dsl_dataset.c revision 273195
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) 2011, 2014 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 mutex_destroy(&ds->ds_sendstream_lock); 274 refcount_destroy(&ds->ds_longholds); 275 276 kmem_free(ds, sizeof (dsl_dataset_t)); 277} 278 279int 280dsl_dataset_get_snapname(dsl_dataset_t *ds) 281{ 282 dsl_dataset_phys_t *headphys; 283 int err; 284 dmu_buf_t *headdbuf; 285 dsl_pool_t *dp = ds->ds_dir->dd_pool; 286 objset_t *mos = dp->dp_meta_objset; 287 288 if (ds->ds_snapname[0]) 289 return (0); 290 if (ds->ds_phys->ds_next_snap_obj == 0) 291 return (0); 292 293 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, 294 FTAG, &headdbuf); 295 if (err != 0) 296 return (err); 297 headphys = headdbuf->db_data; 298 err = zap_value_search(dp->dp_meta_objset, 299 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); 300 dmu_buf_rele(headdbuf, FTAG); 301 return (err); 302} 303 304int 305dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) 306{ 307 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 308 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 309 matchtype_t mt; 310 int err; 311 312 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 313 mt = MT_FIRST; 314 else 315 mt = MT_EXACT; 316 317 err = zap_lookup_norm(mos, snapobj, name, 8, 1, 318 value, mt, NULL, 0, NULL); 319 if (err == ENOTSUP && mt == MT_FIRST) 320 err = zap_lookup(mos, snapobj, name, 8, 1, value); 321 return (err); 322} 323 324int 325dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx, 326 boolean_t adj_cnt) 327{ 328 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 329 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; 330 matchtype_t mt; 331 int err; 332 333 dsl_dir_snap_cmtime_update(ds->ds_dir); 334 335 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) 336 mt = MT_FIRST; 337 else 338 mt = MT_EXACT; 339 340 err = zap_remove_norm(mos, snapobj, name, mt, tx); 341 if (err == ENOTSUP && mt == MT_FIRST) 342 err = zap_remove(mos, snapobj, name, tx); 343 344 if (err == 0 && adj_cnt) 345 dsl_fs_ss_count_adjust(ds->ds_dir, -1, 346 DD_FIELD_SNAPSHOT_COUNT, tx); 347 348 return (err); 349} 350 351int 352dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, 353 dsl_dataset_t **dsp) 354{ 355 objset_t *mos = dp->dp_meta_objset; 356 dmu_buf_t *dbuf; 357 dsl_dataset_t *ds; 358 int err; 359 dmu_object_info_t doi; 360 361 ASSERT(dsl_pool_config_held(dp)); 362 363 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); 364 if (err != 0) 365 return (err); 366 367 /* Make sure dsobj has the correct object type. */ 368 dmu_object_info_from_db(dbuf, &doi); 369 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) { 370 dmu_buf_rele(dbuf, tag); 371 return (SET_ERROR(EINVAL)); 372 } 373 374 ds = dmu_buf_get_user(dbuf); 375 if (ds == NULL) { 376 dsl_dataset_t *winner = NULL; 377 378 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); 379 ds->ds_dbuf = dbuf; 380 ds->ds_object = dsobj; 381 ds->ds_phys = dbuf->db_data; 382 383 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); 384 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); 385 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL); 386 refcount_create(&ds->ds_longholds); 387 388 bplist_create(&ds->ds_pending_deadlist); 389 dsl_deadlist_open(&ds->ds_deadlist, 390 mos, ds->ds_phys->ds_deadlist_obj); 391 392 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t), 393 offsetof(dmu_sendarg_t, dsa_link)); 394 395 if (err == 0) { 396 err = dsl_dir_hold_obj(dp, 397 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); 398 } 399 if (err != 0) { 400 mutex_destroy(&ds->ds_lock); 401 mutex_destroy(&ds->ds_opening_lock); 402 mutex_destroy(&ds->ds_sendstream_lock); 403 refcount_destroy(&ds->ds_longholds); 404 bplist_destroy(&ds->ds_pending_deadlist); 405 dsl_deadlist_close(&ds->ds_deadlist); 406 kmem_free(ds, sizeof (dsl_dataset_t)); 407 dmu_buf_rele(dbuf, tag); 408 return (err); 409 } 410 411 if (!dsl_dataset_is_snapshot(ds)) { 412 ds->ds_snapname[0] = '\0'; 413 if (ds->ds_phys->ds_prev_snap_obj != 0) { 414 err = dsl_dataset_hold_obj(dp, 415 ds->ds_phys->ds_prev_snap_obj, 416 ds, &ds->ds_prev); 417 } 418 if (doi.doi_type == DMU_OTN_ZAP_METADATA) { 419 int zaperr = zap_lookup(mos, ds->ds_object, 420 DS_FIELD_BOOKMARK_NAMES, 421 sizeof (ds->ds_bookmarks), 1, 422 &ds->ds_bookmarks); 423 if (zaperr != ENOENT) 424 VERIFY0(zaperr); 425 } 426 } else { 427 if (zfs_flags & ZFS_DEBUG_SNAPNAMES) 428 err = dsl_dataset_get_snapname(ds); 429 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) { 430 err = zap_count( 431 ds->ds_dir->dd_pool->dp_meta_objset, 432 ds->ds_phys->ds_userrefs_obj, 433 &ds->ds_userrefs); 434 } 435 } 436 437 if (err == 0 && !dsl_dataset_is_snapshot(ds)) { 438 err = dsl_prop_get_int_ds(ds, 439 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 440 &ds->ds_reserved); 441 if (err == 0) { 442 err = dsl_prop_get_int_ds(ds, 443 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 444 &ds->ds_quota); 445 } 446 } else { 447 ds->ds_reserved = ds->ds_quota = 0; 448 } 449 450 if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds, 451 &ds->ds_phys, dsl_dataset_evict)) != NULL) { 452 bplist_destroy(&ds->ds_pending_deadlist); 453 dsl_deadlist_close(&ds->ds_deadlist); 454 if (ds->ds_prev) 455 dsl_dataset_rele(ds->ds_prev, ds); 456 dsl_dir_rele(ds->ds_dir, ds); 457 mutex_destroy(&ds->ds_lock); 458 mutex_destroy(&ds->ds_opening_lock); 459 mutex_destroy(&ds->ds_sendstream_lock); 460 refcount_destroy(&ds->ds_longholds); 461 kmem_free(ds, sizeof (dsl_dataset_t)); 462 if (err != 0) { 463 dmu_buf_rele(dbuf, tag); 464 return (err); 465 } 466 ds = winner; 467 } else { 468 ds->ds_fsid_guid = 469 unique_insert(ds->ds_phys->ds_fsid_guid); 470 } 471 } 472 ASSERT3P(ds->ds_dbuf, ==, dbuf); 473 ASSERT3P(ds->ds_phys, ==, dbuf->db_data); 474 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 || 475 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || 476 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); 477 *dsp = ds; 478 return (0); 479} 480 481int 482dsl_dataset_hold(dsl_pool_t *dp, const char *name, 483 void *tag, dsl_dataset_t **dsp) 484{ 485 dsl_dir_t *dd; 486 const char *snapname; 487 uint64_t obj; 488 int err = 0; 489 490 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname); 491 if (err != 0) 492 return (err); 493 494 ASSERT(dsl_pool_config_held(dp)); 495 obj = dd->dd_phys->dd_head_dataset_obj; 496 if (obj != 0) 497 err = dsl_dataset_hold_obj(dp, obj, tag, dsp); 498 else 499 err = SET_ERROR(ENOENT); 500 501 /* we may be looking for a snapshot */ 502 if (err == 0 && snapname != NULL) { 503 dsl_dataset_t *ds; 504 505 if (*snapname++ != '@') { 506 dsl_dataset_rele(*dsp, tag); 507 dsl_dir_rele(dd, FTAG); 508 return (SET_ERROR(ENOENT)); 509 } 510 511 dprintf("looking for snapshot '%s'\n", snapname); 512 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); 513 if (err == 0) 514 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 515 dsl_dataset_rele(*dsp, tag); 516 517 if (err == 0) { 518 mutex_enter(&ds->ds_lock); 519 if (ds->ds_snapname[0] == 0) 520 (void) strlcpy(ds->ds_snapname, snapname, 521 sizeof (ds->ds_snapname)); 522 mutex_exit(&ds->ds_lock); 523 *dsp = ds; 524 } 525 } 526 527 dsl_dir_rele(dd, FTAG); 528 return (err); 529} 530 531int 532dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, 533 void *tag, dsl_dataset_t **dsp) 534{ 535 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); 536 if (err != 0) 537 return (err); 538 if (!dsl_dataset_tryown(*dsp, tag)) { 539 dsl_dataset_rele(*dsp, tag); 540 *dsp = NULL; 541 return (SET_ERROR(EBUSY)); 542 } 543 return (0); 544} 545 546int 547dsl_dataset_own(dsl_pool_t *dp, const char *name, 548 void *tag, dsl_dataset_t **dsp) 549{ 550 int err = dsl_dataset_hold(dp, name, tag, dsp); 551 if (err != 0) 552 return (err); 553 if (!dsl_dataset_tryown(*dsp, tag)) { 554 dsl_dataset_rele(*dsp, tag); 555 return (SET_ERROR(EBUSY)); 556 } 557 return (0); 558} 559 560/* 561 * See the comment above dsl_pool_hold() for details. In summary, a long 562 * hold is used to prevent destruction of a dataset while the pool hold 563 * is dropped, allowing other concurrent operations (e.g. spa_sync()). 564 * 565 * The dataset and pool must be held when this function is called. After it 566 * is called, the pool hold may be released while the dataset is still held 567 * and accessed. 