dsl_dataset.c revision 265744
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 * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org> 24 * Copyright (c) 2013 by Delphix. All rights reserved. 25 * Copyright (c) 2014, Joyent, Inc. All rights reserved. 26 * Copyright (c) 2014 RackTop Systems. 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#define SWITCH64(x, y) \ 55 { \ 56 uint64_t __tmp = (x); \ 57 (x) = (y); \ 58 (y) = __tmp; \ 59 } 60 61#define DS_REF_MAX (1ULL << 62) 62 63#define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE 64 65/* 66 * Figure out how much of this delta should be propogated to the dsl_dir 67 * layer. If there's a refreservation, that space has already been 68 * partially accounted for in our ancestors. 69 */ 70static int64_t 71parent_delta(dsl_dataset_t *ds, int64_t delta) 72{ 73 uint64_t old_bytes, new_bytes; 74 75 if (ds->ds_reserved == 0) 76 return (delta); 77 78 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 79 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved); 80 81 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); 82 return (new_bytes - old_bytes); 83} 84 85void 86dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) 87{ 88 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 89 int compressed = BP_GET_PSIZE(bp); 90 int uncompressed = BP_GET_UCSIZE(bp); 91 int64_t delta; 92 93 dprintf_bp(bp, "ds=%p", ds); 94 95 ASSERT(dmu_tx_is_syncing(tx)); 96 /* It could have been compressed away to nothing */ 97 if (BP_IS_HOLE(bp)) 98 return; 99 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); 100 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp))); 101 if (ds == NULL) { 102 dsl_pool_mos_diduse_space(tx->tx_pool, 103 used, compressed, uncompressed); 104 return; 105 } 106 107 dmu_buf_will_dirty(ds->ds_dbuf, tx); 108 mutex_enter(&ds->ds_lock); 109 delta = parent_delta(ds, used); 110 ds->ds_phys->ds_referenced_bytes += used; 111 ds->ds_phys->ds_compressed_bytes += compressed; 112 ds->ds_phys->ds_uncompressed_bytes += uncompressed; 113 ds->ds_phys->ds_unique_bytes += used; 114 mutex_exit(&ds->ds_lock); 115 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, 116 compressed, uncompressed, tx); 117 dsl_dir_transfer_space(ds->ds_dir, used - delta, 118 DD_USED_REFRSRV, DD_USED_HEAD, tx); 119} 120 121int 122dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, 123 boolean_t async) 124{ 125 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); 126 int compressed = BP_GET_PSIZE(bp); 127 int uncompressed = BP_GET_UCSIZE(bp); 128 129 if (BP_IS_HOLE(bp)) 130 return (0); 131 132 ASSERT(dmu_tx_is_syncing(tx)); 133 ASSERT(bp->blk_birth <= tx->tx_txg); 134 135 if (ds == NULL) { 136 dsl_free(tx->tx_pool, tx->tx_txg, bp); 137 dsl_pool_mos_diduse_space(tx->tx_pool, 138 -used, -compressed, -uncompressed); 139 return (used); 140 } 141 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); 142 143 ASSERT(!dsl_dataset_is_snapshot(ds)); 144 dmu_buf_will_dirty(ds->ds_dbuf, tx); 145 146 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { 147 int64_t delta; 148 149 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); 150 dsl_free(tx->tx_pool, tx->tx_txg, bp); 151 152 mutex_enter(&ds->ds_lock); 153 ASSERT(ds->ds_phys->ds_unique_bytes >= used || 154 !DS_UNIQUE_IS_ACCURATE(ds)); 155 delta = parent_delta(ds, -used); 156 ds->ds_phys->ds_unique_bytes -= used; 157 mutex_exit(&ds->ds_lock); 158 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, 159 delta, -compressed, -uncompressed, tx); 160 dsl_dir_transfer_space(ds->ds_dir, -used - delta, 161 DD_USED_REFRSRV, DD_USED_HEAD, tx); 162 } else { 163 dprintf_bp(bp, "putting on dead list: %s", ""); 164 if (async) { 165 /* 166 * We are here as part of zio's write done callback, 167 * which means we're a zio interrupt thread. We can't 168 * call dsl_deadlist_insert() now because it may block 169 * waiting for I/O. Instead, put bp on the deferred 170 * queue and let dsl_pool_sync() finish the job. 171 */ 172 bplist_append(&ds->ds_pending_deadlist, bp); 173 } else { 174 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); 175 } 176 ASSERT3U(ds->ds_prev->ds_object, ==, 177 ds->ds_phys->ds_prev_snap_obj); 178 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); 179 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ 180 if (ds->ds_prev->ds_phys->ds_next_snap_obj == 181 ds->ds_object && bp->blk_birth > 182 ds->ds_prev->ds_phys->ds_prev_snap_txg) { 183 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 184 mutex_enter(&ds->ds_prev->ds_lock); 185 ds->ds_prev->ds_phys->ds_unique_bytes += used; 186 mutex_exit(&ds->ds_prev->ds_lock); 187 } 188 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { 189 dsl_dir_transfer_space(ds->ds_dir, used, 190 DD_USED_HEAD, DD_USED_SNAP, tx); 191 } 192 } 193 mutex_enter(&ds->ds_lock); 194 ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used); 195 ds->ds_phys->ds_referenced_bytes -= used; 196 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); 197 ds->ds_phys->ds_compressed_bytes -= compressed; 198 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); 199 ds->ds_phys->ds_uncompressed_bytes -= uncompressed; 200 mutex_exit(&ds->ds_lock); 201 202 return (used); 203} 204 205uint64_t 206dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) 207{ 208 uint64_t trysnap = 0; 209 210 if (ds == NULL) 211 return (0); 212 /* 213 * The snapshot creation could fail, but that would cause an 214 * incorrect FALSE return, which would only result in an 215 * overestimation of the amount of space that an operation would 216 * consume, which is OK. 217 * 218 * There's also a small window where we could miss a pending 219 * snapshot, because we could set the sync task in the quiescing 220 * phase. So this should only be used as a guess. 221 */ 222 if (ds->ds_trysnap_txg > 223 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) 224 trysnap = ds->ds_trysnap_txg; 225 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap)); 226} 227 228boolean_t 229dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp, 230 uint64_t blk_birth) 231{ 232 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) || 233 (bp != NULL && BP_IS_HOLE(bp))) 234 return (B_FALSE); 235 236 ddt_prefetch(dsl_dataset_get_spa(ds), bp); 237 238 return (B_TRUE); 239} 240 241/* ARGSUSED */ 242static void 243dsl_dataset_evict(dmu_buf_t *db, void *dsv) 244{ 245 dsl_dataset_t *ds = dsv; 246 247 ASSERT(ds->ds_owner == NULL); 248 249 unique_remove(ds->ds_fsid_guid); 250 251 if (ds->ds_objset != NULL) 252 dmu_objset_evict(ds->ds_objset); 253 254 if (ds->ds_prev) { 255 dsl_dataset_rele(ds->ds_prev, ds); 256 ds->ds_prev = NULL; 257 } 258 259 bplist_destroy(&ds->ds_pending_deadlist); 260 if (ds->ds_phys->ds_deadlist_obj != 0) 261 dsl_deadlist_close(&ds->ds_deadlist); 262 if (ds->ds_dir) 263 dsl_dir_rele(ds->ds_dir, ds); 264 265 ASSERT(!list_link_active(&ds->ds_synced_link)); 266 267 if (mutex_owned(&ds->ds_lock)) 268 mutex_exit(&ds->ds_lock); 269 mutex_destroy(&ds->ds_lock); 270 if (mutex_owned(&ds->ds_opening_lock)) 271 mutex_exit(&ds->ds_opening_lock); 272 mutex_destroy(&ds->ds_opening_lock); 273 refcount_destroy(&ds->ds_longholds); 274 275 kmem_free(ds, sizeof (dsl_dataset_t)); 276} 277 278int 279dsl_dataset_get_snapname(dsl_dataset_t *ds) 280{ 281 dsl_dataset_phys_t *headphys; 282 int err; 283 dmu_buf_t *headdbuf; 284 dsl_pool_t *dp = ds->ds_dir->dd_pool; 285 objset_t *mos = dp->dp_meta_objset; 286 287 if (ds->ds_snapname[0]) 288 return (0); 289 if (ds->ds_phys->ds_next_snap_obj == 0) 290 return (0); 291 292 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, 293 FTAG, &headdbuf); 294 if (err != 0) 295 return (err); 296 headphys = headdbuf->db_data; 297 err = zap_value_search(dp->dp_meta_objset, 298 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 299 dmu_buf_rele(headdbuf, FTAG); 300 return (err); 301} 302 303int 304dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 305{ 306 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 307 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 308 matchtype_t mt; 309 int err; 310 311 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 312 mt = MT_FIRST; 313 else 314 mt = MT_EXACT; 315 316 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 317 value, mt, NULL, 0, NULL); 318 if (err == ENOTSUP && mt == MT_FIRST) 319 err = zap_lookup(mos, snapobj, name, 8, 1, value); 320 return (err); 321} 322 323int 324dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 325 boolean_t adj_cnt) 326{ 327 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 328 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 329 matchtype_t mt; 330 int err; 331 332 dsl_dir_snap_cmtime_update(ds->ds_dir); 333 334 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 335 mt = MT_FIRST; 336 else 337 mt = MT_EXACT; 338 339 err = zap_remove_norm(mos, snapobj, name, mt, tx); 340 if (err == ENOTSUP && mt == MT_FIRST) 341 err = zap_remove(mos, snapobj, name, tx); 342 343 if (err == 0 && adj_cnt) 344 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 345 DD_FIELD_SNAPSHOT_COUNT, tx); 346 347 return (err); 348} 349 350int 351dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 352 dsl_dataset_t **dsp) 353{ 354 objset_t *mos = dp->dp_meta_objset; 355 dmu_buf_t *dbuf; 356 dsl_dataset_t *ds; 357 int err; 358 dmu_object_info_t doi; 359 360 ASSERT(dsl_pool_config_held(dp)); 361 362 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 363 if (err != 0) 364 return (err); 365 366 /* Make sure dsobj has the correct object type. */ 367 dmu_object_info_from_db(dbuf, &doi); 368 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 369 dmu_buf_rele(dbuf, tag); 370 return (SET_ERROR(EINVAL)); 371 } 372 373 ds = dmu_buf_get_user(dbuf); 374 if (ds == NULL) { 375 dsl_dataset_t *winner = NULL; 376 377 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 378 ds->ds_dbuf = dbuf; 379 ds->ds_object = dsobj; 380 ds->ds_phys = dbuf->db_data; 381 382 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 383 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 384 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 385 refcount_create(&ds->ds_longholds); 386 387 bplist_create(&ds->ds_pending_deadlist); 388 dsl_deadlist_open(&ds->ds_deadlist, 389 mos, ds->ds_phys->ds_deadlist_obj); 390 391 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 392 offsetof(dmu_sendarg_t, dsa_link)); 393 394 if (err == 0) { 395 err = dsl_dir_hold_obj(dp, 396 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); 397 } 398 if (err != 0) { 399 mutex_destroy(&ds->ds_lock); 400 mutex_destroy(&ds->ds_opening_lock); 401 refcount_destroy(&ds->ds_longholds); 402 bplist_destroy(&ds->ds_pending_deadlist); 403 dsl_deadlist_close(&ds->ds_deadlist); 404 kmem_free(ds, sizeof (dsl_dataset_t)); 405 dmu_buf_rele(dbuf, tag); 406 return (err); 407 } 408 409 if (!dsl_dataset_is_snapshot(ds)) { 410 ds->ds_snapname[0] = '\0'; 411 if (ds->ds_phys->ds_prev_snap_obj != 0) { 412 err = dsl_dataset_hold_obj(dp, 413 ds->ds_phys->ds_prev_snap_obj, 414 ds, &ds->ds_prev); 415 } 416 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 417 int zaperr = zap_lookup(mos, ds->ds_object, 418 DS_FIELD_BOOKMARK_NAMES, 419 sizeof (ds->ds_bookmarks), 1, 420 &ds->ds_bookmarks); 421 if (zaperr != ENOENT) 422 VERIFY0(zaperr); 423 } 424 } else { 425 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 426 err = dsl_dataset_get_snapname(ds); 427 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) { 428 err = zap_count( 429 ds->ds_dir->dd_pool->dp_meta_objset, 430 ds->ds_phys->ds_userrefs_obj, 431 &ds->ds_userrefs); 432 } 433 } 434 435 if (err == 0 && !dsl_dataset_is_snapshot(ds)) { 436 err = dsl_prop_get_int_ds(ds, 437 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 438 &ds->ds_reserved); 439 if (err == 0) { 440 err = dsl_prop_get_int_ds(ds, 441 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 442 &ds->ds_quota); 443 } 444 } else { 445 ds->ds_reserved = ds->ds_quota = 0; 446 } 447 448 if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds, 449 &ds->ds_phys, dsl_dataset_evict)) != NULL) { 450 bplist_destroy(&ds->ds_pending_deadlist); 451 dsl_deadlist_close(&ds->ds_deadlist); 452 if (ds->ds_prev) 453 