ds = snap->ds; 2282 dd = hds->ds_dir; 2283 2284 snap = list_head(&ddpa->origin_snaps); 2285 origin_head = snap->ds; 2286 2287 /* 2288 * We need to explicitly open odd, since origin_ds's dd will be 2289 * changing. 2290 */ 2291 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2292 NULL, FTAG, &odd)); 2293 2294 /* change origin's next snap */ 2295 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2296 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj; 2297 snap = list_tail(&ddpa->clone_snaps); 2298 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==, 2299 origin_ds->ds_object); 2300 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object; 2301 2302 /* change the origin's next clone */ 2303 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) { 2304 dsl_dataset_remove_from_next_clones(origin_ds, 2305 snap->ds->ds_object, tx); 2306 VERIFY0(zap_add_int(dp->dp_meta_objset, 2307 dsl_dataset_phys(origin_ds)->ds_next_clones_obj, 2308 oldnext_obj, tx)); 2309 } 2310 2311 /* change origin */ 2312 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2313 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object); 2314 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj; 2315 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2316 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2317 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object; 2318 origin_head->ds_dir->dd_origin_txg = 2319 dsl_dataset_phys(origin_ds)->ds_creation_txg; 2320 2321 /* change dd_clone entries */ 2322 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2323 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2324 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx)); 2325 VERIFY0(zap_add_int(dp->dp_meta_objset, 2326 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2327 hds->ds_object, tx)); 2328 2329 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2330 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones, 2331 origin_head->ds_object, tx)); 2332 if (dsl_dir_phys(dd)->dd_clones == 0) { 2333 dsl_dir_phys(dd)->dd_clones = 2334 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES, 2335 DMU_OT_NONE, 0, tx); 2336 } 2337 VERIFY0(zap_add_int(dp->dp_meta_objset, 2338 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx)); 2339 } 2340 2341 /* move snapshots to this dir */ 2342 for (snap = list_head(&ddpa->shared_snaps); snap; 2343 snap = list_next(&ddpa->shared_snaps, snap)) { 2344 dsl_dataset_t *ds = snap->ds; 2345 2346 /* 2347 * Property callbacks are registered to a particular 2348 * dsl_dir. Since ours is changing, evict the objset 2349 * so that they will be unregistered from the old dsl_dir. 2350 */ 2351 if (ds->ds_objset) { 2352 dmu_objset_evict(ds->ds_objset); 2353 ds->ds_objset = NULL; 2354 } 2355 2356 /* move snap name entry */ 2357 VERIFY0(dsl_dataset_get_snapname(ds)); 2358 VERIFY0(dsl_dataset_snap_remove(origin_head, 2359 ds->ds_snapname, tx, B_TRUE)); 2360 VERIFY0(zap_add(dp->dp_meta_objset, 2361 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname, 2362 8, 1, &ds->ds_object, tx)); 2363 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 2364 DD_FIELD_SNAPSHOT_COUNT, tx); 2365 2366 /* change containing dsl_dir */ 2367 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2368 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object); 2369 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object; 2370 ASSERT3P(ds->ds_dir, ==, odd); 2371 dsl_dir_rele(ds->ds_dir, ds); 2372 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 2373 NULL, ds, &ds->ds_dir)); 2374 2375 /* move any clone references */ 2376 if (dsl_dataset_phys(ds)->ds_next_clones_obj && 2377 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2378 zap_cursor_t zc; 2379 zap_attribute_t za; 2380 2381 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2382 dsl_dataset_phys(ds)->ds_next_clones_obj); 2383 zap_cursor_retrieve(&zc, &za) == 0; 2384 zap_cursor_advance(&zc)) { 2385 dsl_dataset_t *cnds; 2386 uint64_t o; 2387 2388 if (za.za_first_integer == oldnext_obj) { 2389 /* 2390 * We've already moved the 2391 * origin's reference. 