568 */ 569void 570dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag) 571{ 572 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 573 (void) refcount_add(&ds->ds_longholds, tag); 574} 575 576void 577dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag) 578{ 579 (void) refcount_remove(&ds->ds_longholds, tag); 580} 581 582/* Return B_TRUE if there are any long holds on this dataset. */ 583boolean_t 584dsl_dataset_long_held(dsl_dataset_t *ds) 585{ 586 return (!refcount_is_zero(&ds->ds_longholds)); 587} 588 589void 590dsl_dataset_name(dsl_dataset_t *ds, char *name) 591{ 592 if (ds == NULL) { 593 (void) strcpy(name, "mos"); 594 } else { 595 dsl_dir_name(ds->ds_dir, name); 596 VERIFY0(dsl_dataset_get_snapname(ds)); 597 if (ds->ds_snapname[0]) { 598 (void) strcat(name, "@"); 599 /* 600 * We use a "recursive" mutex so that we 601 * can call dprintf_ds() with ds_lock held. 602 */ 603 if (!MUTEX_HELD(&ds->ds_lock)) { 604 mutex_enter(&ds->ds_lock); 605 (void) strcat(name, ds->ds_snapname); 606 mutex_exit(&ds->ds_lock); 607 } else { 608 (void) strcat(name, ds->ds_snapname); 609 } 610 } 611 } 612} 613 614void 615dsl_dataset_rele(dsl_dataset_t *ds, void *tag) 616{ 617 dmu_buf_rele(ds->ds_dbuf, tag); 618} 619 620void 621dsl_dataset_disown(dsl_dataset_t *ds, void *tag) 622{ 623 ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL); 624 625 mutex_enter(&ds->ds_lock); 626 ds->ds_owner = NULL; 627 mutex_exit(&ds->ds_lock); 628 dsl_dataset_long_rele(ds, tag); 629 if (ds->ds_dbuf != NULL) 630 dsl_dataset_rele(ds, tag); 631 else 632 dsl_dataset_evict(NULL, ds); 633} 634 635boolean_t 636dsl_dataset_tryown(dsl_dataset_t *ds, void *tag) 637{ 638 boolean_t gotit = FALSE; 639 640 mutex_enter(&ds->ds_lock); 641 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) { 642 ds->ds_owner = tag; 643 dsl_dataset_long_hold(ds, tag); 644 gotit = TRUE; 645 } 646 mutex_exit(&ds->ds_lock); 647 return (gotit); 648} 649 650uint64_t 651dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, 652 uint64_t flags, dmu_tx_t *tx) 653{ 654 dsl_pool_t *dp = dd->dd_pool; 655 dmu_buf_t *dbuf; 656 dsl_dataset_phys_t *dsphys; 657 uint64_t dsobj; 658 objset_t *mos = dp->dp_meta_objset; 659 660 if (origin == NULL) 661 origin = dp->dp_origin_snap; 662 663 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); 664 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0); 665 ASSERT(dmu_tx_is_syncing(tx)); 666 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 667 668 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 669 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 670 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 671 dmu_buf_will_dirty(dbuf, tx); 672 dsphys = dbuf->db_data; 673 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 674 dsphys->ds_dir_obj = dd->dd_object; 675 dsphys->ds_flags = flags; 676 dsphys->ds_fsid_guid = unique_create(); 677 do { 678 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 679 sizeof (dsphys->ds_guid)); 680 } while (dsphys->ds_guid == 0); 681 dsphys->ds_snapnames_zapobj = 682 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, 683 DMU_OT_NONE, 0, tx); 684 dsphys->ds_creation_time = gethrestime_sec(); 685 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; 686 687 if (origin == NULL) { 688 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); 689 } else { 690 dsl_dataset_t *ohds; /* head of the origin snapshot */ 691 692 dsphys->ds_prev_snap_obj = origin->ds_object; 693 dsphys->ds_prev_snap_txg = 694 origin->ds_phys->ds_creation_txg; 695 dsphys->ds_referenced_bytes = 696 origin->ds_phys->ds_referenced_bytes; 697 dsphys->ds_compressed_bytes = 698 origin->ds_phys->ds_compressed_bytes; 699 dsphys->ds_uncompressed_bytes = 700 origin->ds_phys->ds_uncompressed_bytes; 701 dsphys->ds_bp = origin->ds_phys->ds_bp; 702 703 /* 704 * Inherit flags that describe the dataset's contents 705 * (INCONSISTENT) or properties (Case Insensitive). 706 */ 707 dsphys->ds_flags |= origin->ds_phys->ds_flags & 708 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET); 709 710 dmu_buf_will_dirty(origin->ds_dbuf, tx); 711 origin->ds_phys->ds_num_children++; 712 713 VERIFY0(dsl_dataset_hold_obj(dp, 714 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds)); 715 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, 716 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); 717 dsl_dataset_rele(ohds, FTAG); 718 719 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { 720 if (origin->ds_phys->ds_next_clones_obj == 0) { 721 origin->ds_phys->ds_next_clones_obj = 722 zap_create(mos, 723 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 724 } 725 VERIFY0(zap_add_int(mos, 726 origin->ds_phys->ds_next_clones_obj, dsobj, tx)); 727 } 728 729 dmu_buf_will_dirty(dd->dd_dbuf, tx); 730 dd->dd_phys->dd_origin_obj = origin->ds_object; 731 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 732 if (origin->ds_dir->dd_phys->dd_clones == 0) { 733 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 734 origin->ds_dir->dd_phys->dd_clones = 735 zap_create(mos, 736 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 737 } 738 VERIFY0(zap_add_int(mos, 739 origin->ds_dir->dd_phys->dd_clones, dsobj, tx)); 740 } 741 } 742 743 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 744 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 745 746 dmu_buf_rele(dbuf, FTAG); 747 748 dmu_buf_will_dirty(dd->dd_dbuf, tx); 749 dd->dd_phys->dd_head_dataset_obj = dsobj; 750 751 return (dsobj); 752} 753 754static void 755dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx) 756{ 757 objset_t *os; 758 759 VERIFY0(dmu_objset_from_ds(ds, &os)); 760 bzero(&os->os_zil_header, sizeof (os->os_zil_header)); 761 dsl_dataset_dirty(ds, tx); 762} 763 764uint64_t 765dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, 766 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) 767{ 768 dsl_pool_t *dp = pdd->dd_pool; 769 uint64_t dsobj, ddobj; 770 dsl_dir_t *dd; 771 772 ASSERT(dmu_tx_is_syncing(tx)); 773 ASSERT(lastname[0] != '@'); 774 775 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); 776 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd)); 777 778 dsobj = dsl_dataset_create_sync_dd(dd, origin, 779 flags & ~DS_CREATE_FLAG_NODIRTY, tx); 780 781 dsl_deleg_set_create_perms(dd, tx, cr); 782 783 /* 784 * Since we're creating a new node we know it's a leaf, so we can 785 * initialize the counts if the limit feature is active. 786 */ 787 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) { 788 uint64_t cnt = 0; 789 objset_t *os = dd->dd_pool->dp_meta_objset; 790 791 dsl_dir_zapify(dd, tx); 792 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT, 793 sizeof (cnt), 1, &cnt, tx)); 794 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT, 795 sizeof (cnt), 1, &cnt, tx)); 796 } 797 798 dsl_dir_rele(dd, FTAG); 799 800 /* 801 * If we are creating a clone, make sure we zero out any stale 802 * data from the origin snapshots zil header. 803 */ 804 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) { 805 dsl_dataset_t *ds; 806 807 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 808 dsl_dataset_zero_zil(ds, tx); 809 dsl_dataset_rele(ds, FTAG); 810 } 811 812 return (dsobj); 813} 814 815#ifdef __FreeBSD__ 816/* FreeBSD ioctl compat begin */ 817struct destroyarg { 818 nvlist_t *nvl; 819 const char *snapname; 820}; 821 822static int 823dsl_check_snap_cb(const char *name, void *arg) 824{ 825 struct destroyarg *da = arg; 826 dsl_dataset_t *ds; 827 char *dsname; 828 829 dsname = kmem_asprintf("%s@%s", name, da->snapname); 830 fnvlist_add_boolean(da->nvl, dsname); 831 kmem_free(dsname, strlen(dsname) + 1); 832 833 return (0); 834} 835 836int 837dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname, 838 nvlist_t *snaps) 839{ 840 struct destroyarg *da; 841 int err; 842 843 da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP); 844 da->nvl = snaps; 845 da->snapname = snapname; 846 err = dmu_objset_find(fsname, dsl_check_snap_cb, da, 847 DS_FIND_CHILDREN); 848 kmem_free(da, sizeof (struct destroyarg)); 849 850 return (err); 851} 852/* FreeBSD ioctl compat end */ 853#endif /* __FreeBSD__ */ 854 855/* 856 * The unique space in the head dataset can be calculated by subtracting 857 * the space used in the most recent snapshot, that is still being used 858 * in this file system, from the space currently in use. To figure out 859 * the space in the most recent snapshot still in use, we need to take 860 * the total space used in the snapshot and subtract out the space that 861 * has been freed up since the snapshot was taken. 862 */ 863void 864dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) 865{ 866 uint64_t mrs_used; 867 uint64_t dlused, dlcomp, dluncomp; 868 869 ASSERT(!dsl_dataset_is_snapshot(ds)); 870 871 if (ds->ds_phys->ds_prev_snap_obj != 0) 872 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes; 873 else 874 mrs_used = 0; 875 876 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); 877 878 ASSERT3U(dlused, <=, mrs_used); 879 ds->ds_phys->ds_unique_bytes = 880 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused); 881 882 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= 883 SPA_VERSION_UNIQUE_ACCURATE) 884 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 885} 886 887void 888dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, 889 dmu_tx_t *tx) 890{ 891 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 892 uint64_t count; 893 int err; 894 895 ASSERT(ds->ds_phys->ds_num_children >= 2); 896 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx); 897 /* 898 * The err should not be ENOENT, but a bug in a previous version 899 * of the code could cause upgrade_clones_cb() to not set 900 * ds_next_snap_obj when it should, leading to a missing entry. 