dsl_dataset_rele(ds->ds_prev, ds); 454 dsl_dir_rele(ds->ds_dir, ds); 455 mutex_destroy(&ds->ds_lock); 456 mutex_destroy(&ds->ds_opening_lock); 457 refcount_destroy(&ds->ds_longholds); 458 kmem_free(ds, sizeof (dsl_dataset_t)); 459 if (err != 0) { 460 dmu_buf_rele(dbuf, tag); 461 return (err); 462 } 463 ds = winner; 464 } else { 465 ds->ds_fsid_guid = 466 unique_insert(ds->ds_phys->ds_fsid_guid); 467 } 468 } 469 ASSERT3P(ds->ds_dbuf, ==, dbuf); 470 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); 471 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 || 472 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 473 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 474 *dsp = ds; 475 return (0); 476} 477 478int 479dsl_dataset_hold(dsl_pool_t *dp, const char *name, 480 void *tag, dsl_dataset_t **dsp) 481{ 482 dsl_dir_t *dd; 483 const char *snapname; 484 uint64_t obj; 485 int err = 0; 486 487 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 488 if (err != 0) 489 return (err); 490 491 ASSERT(dsl_pool_config_held(dp)); 492 obj = dd->dd_phys->dd_head_dataset_obj; 493 if (obj != 0) 494 err = dsl_dataset_hold_obj(dp, obj, tag, dsp); 495 else 496 err = SET_ERROR(ENOENT); 497 498 /* we may be looking for a snapshot */ 499 if (err == 0 && snapname != NULL) { 500 dsl_dataset_t *ds; 501 502 if (*snapname++ != '@') { 503 dsl_dataset_rele(*dsp, tag); 504 dsl_dir_rele(dd, FTAG); 505 return (SET_ERROR(ENOENT)); 506 } 507 508 dprintf("looking for snapshot '%s'\n", snapname); 509 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); 510 if (err == 0) 511 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 512 dsl_dataset_rele(*dsp, tag); 513 514 if (err == 0) { 515 mutex_enter(&ds->ds_lock); 516 if (ds->ds_snapname[0] == 0) 517 (void) strlcpy(ds->ds_snapname, snapname, 518 sizeof (ds->ds_snapname)); 519 mutex_exit(&ds->ds_lock); 520 *dsp = ds; 521 } 522 } 523 524 dsl_dir_rele(dd, FTAG); 525 return (err); 526} 527 528int 529dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 530 void *tag, dsl_dataset_t **dsp) 531{ 532 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 533 if (err != 0) 534 return (err); 535 if (!dsl_dataset_tryown(*dsp, tag)) { 536 dsl_dataset_rele(*dsp, tag); 537 *dsp = NULL; 538 return (SET_ERROR(EBUSY)); 539 } 540 return (0); 541} 542 543int 544dsl_dataset_own(dsl_pool_t *dp, const char *name, 545 void *tag, dsl_dataset_t **dsp) 546{ 547 int err = dsl_dataset_hold(dp, name, tag, dsp); 548 if (err != 0) 549 return (err); 550 if (!dsl_dataset_tryown(*dsp, tag)) { 551 dsl_dataset_rele(*dsp, tag); 552 return (SET_ERROR(EBUSY)); 553 } 554 return (0); 555} 556 557/* 558 * See the comment above dsl_pool_hold() for details. In summary, a long 559 * hold is used to prevent destruction of a dataset while the pool hold 560 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 561 * 562 * The dataset and pool must be held when this function is called. After it 563 * is called, the pool hold may be released while the dataset is still held 564 * and accessed. 565 */ 566void 567dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 568{ 569 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 570 (void) refcount_add(&ds->ds_longholds, tag); 571} 572 573void 574dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 575{ 576 (void) refcount_remove(&ds->ds_longholds, tag); 577} 578 579/* Return B_TRUE if there are any long holds on this dataset. */ 580boolean_t 581dsl_dataset_long_held(dsl_dataset_t *ds) 582{ 583 return (!refcount_is_zero(&ds->ds_longholds)); 584} 585 586void 587dsl_dataset_name(dsl_dataset_t *ds, char *name) 588{ 589 if (ds == NULL) { 590 (void) strcpy(name, "mos"); 591 } else { 592 dsl_dir_name(ds->ds_dir, name); 593 VERIFY0(dsl_dataset_get_snapname(ds)); 594 if (ds->ds_snapname[0]) { 595 (void) strcat(name, "@"); 596 /* 597 * We use a "recursive" mutex so that we 598 * can call dprintf_ds() with ds_lock held. 599 */ 600 if (!MUTEX_HELD(&ds->ds_lock)) { 601 mutex_enter(&ds->ds_lock); 602 (void) strcat(name, ds->ds_snapname); 603 mutex_exit(&ds->ds_lock); 604 } else { 605 (void) strcat(name, ds->ds_snapname); 606 } 607 } 608 } 609} 610 611void 612dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 613{ 614 dmu_buf_rele(ds->ds_dbuf, tag); 615} 616 617void 618dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 619{ 620 ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL); 621 622 mutex_enter(&ds->ds_lock); 623 ds->ds_owner = NULL; 624 mutex_exit(&ds->ds_lock); 625 dsl_dataset_long_rele(ds, tag); 626 if (ds->ds_dbuf != NULL) 627 dsl_dataset_rele(ds, tag); 628 else 629 dsl_dataset_evict(NULL, ds); 630} 631 632boolean_t 633dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 634{ 635 boolean_t gotit = FALSE; 636 637 mutex_enter(&ds->ds_lock); 638 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 639 ds->ds_owner = tag; 640 dsl_dataset_long_hold(ds, tag); 641 gotit = TRUE; 642 } 643 mutex_exit(&ds->ds_lock); 644 return (gotit); 645} 646 647uint64_t 648dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 649 uint64_t flags, dmu_tx_t *tx) 650{ 651 dsl_pool_t *dp = dd->dd_pool; 652 dmu_buf_t *dbuf; 653 dsl_dataset_phys_t *dsphys; 654 uint64_t dsobj; 655 objset_t *mos = dp->dp_meta_objset; 656 657 if (origin == NULL) 658 origin = dp->dp_origin_snap; 659 660 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 661 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0); 662 ASSERT(dmu_tx_is_syncing(tx)); 663 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 664 665 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 666 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 667 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 668 dmu_buf_will_dirty(dbuf, tx); 669 dsphys = dbuf->db_data; 670 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 671 dsphys->ds_dir_obj = dd->dd_object; 672 dsphys->ds_flags = flags; 673 dsphys->ds_fsid_guid = unique_create(); 674 do { 675 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 676 sizeof (dsphys->ds_guid)); 677 } while (dsphys->ds_guid == 0); 678 dsphys->ds_snapnames_zapobj = 679 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 680 DMU_OT_NONE, 0, tx); 681 dsphys->ds_creation_time = gethrestime_sec(); 682 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 683 684 if (origin == NULL) { 685 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 686 } else { 687 dsl_dataset_t *ohds; /* head of the origin snapshot */ 688 689 dsphys->ds_prev_snap_obj = origin->ds_object; 690 dsphys->ds_prev_snap_txg = 691 origin->ds_phys->ds_creation_txg; 692 dsphys->ds_referenced_bytes = 693 origin->ds_phys->ds_referenced_bytes; 694 dsphys->ds_compressed_bytes = 695 origin->ds_phys->ds_compressed_bytes; 696 dsphys->ds_uncompressed_bytes = 697 origin->ds_phys->ds_uncompressed_bytes; 698 dsphys->ds_bp = origin->ds_phys->ds_bp; 699 dsphys->ds_flags |= origin->ds_phys->ds_flags; 700 701 dmu_buf_will_dirty(origin->ds_dbuf, tx); 702 origin->ds_phys->ds_num_children++; 703 704 VERIFY0(dsl_dataset_hold_obj(dp, 705 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds)); 706 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 707 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 708 dsl_dataset_rele(ohds, FTAG); 709 710 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 711 if (origin->ds_phys->ds_next_clones_obj == 0) { 712 origin->ds_phys->ds_next_clones_obj = 713 zap_create(mos, 714 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 715 } 716 VERIFY0(zap_add_int(mos, 717 origin->ds_phys->ds_next_clones_obj, dsobj, tx)); 718 } 719 720 dmu_buf_will_dirty(dd->dd_dbuf, tx); 721 dd->dd_phys->dd_origin_obj = origin->ds_object; 722 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 723 if (origin->ds_dir->dd_phys->dd_clones == 0) { 724 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 725 origin->ds_dir->dd_phys->dd_clones = 726 zap_create(mos, 727 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 728 } 729 VERIFY0(zap_add_int(mos, 730 origin->ds_dir->dd_phys->dd_clones, dsobj, tx)); 731 } 732 } 733 734 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 735 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 736 737 dmu_buf_rele(dbuf, FTAG); 738 739 dmu_buf_will_dirty(dd->dd_dbuf, tx); 740 dd->dd_phys->dd_head_dataset_obj = dsobj; 741 742 return (dsobj); 743} 744 745static void 746dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 747{ 748 objset_t *os; 749 750 VERIFY0(dmu_objset_from_ds(ds, &os)); 751 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 752 dsl_dataset_dirty(ds, tx); 753} 754 755uint64_t 756dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 757 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 758{ 759 dsl_pool_t *dp = pdd->dd_pool; 760 uint64_t dsobj, ddobj; 761 dsl_dir_t *dd; 762 763 ASSERT(dmu_tx_is_syncing(tx)); 764 ASSERT(lastname[0] != '@'); 765 766 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 767 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 768 769 dsobj = dsl_dataset_create_sync_dd(dd, origin, 770 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 771 772 dsl_deleg_set_create_perms(dd, tx, cr); 773 774 /* 775 * Since we're creating a new node we know it's a leaf, so we can 776 * initialize the counts if the limit feature is active. 777 */ 778 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 779 uint64_t cnt = 0; 780 objset_t *os = dd->dd_pool->dp_meta_objset; 781 782 dsl_dir_zapify(dd, tx); 783 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 784 sizeof (cnt), 1, &cnt, tx)); 785 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 786 sizeof (cnt), 1, &cnt, tx)); 787 } 788 789 dsl_dir_rele(dd, FTAG); 790 791 /* 792 * If we are creating a clone, make sure we zero out any stale 793 * data from the origin snapshots zil header. 794 */ 795 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 796 dsl_dataset_t *ds; 797 798 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 799 dsl_dataset_zero_zil(ds, tx); 800 dsl_dataset_rele(ds, FTAG); 801 } 802 803 return (dsobj); 804} 805 806#ifdef __FreeBSD__ 807/* FreeBSD ioctl compat begin */ 808struct destroyarg { 809 nvlist_t *nvl; 810 const char *snapname; 811}; 812 813static int 814dsl_check_snap_cb(const char *name, void *arg) 815{ 816 struct destroyarg *da = arg; 817 dsl_dataset_t *ds; 818 char *dsname; 819 820 dsname = kmem_asprintf("%s@%s", name, da->snapname); 821 fnvlist_add_boolean(da->nvl, dsname); 822 kmem_free(dsname, strlen(dsname) + 1); 823 824 return (0); 825} 826 827int 828dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname, 829 nvlist_t *snaps) 830{ 831 struct destroyarg *da; 832 int err; 833 834 da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP); 835 da->nvl = snaps; 836 da->snapname = snapname; 837 err = dmu_objset_find(fsname, dsl_check_snap_cb, da, 838 DS_FIND_CHILDREN); 839 kmem_free(da, sizeof (struct destroyarg)); 840 841 return (err); 842} 843/* FreeBSD ioctl compat end */ 844#endif /* __FreeBSD__ */ 845 846/* 847 * The unique space in the head dataset can be calculated by subtracting 848 * the space used in the most recent snapshot, that is still being used 849 * in this file system, from the space currently in use. To figure out 850 * the space in the most recent snapshot still in use, we need to take 851 * the total space used in the snapshot and subtract out the space that 852 * has been freed up since the snapshot was taken. 853 */ 854void 855dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 856{ 857 uint64_t mrs_used; 858 uint64_t dlused, dlcomp, dluncomp; 859 860 ASSERT(!dsl_dataset_is_snapshot(ds)); 861 862 if (ds->ds_phys->ds_prev_snap_obj != 0) 863 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes; 864 else 865 mrs_used = 0; 866 867 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 868 869 ASSERT3U(dlused, <=, mrs_used); 870 ds->ds_phys->ds_unique_bytes = 871 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused); 872 873 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 874 SPA_VERSION_UNIQUE_ACCURATE) 875 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 876} 877 878void 879dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 880 dmu_tx_t *tx) 881{ 882 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 883 uint64_t count; 884 int err; 885 886 ASSERT(ds->ds_phys->ds_num_children >= 2); 887 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx); 888 /* 889 * The err should not be ENOENT, but a bug in a previous version 890 * of the code could cause upgrade_clones_cb() to not set 891 * ds_next_snap_obj when it should, leading to a missing entry. 