2392 */ 2393 continue; 2394 } 2395 2396 VERIFY0(dsl_dataset_hold_obj(dp, 2397 za.za_first_integer, FTAG, &cnds)); 2398 o = dsl_dir_phys(cnds->ds_dir)-> 2399 dd_head_dataset_obj; 2400 2401 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2402 dsl_dir_phys(odd)->dd_clones, o, tx)); 2403 VERIFY0(zap_add_int(dp->dp_meta_objset, 2404 dsl_dir_phys(dd)->dd_clones, o, tx)); 2405 dsl_dataset_rele(cnds, FTAG); 2406 } 2407 zap_cursor_fini(&zc); 2408 } 2409 2410 ASSERT(!dsl_prop_hascb(ds)); 2411 } 2412 2413 /* 2414 * Change space accounting. 2415 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2416 * both be valid, or both be 0 (resulting in delta == 0). This 2417 * is true for each of {clone,origin} independently. 2418 */ 2419 2420 delta = ddpa->cloneusedsnap - 2421 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP]; 2422 ASSERT3S(delta, >=, 0); 2423 ASSERT3U(ddpa->used, >=, delta); 2424 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2425 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2426 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 2427 2428 delta = ddpa->originusedsnap - 2429 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP]; 2430 ASSERT3S(delta, <=, 0); 2431 ASSERT3U(ddpa->used, >=, -delta); 2432 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2433 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2434 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 2435 2436 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique; 2437 2438 /* log history record */ 2439 spa_history_log_internal_ds(hds, "promote", tx, ""); 2440 2441 dsl_dir_rele(odd, FTAG); 2442 promote_rele(ddpa, FTAG); 2443 } 2444 2445 /* 2446 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2447 * (exclusive) and last_obj (inclusive). The list will be in reverse 2448 * order (last_obj will be the list_head()). If first_obj == 0, do all 2449 * snapshots back to this dataset's origin. 2450 */ 2451 static int 2452 snaplist_make(dsl_pool_t *dp, 2453 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 2454 { 2455 uint64_t obj = last_obj; 2456 2457 list_create(l, sizeof (struct promotenode), 2458 offsetof(struct promotenode, link)); 2459 2460 while (obj != first_obj) { 2461 dsl_dataset_t *ds; 2462 struct promotenode *snap; 2463 int err; 2464 2465 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 2466 ASSERT(err != ENOENT); 2467 if (err != 0) 2468 return (err); 2469 2470 if (first_obj == 0) 2471 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj; 2472 2473 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 2474 snap->ds = ds; 2475 list_insert_tail(l, snap); 2476 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 2477 } 2478 2479 return (0); 2480 } 2481 2482 static int 2483 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2484 { 2485 struct promotenode *snap; 2486 2487 *spacep = 0; 2488 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2489 uint64_t used, comp, uncomp; 2490 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2491 mintxg, UINT64_MAX, &used, &comp, &uncomp); 2492 *spacep += used; 2493 } 2494 return (0); 2495 } 2496 2497 static void 2498 snaplist_destroy(list_t *l, void *tag) 2499 { 2500 struct promotenode *snap; 2501 2502 if (l == NULL || !list_link_active(&l->list_head)) 2503 return; 2504 2505 while ((snap = list_tail(l)) != NULL) { 2506 list_remove(l, snap); 2507 dsl_dataset_rele(snap->ds, tag); 2508 kmem_free(snap, sizeof (*snap)); 2509 } 2510 list_destroy(l); 2511 } 2512 2513 static int 2514 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 2515 { 2516 int error; 2517 dsl_dir_t *dd; 2518 struct promotenode *snap; 2519 2520 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 2521 &ddpa->ddpa_clone); 2522 if (error != 0) 2523 return (error); 2524 dd = ddpa->ddpa_clone->ds_dir; 2525 2526 if (ddpa->ddpa_clone->ds_is_snapshot || 2527 !dsl_dir_is_clone(dd)) { 2528 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2529 return (SET_ERROR(EINVAL)); 2530 } 2531 2532 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj, 2533 &ddpa->shared_snaps, tag); 2534 if (error != 0) 2535 goto out; 2536 2537 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 2538 &ddpa->clone_snaps, tag); 2539 if (error != 0) 2540 goto out; 2541 2542 snap = list_head(&ddpa->shared_snaps); 2543 