901 * If we knew that the pool was created after 902 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't 903 * ENOENT. However, at least we can check that we don't have 904 * too many entries in the next_clones_obj even after failing to 905 * remove this one. 906 */ 907 if (err != ENOENT) 908 VERIFY0(err); 909 ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj, 910 &count)); 911 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2); 912} 913 914 915blkptr_t * 916dsl_dataset_get_blkptr(dsl_dataset_t *ds) 917{ 918 return (&ds->ds_phys->ds_bp); 919} 920 921void 922dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) 923{ 924 ASSERT(dmu_tx_is_syncing(tx)); 925 /* If it's the meta-objset, set dp_meta_rootbp */ 926 if (ds == NULL) { 927 tx->tx_pool->dp_meta_rootbp = *bp; 928 } else { 929 dmu_buf_will_dirty(ds->ds_dbuf, tx); 930 ds->ds_phys->ds_bp = *bp; 931 } 932} 933 934spa_t * 935dsl_dataset_get_spa(dsl_dataset_t *ds) 936{ 937 return (ds->ds_dir->dd_pool->dp_spa); 938} 939 940void 941dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) 942{ 943 dsl_pool_t *dp; 944 945 if (ds == NULL) /* this is the meta-objset */ 946 return; 947 948 ASSERT(ds->ds_objset != NULL); 949 950 if (ds->ds_phys->ds_next_snap_obj != 0) 951 panic("dirtying snapshot!"); 952 953 dp = ds->ds_dir->dd_pool; 954 955 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 956 /* up the hold count until we can be written out */ 957 dmu_buf_add_ref(ds->ds_dbuf, ds); 958 } 959} 960 961boolean_t 962dsl_dataset_is_dirty(dsl_dataset_t *ds) 963{ 964 for (int t = 0; t < TXG_SIZE; t++) { 965 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets, 966 ds, t)) 967 return (B_TRUE); 968 } 969 return (B_FALSE); 970} 971 972static int 973dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) 974{ 975 uint64_t asize; 976 977 if (!dmu_tx_is_syncing(tx)) 978 return (0); 979 980 /* 981 * If there's an fs-only reservation, any blocks that might become 982 * owned by the snapshot dataset must be accommodated by space 983 * outside of the reservation. 984 */ 985 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); 986 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 987 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) 988 return (SET_ERROR(ENOSPC)); 989 990 /* 991 * Propagate any reserved space for this snapshot to other 992 * snapshot checks in this sync group. 993 */ 994 if (asize > 0) 995 dsl_dir_willuse_space(ds->ds_dir, asize, tx); 996 997 return (0); 998} 999 1000typedef struct dsl_dataset_snapshot_arg { 1001 nvlist_t *ddsa_snaps; 1002 nvlist_t *ddsa_props; 1003 nvlist_t *ddsa_errors; 1004 cred_t *ddsa_cr; 1005} dsl_dataset_snapshot_arg_t; 1006 1007int 1008dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname, 1009 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr) 1010{ 1011 int error; 1012 uint64_t value; 1013 1014 ds->ds_trysnap_txg = tx->tx_txg; 1015 1016 if (!dmu_tx_is_syncing(tx)) 1017 return (0); 1018 1019 /* 1020 * We don't allow multiple snapshots of the same txg. If there 1021 * is already one, try again. 1022 */ 1023 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) 1024 return (SET_ERROR(EAGAIN)); 1025 1026 /* 1027 * Check for conflicting snapshot name. 1028 */ 1029 error = dsl_dataset_snap_lookup(ds, snapname, &value); 1030 if (error == 0) 1031 return (SET_ERROR(EEXIST)); 1032 if (error != ENOENT) 1033 return (error); 1034 1035 /* 1036 * We don't allow taking snapshots of inconsistent datasets, such as 1037 * those into which we are currently receiving. However, if we are 1038 * creating this snapshot as part of a receive, this check will be 1039 * executed atomically with respect to the completion of the receive 1040 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this 1041 * case we ignore this, knowing it will be fixed up for us shortly in 1042 * dmu_recv_end_sync(). 1043 */ 1044 if (!recv && DS_IS_INCONSISTENT(ds)) 1045 return (SET_ERROR(EBUSY)); 1046 1047 /* 1048 * Skip the check for temporary snapshots or if we have already checked 1049 * the counts in dsl_dataset_snapshot_check. This means we really only 1050 * check the count here when we're receiving a stream. 1051 */ 1052 if (cnt != 0 && cr != NULL) { 1053 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1054 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr); 1055 if (error != 0) 1056 return (error); 1057 } 1058 1059 error = dsl_dataset_snapshot_reserve_space(ds, tx); 1060 if (error != 0) 1061 return (error); 1062 1063 return (0); 1064} 1065 1066static int 1067dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx) 1068{ 1069 dsl_dataset_snapshot_arg_t *ddsa = arg; 1070 dsl_pool_t *dp = dmu_tx_pool(tx); 1071 nvpair_t *pair; 1072 int rv = 0; 1073 1074 /* 1075 * Pre-compute how many total new snapshots will be created for each 1076 * level in the tree and below. This is needed for validating the 1077 * snapshot limit when either taking a recursive snapshot or when 1078 * taking multiple snapshots. 1079 * 1080 * The problem is that the counts are not actually adjusted when 1081 * we are checking, only when we finally sync. For a single snapshot, 1082 * this is easy, the count will increase by 1 at each node up the tree, 1083 * but its more complicated for the recursive/multiple snapshot case. 1084 * 1085 * The dsl_fs_ss_limit_check function does recursively check the count 1086 * at each level up the tree but since it is validating each snapshot 1087 * independently we need to be sure that we are validating the complete 1088 * count for the entire set of snapshots. We do this by rolling up the 1089 * counts for each component of the name into an nvlist and then 1090 * checking each of those cases with the aggregated count. 1091 * 1092 * This approach properly handles not only the recursive snapshot 1093 * case (where we get all of those on the ddsa_snaps list) but also 1094 * the sibling case (e.g. snapshot a/b and a/c so that we will also 1095 * validate the limit on 'a' using a count of 2). 1096 * 1097 * We validate the snapshot names in the third loop and only report 1098 * name errors once. 1099 */ 1100 if (dmu_tx_is_syncing(tx)) { 1101 nvlist_t *cnt_track = NULL; 1102 cnt_track = fnvlist_alloc(); 1103 1104 /* Rollup aggregated counts into the cnt_track list */ 1105 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1106 pair != NULL; 1107 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1108 char *pdelim; 1109 uint64_t val; 1110 char nm[MAXPATHLEN]; 1111 1112 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm)); 1113 pdelim = strchr(nm, '@'); 1114 if (pdelim == NULL) 1115 continue; 1116 *pdelim = '\0'; 1117 1118 do { 1119 if (nvlist_lookup_uint64(cnt_track, nm, 1120 &val) == 0) { 1121 /* update existing entry */ 1122 fnvlist_add_uint64(cnt_track, nm, 1123 val + 1); 1124 } else { 1125 /* add to list */ 1126 fnvlist_add_uint64(cnt_track, nm, 1); 1127 } 1128 1129 pdelim = strrchr(nm, '/'); 1130 if (pdelim != NULL) 1131 *pdelim = '\0'; 1132 } while (pdelim != NULL); 1133 } 1134 1135 /* Check aggregated counts at each level */ 1136 for (pair = nvlist_next_nvpair(cnt_track, NULL); 1137 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) { 1138 int error = 0; 1139 char *name; 1140 uint64_t cnt = 0; 1141 dsl_dataset_t *ds; 1142 1143 name = nvpair_name(pair); 1144 cnt = fnvpair_value_uint64(pair); 1145 ASSERT(cnt > 0); 1146 1147 error = dsl_dataset_hold(dp, name, FTAG, &ds); 1148 if (error == 0) { 1149 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt, 1150 ZFS_PROP_SNAPSHOT_LIMIT, NULL, 1151 ddsa->ddsa_cr); 1152 dsl_dataset_rele(ds, FTAG); 1153 } 1154 1155 if (error != 0) { 1156 if (ddsa->ddsa_errors != NULL) 1157 fnvlist_add_int32(ddsa->ddsa_errors, 1158 name, error); 1159 rv = error; 1160 /* only report one error for this check */ 1161 break; 1162 } 1163 } 1164 nvlist_free(cnt_track); 1165 } 1166 1167 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1168 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1169 int error = 0; 1170 dsl_dataset_t *ds; 1171 char *name, *atp; 1172 char dsname[MAXNAMELEN]; 1173 1174 name = nvpair_name(pair); 1175 if (strlen(name) >= MAXNAMELEN) 1176 error = SET_ERROR(ENAMETOOLONG); 1177 if (error == 0) { 1178 atp = strchr(name, '@'); 1179 if (atp == NULL) 1180 error = SET_ERROR(EINVAL); 1181 if (error == 0) 1182 (void) strlcpy(dsname, name, atp - name + 1); 1183 } 1184 if (error == 0) 1185 error = dsl_dataset_hold(dp, dsname, FTAG, &ds); 1186 if (error == 0) { 1187 /* passing 0/NULL skips dsl_fs_ss_limit_check */ 1188 error = dsl_dataset_snapshot_check_impl(ds, 1189 atp + 1, tx, B_FALSE, 0, NULL); 1190 dsl_dataset_rele(ds, FTAG); 1191 } 1192 1193 if (error != 0) { 1194 if (ddsa->ddsa_errors != NULL) { 1195 fnvlist_add_int32(ddsa->ddsa_errors, 1196 