892 * If we knew that the pool was created after 893 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 894 * ENOENT. However, at least we can check that we don't have 895 * too many entries in the next_clones_obj even after failing to 896 * remove this one. 897 */ 898 if (err != ENOENT) 899 VERIFY0(err); 900 ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj, 901 &count)); 902 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2); 903} 904 905 906blkptr_t * 907dsl_dataset_get_blkptr(dsl_dataset_t *ds) 908{ 909 return (&ds->ds_phys->ds_bp); 910} 911 912void 913dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 914{ 915 ASSERT(dmu_tx_is_syncing(tx)); 916 /* If it's the meta-objset, set dp_meta_rootbp */ 917 if (ds == NULL) { 918 tx->tx_pool->dp_meta_rootbp = *bp; 919 } else { 920 dmu_buf_will_dirty(ds->ds_dbuf, tx); 921 ds->ds_phys->ds_bp = *bp; 922 } 923} 924 925spa_t * 926dsl_dataset_get_spa(dsl_dataset_t *ds) 927{ 928 return (ds->ds_dir->dd_pool->dp_spa); 929} 930 931void 932dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 933{ 934 dsl_pool_t *dp; 935 936 if (ds == NULL) /* this is the meta-objset */ 937 return; 938 939 ASSERT(ds->ds_objset != NULL); 940 941 if (ds->ds_phys->ds_next_snap_obj != 0) 942 panic("dirtying snapshot!"); 943 944 dp = ds->ds_dir->dd_pool; 945 946 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 947 /* up the hold count until we can be written out */ 948 dmu_buf_add_ref(ds->ds_dbuf, ds); 949 } 950} 951 952boolean_t 953dsl_dataset_is_dirty(dsl_dataset_t *ds) 954{ 955 for (int t = 0; t < TXG_SIZE; t++) { 956 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 957 ds, t)) 958 return (B_TRUE); 959 } 960 return (B_FALSE); 961} 962 963static int 964dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 965{ 966 uint64_t asize; 967 968 if (!dmu_tx_is_syncing(tx)) 969 return (0); 970 971 /* 972 * If there's an fs-only reservation, any blocks that might become 973 * owned by the snapshot dataset must be accommodated by space 974 * outside of the reservation. 975 */ 976 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 977 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 978 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 979 return (SET_ERROR(ENOSPC)); 980 981 /* 982 * Propagate any reserved space for this snapshot to other 983 * snapshot checks in this sync group. 984 */ 985 if (asize > 0) 986 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 987 988 return (0); 989} 990 991typedef struct dsl_dataset_snapshot_arg { 992 nvlist_t *ddsa_snaps; 993 nvlist_t *ddsa_props; 994 nvlist_t *ddsa_errors; 995 cred_t *ddsa_cr; 996} dsl_dataset_snapshot_arg_t; 997 998int 999dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1000 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 1001{ 1002 int error; 1003 uint64_t value; 1004 1005 ds->ds_trysnap_txg = tx->tx_txg; 1006 1007 if (!dmu_tx_is_syncing(tx)) 1008 return (0); 1009 1010 /* 1011 * We don't allow multiple snapshots of the same txg. If there 1012 * is already one, try again. 1013 */ 1014 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) 1015 return (SET_ERROR(EAGAIN)); 1016 1017 /* 1018 * Check for conflicting snapshot name. 1019 */ 1020 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1021 if (error == 0) 1022 return (SET_ERROR(EEXIST)); 1023 if (error != ENOENT) 1024 return (error); 1025 1026 /* 1027 * We don't allow taking snapshots of inconsistent datasets, such as 1028 * those into which we are currently receiving. However, if we are 1029 * creating this snapshot as part of a receive, this check will be 1030 * executed atomically with respect to the completion of the receive 1031 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1032 * case we ignore this, knowing it will be fixed up for us shortly in 1033 * dmu_recv_end_sync(). 1034 */ 1035 if (!recv && DS_IS_INCONSISTENT(ds)) 1036 return (SET_ERROR(EBUSY)); 1037 1038 /* 1039 * Skip the check for temporary snapshots or if we have already checked 1040 * the counts in dsl_dataset_snapshot_check. This means we really only 1041 * check the count here when we're receiving a stream. 1042 */ 1043 if (cnt != 0 && cr != NULL) { 1044 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1045 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 1046 if (error != 0) 1047 return (error); 1048 } 1049 1050 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1051 if (error != 0) 1052 return (error); 1053 1054 return (0); 1055} 1056 1057static int 1058dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1059{ 1060 dsl_dataset_snapshot_arg_t *ddsa = arg; 1061 dsl_pool_t *dp = dmu_tx_pool(tx); 1062 nvpair_t *pair; 1063 int rv = 0; 1064 1065 /* 1066 * Pre-compute how many total new snapshots will be created for each 1067 * level in the tree and below. This is needed for validating the 1068 * snapshot limit when either taking a recursive snapshot or when 1069 * taking multiple snapshots. 1070 * 1071 * The problem is that the counts are not actually adjusted when 1072 * we are checking, only when we finally sync. For a single snapshot, 1073 * this is easy, the count will increase by 1 at each node up the tree, 1074 * but its more complicated for the recursive/multiple snapshot case. 1075 * 1076 * The dsl_fs_ss_limit_check function does recursively check the count 1077 * at each level up the tree but since it is validating each snapshot 1078 * independently we need to be sure that we are validating the complete 1079 * count for the entire set of snapshots. We do this by rolling up the 1080 * counts for each component of the name into an nvlist and then 1081 * checking each of those cases with the aggregated count. 1082 * 1083 * This approach properly handles not only the recursive snapshot 1084 * case (where we get all of those on the ddsa_snaps list) but also 1085 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1086 * validate the limit on 'a' using a count of 2). 1087 * 1088 * We validate the snapshot names in the third loop and only report 1089 * name errors once. 1090 */ 1091 if (dmu_tx_is_syncing(tx)) { 1092 nvlist_t *cnt_track = NULL; 1093 cnt_track = fnvlist_alloc(); 1094 1095 /* Rollup aggregated counts into the cnt_track list */ 1096 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1097 pair != NULL; 1098 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1099 char *pdelim; 1100 uint64_t val; 1101 char nm[MAXPATHLEN]; 1102 1103 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1104 pdelim = strchr(nm, '@'); 1105 if (pdelim == NULL) 1106 continue; 1107 *pdelim = '\0'; 1108 1109 do { 1110 if (nvlist_lookup_uint64(cnt_track, nm, 1111 &val) == 0) { 1112 /* update existing entry */ 1113 fnvlist_add_uint64(cnt_track, nm, 1114 val + 1); 1115 } else { 1116 /* add to list */ 1117 fnvlist_add_uint64(cnt_track, nm, 1); 1118 } 1119 1120 pdelim = strrchr(nm, '/'); 1121 if (pdelim != NULL) 1122 *pdelim = '\0'; 1123 } while (pdelim != NULL); 1124 } 1125 1126 /* Check aggregated counts at each level */ 1127 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1128 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1129 int error = 0; 1130 char *name; 1131 uint64_t cnt = 0; 1132 dsl_dataset_t *ds; 1133 1134 name = nvpair_name(pair); 1135 cnt = fnvpair_value_uint64(pair); 1136 ASSERT(cnt > 0); 1137 1138 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1139 if (error == 0) { 1140 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1141 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1142 ddsa->ddsa_cr); 1143 dsl_dataset_rele(ds, FTAG); 1144 } 1145 1146 if (error != 0) { 1147 if (ddsa->ddsa_errors != NULL) 1148 fnvlist_add_int32(ddsa->ddsa_errors, 1149 name, error); 1150 rv = error; 1151 /* only report one error for this check */ 1152 break; 1153 } 1154 } 1155 nvlist_free(cnt_track); 1156 } 1157 1158 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1159 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1160 int error = 0; 1161 dsl_dataset_t *ds; 1162 char *name, *atp; 1163 char dsname[MAXNAMELEN]; 1164 1165 name = nvpair_name(pair); 1166 if (strlen(name) >= MAXNAMELEN) 1167 error = SET_ERROR(ENAMETOOLONG); 1168 if (error == 0) { 1169 atp = strchr(name, '@'); 1170 if (atp == NULL) 1171 error = SET_ERROR(EINVAL); 1172 if (error == 0) 1173 (void) strlcpy(dsname, name, atp - name + 1); 1174 } 1175 if (error == 0) 1176 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1177 if (error == 0) { 1178 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1179 error = dsl_dataset_snapshot_check_impl(ds, 1180 atp + 1, tx, B_FALSE, 0, NULL); 1181 dsl_dataset_rele(ds, FTAG); 1182 } 1183 1184 if (error != 0) { 1185 if (ddsa->ddsa_errors != NULL) { 1186 fnvlist_add_int32(ddsa->ddsa_errors, 1187 name, error); 1188 } 1189 rv = error; 1190 } 1191 } 1192 1193 return (rv); 1194} 1195 1196void 1197dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1198 dmu_tx_t *tx) 1199{ 1200 static zil_header_t zero_zil; 1201 1202 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1203 dmu_buf_t *dbuf; 1204 dsl_dataset_phys_t *dsphys; 1205 uint64_t dsobj, crtxg; 1206 objset_t *mos = dp->dp_meta_objset; 1207 objset_t *os; 1208 1209 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1210 1211 /* 1212 * If we are on an old pool, the zil must not be active, in which 1213 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1214 */ 1215 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1216 dmu_objset_from_ds(ds, &os) != 0 || 1217 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1218 sizeof (zero_zil)) == 0); 1219 1220 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1221 1222 /* 1223 * The origin's ds_creation_txg has to be < TXG_INITIAL 1224 */ 1225 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1226 crtxg = 1; 1227 else 1228 crtxg = tx->tx_txg; 1229 1230 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1231 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1232 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1233 dmu_buf_will_dirty(dbuf, tx); 1234 dsphys = dbuf->db_data; 1235 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1236 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1237 dsphys->ds_fsid_guid = unique_create(); 1238 do { 1239 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1240 sizeof (dsphys->ds_guid)); 1241 } while (dsphys->ds_guid == 0); 1242 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; 1243 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; 1244 dsphys->ds_next_snap_obj = ds->ds_object; 1245 dsphys->ds_num_children = 1; 1246 dsphys->ds_creation_time = gethrestime_sec(); 1247 dsphys->ds_creation_txg = crtxg; 1248 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; 1249 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes; 1250 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; 1251 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; 1252 dsphys->ds_flags = ds->ds_phys->ds_flags; 1253 dsphys->ds_bp = ds->ds_phys->ds_bp; 1254 dmu_buf_rele(dbuf, FTAG); 1255 1256 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); 1257 if (ds->ds_prev) { 1258 uint64_t next_clones_obj = 1259 ds->ds_prev->ds_phys->ds_next_clones_obj; 1260 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == 1261 ds->ds_object || 1262 ds->ds_prev->ds_phys->ds_num_children > 1); 1263 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { 1264 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1265 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1266 ds->ds_prev->ds_phys->ds_creation_txg); 1267 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; 1268 } else if (next_clones_obj != 0) { 1269 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1270 dsphys->ds_next_snap_obj, tx); 1271 VERIFY0(zap_add_int(mos, 1272 next_clones_obj, dsobj, tx)); 1273 } 1274 } 1275 1276 /* 1277 * If we have a reference-reservation on this dataset, we will 1278 * need to increase the amount of refreservation being charged 1279 * since our unique space is going to zero. 1280 */ 1281 if (ds->ds_reserved) { 1282 int64_t delta; 1283 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1284 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 1285 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1286 delta, 0, 0, tx); 1287 } 1288 1289 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1290 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist, 1291 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx); 1292 dsl_deadlist_close(&ds->ds_deadlist); 1293 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); 1294 dsl_deadlist_add_key(&ds->ds_deadlist, 1295 ds->ds_phys->ds_prev_snap_txg, tx); 1296 1297 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg); 1298 ds->ds_phys->ds_prev_snap_obj = dsobj; 1299 ds->ds_phys->ds_prev_snap_txg = crtxg; 1300 ds->ds_phys->ds_unique_bytes = 0; 1301 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1302 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1303 1304 VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, 1305 snapname, 8, 1, &dsobj, tx)); 1306 1307 if (ds->ds_prev) 1308 dsl_dataset_rele(ds->ds_prev, ds); 1309 VERIFY0(dsl_dataset_hold_obj(dp, 1310 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 1311 1312 dsl_scan_ds_snapshotted(ds, tx); 1313 1314 dsl_dir_snap_cmtime_update(ds->ds_dir); 1315 1316 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1317} 1318 1319static void 1320dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1321{ 1322 dsl_dataset_snapshot_arg_t *ddsa = arg; 1323 dsl_pool_t *dp = dmu_tx_pool(tx); 1324 nvpair_t *pair; 1325 1326 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1327 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1328 dsl_dataset_t *ds; 1329 char *name, *atp; 1330 char dsname[MAXNAMELEN]; 1331 1332 name = nvpair_name(pair); 1333 atp = strchr(name, '@'); 1334 (void) strlcpy(dsname, name, atp - name + 1); 1335 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1336 1337 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1338 if (ddsa->ddsa_props != NULL) { 1339 dsl_props_set_sync_impl(ds->ds_prev, 1340 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1341 } 1342 dsl_dataset_rele(ds, FTAG); 1343 } 1344} 1345 1346/* 1347 * The snapshots must all be in the same pool. 1348 * All-or-nothing: if there are any failures, nothing will be modified. 1349 */ 1350int 1351dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1352{ 1353 dsl_dataset_snapshot_arg_t ddsa; 1354 nvpair_t *pair; 1355 boolean_t needsuspend; 1356 int error; 1357 spa_t *spa; 1358 char *firstname; 1359 nvlist_t *suspended = NULL; 1360 1361 pair = nvlist_next_nvpair(snaps, NULL); 1362 if (pair == NULL) 1363 return (0); 1364 firstname = nvpair_name(pair); 1365 1366 error = spa_open(firstname, &spa, FTAG); 1367 if (error != 0) 1368 return (error); 1369 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1370 spa_close(spa, FTAG); 1371 1372 if (needsuspend) { 1373 suspended = fnvlist_alloc(); 1374 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1375 pair = nvlist_next_nvpair(snaps, pair)) { 1376 char fsname[MAXNAMELEN]; 1377 char *snapname = nvpair_name(pair); 1378 char *atp; 1379 void *cookie; 1380 1381 atp = strchr(snapname, '@'); 1382 if (atp == NULL) { 1383 error = SET_ERROR(EINVAL); 1384 break; 1385 } 1386 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1387 1388 error = zil_suspend(fsname, &cookie); 1389 if (error != 0) 1390 break; 1391 fnvlist_add_uint64(suspended, fsname, 1392 (uintptr_t)cookie); 1393 } 1394 } 1395 1396 ddsa.ddsa_snaps = snaps; 1397 ddsa.ddsa_props = props; 1398 ddsa.ddsa_errors = errors; 1399 ddsa.ddsa_cr = CRED(); 1400 1401 if (error == 0) { 1402 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1403 dsl_dataset_snapshot_sync, &ddsa, 1404 fnvlist_num_pairs(snaps) * 3); 1405 } 1406 1407 if (suspended != NULL) { 1408 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1409 pair = nvlist_next_nvpair(suspended, pair)) { 1410 zil_resume((void *)(uintptr_t) 1411 fnvpair_value_uint64(pair)); 1412 } 1413 fnvlist_free(suspended); 1414 } 1415 1416#ifdef __FreeBSD__ 1417#ifdef _KERNEL 1418 if (error == 0) { 1419 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1420 pair = nvlist_next_nvpair(snaps, pair)) { 1421 char *snapname = nvpair_name(pair); 1422 zvol_create_minors(snapname); 1423 } 1424 } 1425#endif 1426#endif 1427 return (error); 1428} 1429 1430typedef struct dsl_dataset_snapshot_tmp_arg { 1431 const char *ddsta_fsname; 1432 const char *ddsta_snapname; 1433 minor_t ddsta_cleanup_minor; 1434 const char *ddsta_htag; 1435} dsl_dataset_snapshot_tmp_arg_t; 1436 1437static int 1438dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1439{ 1440 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1441 dsl_pool_t *dp = dmu_tx_pool(tx); 1442 dsl_dataset_t *ds; 1443 int error; 1444 1445 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1446 if (error != 0) 1447 return (error); 1448 1449 /* NULL cred means no limit check for tmp snapshot */ 1450 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1451 tx, B_FALSE, 0, NULL); 1452 if (error != 0) { 1453 dsl_dataset_rele(ds, FTAG); 1454 return (error); 1455 } 1456 1457 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1458 dsl_dataset_rele(ds, FTAG); 1459 return (SET_ERROR(ENOTSUP)); 1460 } 1461 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1462 B_TRUE, tx); 1463 if (error != 0) { 1464 dsl_dataset_rele(ds, FTAG); 1465 return (error); 1466 } 1467 1468 dsl_dataset_rele(ds, FTAG); 1469 return (0); 1470} 1471 1472static void 1473dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1474{ 1475 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1476 dsl_pool_t *dp = dmu_tx_pool(tx); 1477 dsl_dataset_t *ds; 1478 1479 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1480 1481 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1482 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1483 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1484 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1485 1486 dsl_dataset_rele(ds, FTAG); 1487} 1488 1489int 1490dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1491 minor_t cleanup_minor, const char *htag) 1492{ 1493 dsl_dataset_snapshot_tmp_arg_t ddsta; 1494 int error; 1495 spa_t *spa; 1496 boolean_t needsuspend; 1497 void *cookie; 1498 1499 ddsta.ddsta_fsname = fsname; 1500 ddsta.ddsta_snapname = snapname; 1501 ddsta.ddsta_cleanup_minor = cleanup_minor; 1502 ddsta.ddsta_htag = htag; 1503 1504 error = spa_open(fsname, &spa, FTAG); 1505 if (error != 0) 1506 return (error); 1507 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1508 spa_close(spa, FTAG); 1509 1510 if (needsuspend) { 1511 error = zil_suspend(fsname, &cookie); 1512 if (error != 0) 1513 return (error); 1514 } 1515 1516 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1517 dsl_dataset_snapshot_tmp_sync, &ddsta, 3); 1518 1519 if (needsuspend) 1520 zil_resume(cookie); 1521 return (error); 1522} 1523 1524 1525void 1526dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1527{ 1528 ASSERT(dmu_tx_is_syncing(tx)); 1529 ASSERT(ds->ds_objset != NULL); 1530 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); 1531 1532 /* 1533 * in case we had to change ds_fsid_guid when we opened it, 1534 * sync it out now. 1535 */ 1536 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1537 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; 1538 1539 dmu_objset_sync(ds->ds_objset, zio, tx); 1540} 1541 1542static void 1543get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1544{ 1545 uint64_t count = 0; 1546 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1547 zap_cursor_t zc; 1548 zap_attribute_t za; 1549 nvlist_t *propval = fnvlist_alloc(); 1550 nvlist_t *val = fnvlist_alloc(); 1551 1552 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1553 1554 /* 1555 * There may be missing entries in ds_next_clones_obj 1556 * due to a bug in a previous version of the code. 1557 * Only trust it if it has the right number of entries. 1558 */ 1559 if (ds->ds_phys->ds_next_clones_obj != 0) { 1560 VERIFY0(zap_count(mos, ds->ds_phys->ds_next_clones_obj, 1561 &count)); 1562 } 1563 if (count != ds->ds_phys->ds_num_children - 1) 1564 goto fail; 1565 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj); 1566 zap_cursor_retrieve(&zc, &za) == 0; 1567 zap_cursor_advance(&zc)) { 1568 dsl_dataset_t *clone; 1569 char buf[ZFS_MAXNAMELEN]; 1570 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1571 za.za_first_integer, FTAG, &clone)); 1572 dsl_dir_name(clone->ds_dir, buf); 1573 fnvlist_add_boolean(val, buf); 1574 dsl_dataset_rele(clone, FTAG); 1575 } 1576 zap_cursor_fini(&zc); 1577 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1578 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); 1579fail: 1580 nvlist_free(val); 1581 nvlist_free(propval); 1582} 1583 1584void 1585dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1586{ 1587 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1588 uint64_t refd, avail, uobjs, aobjs, ratio; 1589 1590 ASSERT(dsl_pool_config_held(dp)); 1591 1592 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 : 1593 (ds->ds_phys->ds_uncompressed_bytes * 100 / 1594 ds->ds_phys->ds_compressed_bytes); 1595 1596 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); 1597 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 1598 ds->ds_phys->ds_uncompressed_bytes); 1599 1600 if (dsl_dataset_is_snapshot(ds)) { 1601 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); 1602 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 1603 ds->ds_phys->ds_unique_bytes); 1604 get_clones_stat(ds, nv); 1605 } else { 1606 dsl_dir_stats(ds->ds_dir, nv); 1607 } 1608 1609 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 1610 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 1611 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 1612 1613 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 1614 ds->ds_phys->ds_creation_time); 1615 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 1616 ds->ds_phys->ds_creation_txg); 1617 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 1618 ds->ds_quota); 1619 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 1620 ds->ds_reserved); 1621 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 1622 ds->ds_phys->ds_guid); 1623 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 1624 ds->ds_phys->ds_unique_bytes); 1625 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 1626 ds->ds_object); 1627 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 1628 ds->ds_userrefs); 1629 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 1630 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 1631 1632 if (ds->ds_phys->ds_prev_snap_obj != 0) { 1633 uint64_t written, comp, uncomp; 1634 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1635 dsl_dataset_t *prev; 1636 1637 int err = dsl_dataset_hold_obj(dp, 1638 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev); 1639 if (err == 0) { 1640 err = dsl_dataset_space_written(prev, ds, &written, 1641 &comp, &uncomp); 1642 dsl_dataset_rele(prev, FTAG); 1643 if (err == 0) { 1644 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 1645 written); 1646 } 1647 } 1648 } 1649} 1650 1651void 1652dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 1653{ 1654 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1655 ASSERT(dsl_pool_config_held(dp)); 1656 1657 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; 1658 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; 1659 stat->dds_guid = ds->ds_phys->ds_guid; 1660 stat->dds_origin[0] = '\0'; 1661 if (dsl_dataset_is_snapshot(ds)) { 1662 stat->dds_is_snapshot = B_TRUE; 1663 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; 1664 } else { 1665 stat->dds_is_snapshot = B_FALSE; 1666 stat->dds_num_clones = 0; 1667 1668 if (dsl_dir_is_clone(ds->ds_dir)) { 1669 dsl_dataset_t *ods; 1670 1671 VERIFY0(dsl_dataset_hold_obj(dp, 1672 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods)); 1673 dsl_dataset_name(ods, stat->dds_origin); 1674 dsl_dataset_rele(ods, FTAG); 1675 } 1676 } 1677} 1678 1679uint64_t 1680dsl_dataset_fsid_guid(dsl_dataset_t *ds) 1681{ 1682 return (ds->ds_fsid_guid); 1683} 1684 1685void 1686dsl_dataset_space(dsl_dataset_t *ds, 1687 uint64_t *refdbytesp, uint64_t *availbytesp, 1688 uint64_t *usedobjsp, uint64_t *availobjsp) 1689{ 1690 *refdbytesp = ds->ds_phys->ds_referenced_bytes; 1691 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 1692 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) 1693 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes; 1694 if (ds->ds_quota != 0) { 1695 /* 1696 * Adjust available bytes according to refquota 1697 */ 1698 if (*refdbytesp < ds->ds_quota) 1699 *availbytesp = MIN(*availbytesp, 1700 ds->ds_quota - *refdbytesp); 1701 else 1702 *availbytesp = 0; 1703 } 1704 *usedobjsp = ds->ds_phys->ds_bp.blk_fill; 1705 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 1706} 1707 1708boolean_t 1709dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 1710{ 1711 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1712 1713 ASSERT(dsl_pool_config_held(dp)); 1714 if (snap == NULL) 1715 return (B_FALSE); 1716 if (ds->ds_phys->ds_bp.blk_birth > 1717 snap->ds_phys->ds_creation_txg) { 1718 objset_t *os, *os_snap; 1719 /* 1720 * It may be that only the ZIL differs, because it was 1721 * reset in the head. Don't count that as being 1722 * modified. 