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj); 2544 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj, 2545 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj, 2546 &ddpa->origin_snaps, tag); 2547 if (error != 0) 2548 goto out; 2549 2550 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) { 2551 error = dsl_dataset_hold_obj(dp, 2552 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj, 2553 tag, &ddpa->origin_origin); 2554 if (error != 0) 2555 goto out; 2556 } 2557 out: 2558 if (error != 0) 2559 promote_rele(ddpa, tag); 2560 return (error); 2561 } 2562 2563 static void 2564 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 2565 { 2566 snaplist_destroy(&ddpa->shared_snaps, tag); 2567 snaplist_destroy(&ddpa->clone_snaps, tag); 2568 snaplist_destroy(&ddpa->origin_snaps, tag); 2569 if (ddpa->origin_origin != NULL) 2570 dsl_dataset_rele(ddpa->origin_origin, tag); 2571 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2572 } 2573 2574 /* 2575 * Promote a clone. 2576 * 2577 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 2578 * in with the name. (It must be at least MAXNAMELEN bytes long.) 2579 */ 2580 int 2581 dsl_dataset_promote(const char *name, char *conflsnap) 2582 { 2583 dsl_dataset_promote_arg_t ddpa = { 0 }; 2584 uint64_t numsnaps; 2585 int error; 2586 objset_t *os; 2587 2588 /* 2589 * We will modify space proportional to the number of 2590 * snapshots. Compute numsnaps. 2591 */ 2592 error = dmu_objset_hold(name, FTAG, &os); 2593 if (error != 0) 2594 return (error); 2595 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 2596 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj, 2597 &numsnaps); 2598 dmu_objset_rele(os, FTAG); 2599 if (error != 0) 2600 return (error); 2601 2602 ddpa.ddpa_clonename = name; 2603 ddpa.err_ds = conflsnap; 2604 ddpa.cr = CRED(); 2605 2606 return (dsl_sync_task(name, dsl_dataset_promote_check, 2607 dsl_dataset_promote_sync, &ddpa, 2608 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED)); 2609 } 2610 2611 int 2612 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 2613 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 2614 { 2615 int64_t unused_refres_delta; 2616 2617 /* they should both be heads */ 2618 if (clone->ds_is_snapshot || 2619 origin_head->ds_is_snapshot) 2620 return (SET_ERROR(EINVAL)); 2621 2622 /* if we are not forcing, the branch point should be just before them */ 2623 if (!force && clone->ds_prev != origin_head->ds_prev) 2624 return (SET_ERROR(EINVAL)); 2625 2626 /* clone should be the clone (unless they are unrelated) */ 2627 if (clone->ds_prev != NULL && 2628 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 2629 origin_head->ds_dir != clone->ds_prev->ds_dir) 2630 return (SET_ERROR(EINVAL)); 2631 2632 /* the clone should be a child of the origin */ 2633 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 2634 return (SET_ERROR(EINVAL)); 2635 2636 /* origin_head shouldn't be modified unless 'force' */ 2637 if (!force && 2638 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 2639 return (SET_ERROR(ETXTBSY)); 2640 2641 /* origin_head should have no long holds (e.g. is not mounted) */ 2642 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 2643 return (SET_ERROR(EBUSY)); 2644 2645 /* check amount of any unconsumed refreservation */ 2646 unused_refres_delta = 2647 (int64_t)MIN(origin_head->ds_reserved, 2648 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 2649 (int64_t)MIN(origin_head->ds_reserved, 2650 dsl_dataset_phys(clone)->ds_unique_bytes); 2651 2652 if (unused_refres_delta > 0 && 2653 unused_refres_delta > 2654 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 2655 return (SET_ERROR(ENOSPC)); 2656 2657 /* clone can't be over the head's refquota */ 2658 if (origin_head->ds_quota != 0 && 2659 dsl_dataset_phys(clone)->ds_referenced_bytes > 2660 origin_head->ds_quota) 2661 return (SET_ERROR(EDQUOT)); 2662 2663 return (0); 2664 } 2665 2666 void 2667 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 2668 dsl_dataset_t *origin_head, dmu_tx_t *tx) 2669 { 2670 dsl_pool_t *dp = dmu_tx_pool(tx); 2671 int64_t unused_refres_delta; 2672 2673 ASSERT(clone->ds_reserved == 0); 2674 ASSERT(origin_head->ds_quota == 0 || 2675 