name, error); 1197 } 1198 rv = error; 1199 } 1200 } 1201 1202 return (rv); 1203} 1204 1205void 1206dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname, 1207 dmu_tx_t *tx) 1208{ 1209 static zil_header_t zero_zil; 1210 1211 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1212 dmu_buf_t *dbuf; 1213 dsl_dataset_phys_t *dsphys; 1214 uint64_t dsobj, crtxg; 1215 objset_t *mos = dp->dp_meta_objset; 1216 objset_t *os; 1217 1218 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1219 1220 /* 1221 * If we are on an old pool, the zil must not be active, in which 1222 * case it will be zeroed. Usually zil_suspend() accomplishes this. 1223 */ 1224 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP || 1225 dmu_objset_from_ds(ds, &os) != 0 || 1226 bcmp(&os->os_phys->os_zil_header, &zero_zil, 1227 sizeof (zero_zil)) == 0); 1228 1229 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx); 1230 1231 /* 1232 * The origin's ds_creation_txg has to be < TXG_INITIAL 1233 */ 1234 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) 1235 crtxg = 1; 1236 else 1237 crtxg = tx->tx_txg; 1238 1239 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, 1240 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); 1241 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); 1242 dmu_buf_will_dirty(dbuf, tx); 1243 dsphys = dbuf->db_data; 1244 bzero(dsphys, sizeof (dsl_dataset_phys_t)); 1245 dsphys->ds_dir_obj = ds->ds_dir->dd_object; 1246 dsphys->ds_fsid_guid = unique_create(); 1247 do { 1248 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, 1249 sizeof (dsphys->ds_guid)); 1250 } while (dsphys->ds_guid == 0); 1251 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; 1252 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; 1253 dsphys->ds_next_snap_obj = ds->ds_object; 1254 dsphys->ds_num_children = 1; 1255 dsphys->ds_creation_time = gethrestime_sec(); 1256 dsphys->ds_creation_txg = crtxg; 1257 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; 1258 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes; 1259 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; 1260 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; 1261 dsphys->ds_flags = ds->ds_phys->ds_flags; 1262 dsphys->ds_bp = ds->ds_phys->ds_bp; 1263 dmu_buf_rele(dbuf, FTAG); 1264 1265 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); 1266 if (ds->ds_prev) { 1267 uint64_t next_clones_obj = 1268 ds->ds_prev->ds_phys->ds_next_clones_obj; 1269 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == 1270 ds->ds_object || 1271 ds->ds_prev->ds_phys->ds_num_children > 1); 1272 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { 1273 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); 1274 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, 1275 ds->ds_prev->ds_phys->ds_creation_txg); 1276 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; 1277 } else if (next_clones_obj != 0) { 1278 dsl_dataset_remove_from_next_clones(ds->ds_prev, 1279 dsphys->ds_next_snap_obj, tx); 1280 VERIFY0(zap_add_int(mos, 1281 next_clones_obj, dsobj, tx)); 1282 } 1283 } 1284 1285 /* 1286 * If we have a reference-reservation on this dataset, we will 1287 * need to increase the amount of refreservation being charged 1288 * since our unique space is going to zero. 1289 */ 1290 if (ds->ds_reserved) { 1291 int64_t delta; 1292 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 1293 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); 1294 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, 1295 delta, 0, 0, tx); 1296 } 1297 1298 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1299 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist, 1300 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx); 1301 dsl_deadlist_close(&ds->ds_deadlist); 1302 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); 1303 dsl_deadlist_add_key(&ds->ds_deadlist, 1304 ds->ds_phys->ds_prev_snap_txg, tx); 1305 1306 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg); 1307 ds->ds_phys->ds_prev_snap_obj = dsobj; 1308 ds->ds_phys->ds_prev_snap_txg = crtxg; 1309 ds->ds_phys->ds_unique_bytes = 0; 1310 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) 1311 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; 1312 1313 VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, 1314 snapname, 8, 1, &dsobj, tx)); 1315 1316 if (ds->ds_prev) 1317 dsl_dataset_rele(ds->ds_prev, ds); 1318 VERIFY0(dsl_dataset_hold_obj(dp, 1319 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 1320 1321 dsl_scan_ds_snapshotted(ds, tx); 1322 1323 dsl_dir_snap_cmtime_update(ds->ds_dir); 1324 1325 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, ""); 1326} 1327 1328static void 1329dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx) 1330{ 1331 dsl_dataset_snapshot_arg_t *ddsa = arg; 1332 dsl_pool_t *dp = dmu_tx_pool(tx); 1333 nvpair_t *pair; 1334 1335 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL); 1336 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) { 1337 dsl_dataset_t *ds; 1338 char *name, *atp; 1339 char dsname[MAXNAMELEN]; 1340 1341 name = nvpair_name(pair); 1342 atp = strchr(name, '@'); 1343 (void) strlcpy(dsname, name, atp - name + 1); 1344 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds)); 1345 1346 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx); 1347 if (ddsa->ddsa_props != NULL) { 1348 dsl_props_set_sync_impl(ds->ds_prev, 1349 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx); 1350 } 1351 dsl_dataset_rele(ds, FTAG); 1352 } 1353} 1354 1355/* 1356 * The snapshots must all be in the same pool. 1357 * All-or-nothing: if there are any failures, nothing will be modified. 1358 */ 1359int 1360dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors) 1361{ 1362 dsl_dataset_snapshot_arg_t ddsa; 1363 nvpair_t *pair; 1364 boolean_t needsuspend; 1365 int error; 1366 spa_t *spa; 1367 char *firstname; 1368 nvlist_t *suspended = NULL; 1369 1370 pair = nvlist_next_nvpair(snaps, NULL); 1371 if (pair == NULL) 1372 return (0); 1373 firstname = nvpair_name(pair); 1374 1375 error = spa_open(firstname, &spa, FTAG); 1376 if (error != 0) 1377 return (error); 1378 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1379 spa_close(spa, FTAG); 1380 1381 if (needsuspend) { 1382 suspended = fnvlist_alloc(); 1383 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1384 pair = nvlist_next_nvpair(snaps, pair)) { 1385 char fsname[MAXNAMELEN]; 1386 char *snapname = nvpair_name(pair); 1387 char *atp; 1388 void *cookie; 1389 1390 atp = strchr(snapname, '@'); 1391 if (atp == NULL) { 1392 error = SET_ERROR(EINVAL); 1393 break; 1394 } 1395 (void) strlcpy(fsname, snapname, atp - snapname + 1); 1396 1397 error = zil_suspend(fsname, &cookie); 1398 if (error != 0) 1399 break; 1400 fnvlist_add_uint64(suspended, fsname, 1401 (uintptr_t)cookie); 1402 } 1403 } 1404 1405 ddsa.ddsa_snaps = snaps; 1406 ddsa.ddsa_props = props; 1407 ddsa.ddsa_errors = errors; 1408 ddsa.ddsa_cr = CRED(); 1409 1410 if (error == 0) { 1411 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check, 1412 dsl_dataset_snapshot_sync, &ddsa, 1413 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL); 1414 } 1415 1416 if (suspended != NULL) { 1417 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL; 1418 pair = nvlist_next_nvpair(suspended, pair)) { 1419 zil_resume((void *)(uintptr_t) 1420 fnvpair_value_uint64(pair)); 1421 } 1422 fnvlist_free(suspended); 1423 } 1424 1425#ifdef __FreeBSD__ 1426#ifdef _KERNEL 1427 if (error == 0) { 1428 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL; 1429 pair = nvlist_next_nvpair(snaps, pair)) { 1430 char *snapname = nvpair_name(pair); 1431 zvol_create_minors(snapname); 1432 } 1433 } 1434#endif 1435#endif 1436 return (error); 1437} 1438 1439typedef struct dsl_dataset_snapshot_tmp_arg { 1440 const char *ddsta_fsname; 1441 const char *ddsta_snapname; 1442 minor_t ddsta_cleanup_minor; 1443 const char *ddsta_htag; 1444} dsl_dataset_snapshot_tmp_arg_t; 1445 1446static int 1447dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx) 1448{ 1449 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1450 dsl_pool_t *dp = dmu_tx_pool(tx); 1451 dsl_dataset_t *ds; 1452 int error; 1453 1454 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds); 1455 if (error != 0) 1456 return (error); 1457 1458 /* NULL cred means no limit check for tmp snapshot */ 1459 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, 1460 tx, B_FALSE, 0, NULL); 1461 if (error != 0) { 1462 dsl_dataset_rele(ds, FTAG); 1463 return (error); 1464 } 1465 1466 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) { 1467 dsl_dataset_rele(ds, FTAG); 1468 return (SET_ERROR(ENOTSUP)); 1469 } 1470 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag, 1471 B_TRUE, tx); 1472 if (error != 0) { 1473 dsl_dataset_rele(ds, FTAG); 1474 return (error); 1475 } 1476 1477 dsl_dataset_rele(ds, FTAG); 1478 return (0); 1479} 1480 1481static void 1482dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx) 