1723 */ 1724 if (dmu_objset_from_ds(ds, &os) != 0) 1725 return (B_TRUE); 1726 if (dmu_objset_from_ds(snap, &os_snap) != 0) 1727 return (B_TRUE); 1728 return (bcmp(&os->os_phys->os_meta_dnode, 1729 &os_snap->os_phys->os_meta_dnode, 1730 sizeof (os->os_phys->os_meta_dnode)) != 0); 1731 } 1732 return (B_FALSE); 1733} 1734 1735typedef struct dsl_dataset_rename_snapshot_arg { 1736 const char *ddrsa_fsname; 1737 const char *ddrsa_oldsnapname; 1738 const char *ddrsa_newsnapname; 1739 boolean_t ddrsa_recursive; 1740 dmu_tx_t *ddrsa_tx; 1741} dsl_dataset_rename_snapshot_arg_t; 1742 1743/* ARGSUSED */ 1744static int 1745dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 1746 dsl_dataset_t *hds, void *arg) 1747{ 1748 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1749 int error; 1750 uint64_t val; 1751 1752 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1753 if (error != 0) { 1754 /* ignore nonexistent snapshots */ 1755 return (error == ENOENT ? 0 : error); 1756 } 1757 1758 /* new name should not exist */ 1759 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 1760 if (error == 0) 1761 error = SET_ERROR(EEXIST); 1762 else if (error == ENOENT) 1763 error = 0; 1764 1765 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 1766 if (dsl_dir_namelen(hds->ds_dir) + 1 + 1767 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN) 1768 error = SET_ERROR(ENAMETOOLONG); 1769 1770 return (error); 1771} 1772 1773static int 1774dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 1775{ 1776 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1777 dsl_pool_t *dp = dmu_tx_pool(tx); 1778 dsl_dataset_t *hds; 1779 int error; 1780 1781 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 1782 if (error != 0) 1783 return (error); 1784 1785 if (ddrsa->ddrsa_recursive) { 1786 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1787 dsl_dataset_rename_snapshot_check_impl, ddrsa, 1788 DS_FIND_CHILDREN); 1789 } else { 1790 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 1791 } 1792 dsl_dataset_rele(hds, FTAG); 1793 return (error); 1794} 1795 1796static int 1797dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 1798 dsl_dataset_t *hds, void *arg) 1799{ 1800#ifdef __FreeBSD__ 1801#ifdef _KERNEL 1802 char *oldname, *newname; 1803#endif 1804#endif 1805 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1806 dsl_dataset_t *ds; 1807 uint64_t val; 1808 dmu_tx_t *tx = ddrsa->ddrsa_tx; 1809 int error; 1810 1811 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1812 ASSERT(error == 0 || error == ENOENT); 1813 if (error == ENOENT) { 1814 /* ignore nonexistent snapshots */ 1815 return (0); 1816 } 1817 1818 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 1819 1820 /* log before we change the name */ 1821 spa_history_log_internal_ds(ds, "rename", tx, 1822 "-> @%s", ddrsa->ddrsa_newsnapname); 1823 1824 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 1825 B_FALSE)); 1826 mutex_enter(&ds->ds_lock); 1827 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 1828 mutex_exit(&ds->ds_lock); 1829 VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj, 1830 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 1831 1832#ifdef __FreeBSD__ 1833#ifdef _KERNEL 1834 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1835 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1836 snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname, 1837 ddrsa->ddrsa_oldsnapname); 1838 snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname, 1839 ddrsa->ddrsa_newsnapname); 1840 zfsvfs_update_fromname(oldname, newname); 1841 zvol_rename_minors(oldname, newname); 1842 kmem_free(newname, MAXPATHLEN); 1843 kmem_free(oldname, MAXPATHLEN); 1844#endif 1845#endif 1846 dsl_dataset_rele(ds, FTAG); 1847 1848 return (0); 1849} 1850 1851static void 1852dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 1853{ 1854 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1855 dsl_pool_t *dp = dmu_tx_pool(tx); 1856 dsl_dataset_t *hds; 1857 1858 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 1859 ddrsa->ddrsa_tx = tx; 1860 if (ddrsa->ddrsa_recursive) { 1861 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1862 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 1863 DS_FIND_CHILDREN)); 1864 } else { 1865 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 1866 } 1867 dsl_dataset_rele(hds, FTAG); 1868} 1869 1870int 1871dsl_dataset_rename_snapshot(const char *fsname, 1872 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 1873{ 1874 dsl_dataset_rename_snapshot_arg_t ddrsa; 1875 1876 ddrsa.ddrsa_fsname = fsname; 1877 ddrsa.ddrsa_oldsnapname = oldsnapname; 1878 ddrsa.ddrsa_newsnapname = newsnapname; 1879 ddrsa.ddrsa_recursive = recursive; 1880 1881 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 1882 dsl_dataset_rename_snapshot_sync, &ddrsa, 1)); 1883} 1884 1885/* 1886 * If we're doing an ownership handoff, we need to make sure that there is 1887 * only one long hold on the dataset. We're not allowed to change anything here 1888 * so we don't permanently release the long hold or regular hold here. We want 1889 * to do this only when syncing to avoid the dataset unexpectedly going away 1890 * when we release the long hold. 1891 */ 1892static int 1893dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 1894{ 1895 boolean_t held; 1896 1897 if (!dmu_tx_is_syncing(tx)) 1898 return (0); 1899 1900 if (owner != NULL) { 1901 VERIFY3P(ds->ds_owner, ==, owner); 1902 dsl_dataset_long_rele(ds, owner); 1903 } 1904 1905 held = dsl_dataset_long_held(ds); 1906 1907 if (owner != NULL) 1908 dsl_dataset_long_hold(ds, owner); 1909 1910 if (held) 1911 return (SET_ERROR(EBUSY)); 1912 1913 return (0); 1914} 1915 1916typedef struct dsl_dataset_rollback_arg { 1917 const char *ddra_fsname; 1918 void *ddra_owner; 1919 nvlist_t *ddra_result; 1920} dsl_dataset_rollback_arg_t; 1921 1922static int 1923dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 1924{ 1925 dsl_dataset_rollback_arg_t *ddra = arg; 1926 dsl_pool_t *dp = dmu_tx_pool(tx); 1927 dsl_dataset_t *ds; 1928 int64_t unused_refres_delta; 1929 int error; 1930 1931 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 1932 if (error != 0) 1933 return (error); 1934 1935 /* must not be a snapshot */ 1936 if (dsl_dataset_is_snapshot(ds)) { 1937 dsl_dataset_rele(ds, FTAG); 1938 return (SET_ERROR(EINVAL)); 1939 } 1940 1941 /* must have a most recent snapshot */ 1942 if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) { 1943 dsl_dataset_rele(ds, FTAG); 1944 return (SET_ERROR(EINVAL)); 1945 } 1946 1947 /* must not have any bookmarks after the most recent snapshot */ 1948 nvlist_t *proprequest = fnvlist_alloc(); 1949 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 1950 nvlist_t *bookmarks = fnvlist_alloc(); 1951 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 1952 fnvlist_free(proprequest); 1953 if (error != 0) 1954 return (error); 1955 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 1956 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 1957 nvlist_t *valuenv = 1958 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 1959 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 1960 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 1961 if (createtxg > ds->ds_phys->ds_prev_snap_txg) { 1962 fnvlist_free(bookmarks); 1963 dsl_dataset_rele(ds, FTAG); 1964 return (SET_ERROR(EEXIST)); 1965 } 1966 } 1967 fnvlist_free(bookmarks); 1968 1969 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 1970 if (error != 0) { 1971 dsl_dataset_rele(ds, FTAG); 1972 return (error); 1973 } 1974 1975 /* 1976 * Check if the snap we are rolling back to uses more than 1977 * the refquota. 1978 */ 1979 if (ds->ds_quota != 0 && 1980 ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) { 1981 dsl_dataset_rele(ds, FTAG); 1982 return (SET_ERROR(EDQUOT)); 1983 } 1984 1985 /* 1986 * When we do the clone swap, we will temporarily use more space 1987 * due to the refreservation (the head will no longer have any 1988 * unique space, so the entire amount of the refreservation will need 1989 * to be free). We will immediately destroy the clone, freeing 1990 * this space, but the freeing happens over many txg's. 1991 */ 1992 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 1993 ds->ds_phys->ds_unique_bytes); 1994 1995 if (unused_refres_delta > 0 && 1996 unused_refres_delta > 1997 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 1998 dsl_dataset_rele(ds, FTAG); 1999 return (SET_ERROR(ENOSPC)); 2000 } 2001 2002 dsl_dataset_rele(ds, FTAG); 2003 return (0); 2004} 2005 2006static void 2007dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 2008{ 2009 dsl_dataset_rollback_arg_t *ddra = arg; 2010 dsl_pool_t *dp = dmu_tx_pool(tx); 2011 dsl_dataset_t *ds, *clone; 2012 uint64_t cloneobj; 2013 char namebuf[ZFS_MAXNAMELEN]; 2014 2015 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 2016 2017 dsl_dataset_name(ds->ds_prev, namebuf); 2018 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 2019 2020 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 2021 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 2022 2023 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 2024 2025 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 2026 dsl_dataset_zero_zil(ds, tx); 2027 2028 dsl_destroy_head_sync_impl(clone, tx); 2029 2030 dsl_dataset_rele(clone, FTAG); 2031 dsl_dataset_rele(ds, FTAG); 2032} 2033 2034/* 2035 * Rolls back the given filesystem or volume to the most recent snapshot. 2036 * The name of the most recent snapshot will be returned under key "target" 2037 * in the result nvlist. 2038 * 2039 * If owner != NULL: 2040 * - The existing dataset MUST be owned by the specified owner at entry 2041 * - Upon return, dataset will still be held by the same owner, whether we 2042 * succeed or not. 2043 * 2044 * This mode is required any time the existing filesystem is mounted. See 2045 * notes above zfs_suspend_fs() for further details. 