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota); 2676 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 2677 2678 dmu_buf_will_dirty(clone->ds_dbuf, tx); 2679 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 2680 2681 if (clone->ds_objset != NULL) { 2682 dmu_objset_evict(clone->ds_objset); 2683 clone->ds_objset = NULL; 2684 } 2685 2686 if (origin_head->ds_objset != NULL) { 2687 dmu_objset_evict(origin_head->ds_objset); 2688 origin_head->ds_objset = NULL; 2689 } 2690 2691 unused_refres_delta = 2692 (int64_t)MIN(origin_head->ds_reserved, 2693 dsl_dataset_phys(origin_head)->ds_unique_bytes) - 2694 (int64_t)MIN(origin_head->ds_reserved, 2695 dsl_dataset_phys(clone)->ds_unique_bytes); 2696 2697 /* 2698 * Reset origin's unique bytes, if it exists. 2699 */ 2700 if (clone->ds_prev) { 2701 dsl_dataset_t *origin = clone->ds_prev; 2702 uint64_t comp, uncomp; 2703 2704 dmu_buf_will_dirty(origin->ds_dbuf, tx); 2705 dsl_deadlist_space_range(&clone->ds_deadlist, 2706 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX, 2707 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp); 2708 } 2709 2710 /* swap blkptrs */ 2711 { 2712 blkptr_t tmp; 2713 tmp = dsl_dataset_phys(origin_head)->ds_bp; 2714 dsl_dataset_phys(origin_head)->ds_bp = 2715 dsl_dataset_phys(clone)->ds_bp; 2716 dsl_dataset_phys(clone)->ds_bp = tmp; 2717 } 2718 2719 /* set dd_*_bytes */ 2720 { 2721 int64_t dused, dcomp, duncomp; 2722 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2723 uint64_t odl_used, odl_comp, odl_uncomp; 2724 2725 ASSERT3U(dsl_dir_phys(clone->ds_dir)-> 2726 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2727 2728 dsl_deadlist_space(&clone->ds_deadlist, 2729 &cdl_used, &cdl_comp, &cdl_uncomp); 2730 dsl_deadlist_space(&origin_head->ds_deadlist, 2731 &odl_used, &odl_comp, &odl_uncomp); 2732 2733 dused = dsl_dataset_phys(clone)->ds_referenced_bytes + 2734 cdl_used - 2735 (dsl_dataset_phys(origin_head)->ds_referenced_bytes + 2736 odl_used); 2737 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes + 2738 cdl_comp - 2739 (dsl_dataset_phys(origin_head)->ds_compressed_bytes + 2740 odl_comp); 2741 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes + 2742 cdl_uncomp - 2743 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes + 2744 odl_uncomp); 2745 2746 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 2747 dused, dcomp, duncomp, tx); 2748 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 2749 -dused, -dcomp, -duncomp, tx); 2750 2751 /* 2752 * The difference in the space used by snapshots is the 2753 * difference in snapshot space due to the head's 2754 * deadlist (since that's the only thing that's 2755 * changing that affects the snapused). 2756 */ 2757 dsl_deadlist_space_range(&clone->ds_deadlist, 2758 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2759 &cdl_used, &cdl_comp, &cdl_uncomp); 2760 dsl_deadlist_space_range(&origin_head->ds_deadlist, 2761 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2762 &odl_used, &odl_comp, &odl_uncomp); 2763 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 2764 DD_USED_HEAD, DD_USED_SNAP, tx); 2765 } 2766 2767 /* swap ds_*_bytes */ 2768 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes, 2769 dsl_dataset_phys(clone)->ds_referenced_bytes); 2770 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes, 2771 dsl_dataset_phys(clone)->ds_compressed_bytes); 2772 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes, 2773 dsl_dataset_phys(clone)->ds_uncompressed_bytes); 2774 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes, 2775 dsl_dataset_phys(clone)->ds_unique_bytes); 2776 2777 /* apply any parent delta for change in unconsumed refreservation */ 2778 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 2779 unused_refres_delta, 0, 0, tx); 2780 2781 /* 2782 * Swap deadlists. 2783 */ 2784 dsl_deadlist_close(&clone->ds_deadlist); 2785 dsl_deadlist_close(&origin_head->ds_deadlist); 2786 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj, 2787 dsl_dataset_phys(clone)->ds_deadlist_obj); 2788 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 2789 dsl_dataset_phys(clone)->ds_deadlist_obj); 2790 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 2791 dsl_dataset_phys(origin_head)->ds_deadlist_obj); 2792 2793 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 2794 2795 spa_history_log_internal_ds(clone, "clone swap", tx, 2796 "parent=%s", origin_head->ds_dir->dd_myname); 2797 } 2798 2799 /* 2800 * Given a pool name and a dataset object number in that pool, 2801 * return the name of that dataset. 