1483{ 1484 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg; 1485 dsl_pool_t *dp = dmu_tx_pool(tx); 1486 dsl_dataset_t *ds; 1487 1488 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds)); 1489 1490 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx); 1491 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag, 1492 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx); 1493 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx); 1494 1495 dsl_dataset_rele(ds, FTAG); 1496} 1497 1498int 1499dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname, 1500 minor_t cleanup_minor, const char *htag) 1501{ 1502 dsl_dataset_snapshot_tmp_arg_t ddsta; 1503 int error; 1504 spa_t *spa; 1505 boolean_t needsuspend; 1506 void *cookie; 1507 1508 ddsta.ddsta_fsname = fsname; 1509 ddsta.ddsta_snapname = snapname; 1510 ddsta.ddsta_cleanup_minor = cleanup_minor; 1511 ddsta.ddsta_htag = htag; 1512 1513 error = spa_open(fsname, &spa, FTAG); 1514 if (error != 0) 1515 return (error); 1516 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP); 1517 spa_close(spa, FTAG); 1518 1519 if (needsuspend) { 1520 error = zil_suspend(fsname, &cookie); 1521 if (error != 0) 1522 return (error); 1523 } 1524 1525 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check, 1526 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED); 1527 1528 if (needsuspend) 1529 zil_resume(cookie); 1530 return (error); 1531} 1532 1533 1534void 1535dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) 1536{ 1537 ASSERT(dmu_tx_is_syncing(tx)); 1538 ASSERT(ds->ds_objset != NULL); 1539 ASSERT(ds->ds_phys->ds_next_snap_obj == 0); 1540 1541 /* 1542 * in case we had to change ds_fsid_guid when we opened it, 1543 * sync it out now. 1544 */ 1545 dmu_buf_will_dirty(ds->ds_dbuf, tx); 1546 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; 1547 1548 dmu_objset_sync(ds->ds_objset, zio, tx); 1549} 1550 1551static void 1552get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv) 1553{ 1554 uint64_t count = 0; 1555 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 1556 zap_cursor_t zc; 1557 zap_attribute_t za; 1558 nvlist_t *propval = fnvlist_alloc(); 1559 nvlist_t *val = fnvlist_alloc(); 1560 1561 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 1562 1563 /* 1564 * There may be missing entries in ds_next_clones_obj 1565 * due to a bug in a previous version of the code. 1566 * Only trust it if it has the right number of entries. 1567 */ 1568 if (ds->ds_phys->ds_next_clones_obj != 0) { 1569 VERIFY0(zap_count(mos, ds->ds_phys->ds_next_clones_obj, 1570 &count)); 1571 } 1572 if (count != ds->ds_phys->ds_num_children - 1) 1573 goto fail; 1574 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj); 1575 zap_cursor_retrieve(&zc, &za) == 0; 1576 zap_cursor_advance(&zc)) { 1577 dsl_dataset_t *clone; 1578 char buf[ZFS_MAXNAMELEN]; 1579 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool, 1580 za.za_first_integer, FTAG, &clone)); 1581 dsl_dir_name(clone->ds_dir, buf); 1582 fnvlist_add_boolean(val, buf); 1583 dsl_dataset_rele(clone, FTAG); 1584 } 1585 zap_cursor_fini(&zc); 1586 fnvlist_add_nvlist(propval, ZPROP_VALUE, val); 1587 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval); 1588fail: 1589 nvlist_free(val); 1590 nvlist_free(propval); 1591} 1592 1593void 1594dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) 1595{ 1596 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1597 uint64_t refd, avail, uobjs, aobjs, ratio; 1598 1599 ASSERT(dsl_pool_config_held(dp)); 1600 1601 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 : 1602 (ds->ds_phys->ds_uncompressed_bytes * 100 / 1603 ds->ds_phys->ds_compressed_bytes); 1604 1605 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); 1606 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED, 1607 ds->ds_phys->ds_uncompressed_bytes); 1608 1609 if (dsl_dataset_is_snapshot(ds)) { 1610 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); 1611 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 1612 ds->ds_phys->ds_unique_bytes); 1613 get_clones_stat(ds, nv); 1614 } else { 1615 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) { 1616 char buf[MAXNAMELEN]; 1617 dsl_dataset_name(ds->ds_prev, buf); 1618 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf); 1619 } 1620 1621 dsl_dir_stats(ds->ds_dir, nv); 1622 } 1623 1624 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); 1625 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); 1626 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); 1627 1628 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, 1629 ds->ds_phys->ds_creation_time); 1630 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, 1631 ds->ds_phys->ds_creation_txg); 1632 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, 1633 ds->ds_quota); 1634 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, 1635 ds->ds_reserved); 1636 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, 1637 ds->ds_phys->ds_guid); 1638 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, 1639 ds->ds_phys->ds_unique_bytes); 1640 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, 1641 ds->ds_object); 1642 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, 1643 ds->ds_userrefs); 1644 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, 1645 DS_IS_DEFER_DESTROY(ds) ? 1 : 0); 1646 1647 if (ds->ds_phys->ds_prev_snap_obj != 0) { 1648 uint64_t written, comp, uncomp; 1649 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1650 dsl_dataset_t *prev; 1651 1652 int err = dsl_dataset_hold_obj(dp, 1653 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev); 1654 if (err == 0) { 1655 err = dsl_dataset_space_written(prev, ds, &written, 1656 &comp, &uncomp); 1657 dsl_dataset_rele(prev, FTAG); 1658 if (err == 0) { 1659 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN, 1660 written); 1661 } 1662 } 1663 } 1664} 1665 1666void 1667dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) 1668{ 1669 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1670 ASSERT(dsl_pool_config_held(dp)); 1671 1672 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; 1673 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; 1674 stat->dds_guid = ds->ds_phys->ds_guid; 1675 stat->dds_origin[0] = '\0'; 1676 if (dsl_dataset_is_snapshot(ds)) { 1677 stat->dds_is_snapshot = B_TRUE; 1678 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; 1679 } else { 1680 stat->dds_is_snapshot = B_FALSE; 1681 stat->dds_num_clones = 0; 1682 1683 if (dsl_dir_is_clone(ds->ds_dir)) { 1684 dsl_dataset_t *ods; 1685 1686 VERIFY0(dsl_dataset_hold_obj(dp, 1687 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods)); 1688 dsl_dataset_name(ods, stat->dds_origin); 1689 dsl_dataset_rele(ods, FTAG); 1690 } 1691 } 1692} 1693 1694uint64_t 1695dsl_dataset_fsid_guid(dsl_dataset_t *ds) 1696{ 1697 return (ds->ds_fsid_guid); 1698} 1699 1700void 1701dsl_dataset_space(dsl_dataset_t *ds, 1702 uint64_t *refdbytesp, uint64_t *availbytesp, 1703 uint64_t *usedobjsp, uint64_t *availobjsp) 1704{ 1705 *refdbytesp = ds->ds_phys->ds_referenced_bytes; 1706 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); 1707 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) 1708 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes; 1709 if (ds->ds_quota != 0) { 1710 /* 1711 * Adjust available bytes according to refquota 1712 */ 1713 if (*refdbytesp < ds->ds_quota) 1714 *availbytesp = MIN(*availbytesp, 1715 ds->ds_quota - *refdbytesp); 1716 else 1717 *availbytesp = 0; 1718 } 1719 *usedobjsp = BP_GET_FILL(&ds->ds_phys->ds_bp); 1720 *availobjsp = DN_MAX_OBJECT - *usedobjsp; 1721} 1722 1723boolean_t 1724dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap) 1725{ 1726 dsl_pool_t *dp = ds->ds_dir->dd_pool; 1727 1728 ASSERT(dsl_pool_config_held(dp)); 1729 if (snap == NULL) 1730 return (B_FALSE); 1731 if (ds->ds_phys->ds_bp.blk_birth > 1732 snap->ds_phys->ds_creation_txg) { 1733 objset_t *os, *os_snap; 1734 /* 1735 * It may be that only the ZIL differs, because it was 1736 * reset in the head. Don't count that as being 1737 * modified. 1738 */ 1739 if (dmu_objset_from_ds(ds, &os) != 0) 1740 return (B_TRUE); 1741 if (dmu_objset_from_ds(snap, &os_snap) != 0) 1742 return (B_TRUE); 1743 return (bcmp(&os->os_phys->os_meta_dnode, 1744 &os_snap->os_phys->os_meta_dnode, 1745 sizeof (os->os_phys->os_meta_dnode)) != 0); 1746 } 1747 return (B_FALSE); 1748} 1749 1750typedef struct dsl_dataset_rename_snapshot_arg { 1751 const char *ddrsa_fsname; 1752 const char *ddrsa_oldsnapname; 1753 const char *ddrsa_newsnapname; 1754 boolean_t ddrsa_recursive; 1755 dmu_tx_t *ddrsa_tx; 1756} dsl_dataset_rename_snapshot_arg_t; 1757 1758/* ARGSUSED */ 1759static int 1760dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp, 1761 dsl_dataset_t *hds, void *arg) 1762{ 1763 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1764 int error; 1765 uint64_t val; 1766 1767 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1768 if (error != 0) { 1769 /* ignore nonexistent snapshots */ 1770 return (error == ENOENT ? 