2046 */ 2047int 2048dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result) 2049{ 2050 dsl_dataset_rollback_arg_t ddra; 2051 2052 ddra.ddra_fsname = fsname; 2053 ddra.ddra_owner = owner; 2054 ddra.ddra_result = result; 2055 2056 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 2057 dsl_dataset_rollback_sync, &ddra, 1)); 2058} 2059 2060struct promotenode { 2061 list_node_t link; 2062 dsl_dataset_t *ds; 2063}; 2064 2065typedef struct dsl_dataset_promote_arg { 2066 const char *ddpa_clonename; 2067 dsl_dataset_t *ddpa_clone; 2068 list_t shared_snaps, origin_snaps, clone_snaps; 2069 dsl_dataset_t *origin_origin; /* origin of the origin */ 2070 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2071 char *err_ds; 2072 cred_t *cr; 2073} dsl_dataset_promote_arg_t; 2074 2075static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2076static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2077 void *tag); 2078static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2079 2080static int 2081dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2082{ 2083 dsl_dataset_promote_arg_t *ddpa = arg; 2084 dsl_pool_t *dp = dmu_tx_pool(tx); 2085 dsl_dataset_t *hds; 2086 struct promotenode *snap; 2087 dsl_dataset_t *origin_ds; 2088 int err; 2089 uint64_t unused; 2090 uint64_t ss_mv_cnt; 2091 2092 err = promote_hold(ddpa, dp, FTAG); 2093 if (err != 0) 2094 return (err); 2095 2096 hds = ddpa->ddpa_clone; 2097 2098 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) { 2099 promote_rele(ddpa, FTAG); 2100 return (SET_ERROR(EXDEV)); 2101 } 2102 2103 /* 2104 * Compute and check the amount of space to transfer. Since this is 2105 * so expensive, don't do the preliminary check. 2106 */ 2107 if (!dmu_tx_is_syncing(tx)) { 2108 promote_rele(ddpa, FTAG); 2109 return (0); 2110 } 2111 2112 snap = list_head(&ddpa->shared_snaps); 2113 origin_ds = snap->ds; 2114 2115 /* compute origin's new unique space */ 2116 snap = list_tail(&ddpa->clone_snaps); 2117 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2118 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2119 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2120 &ddpa->unique, &unused, &unused); 2121 2122 /* 2123 * Walk the snapshots that we are moving 2124 * 2125 * Compute space to transfer. Consider the incremental changes 2126 * to used by each snapshot: 2127 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2128 * So each snapshot gave birth to: 2129 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2130 * So a sequence would look like: 2131 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2132 * Which simplifies to: 2133 * uN + kN + kN-1 + ... + k1 + k0 2134 * Note however, if we stop before we reach the ORIGIN we get: 2135 * uN + kN + kN-1 + ... + kM - uM-1 2136 */ 2137 ss_mv_cnt = 0; 2138 ddpa->used = origin_ds->ds_phys->ds_referenced_bytes; 2139 ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes; 2140 ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; 2141 for (snap = list_head(&ddpa->shared_snaps); snap; 2142 snap = list_next(&ddpa->shared_snaps, snap)) { 2143 uint64_t val, dlused, dlcomp, dluncomp; 2144 dsl_dataset_t *ds = snap->ds; 2145 2146 ss_mv_cnt++; 2147 2148 /* 2149 * If there are long holds, we won't be able to evict 2150 * the objset. 2151 */ 2152 if (dsl_dataset_long_held(ds)) { 2153 err = SET_ERROR(EBUSY); 2154 goto out; 2155 } 2156 2157 /* Check that the snapshot name does not conflict */ 2158 VERIFY0(dsl_dataset_get_snapname(ds)); 2159 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2160 if (err == 0) { 2161 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname); 2162 err = SET_ERROR(EEXIST); 2163 goto out; 2164 } 2165 if (err != ENOENT) 2166 goto out; 2167 2168 /* The very first snapshot does not have a deadlist */ 2169 if (ds->ds_phys->ds_prev_snap_obj == 0) 2170 continue; 2171 2172 dsl_deadlist_space(&ds->ds_deadlist, 2173 &dlused, &dlcomp, &dluncomp); 2174 ddpa->used += dlused; 2175 ddpa->comp += dlcomp; 2176 ddpa->uncomp += dluncomp; 2177 } 2178 2179 /* 2180 * If we are a clone of a clone then we never reached ORIGIN, 2181 * so we need to subtract out the clone origin's used space. 2182 */ 2183 if (ddpa->origin_origin) { 2184 ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes; 2185 ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes; 2186 ddpa->uncomp -= 2187 ddpa->origin_origin->ds_phys->ds_uncompressed_bytes; 2188 } 2189 2190 /* Check that there is enough space and limit headroom here */ 2191 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2192 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2193 if (err != 0) 2194 goto out; 2195 2196 /* 2197 * Compute the amounts of space that will be used by snapshots 2198 * after the promotion (for both origin and clone). For each, 2199 * it is the amount of space that will be on all of their 2200 * deadlists (that was not born before their new origin). 2201 */ 2202 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2203 uint64_t space; 2204 2205 /* 2206 * Note, typically this will not be a clone of a clone, 2207 * so dd_origin_txg will be < TXG_INITIAL, so 2208 * these snaplist_space() -> dsl_deadlist_space_range() 2209 * calls will be fast because they do not have to 2210 * iterate over all bps. 2211 */ 2212 snap = list_head(&ddpa->origin_snaps); 2213 err = snaplist_space(&ddpa->shared_snaps, 2214 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2215 if (err != 0) 2216 goto out; 2217 2218 err = snaplist_space(&ddpa->clone_snaps, 2219 snap->ds->ds_dir->dd_origin_txg, &space); 2220 if (err != 0) 2221 goto out; 2222 ddpa->cloneusedsnap += space; 2223 } 2224 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2225 err = snaplist_space(&ddpa->origin_snaps, 2226 origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap); 2227 if (err != 0) 2228 goto out; 2229 } 2230 2231out: 2232 promote_rele(ddpa, FTAG); 2233 return (err); 2234} 2235 2236static void 2237dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2238{ 2239 dsl_dataset_promote_arg_t *ddpa = arg; 2240 dsl_pool_t *dp = dmu_tx_pool(tx); 2241 dsl_dataset_t *hds; 2242 struct promotenode *snap; 2243 dsl_dataset_t *origin_ds; 2244 dsl_dataset_t *origin_head; 2245 dsl_dir_t *dd; 2246 dsl_dir_t *odd = NULL; 2247 uint64_t oldnext_obj; 2248 int64_t delta; 2249 2250 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2251 hds = ddpa->ddpa_clone; 2252 2253 ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE); 2254 2255 snap = list_head(&ddpa->shared_snaps); 2256 origin_ds = snap->ds; 2257 dd = hds->ds_dir; 2258 2259 snap = list_head(&ddpa->origin_snaps); 2260 origin_head = snap->ds; 2261 2262 /* 2263 * We need to explicitly open odd, since origin_ds's dd will be 2264 * changing. 2265 */ 2266 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2267 NULL, FTAG, &odd)); 2268 2269 /* change origin's next snap */ 2270 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2271 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj; 2272 snap = list_tail(&ddpa->clone_snaps); 2273 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2274 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object; 2275 2276 /* change the origin's next clone */ 2277 if (origin_ds->ds_phys->ds_next_clones_obj) { 2278 dsl_dataset_remove_from_next_clones(origin_ds, 2279 snap->ds->ds_object, tx); 2280 VERIFY0(zap_add_int(dp->dp_meta_objset, 2281 origin_ds->ds_phys->ds_next_clones_obj, 2282 oldnext_obj, tx)); 2283 } 2284 2285 /* change origin */ 2286 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2287 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object); 2288 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj; 2289 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2290 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2291 odd->dd_phys->dd_origin_obj = origin_ds->ds_object; 2292 origin_head->ds_dir->dd_origin_txg = 2293 origin_ds->ds_phys->ds_creation_txg; 2294 2295 /* change dd_clone entries */ 2296 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2297 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2298 odd->dd_phys->dd_clones, hds->ds_object, tx)); 2299 VERIFY0(zap_add_int(dp->dp_meta_objset, 2300 ddpa->origin_origin->ds_dir->dd_phys->dd_clones, 2301 hds->ds_object, tx)); 2302 2303 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2304 ddpa->origin_origin->ds_dir->dd_phys->dd_clones, 2305 origin_head->ds_object, tx)); 2306 if (dd->dd_phys->dd_clones == 0) { 2307 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset, 2308 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 2309 } 2310 VERIFY0(zap_add_int(dp->dp_meta_objset, 2311 dd->dd_phys->dd_clones, origin_head->ds_object, tx)); 2312 } 2313 2314 /* move snapshots to this dir */ 2315 for (snap = list_head(&ddpa->shared_snaps); snap; 2316 snap = list_next(&ddpa->shared_snaps, snap)) { 2317 dsl_dataset_t *ds = snap->ds; 2318 2319 /* 2320 * Property callbacks are registered to a particular 2321 * dsl_dir. Since ours is changing, evict the objset 2322 * so that they will be unregistered from the old dsl_dir. 2323 */ 2324 if (ds->ds_objset) { 2325 dmu_objset_evict(ds->ds_objset); 2326 ds->ds_objset = NULL; 2327 } 2328 2329 /* move snap name entry */ 2330 VERIFY0(dsl_dataset_get_snapname(ds)); 2331 VERIFY0(dsl_dataset_snap_remove(origin_head, 2332 ds->ds_snapname, tx, B_TRUE)); 2333 VERIFY0(zap_add(dp->dp_meta_objset, 2334 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, 2335 8, 1, &ds->ds_object, tx)); 2336 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 2337 DD_FIELD_SNAPSHOT_COUNT, tx); 2338 2339 /* change containing dsl_dir */ 2340 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2341 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object); 2342 ds->ds_phys->ds_dir_obj = dd->dd_object; 2343 ASSERT3P(ds->ds_dir, ==, odd); 2344 dsl_dir_rele(ds->ds_dir, ds); 2345 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 2346 NULL, ds, &ds->ds_dir)); 2347 2348 /* move any clone references */ 2349 if (ds->ds_phys->ds_next_clones_obj && 2350 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2351 zap_cursor_t zc; 2352 zap_attribute_t za; 2353 2354 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2355 ds->ds_phys->ds_next_clones_obj); 2356 zap_cursor_retrieve(&zc, &za) == 0; 2357 zap_cursor_advance(&zc)) { 2358 dsl_dataset_t *cnds; 2359 uint64_t o; 2360 2361 if (za.za_first_integer == oldnext_obj) { 2362 /* 2363 * We've already moved the 2364 * origin's reference. 2365 */ 2366 continue; 2367 } 2368 2369 VERIFY0(dsl_dataset_hold_obj(dp, 2370 za.za_first_integer, FTAG, &cnds)); 2371 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj; 2372 2373 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2374 odd->dd_phys->dd_clones, o, tx)); 2375 VERIFY0(zap_add_int(dp->dp_meta_objset, 2376 dd->dd_phys->dd_clones, o, tx)); 2377 dsl_dataset_rele(cnds, FTAG); 2378 } 2379 zap_cursor_fini(&zc); 2380 } 2381 2382 ASSERT(!dsl_prop_hascb(ds)); 2383 } 2384 2385 /* 2386 * Change space accounting. 2387 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2388 * both be valid, or both be 0 (resulting in delta == 0). This 2389 * is true for each of {clone,origin} independently. 2390 */ 2391 2392 delta = ddpa->cloneusedsnap - 2393 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2394 ASSERT3S(delta, >=, 0); 2395 ASSERT3U(ddpa->used, >=, delta); 2396 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2397 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2398 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 2399 2400 delta = ddpa->originusedsnap - 2401 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2402 ASSERT3S(delta, <=, 0); 2403 ASSERT3U(ddpa->used, >=, -delta); 2404 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2405 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2406 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 2407 2408 origin_ds->ds_phys->ds_unique_bytes = ddpa->unique; 2409 2410 /* log history record */ 2411 spa_history_log_internal_ds(hds, "promote", tx, ""); 2412 2413 dsl_dir_rele(odd, FTAG); 2414 promote_rele(ddpa, FTAG); 2415} 2416 2417/* 2418 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2419 * (exclusive) and last_obj (inclusive). The list will be in reverse 2420 * order (last_obj will be the list_head()). If first_obj == 0, do all 2421 * snapshots back to this dataset's origin. 