2802 */ 2803 int 2804 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 2805 { 2806 dsl_pool_t *dp; 2807 dsl_dataset_t *ds; 2808 int error; 2809 2810 error = dsl_pool_hold(pname, FTAG, &dp); 2811 if (error != 0) 2812 return (error); 2813 2814 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 2815 if (error == 0) { 2816 dsl_dataset_name(ds, buf); 2817 dsl_dataset_rele(ds, FTAG); 2818 } 2819 dsl_pool_rele(dp, FTAG); 2820 2821 return (error); 2822 } 2823 2824 int 2825 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 2826 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 2827 { 2828 int error = 0; 2829 2830 ASSERT3S(asize, >, 0); 2831 2832 /* 2833 * *ref_rsrv is the portion of asize that will come from any 2834 * unconsumed refreservation space. 2835 */ 2836 *ref_rsrv = 0; 2837 2838 mutex_enter(&ds->ds_lock); 2839 /* 2840 * Make a space adjustment for reserved bytes. 2841 */ 2842 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) { 2843 ASSERT3U(*used, >=, 2844 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 2845 *used -= 2846 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes); 2847 *ref_rsrv = 2848 asize - MIN(asize, parent_delta(ds, asize + inflight)); 2849 } 2850 2851 if (!check_quota || ds->ds_quota == 0) { 2852 mutex_exit(&ds->ds_lock); 2853 return (0); 2854 } 2855 /* 2856 * If they are requesting more space, and our current estimate 2857 * is over quota, they get to try again unless the actual 2858 * on-disk is over quota and there are no pending changes (which 2859 * may free up space for us). 2860 */ 2861 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >= 2862 ds->ds_quota) { 2863 if (inflight > 0 || 2864 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota) 2865 error = SET_ERROR(ERESTART); 2866 else 2867 error = SET_ERROR(EDQUOT); 2868 } 2869 mutex_exit(&ds->ds_lock); 2870 2871 return (error); 2872 } 2873 2874 typedef struct dsl_dataset_set_qr_arg { 2875 const char *ddsqra_name; 2876 zprop_source_t ddsqra_source; 2877 uint64_t ddsqra_value; 2878 } dsl_dataset_set_qr_arg_t; 2879 2880 2881 /* ARGSUSED */ 2882 static int 2883 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 2884 { 2885 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2886 dsl_pool_t *dp = dmu_tx_pool(tx); 2887 dsl_dataset_t *ds; 2888 int error; 2889 uint64_t newval; 2890 2891 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 2892 return (SET_ERROR(ENOTSUP)); 2893 2894 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 2895 if (error != 0) 2896 return (error); 2897 2898 if (ds->ds_is_snapshot) { 2899 dsl_dataset_rele(ds, FTAG); 2900 return (SET_ERROR(EINVAL)); 2901 } 2902 2903 error = dsl_prop_predict(ds->ds_dir, 2904 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 2905 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 2906 if (error != 0) { 2907 dsl_dataset_rele(ds, FTAG); 2908 return (error); 2909 } 2910 2911 if (newval == 0) { 2912 dsl_dataset_rele(ds, FTAG); 2913 return (0); 2914 } 2915 2916 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes || 2917 newval < ds->ds_reserved) { 2918 dsl_dataset_rele(ds, FTAG); 2919 return (SET_ERROR(ENOSPC)); 2920 } 2921 2922 dsl_dataset_rele(ds, FTAG); 2923 return (0); 2924 } 2925 2926 static void 2927 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 2928 { 2929 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2930 dsl_pool_t *dp = dmu_tx_pool(tx); 2931 dsl_dataset_t *ds; 2932 uint64_t newval; 2933 2934 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 2935 2936 dsl_prop_set_sync_impl(ds, 2937 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 2938 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 2939 &ddsqra->ddsqra_value, tx); 2940 2941 VERIFY0(dsl_prop_get_int_ds(ds, 2942 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 2943 2944 if (ds->ds_quota != newval) { 2945 