0 : error); 1771 } 1772 1773 /* new name should not exist */ 1774 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val); 1775 if (error == 0) 1776 error = SET_ERROR(EEXIST); 1777 else if (error == ENOENT) 1778 error = 0; 1779 1780 /* dataset name + 1 for the "@" + the new snapshot name must fit */ 1781 if (dsl_dir_namelen(hds->ds_dir) + 1 + 1782 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN) 1783 error = SET_ERROR(ENAMETOOLONG); 1784 1785 return (error); 1786} 1787 1788static int 1789dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx) 1790{ 1791 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1792 dsl_pool_t *dp = dmu_tx_pool(tx); 1793 dsl_dataset_t *hds; 1794 int error; 1795 1796 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds); 1797 if (error != 0) 1798 return (error); 1799 1800 if (ddrsa->ddrsa_recursive) { 1801 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1802 dsl_dataset_rename_snapshot_check_impl, ddrsa, 1803 DS_FIND_CHILDREN); 1804 } else { 1805 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa); 1806 } 1807 dsl_dataset_rele(hds, FTAG); 1808 return (error); 1809} 1810 1811static int 1812dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp, 1813 dsl_dataset_t *hds, void *arg) 1814{ 1815#ifdef __FreeBSD__ 1816#ifdef _KERNEL 1817 char *oldname, *newname; 1818#endif 1819#endif 1820 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1821 dsl_dataset_t *ds; 1822 uint64_t val; 1823 dmu_tx_t *tx = ddrsa->ddrsa_tx; 1824 int error; 1825 1826 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val); 1827 ASSERT(error == 0 || error == ENOENT); 1828 if (error == ENOENT) { 1829 /* ignore nonexistent snapshots */ 1830 return (0); 1831 } 1832 1833 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds)); 1834 1835 /* log before we change the name */ 1836 spa_history_log_internal_ds(ds, "rename", tx, 1837 "-> @%s", ddrsa->ddrsa_newsnapname); 1838 1839 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx, 1840 B_FALSE)); 1841 mutex_enter(&ds->ds_lock); 1842 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname); 1843 mutex_exit(&ds->ds_lock); 1844 VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj, 1845 ds->ds_snapname, 8, 1, &ds->ds_object, tx)); 1846 1847#ifdef __FreeBSD__ 1848#ifdef _KERNEL 1849 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1850 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 1851 snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname, 1852 ddrsa->ddrsa_oldsnapname); 1853 snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname, 1854 ddrsa->ddrsa_newsnapname); 1855 zfsvfs_update_fromname(oldname, newname); 1856 zvol_rename_minors(oldname, newname); 1857 kmem_free(newname, MAXPATHLEN); 1858 kmem_free(oldname, MAXPATHLEN); 1859#endif 1860#endif 1861 dsl_dataset_rele(ds, FTAG); 1862 1863 return (0); 1864} 1865 1866static void 1867dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx) 1868{ 1869 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg; 1870 dsl_pool_t *dp = dmu_tx_pool(tx); 1871 dsl_dataset_t *hds; 1872 1873 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds)); 1874 ddrsa->ddrsa_tx = tx; 1875 if (ddrsa->ddrsa_recursive) { 1876 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object, 1877 dsl_dataset_rename_snapshot_sync_impl, ddrsa, 1878 DS_FIND_CHILDREN)); 1879 } else { 1880 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa)); 1881 } 1882 dsl_dataset_rele(hds, FTAG); 1883} 1884 1885int 1886dsl_dataset_rename_snapshot(const char *fsname, 1887 const char *oldsnapname, const char *newsnapname, boolean_t recursive) 1888{ 1889 dsl_dataset_rename_snapshot_arg_t ddrsa; 1890 1891 ddrsa.ddrsa_fsname = fsname; 1892 ddrsa.ddrsa_oldsnapname = oldsnapname; 1893 ddrsa.ddrsa_newsnapname = newsnapname; 1894 ddrsa.ddrsa_recursive = recursive; 1895 1896 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check, 1897 dsl_dataset_rename_snapshot_sync, &ddrsa, 1898 1, ZFS_SPACE_CHECK_RESERVED)); 1899} 1900 1901/* 1902 * If we're doing an ownership handoff, we need to make sure that there is 1903 * only one long hold on the dataset. We're not allowed to change anything here 1904 * so we don't permanently release the long hold or regular hold here. We want 1905 * to do this only when syncing to avoid the dataset unexpectedly going away 1906 * when we release the long hold. 1907 */ 1908static int 1909dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx) 1910{ 1911 boolean_t held; 1912 1913 if (!dmu_tx_is_syncing(tx)) 1914 return (0); 1915 1916 if (owner != NULL) { 1917 VERIFY3P(ds->ds_owner, ==, owner); 1918 dsl_dataset_long_rele(ds, owner); 1919 } 1920 1921 held = dsl_dataset_long_held(ds); 1922 1923 if (owner != NULL) 1924 dsl_dataset_long_hold(ds, owner); 1925 1926 if (held) 1927 return (SET_ERROR(EBUSY)); 1928 1929 return (0); 1930} 1931 1932typedef struct dsl_dataset_rollback_arg { 1933 const char *ddra_fsname; 1934 void *ddra_owner; 1935 nvlist_t *ddra_result; 1936} dsl_dataset_rollback_arg_t; 1937 1938static int 1939dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx) 1940{ 1941 dsl_dataset_rollback_arg_t *ddra = arg; 1942 dsl_pool_t *dp = dmu_tx_pool(tx); 1943 dsl_dataset_t *ds; 1944 int64_t unused_refres_delta; 1945 int error; 1946 1947 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds); 1948 if (error != 0) 1949 return (error); 1950 1951 /* must not be a snapshot */ 1952 if (dsl_dataset_is_snapshot(ds)) { 1953 dsl_dataset_rele(ds, FTAG); 1954 return (SET_ERROR(EINVAL)); 1955 } 1956 1957 /* must have a most recent snapshot */ 1958 if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) { 1959 dsl_dataset_rele(ds, FTAG); 1960 return (SET_ERROR(EINVAL)); 1961 } 1962 1963 /* must not have any bookmarks after the most recent snapshot */ 1964 nvlist_t *proprequest = fnvlist_alloc(); 1965 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG)); 1966 nvlist_t *bookmarks = fnvlist_alloc(); 1967 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks); 1968 fnvlist_free(proprequest); 1969 if (error != 0) 1970 return (error); 1971 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL); 1972 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) { 1973 nvlist_t *valuenv = 1974 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair), 1975 zfs_prop_to_name(ZFS_PROP_CREATETXG)); 1976 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value"); 1977 if (createtxg > ds->ds_phys->ds_prev_snap_txg) { 1978 fnvlist_free(bookmarks); 1979 dsl_dataset_rele(ds, FTAG); 1980 return (SET_ERROR(EEXIST)); 1981 } 1982 } 1983 fnvlist_free(bookmarks); 1984 1985 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx); 1986 if (error != 0) { 1987 dsl_dataset_rele(ds, FTAG); 1988 return (error); 1989 } 1990 1991 /* 1992 * Check if the snap we are rolling back to uses more than 1993 * the refquota. 1994 */ 1995 if (ds->ds_quota != 0 && 1996 ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) { 1997 dsl_dataset_rele(ds, FTAG); 1998 return (SET_ERROR(EDQUOT)); 1999 } 2000 2001 /* 2002 * When we do the clone swap, we will temporarily use more space 2003 * due to the refreservation (the head will no longer have any 2004 * unique space, so the entire amount of the refreservation will need 2005 * to be free). We will immediately destroy the clone, freeing 2006 * this space, but the freeing happens over many txg's. 2007 */ 2008 unused_refres_delta = (int64_t)MIN(ds->ds_reserved, 2009 ds->ds_phys->ds_unique_bytes); 2010 2011 if (unused_refres_delta > 0 && 2012 unused_refres_delta > 2013 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) { 2014 dsl_dataset_rele(ds, FTAG); 2015 return (SET_ERROR(ENOSPC)); 2016 } 2017 2018 dsl_dataset_rele(ds, FTAG); 2019 return (0); 2020} 2021 2022static void 2023dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx) 2024{ 2025 dsl_dataset_rollback_arg_t *ddra = arg; 2026 dsl_pool_t *dp = dmu_tx_pool(tx); 2027 dsl_dataset_t *ds, *clone; 2028 uint64_t cloneobj; 2029 char namebuf[ZFS_MAXNAMELEN]; 2030 2031 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds)); 2032 2033 dsl_dataset_name(ds->ds_prev, namebuf); 2034 fnvlist_add_string(ddra->ddra_result, "target", namebuf); 2035 2036 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback", 2037 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx); 2038 2039 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone)); 2040 2041 dsl_dataset_clone_swap_sync_impl(clone, ds, tx); 2042 dsl_dataset_zero_zil(ds, tx); 2043 2044 dsl_destroy_head_sync_impl(clone, tx); 2045 2046 dsl_dataset_rele(clone, FTAG); 2047 dsl_dataset_rele(ds, FTAG); 2048} 2049 2050/* 2051 * Rolls back the given filesystem or volume to the most recent snapshot. 2052 * The name of the most recent snapshot will be returned under key "target" 2053 * in the result nvlist. 