2422 */ 2423static int 2424snaplist_make(dsl_pool_t *dp, 2425 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 2426{ 2427 uint64_t obj = last_obj; 2428 2429 list_create(l, sizeof (struct promotenode), 2430 offsetof(struct promotenode, link)); 2431 2432 while (obj != first_obj) { 2433 dsl_dataset_t *ds; 2434 struct promotenode *snap; 2435 int err; 2436 2437 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 2438 ASSERT(err != ENOENT); 2439 if (err != 0) 2440 return (err); 2441 2442 if (first_obj == 0) 2443 first_obj = ds->ds_dir->dd_phys->dd_origin_obj; 2444 2445 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 2446 snap->ds = ds; 2447 list_insert_tail(l, snap); 2448 obj = ds->ds_phys->ds_prev_snap_obj; 2449 } 2450 2451 return (0); 2452} 2453 2454static int 2455snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2456{ 2457 struct promotenode *snap; 2458 2459 *spacep = 0; 2460 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2461 uint64_t used, comp, uncomp; 2462 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2463 mintxg, UINT64_MAX, &used, &comp, &uncomp); 2464 *spacep += used; 2465 } 2466 return (0); 2467} 2468 2469static void 2470snaplist_destroy(list_t *l, void *tag) 2471{ 2472 struct promotenode *snap; 2473 2474 if (l == NULL || !list_link_active(&l->list_head)) 2475 return; 2476 2477 while ((snap = list_tail(l)) != NULL) { 2478 list_remove(l, snap); 2479 dsl_dataset_rele(snap->ds, tag); 2480 kmem_free(snap, sizeof (*snap)); 2481 } 2482 list_destroy(l); 2483} 2484 2485static int 2486promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 2487{ 2488 int error; 2489 dsl_dir_t *dd; 2490 struct promotenode *snap; 2491 2492 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 2493 &ddpa->ddpa_clone); 2494 if (error != 0) 2495 return (error); 2496 dd = ddpa->ddpa_clone->ds_dir; 2497 2498 if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) || 2499 !dsl_dir_is_clone(dd)) { 2500 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2501 return (SET_ERROR(EINVAL)); 2502 } 2503 2504 error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj, 2505 &ddpa->shared_snaps, tag); 2506 if (error != 0) 2507 goto out; 2508 2509 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 2510 &ddpa->clone_snaps, tag); 2511 if (error != 0) 2512 goto out; 2513 2514 snap = list_head(&ddpa->shared_snaps); 2515 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); 2516 error = snaplist_make(dp, dd->dd_phys->dd_origin_obj, 2517 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, 2518 &ddpa->origin_snaps, tag); 2519 if (error != 0) 2520 goto out; 2521 2522 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) { 2523 error = dsl_dataset_hold_obj(dp, 2524 snap->ds->ds_dir->dd_phys->dd_origin_obj, 2525 tag, &ddpa->origin_origin); 2526 if (error != 0) 2527 goto out; 2528 } 2529out: 2530 if (error != 0) 2531 promote_rele(ddpa, tag); 2532 return (error); 2533} 2534 2535static void 2536promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 2537{ 2538 snaplist_destroy(&ddpa->shared_snaps, tag); 2539 snaplist_destroy(&ddpa->clone_snaps, tag); 2540 snaplist_destroy(&ddpa->origin_snaps, tag); 2541 if (ddpa->origin_origin != NULL) 2542 dsl_dataset_rele(ddpa->origin_origin, tag); 2543 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2544} 2545 2546/* 2547 * Promote a clone. 2548 * 2549 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 2550 * in with the name. (It must be at least MAXNAMELEN bytes long.) 2551 */ 2552int 2553dsl_dataset_promote(const char *name, char *conflsnap) 2554{ 2555 dsl_dataset_promote_arg_t ddpa = { 0 }; 2556 uint64_t numsnaps; 2557 int error; 2558 objset_t *os; 2559 2560 /* 2561 * We will modify space proportional to the number of 2562 * snapshots. Compute numsnaps. 2563 */ 2564 error = dmu_objset_hold(name, FTAG, &os); 2565 if (error != 0) 2566 return (error); 2567 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 2568 dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps); 2569 dmu_objset_rele(os, FTAG); 2570 if (error != 0) 2571 return (error); 2572 2573 ddpa.ddpa_clonename = name; 2574 ddpa.err_ds = conflsnap; 2575 ddpa.cr = CRED(); 2576 2577 return (dsl_sync_task(name, dsl_dataset_promote_check, 2578 dsl_dataset_promote_sync, &ddpa, 2 + numsnaps)); 2579} 2580 2581int 2582dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 2583 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 2584{ 2585 int64_t unused_refres_delta; 2586 2587 /* they should both be heads */ 2588 if (dsl_dataset_is_snapshot(clone) || 2589 dsl_dataset_is_snapshot(origin_head)) 2590 return (SET_ERROR(EINVAL)); 2591 2592 /* if we are not forcing, the branch point should be just before them */ 2593 if (!force && clone->ds_prev != origin_head->ds_prev) 2594 return (SET_ERROR(EINVAL)); 2595 2596 /* clone should be the clone (unless they are unrelated) */ 2597 if (clone->ds_prev != NULL && 2598 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 2599 origin_head->ds_dir != clone->ds_prev->ds_dir) 2600 return (SET_ERROR(EINVAL)); 2601 2602 /* the clone should be a child of the origin */ 2603 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 2604 return (SET_ERROR(EINVAL)); 2605 2606 /* origin_head shouldn't be modified unless 'force' */ 2607 if (!force && 2608 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 2609 return (SET_ERROR(ETXTBSY)); 2610 2611 /* origin_head should have no long holds (e.g. is not mounted) */ 2612 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 2613 return (SET_ERROR(EBUSY)); 2614 2615 /* check amount of any unconsumed refreservation */ 2616 unused_refres_delta = 2617 (int64_t)MIN(origin_head->ds_reserved, 2618 origin_head->ds_phys->ds_unique_bytes) - 2619 (int64_t)MIN(origin_head->ds_reserved, 2620 clone->ds_phys->ds_unique_bytes); 2621 2622 if (unused_refres_delta > 0 && 2623 unused_refres_delta > 2624 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 2625 return (SET_ERROR(ENOSPC)); 2626 2627 /* clone can't be over the head's refquota */ 2628 if (origin_head->ds_quota != 0 && 2629 clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota) 2630 return (SET_ERROR(EDQUOT)); 2631 2632 return (0); 2633} 2634 2635void 2636dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 2637 dsl_dataset_t *origin_head, dmu_tx_t *tx) 2638{ 2639 dsl_pool_t *dp = dmu_tx_pool(tx); 2640 int64_t unused_refres_delta; 2641 2642 ASSERT(clone->ds_reserved == 0); 2643 ASSERT(origin_head->ds_quota == 0 || 2644 clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota); 2645 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 2646 2647 dmu_buf_will_dirty(clone->ds_dbuf, tx); 2648 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 2649 2650 if (clone->ds_objset != NULL) { 2651 dmu_objset_evict(clone->ds_objset); 2652 clone->ds_objset = NULL; 2653 } 2654 2655 if (origin_head->ds_objset != NULL) { 2656 dmu_objset_evict(origin_head->ds_objset); 2657 origin_head->ds_objset = NULL; 2658 } 2659 2660 unused_refres_delta = 2661 (int64_t)MIN(origin_head->ds_reserved, 2662 origin_head->ds_phys->ds_unique_bytes) - 2663 (int64_t)MIN(origin_head->ds_reserved, 2664 clone->ds_phys->ds_unique_bytes); 2665 2666 /* 2667 * Reset origin's unique bytes, if it exists. 2668 */ 2669 if (clone->ds_prev) { 2670 dsl_dataset_t *origin = clone->ds_prev; 2671 uint64_t comp, uncomp; 2672 2673 dmu_buf_will_dirty(origin->ds_dbuf, tx); 2674 dsl_deadlist_space_range(&clone->ds_deadlist, 2675 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2676 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp); 2677 } 2678 2679 /* swap blkptrs */ 2680 { 2681 blkptr_t tmp; 2682 tmp = origin_head->ds_phys->ds_bp; 2683 origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp; 2684 clone->ds_phys->ds_bp = tmp; 2685 } 2686 2687 /* set dd_*_bytes */ 2688 { 2689 int64_t dused, dcomp, duncomp; 2690 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2691 uint64_t odl_used, odl_comp, odl_uncomp; 2692 2693 ASSERT3U(clone->ds_dir->dd_phys-> 2694 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2695 2696 dsl_deadlist_space(&clone->ds_deadlist, 2697 &cdl_used, &cdl_comp, &cdl_uncomp); 2698 dsl_deadlist_space(&origin_head->ds_deadlist, 2699 &odl_used, &odl_comp, &odl_uncomp); 2700 2701 dused = clone->ds_phys->ds_referenced_bytes + cdl_used - 2702 (origin_head->ds_phys->ds_referenced_bytes + odl_used); 2703 dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp - 2704 (origin_head->ds_phys->ds_compressed_bytes + odl_comp); 2705 duncomp = clone->ds_phys->ds_uncompressed_bytes + 2706 cdl_uncomp - 2707 (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp); 2708 2709 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 2710 dused, dcomp, duncomp, tx); 2711 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 2712 -dused, -dcomp, -duncomp, tx); 2713 2714 /* 2715 * The difference in the space used by snapshots is the 2716 * difference in snapshot space due to the head's 2717 * deadlist (since that's the only thing that's 2718 * changing that affects the snapused). 2719 */ 2720 dsl_deadlist_space_range(&clone->ds_deadlist, 2721 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2722 &cdl_used, &cdl_comp, &cdl_uncomp); 2723 dsl_deadlist_space_range(&origin_head->ds_deadlist, 2724 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2725 &odl_used, &odl_comp, &odl_uncomp); 2726 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 2727 DD_USED_HEAD, DD_USED_SNAP, tx); 2728 } 2729 2730 /* swap ds_*_bytes */ 2731 SWITCH64(origin_head->ds_phys->ds_referenced_bytes, 2732 clone->ds_phys->ds_referenced_bytes); 2733 SWITCH64(origin_head->ds_phys->ds_compressed_bytes, 2734 clone->ds_phys->ds_compressed_bytes); 2735 SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes, 2736 clone->ds_phys->ds_uncompressed_bytes); 2737 SWITCH64(origin_head->ds_phys->ds_unique_bytes, 2738 clone->ds_phys->ds_unique_bytes); 2739 2740 /* apply any parent delta for change in unconsumed refreservation */ 2741 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 2742 unused_refres_delta, 0, 0, tx); 2743 2744 /* 2745 * Swap deadlists. 2746 */ 2747 dsl_deadlist_close(&clone->ds_deadlist); 2748 dsl_deadlist_close(&origin_head->ds_deadlist); 2749 SWITCH64(origin_head->ds_phys->ds_deadlist_obj, 2750 clone->ds_phys->ds_deadlist_obj); 2751 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 2752 clone->ds_phys->ds_deadlist_obj); 2753 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 2754 origin_head->ds_phys->ds_deadlist_obj); 2755 2756 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 2757 2758 spa_history_log_internal_ds(clone, "clone swap", tx, 2759 "parent=%s", origin_head->ds_dir->dd_myname); 2760} 2761 2762/* 2763 * Given a pool name and a dataset object number in that pool, 2764 * return the name of that dataset. 2765 */ 2766int 2767dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 2768{ 2769 dsl_pool_t *dp; 2770 dsl_dataset_t *ds; 2771 int error; 2772 2773 error = dsl_pool_hold(pname, FTAG, &dp); 2774 if (error != 0) 2775 return (error); 2776 2777 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 2778 if (error == 0) { 2779 dsl_dataset_name(ds, buf); 2780 dsl_dataset_rele(ds, FTAG); 2781 } 2782 dsl_pool_rele(dp, FTAG); 2783 2784 return (error); 2785} 2786 2787int 2788dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 2789 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 2790{ 2791 int error = 0; 2792 2793 ASSERT3S(asize, >, 0); 2794 2795 /* 2796 * *ref_rsrv is the portion of asize that will come from any 2797 * unconsumed refreservation space. 2798 */ 2799 *ref_rsrv = 0; 2800 2801 mutex_enter(&ds->ds_lock); 2802 /* 2803 * Make a space adjustment for reserved bytes. 2804 */ 2805 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) { 2806 ASSERT3U(*used, >=, 2807 ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 2808 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 2809 *ref_rsrv = 2810 asize - MIN(asize, parent_delta(ds, asize + inflight)); 2811 } 2812 2813 if (!check_quota || ds->ds_quota == 0) { 2814 mutex_exit(&ds->ds_lock); 2815 return (0); 2816 } 2817 /* 2818 * If they are requesting more space, and our current estimate 2819 * is over quota, they get to try again unless the actual 2820 * on-disk is over quota and there are no pending changes (which 2821 * may free up space for us). 