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2946 ds->ds_quota = newval; 2947 } 2948 dsl_dataset_rele(ds, FTAG); 2949 } 2950 2951 int 2952 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 2953 uint64_t refquota) 2954 { 2955 dsl_dataset_set_qr_arg_t ddsqra; 2956 2957 ddsqra.ddsqra_name = dsname; 2958 ddsqra.ddsqra_source = source; 2959 ddsqra.ddsqra_value = refquota; 2960 2961 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 2962 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 2963 } 2964 2965 static int 2966 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 2967 { 2968 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2969 dsl_pool_t *dp = dmu_tx_pool(tx); 2970 dsl_dataset_t *ds; 2971 int error; 2972 uint64_t newval, unique; 2973 2974 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 2975 return (SET_ERROR(ENOTSUP)); 2976 2977 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 2978 if (error != 0) 2979 return (error); 2980 2981 if (ds->ds_is_snapshot) { 2982 dsl_dataset_rele(ds, FTAG); 2983 return (SET_ERROR(EINVAL)); 2984 } 2985 2986 error = dsl_prop_predict(ds->ds_dir, 2987 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 2988 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 2989 if (error != 0) { 2990 dsl_dataset_rele(ds, FTAG); 2991 return (error); 2992 } 2993 2994 /* 2995 * If we are doing the preliminary check in open context, the 2996 * space estimates may be inaccurate. 2997 */ 2998 if (!dmu_tx_is_syncing(tx)) { 2999 dsl_dataset_rele(ds, FTAG); 3000 return (0); 3001 } 3002 3003 mutex_enter(&ds->ds_lock); 3004 if (!DS_UNIQUE_IS_ACCURATE(ds)) 3005 dsl_dataset_recalc_head_uniq(ds); 3006 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3007 mutex_exit(&ds->ds_lock); 3008 3009 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 3010 uint64_t delta = MAX(unique, newval) - 3011 MAX(unique, ds->ds_reserved); 3012 3013 if (delta > 3014 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 3015 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 3016 dsl_dataset_rele(ds, FTAG); 3017 return (SET_ERROR(ENOSPC)); 3018 } 3019 } 3020 3021 dsl_dataset_rele(ds, FTAG); 3022 return (0); 3023 } 3024 3025 void 3026 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 3027 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 3028 { 3029 uint64_t newval; 3030 uint64_t unique; 3031 int64_t delta; 3032 3033 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3034 source, sizeof (value), 1, &value, tx); 3035 3036 VERIFY0(dsl_prop_get_int_ds(ds, 3037 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 3038 3039 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3040 mutex_enter(&ds->ds_dir->dd_lock); 3041 mutex_enter(&ds->ds_lock); 3042 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3043 unique = dsl_dataset_phys(ds)->ds_unique_bytes; 3044 delta = MAX(0, (int64_t)(newval - unique)) - 3045 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3046 ds->ds_reserved = newval; 3047 mutex_exit(&ds->ds_lock); 3048 3049 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3050 mutex_exit(&ds->ds_dir->dd_lock); 3051 } 3052 3053 static void 3054 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 3055 { 3056 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3057 dsl_pool_t *dp = dmu_tx_pool(tx); 3058 dsl_dataset_t *ds; 3059 3060 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3061 dsl_dataset_set_refreservation_sync_impl(ds, 3062 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 3063 dsl_dataset_rele(ds, FTAG); 3064 } 3065 3066 int 3067 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 3068 uint64_t refreservation) 3069 { 3070 dsl_dataset_set_qr_arg_t ddsqra; 3071 3072 ddsqra.ddsqra_name = dsname; 3073 ddsqra.ddsqra_source = source; 3074 ddsqra.ddsqra_value = refreservation; 3075 3076 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 3077 dsl_dataset_set_refreservation_sync, &ddsqra, 3078 0, ZFS_SPACE_CHECK_NONE)); 3079 } 3080 3081 /* 3082 * Return (in *usedp) the amount of space written in new that is not 3083 * present in oldsnap. New may be a snapshot or the head. Old must be 3084 * a snapshot before new, in new's filesystem (or its origin). If not then 3085 * fail and return EINVAL. 