2054 * 2055 * If owner != NULL: 2056 * - The existing dataset MUST be owned by the specified owner at entry 2057 * - Upon return, dataset will still be held by the same owner, whether we 2058 * succeed or not. 2059 * 2060 * This mode is required any time the existing filesystem is mounted. See 2061 * notes above zfs_suspend_fs() for further details. 2062 */ 2063int 2064dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result) 2065{ 2066 dsl_dataset_rollback_arg_t ddra; 2067 2068 ddra.ddra_fsname = fsname; 2069 ddra.ddra_owner = owner; 2070 ddra.ddra_result = result; 2071 2072 return (dsl_sync_task(fsname, dsl_dataset_rollback_check, 2073 dsl_dataset_rollback_sync, &ddra, 2074 1, ZFS_SPACE_CHECK_RESERVED)); 2075} 2076 2077struct promotenode { 2078 list_node_t link; 2079 dsl_dataset_t *ds; 2080}; 2081 2082typedef struct dsl_dataset_promote_arg { 2083 const char *ddpa_clonename; 2084 dsl_dataset_t *ddpa_clone; 2085 list_t shared_snaps, origin_snaps, clone_snaps; 2086 dsl_dataset_t *origin_origin; /* origin of the origin */ 2087 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; 2088 char *err_ds; 2089 cred_t *cr; 2090} dsl_dataset_promote_arg_t; 2091 2092static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); 2093static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, 2094 void *tag); 2095static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag); 2096 2097static int 2098dsl_dataset_promote_check(void *arg, dmu_tx_t *tx) 2099{ 2100 dsl_dataset_promote_arg_t *ddpa = arg; 2101 dsl_pool_t *dp = dmu_tx_pool(tx); 2102 dsl_dataset_t *hds; 2103 struct promotenode *snap; 2104 dsl_dataset_t *origin_ds; 2105 int err; 2106 uint64_t unused; 2107 uint64_t ss_mv_cnt; 2108 2109 err = promote_hold(ddpa, dp, FTAG); 2110 if (err != 0) 2111 return (err); 2112 2113 hds = ddpa->ddpa_clone; 2114 2115 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) { 2116 promote_rele(ddpa, FTAG); 2117 return (SET_ERROR(EXDEV)); 2118 } 2119 2120 /* 2121 * Compute and check the amount of space to transfer. Since this is 2122 * so expensive, don't do the preliminary check. 2123 */ 2124 if (!dmu_tx_is_syncing(tx)) { 2125 promote_rele(ddpa, FTAG); 2126 return (0); 2127 } 2128 2129 snap = list_head(&ddpa->shared_snaps); 2130 origin_ds = snap->ds; 2131 2132 /* compute origin's new unique space */ 2133 snap = list_tail(&ddpa->clone_snaps); 2134 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2135 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2136 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2137 &ddpa->unique, &unused, &unused); 2138 2139 /* 2140 * Walk the snapshots that we are moving 2141 * 2142 * Compute space to transfer. Consider the incremental changes 2143 * to used by each snapshot: 2144 * (my used) = (prev's used) + (blocks born) - (blocks killed) 2145 * So each snapshot gave birth to: 2146 * (blocks born) = (my used) - (prev's used) + (blocks killed) 2147 * So a sequence would look like: 2148 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) 2149 * Which simplifies to: 2150 * uN + kN + kN-1 + ... + k1 + k0 2151 * Note however, if we stop before we reach the ORIGIN we get: 2152 * uN + kN + kN-1 + ... + kM - uM-1 2153 */ 2154 ss_mv_cnt = 0; 2155 ddpa->used = origin_ds->ds_phys->ds_referenced_bytes; 2156 ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes; 2157 ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; 2158 for (snap = list_head(&ddpa->shared_snaps); snap; 2159 snap = list_next(&ddpa->shared_snaps, snap)) { 2160 uint64_t val, dlused, dlcomp, dluncomp; 2161 dsl_dataset_t *ds = snap->ds; 2162 2163 ss_mv_cnt++; 2164 2165 /* 2166 * If there are long holds, we won't be able to evict 2167 * the objset. 2168 */ 2169 if (dsl_dataset_long_held(ds)) { 2170 err = SET_ERROR(EBUSY); 2171 goto out; 2172 } 2173 2174 /* Check that the snapshot name does not conflict */ 2175 VERIFY0(dsl_dataset_get_snapname(ds)); 2176 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); 2177 if (err == 0) { 2178 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname); 2179 err = SET_ERROR(EEXIST); 2180 goto out; 2181 } 2182 if (err != ENOENT) 2183 goto out; 2184 2185 /* The very first snapshot does not have a deadlist */ 2186 if (ds->ds_phys->ds_prev_snap_obj == 0) 2187 continue; 2188 2189 dsl_deadlist_space(&ds->ds_deadlist, 2190 &dlused, &dlcomp, &dluncomp); 2191 ddpa->used += dlused; 2192 ddpa->comp += dlcomp; 2193 ddpa->uncomp += dluncomp; 2194 } 2195 2196 /* 2197 * If we are a clone of a clone then we never reached ORIGIN, 2198 * so we need to subtract out the clone origin's used space. 2199 */ 2200 if (ddpa->origin_origin) { 2201 ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes; 2202 ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes; 2203 ddpa->uncomp -= 2204 ddpa->origin_origin->ds_phys->ds_uncompressed_bytes; 2205 } 2206 2207 /* Check that there is enough space and limit headroom here */ 2208 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, 2209 0, ss_mv_cnt, ddpa->used, ddpa->cr); 2210 if (err != 0) 2211 goto out; 2212 2213 /* 2214 * Compute the amounts of space that will be used by snapshots 2215 * after the promotion (for both origin and clone). For each, 2216 * it is the amount of space that will be on all of their 2217 * deadlists (that was not born before their new origin). 2218 */ 2219 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2220 uint64_t space; 2221 2222 /* 2223 * Note, typically this will not be a clone of a clone, 2224 * so dd_origin_txg will be < TXG_INITIAL, so 2225 * these snaplist_space() -> dsl_deadlist_space_range() 2226 * calls will be fast because they do not have to 2227 * iterate over all bps. 2228 */ 2229 snap = list_head(&ddpa->origin_snaps); 2230 err = snaplist_space(&ddpa->shared_snaps, 2231 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap); 2232 if (err != 0) 2233 goto out; 2234 2235 err = snaplist_space(&ddpa->clone_snaps, 2236 snap->ds->ds_dir->dd_origin_txg, &space); 2237 if (err != 0) 2238 goto out; 2239 ddpa->cloneusedsnap += space; 2240 } 2241 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 2242 err = snaplist_space(&ddpa->origin_snaps, 2243 origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap); 2244 if (err != 0) 2245 goto out; 2246 } 2247 2248out: 2249 promote_rele(ddpa, FTAG); 2250 return (err); 2251} 2252 2253static void 2254dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx) 2255{ 2256 dsl_dataset_promote_arg_t *ddpa = arg; 2257 dsl_pool_t *dp = dmu_tx_pool(tx); 2258 dsl_dataset_t *hds; 2259 struct promotenode *snap; 2260 dsl_dataset_t *origin_ds; 2261 dsl_dataset_t *origin_head; 2262 dsl_dir_t *dd; 2263 dsl_dir_t *odd = NULL; 2264 uint64_t oldnext_obj; 2265 int64_t delta; 2266#if defined(__FreeBSD__) && defined(_KERNEL) 2267 char *oldname, *newname; 2268#endif 2269 2270 VERIFY0(promote_hold(ddpa, dp, FTAG)); 2271 hds = ddpa->ddpa_clone; 2272 2273 ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE); 2274 2275 snap = list_head(&ddpa->shared_snaps); 2276 origin_ds = snap->ds; 2277 dd = hds->ds_dir; 2278 2279 snap = list_head(&ddpa->origin_snaps); 2280 origin_head = snap->ds; 2281 2282 /* 2283 * We need to explicitly open odd, since origin_ds's dd will be 2284 * changing. 2285 */ 2286 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object, 2287 NULL, FTAG, &odd)); 2288 2289 /* change origin's next snap */ 2290 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); 2291 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj; 2292 snap = list_tail(&ddpa->clone_snaps); 2293 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); 2294 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object; 2295 2296 /* change the origin's next clone */ 2297 if (origin_ds->ds_phys->ds_next_clones_obj) { 2298 dsl_dataset_remove_from_next_clones(origin_ds, 2299 snap->ds->ds_object, tx); 2300 VERIFY0(zap_add_int(dp->dp_meta_objset, 2301 origin_ds->ds_phys->ds_next_clones_obj, 2302 oldnext_obj, tx)); 2303 } 2304 2305 /* change origin */ 2306 dmu_buf_will_dirty(dd->dd_dbuf, tx); 2307 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object); 2308 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj; 2309 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; 2310 dmu_buf_will_dirty(odd->dd_dbuf, tx); 2311 odd->dd_phys->dd_origin_obj = origin_ds->ds_object; 2312 origin_head->ds_dir->dd_origin_txg = 2313 origin_ds->ds_phys->ds_creation_txg; 2314 2315 /* change dd_clone entries */ 2316 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2317 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2318 odd->dd_phys->dd_clones, hds->ds_object, tx)); 2319 VERIFY0(zap_add_int(dp->dp_meta_objset, 2320 ddpa->origin_origin->ds_dir->dd_phys->dd_clones, 2321 hds->ds_object, tx)); 2322 2323 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2324 ddpa->origin_origin->ds_dir->dd_phys->dd_clones, 2325 origin_head->ds_object, tx)); 2326 if (dd->dd_phys->dd_clones == 0) { 2327 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset, 2328 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 2329 } 2330 