2822 */ 2823 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) { 2824 if (inflight > 0 || 2825 ds->ds_phys->ds_referenced_bytes < ds->ds_quota) 2826 error = SET_ERROR(ERESTART); 2827 else 2828 error = SET_ERROR(EDQUOT); 2829 } 2830 mutex_exit(&ds->ds_lock); 2831 2832 return (error); 2833} 2834 2835typedef struct dsl_dataset_set_qr_arg { 2836 const char *ddsqra_name; 2837 zprop_source_t ddsqra_source; 2838 uint64_t ddsqra_value; 2839} dsl_dataset_set_qr_arg_t; 2840 2841 2842/* ARGSUSED */ 2843static int 2844dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 2845{ 2846 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2847 dsl_pool_t *dp = dmu_tx_pool(tx); 2848 dsl_dataset_t *ds; 2849 int error; 2850 uint64_t newval; 2851 2852 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 2853 return (SET_ERROR(ENOTSUP)); 2854 2855 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 2856 if (error != 0) 2857 return (error); 2858 2859 if (dsl_dataset_is_snapshot(ds)) { 2860 dsl_dataset_rele(ds, FTAG); 2861 return (SET_ERROR(EINVAL)); 2862 } 2863 2864 error = dsl_prop_predict(ds->ds_dir, 2865 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 2866 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 2867 if (error != 0) { 2868 dsl_dataset_rele(ds, FTAG); 2869 return (error); 2870 } 2871 2872 if (newval == 0) { 2873 dsl_dataset_rele(ds, FTAG); 2874 return (0); 2875 } 2876 2877 if (newval < ds->ds_phys->ds_referenced_bytes || 2878 newval < ds->ds_reserved) { 2879 dsl_dataset_rele(ds, FTAG); 2880 return (SET_ERROR(ENOSPC)); 2881 } 2882 2883 dsl_dataset_rele(ds, FTAG); 2884 return (0); 2885} 2886 2887static void 2888dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 2889{ 2890 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2891 dsl_pool_t *dp = dmu_tx_pool(tx); 2892 dsl_dataset_t *ds; 2893 uint64_t newval; 2894 2895 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 2896 2897 dsl_prop_set_sync_impl(ds, 2898 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 2899 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 2900 &ddsqra->ddsqra_value, tx); 2901 2902 VERIFY0(dsl_prop_get_int_ds(ds, 2903 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 2904 2905 if (ds->ds_quota != newval) { 2906 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2907 ds->ds_quota = newval; 2908 } 2909 dsl_dataset_rele(ds, FTAG); 2910} 2911 2912int 2913dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 2914 uint64_t refquota) 2915{ 2916 dsl_dataset_set_qr_arg_t ddsqra; 2917 2918 ddsqra.ddsqra_name = dsname; 2919 ddsqra.ddsqra_source = source; 2920 ddsqra.ddsqra_value = refquota; 2921 2922 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 2923 dsl_dataset_set_refquota_sync, &ddsqra, 0)); 2924} 2925 2926static int 2927dsl_dataset_set_refreservation_check(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 int error; 2933 uint64_t newval, unique; 2934 2935 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 2936 return (SET_ERROR(ENOTSUP)); 2937 2938 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 2939 if (error != 0) 2940 return (error); 2941 2942 if (dsl_dataset_is_snapshot(ds)) { 2943 dsl_dataset_rele(ds, FTAG); 2944 return (SET_ERROR(EINVAL)); 2945 } 2946 2947 error = dsl_prop_predict(ds->ds_dir, 2948 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 2949 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 2950 if (error != 0) { 2951 dsl_dataset_rele(ds, FTAG); 2952 return (error); 2953 } 2954 2955 /* 2956 * If we are doing the preliminary check in open context, the 2957 * space estimates may be inaccurate. 2958 */ 2959 if (!dmu_tx_is_syncing(tx)) { 2960 dsl_dataset_rele(ds, FTAG); 2961 return (0); 2962 } 2963 2964 mutex_enter(&ds->ds_lock); 2965 if (!DS_UNIQUE_IS_ACCURATE(ds)) 2966 dsl_dataset_recalc_head_uniq(ds); 2967 unique = ds->ds_phys->ds_unique_bytes; 2968 mutex_exit(&ds->ds_lock); 2969 2970 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 2971 uint64_t delta = MAX(unique, newval) - 2972 MAX(unique, ds->ds_reserved); 2973 2974 if (delta > 2975 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 2976 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 2977 dsl_dataset_rele(ds, FTAG); 2978 return (SET_ERROR(ENOSPC)); 2979 } 2980 } 2981 2982 dsl_dataset_rele(ds, FTAG); 2983 return (0); 2984} 2985 2986void 2987dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 2988 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 2989{ 2990 uint64_t newval; 2991 uint64_t unique; 2992 int64_t delta; 2993 2994 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 2995 source, sizeof (value), 1, &value, tx); 2996 2997 VERIFY0(dsl_prop_get_int_ds(ds, 2998 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 2999 3000 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3001 mutex_enter(&ds->ds_dir->dd_lock); 3002 mutex_enter(&ds->ds_lock); 3003 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3004 unique = ds->ds_phys->ds_unique_bytes; 3005 delta = MAX(0, (int64_t)(newval - unique)) - 3006 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3007 ds->ds_reserved = newval; 3008 mutex_exit(&ds->ds_lock); 3009 3010 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3011 mutex_exit(&ds->ds_dir->dd_lock); 3012} 3013 3014static void 3015dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 3016{ 3017 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3018 dsl_pool_t *dp = dmu_tx_pool(tx); 3019 dsl_dataset_t *ds; 3020 3021 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3022 dsl_dataset_set_refreservation_sync_impl(ds, 3023 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 3024 dsl_dataset_rele(ds, FTAG); 3025} 3026 3027int 3028dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 3029 uint64_t refreservation) 3030{ 3031 dsl_dataset_set_qr_arg_t ddsqra; 3032 3033 ddsqra.ddsqra_name = dsname; 3034 ddsqra.ddsqra_source = source; 3035 ddsqra.ddsqra_value = refreservation; 3036 3037 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 3038 dsl_dataset_set_refreservation_sync, &ddsqra, 0)); 3039} 3040 3041/* 3042 * Return (in *usedp) the amount of space written in new that is not 3043 * present in oldsnap. New may be a snapshot or the head. Old must be 3044 * a snapshot before new, in new's filesystem (or its origin). If not then 3045 * fail and return EINVAL. 3046 * 3047 * The written space is calculated by considering two components: First, we 3048 * ignore any freed space, and calculate the written as new's used space 3049 * minus old's used space. Next, we add in the amount of space that was freed 3050 * between the two snapshots, thus reducing new's used space relative to old's. 3051 * Specifically, this is the space that was born before old->ds_creation_txg, 3052 * and freed before new (ie. on new's deadlist or a previous deadlist). 3053 * 3054 * space freed [---------------------] 3055 * snapshots ---O-------O--------O-------O------ 3056 * oldsnap new 3057 */ 3058int 3059dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 3060 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3061{ 3062 int err = 0; 3063 uint64_t snapobj; 3064 dsl_pool_t *dp = new->ds_dir->dd_pool; 3065 3066 ASSERT(dsl_pool_config_held(dp)); 3067 3068 *usedp = 0; 3069 *usedp += new->ds_phys->ds_referenced_bytes; 3070 *usedp -= oldsnap->ds_phys->ds_referenced_bytes; 3071 3072 *compp = 0; 3073 *compp += new->ds_phys->ds_compressed_bytes; 3074 *compp -= oldsnap->ds_phys->ds_compressed_bytes; 3075 3076 *uncompp = 0; 3077 *uncompp += new->ds_phys->ds_uncompressed_bytes; 3078 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes; 3079 3080 snapobj = new->ds_object; 3081 while (snapobj != oldsnap->ds_object) { 3082 dsl_dataset_t *snap; 3083 uint64_t used, comp, uncomp; 3084 3085 if (snapobj == new->ds_object) { 3086 snap = new; 3087 } else { 3088 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3089 if (err != 0) 3090 break; 3091 } 3092 3093 if (snap->ds_phys->ds_prev_snap_txg == 3094 oldsnap->ds_phys->ds_creation_txg) { 3095 /* 3096 * The blocks in the deadlist can not be born after 3097 * ds_prev_snap_txg, so get the whole deadlist space, 3098 * which is more efficient (especially for old-format 3099 * deadlists). Unfortunately the deadlist code 3100 * doesn't have enough information to make this 3101 * optimization itself. 3102 */ 3103 dsl_deadlist_space(&snap->ds_deadlist, 3104 &used, &comp, &uncomp); 3105 } else { 3106 dsl_deadlist_space_range(&snap->ds_deadlist, 3107 0, oldsnap->ds_phys->ds_creation_txg, 3108 &used, &comp, &uncomp); 3109 } 3110 *usedp += used; 3111 *compp += comp; 3112 *uncompp += uncomp; 3113 3114 /* 3115 * If we get to the beginning of the chain of snapshots 3116 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3117 * was not a snapshot of/before new. 3118 */ 3119 snapobj = snap->ds_phys->ds_prev_snap_obj; 3120 if (snap != new) 3121 dsl_dataset_rele(snap, FTAG); 3122 if (snapobj == 0) { 3123 err = SET_ERROR(EINVAL); 3124 break; 3125 } 3126 3127 } 3128 return (err); 3129} 3130 3131/* 3132 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3133 * lastsnap, and all snapshots in between are deleted. 3134 * 3135 * blocks that would be freed [---------------------------] 3136 * snapshots ---O-------O--------O-------O--------O 3137 * firstsnap lastsnap 3138 * 3139 * This is the set of blocks that were born after the snap before firstsnap, 3140 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3141 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3142 * We calculate this by iterating over the relevant deadlists (from the snap 3143 * after lastsnap, backward to the snap after firstsnap), summing up the 3144 * space on the deadlist that was born after the snap before firstsnap. 3145 */ 3146int 3147dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3148 dsl_dataset_t *lastsnap, 3149 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3150{ 3151 int err = 0; 3152 uint64_t snapobj; 3153 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3154 3155 ASSERT(dsl_dataset_is_snapshot(firstsnap)); 3156 ASSERT(dsl_dataset_is_snapshot(lastsnap)); 3157 3158 /* 3159 * Check that the snapshots are in the same dsl_dir, and firstsnap 3160 * is before lastsnap. 3161 */ 3162 if (firstsnap->ds_dir != lastsnap->ds_dir || 3163 firstsnap->ds_phys->ds_creation_txg > 3164 lastsnap->ds_phys->ds_creation_txg) 3165 return (SET_ERROR(EINVAL)); 3166 3167 *usedp = *compp = *uncompp = 0; 3168 3169 snapobj = lastsnap->ds_phys->ds_next_snap_obj; 3170 while (snapobj != firstsnap->ds_object) { 3171 dsl_dataset_t *ds; 3172 uint64_t used, comp, uncomp; 3173 3174 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3175 if (err != 0) 3176 break; 3177 3178 dsl_deadlist_space_range(&ds->ds_deadlist, 3179 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX, 3180 &used, &comp, &uncomp); 3181 *usedp += used; 3182 *compp += comp; 3183 *uncompp += uncomp; 3184 3185 snapobj = ds->ds_phys->ds_prev_snap_obj; 3186 ASSERT3U(snapobj, !=, 0); 3187 dsl_dataset_rele(ds, FTAG); 3188 } 3189 return (err); 3190} 3191 3192/* 3193 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 3194 * For example, they could both be snapshots of the same filesystem, and 3195 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 3196 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 3197 * filesystem. Or 'earlier' could be the origin's origin. 3198 * 3199 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 3200 */ 3201boolean_t 3202dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 3203 uint64_t earlier_txg) 3204{ 3205 dsl_pool_t *dp = later->ds_dir->dd_pool; 3206 int error; 3207 boolean_t ret; 3208 3209 ASSERT(dsl_pool_config_held(dp)); 3210 ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0); 3211 3212 if (earlier_txg == 0) 3213 earlier_txg = earlier->ds_phys->ds_creation_txg; 3214 3215 if (dsl_dataset_is_snapshot(later) && 3216 earlier_txg >= later->ds_phys->ds_creation_txg) 3217 return (B_FALSE); 3218 3219 if (later->ds_dir == earlier->ds_dir) 3220 return (B_TRUE); 3221 if (!dsl_dir_is_clone(later->ds_dir)) 3222 return (B_FALSE); 3223 3224 if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object) 3225 return (B_TRUE); 3226 dsl_dataset_t *origin; 3227 error = dsl_dataset_hold_obj(dp, 3228 later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin); 3229 if (error != 0) 3230 return (B_FALSE); 3231 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 3232 dsl_dataset_rele(origin, FTAG); 3233 return (ret); 3234} 3235 3236 3237void 3238dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 3239{ 3240 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3241 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 3242} 3243