3086 * 3087 * The written space is calculated by considering two components: First, we 3088 * ignore any freed space, and calculate the written as new's used space 3089 * minus old's used space. Next, we add in the amount of space that was freed 3090 * between the two snapshots, thus reducing new's used space relative to old's. 3091 * Specifically, this is the space that was born before old->ds_creation_txg, 3092 * and freed before new (ie. on new's deadlist or a previous deadlist). 3093 * 3094 * space freed [---------------------] 3095 * snapshots ---O-------O--------O-------O------ 3096 * oldsnap new 3097 */ 3098 int 3099 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 3100 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3101 { 3102 int err = 0; 3103 uint64_t snapobj; 3104 dsl_pool_t *dp = new->ds_dir->dd_pool; 3105 3106 ASSERT(dsl_pool_config_held(dp)); 3107 3108 *usedp = 0; 3109 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes; 3110 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes; 3111 3112 *compp = 0; 3113 *compp += dsl_dataset_phys(new)->ds_compressed_bytes; 3114 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes; 3115 3116 *uncompp = 0; 3117 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes; 3118 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes; 3119 3120 snapobj = new->ds_object; 3121 while (snapobj != oldsnap->ds_object) { 3122 dsl_dataset_t *snap; 3123 uint64_t used, comp, uncomp; 3124 3125 if (snapobj == new->ds_object) { 3126 snap = new; 3127 } else { 3128 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3129 if (err != 0) 3130 break; 3131 } 3132 3133 if (dsl_dataset_phys(snap)->ds_prev_snap_txg == 3134 dsl_dataset_phys(oldsnap)->ds_creation_txg) { 3135 /* 3136 * The blocks in the deadlist can not be born after 3137 * ds_prev_snap_txg, so get the whole deadlist space, 3138 * which is more efficient (especially for old-format 3139 * deadlists). Unfortunately the deadlist code 3140 * doesn't have enough information to make this 3141 * optimization itself. 3142 */ 3143 dsl_deadlist_space(&snap->ds_deadlist, 3144 &used, &comp, &uncomp); 3145 } else { 3146 dsl_deadlist_space_range(&snap->ds_deadlist, 3147 0, dsl_dataset_phys(oldsnap)->ds_creation_txg, 3148 &used, &comp, &uncomp); 3149 } 3150 *usedp += used; 3151 *compp += comp; 3152 *uncompp += uncomp; 3153 3154 /* 3155 * If we get to the beginning of the chain of snapshots 3156 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3157 * was not a snapshot of/before new. 3158 */ 3159 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj; 3160 if (snap != new) 3161 dsl_dataset_rele(snap, FTAG); 3162 if (snapobj == 0) { 3163 err = SET_ERROR(EINVAL); 3164 break; 3165 } 3166 3167 } 3168 return (err); 3169 } 3170 3171 /* 3172 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3173 * lastsnap, and all snapshots in between are deleted. 3174 * 3175 * blocks that would be freed [---------------------------] 3176 * snapshots ---O-------O--------O-------O--------O 3177 * firstsnap lastsnap 3178 * 3179 * This is the set of blocks that were born after the snap before firstsnap, 3180 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3181 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3182 * We calculate this by iterating over the relevant deadlists (from the snap 3183 * after lastsnap, backward to the snap after firstsnap), summing up the 3184 * space on the deadlist that was born after the snap before firstsnap. 3185 */ 3186 int 3187 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3188 dsl_dataset_t *lastsnap, 3189 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3190 { 3191 int err = 0; 3192 uint64_t snapobj; 3193 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3194 3195 ASSERT(firstsnap->ds_is_snapshot); 3196 ASSERT(lastsnap->ds_is_snapshot); 3197 3198 /* 3199 * Check that the snapshots are in the same dsl_dir, and firstsnap 3200 * is before lastsnap. 3201 */ 3202 if (firstsnap->ds_dir != lastsnap->ds_dir || 3203 dsl_dataset_phys(firstsnap)->ds_creation_txg > 3204 dsl_dataset_phys(lastsnap)->ds_creation_txg) 3205 return (SET_ERROR(EINVAL)); 3206 3207 *usedp = *compp = *uncompp = 0; 3208 3209 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj; 3210 while (snapobj != firstsnap->ds_object) { 3211 dsl_dataset_t *ds; 3212 uint64_t used, comp, uncomp; 3213 3214 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3215 if (err != 0) 3216 break; 3217 3218 dsl_deadlist_space_range(&ds->ds_deadlist, 3219 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX, 3220 &used, &comp, &uncomp); 3221 *usedp += used; 3222 *compp += comp; 3223 *uncompp += uncomp; 3224 3225 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj; 3226 ASSERT3U(snapobj, !=, 0); 3227 dsl_dataset_rele(ds, FTAG); 3228 } 3229 return (err); 3230 } 3231 3232 static int 3233 dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx) 3234 { 3235 const char *dsname = arg; 3236 dsl_dataset_t *ds; 3237 dsl_pool_t *dp = dmu_tx_pool(tx); 3238 int error = 0; 3239 3240 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS)) 3241 return (SET_ERROR(ENOTSUP)); 3242 3243 ASSERT(spa_feature_is_enabled(dp->dp_spa, 3244 SPA_FEATURE_EXTENSIBLE_DATASET)); 3245 3246 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 3247 if (error != 0) 3248 return (error); 3249 3250 if (ds->ds_large_blocks) 3251 error = EALREADY; 3252 dsl_dataset_rele(ds, FTAG); 3253 3254 return (error); 3255 } 3256 3257 void 3258 dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx) 3259 { 3260 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 3261 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset; 3262 uint64_t zero = 0; 3263 3264 spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx); 3265 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx); 3266 3267 VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS, 3268 sizeof (zero), 1, &zero, tx)); 3269 } 3270 3271 static void 3272 dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx) 3273 { 3274 const char *dsname = arg; 3275 dsl_dataset_t *ds; 3276 3277 VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds)); 3278 3279 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx); 3280 ASSERT(!ds->ds_large_blocks); 3281 ds->ds_large_blocks = B_TRUE; 3282 dsl_dataset_rele(ds, FTAG); 3283 } 3284 3285 int 3286 dsl_dataset_activate_large_blocks(const char *dsname) 3287 { 3288 int error; 3289 3290 error = dsl_sync_task(dsname, 3291 dsl_dataset_activate_large_blocks_check, 3292 dsl_dataset_activate_large_blocks_sync, (void *)dsname, 3293 1, ZFS_SPACE_CHECK_RESERVED); 3294 3295 /* 3296 * EALREADY indicates that this dataset already supports large blocks. 3297 */ 3298 if (error == EALREADY) 3299 error = 0; 3300 return (error); 3301 } 3302 3303 /* 3304 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 3305 * For example, they could both be snapshots of the same filesystem, and 3306 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 3307 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 3308 * filesystem. Or 'earlier' could be the origin's origin. 3309 * 3310 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 3311 */ 3312 boolean_t 3313 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 3314 uint64_t earlier_txg) 3315 { 3316 dsl_pool_t *dp = later->ds_dir->dd_pool; 3317 int error; 3318 boolean_t ret; 3319 3320 ASSERT(dsl_pool_config_held(dp)); 3321 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0); 3322 3323 if (earlier_txg == 0) 3324 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg; 3325 3326 if (later->ds_is_snapshot && 3327 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg) 3328 return (B_FALSE); 3329 3330 if (later->ds_dir == earlier->ds_dir) 3331 return (B_TRUE); 3332 if (!dsl_dir_is_clone(later->ds_dir)) 3333 return (B_FALSE); 3334 3335 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object) 3336 return (B_TRUE); 3337 dsl_dataset_t *origin; 3338 error = dsl_dataset_hold_obj(dp, 3339 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin); 3340 if (error != 0) 3341 return (B_FALSE); 3342 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 3343 dsl_dataset_rele(origin, FTAG); 3344 return (ret); 3345 } 3346 3347 3348 void 3349 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 3350 { 3351 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3352 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 3353 } 3354