VERIFY0(zap_add_int(dp->dp_meta_objset, 2331 dd->dd_phys->dd_clones, origin_head->ds_object, tx)); 2332 } 2333 2334#if defined(__FreeBSD__) && defined(_KERNEL) 2335 /* Take the spa_namespace_lock early so zvol renames don't deadlock. */ 2336 mutex_enter(&spa_namespace_lock); 2337 2338 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2339 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2340#endif 2341 2342 /* move snapshots to this dir */ 2343 for (snap = list_head(&ddpa->shared_snaps); snap; 2344 snap = list_next(&ddpa->shared_snaps, snap)) { 2345 dsl_dataset_t *ds = snap->ds; 2346 2347 /* 2348 * Property callbacks are registered to a particular 2349 * dsl_dir. Since ours is changing, evict the objset 2350 * so that they will be unregistered from the old dsl_dir. 2351 */ 2352 if (ds->ds_objset) { 2353 dmu_objset_evict(ds->ds_objset); 2354 ds->ds_objset = NULL; 2355 } 2356 2357 /* move snap name entry */ 2358 VERIFY0(dsl_dataset_get_snapname(ds)); 2359 VERIFY0(dsl_dataset_snap_remove(origin_head, 2360 ds->ds_snapname, tx, B_TRUE)); 2361 VERIFY0(zap_add(dp->dp_meta_objset, 2362 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, 2363 8, 1, &ds->ds_object, tx)); 2364 dsl_fs_ss_count_adjust(hds->ds_dir, 1, 2365 DD_FIELD_SNAPSHOT_COUNT, tx); 2366 2367 /* change containing dsl_dir */ 2368 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2369 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object); 2370 ds->ds_phys->ds_dir_obj = dd->dd_object; 2371 ASSERT3P(ds->ds_dir, ==, odd); 2372 dsl_dir_rele(ds->ds_dir, ds); 2373 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object, 2374 NULL, ds, &ds->ds_dir)); 2375 2376#if defined(__FreeBSD__) && defined(_KERNEL) 2377 dsl_dataset_name(ds, newname); 2378 zfsvfs_update_fromname(oldname, newname); 2379 zvol_rename_minors(oldname, newname); 2380#endif 2381 2382 /* move any clone references */ 2383 if (ds->ds_phys->ds_next_clones_obj && 2384 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { 2385 zap_cursor_t zc; 2386 zap_attribute_t za; 2387 2388 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2389 ds->ds_phys->ds_next_clones_obj); 2390 zap_cursor_retrieve(&zc, &za) == 0; 2391 zap_cursor_advance(&zc)) { 2392 dsl_dataset_t *cnds; 2393 uint64_t o; 2394 2395 if (za.za_first_integer == oldnext_obj) { 2396 /* 2397 * We've already moved the 2398 * origin's reference. 2399 */ 2400 continue; 2401 } 2402 2403 VERIFY0(dsl_dataset_hold_obj(dp, 2404 za.za_first_integer, FTAG, &cnds)); 2405 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj; 2406 2407 VERIFY0(zap_remove_int(dp->dp_meta_objset, 2408 odd->dd_phys->dd_clones, o, tx)); 2409 VERIFY0(zap_add_int(dp->dp_meta_objset, 2410 dd->dd_phys->dd_clones, o, tx)); 2411 dsl_dataset_rele(cnds, FTAG); 2412 } 2413 zap_cursor_fini(&zc); 2414 } 2415 2416 ASSERT(!dsl_prop_hascb(ds)); 2417 } 2418 2419#if defined(__FreeBSD__) && defined(_KERNEL) 2420 mutex_exit(&spa_namespace_lock); 2421 2422 kmem_free(newname, MAXPATHLEN); 2423 kmem_free(oldname, MAXPATHLEN); 2424#endif 2425 /* 2426 * Change space accounting. 2427 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either 2428 * both be valid, or both be 0 (resulting in delta == 0). This 2429 * is true for each of {clone,origin} independently. 2430 */ 2431 2432 delta = ddpa->cloneusedsnap - 2433 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2434 ASSERT3S(delta, >=, 0); 2435 ASSERT3U(ddpa->used, >=, delta); 2436 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); 2437 dsl_dir_diduse_space(dd, DD_USED_HEAD, 2438 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx); 2439 2440 delta = ddpa->originusedsnap - 2441 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; 2442 ASSERT3S(delta, <=, 0); 2443 ASSERT3U(ddpa->used, >=, -delta); 2444 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); 2445 dsl_dir_diduse_space(odd, DD_USED_HEAD, 2446 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx); 2447 2448 origin_ds->ds_phys->ds_unique_bytes = ddpa->unique; 2449 2450 /* log history record */ 2451 spa_history_log_internal_ds(hds, "promote", tx, ""); 2452 2453 dsl_dir_rele(odd, FTAG); 2454 promote_rele(ddpa, FTAG); 2455} 2456 2457/* 2458 * Make a list of dsl_dataset_t's for the snapshots between first_obj 2459 * (exclusive) and last_obj (inclusive). The list will be in reverse 2460 * order (last_obj will be the list_head()). If first_obj == 0, do all 2461 * snapshots back to this dataset's origin. 2462 */ 2463static int 2464snaplist_make(dsl_pool_t *dp, 2465 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag) 2466{ 2467 uint64_t obj = last_obj; 2468 2469 list_create(l, sizeof (struct promotenode), 2470 offsetof(struct promotenode, link)); 2471 2472 while (obj != first_obj) { 2473 dsl_dataset_t *ds; 2474 struct promotenode *snap; 2475 int err; 2476 2477 err = dsl_dataset_hold_obj(dp, obj, tag, &ds); 2478 ASSERT(err != ENOENT); 2479 if (err != 0) 2480 return (err); 2481 2482 if (first_obj == 0) 2483 first_obj = ds->ds_dir->dd_phys->dd_origin_obj; 2484 2485 snap = kmem_alloc(sizeof (*snap), KM_SLEEP); 2486 snap->ds = ds; 2487 list_insert_tail(l, snap); 2488 obj = ds->ds_phys->ds_prev_snap_obj; 2489 } 2490 2491 return (0); 2492} 2493 2494static int 2495snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) 2496{ 2497 struct promotenode *snap; 2498 2499 *spacep = 0; 2500 for (snap = list_head(l); snap; snap = list_next(l, snap)) { 2501 uint64_t used, comp, uncomp; 2502 dsl_deadlist_space_range(&snap->ds->ds_deadlist, 2503 mintxg, UINT64_MAX, &used, &comp, &uncomp); 2504 *spacep += used; 2505 } 2506 return (0); 2507} 2508 2509static void 2510snaplist_destroy(list_t *l, void *tag) 2511{ 2512 struct promotenode *snap; 2513 2514 if (l == NULL || !list_link_active(&l->list_head)) 2515 return; 2516 2517 while ((snap = list_tail(l)) != NULL) { 2518 list_remove(l, snap); 2519 dsl_dataset_rele(snap->ds, tag); 2520 kmem_free(snap, sizeof (*snap)); 2521 } 2522 list_destroy(l); 2523} 2524 2525static int 2526promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag) 2527{ 2528 int error; 2529 dsl_dir_t *dd; 2530 struct promotenode *snap; 2531 2532 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag, 2533 &ddpa->ddpa_clone); 2534 if (error != 0) 2535 return (error); 2536 dd = ddpa->ddpa_clone->ds_dir; 2537 2538 if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) || 2539 !dsl_dir_is_clone(dd)) { 2540 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2541 return (SET_ERROR(EINVAL)); 2542 } 2543 2544 error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj, 2545 &ddpa->shared_snaps, tag); 2546 if (error != 0) 2547 goto out; 2548 2549 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object, 2550 &ddpa->clone_snaps, tag); 2551 if (error != 0) 2552 goto out; 2553 2554 snap = list_head(&ddpa->shared_snaps); 2555 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); 2556 error = snaplist_make(dp, dd->dd_phys->dd_origin_obj, 2557 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, 2558 &ddpa->origin_snaps, tag); 2559 if (error != 0) 2560 goto out; 2561 2562 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) { 2563 error = dsl_dataset_hold_obj(dp, 2564 snap->ds->ds_dir->dd_phys->dd_origin_obj, 2565 tag, &ddpa->origin_origin); 2566 if (error != 0) 2567 goto out; 2568 } 2569out: 2570 if (error != 0) 2571 promote_rele(ddpa, tag); 2572 return (error); 2573} 2574 2575static void 2576promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag) 2577{ 2578 snaplist_destroy(&ddpa->shared_snaps, tag); 2579 snaplist_destroy(&ddpa->clone_snaps, tag); 2580 snaplist_destroy(&ddpa->origin_snaps, tag); 2581 if (ddpa->origin_origin != NULL) 2582 dsl_dataset_rele(ddpa->origin_origin, tag); 2583 dsl_dataset_rele(ddpa->ddpa_clone, tag); 2584} 2585 2586/* 2587 * Promote a clone. 2588 * 2589 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled 2590 * in with the name. (It must be at least MAXNAMELEN bytes long.) 2591 */ 2592int 2593dsl_dataset_promote(const char *name, char *conflsnap) 2594{ 2595 dsl_dataset_promote_arg_t ddpa = { 0 }; 2596 uint64_t numsnaps; 2597 int error; 2598 objset_t *os; 2599 2600 /* 2601 * We will modify space proportional to the number of 2602 * snapshots. Compute numsnaps. 2603 */ 2604 error = dmu_objset_hold(name, FTAG, &os); 2605 if (error != 0) 2606 return (error); 2607 error = zap_count(dmu_objset_pool(os)->dp_meta_objset, 2608 dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps); 2609 dmu_objset_rele(os, FTAG); 2610 if (error != 0) 2611 return (error); 2612 2613 ddpa.ddpa_clonename = name; 2614 ddpa.err_ds = conflsnap; 2615 ddpa.cr = CRED(); 2616 2617 return (dsl_sync_task(name, dsl_dataset_promote_check, 2618 dsl_dataset_promote_sync, &ddpa, 2619 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED)); 2620} 2621 2622int 2623dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone, 2624 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx) 2625{ 2626 int64_t unused_refres_delta; 2627 2628 /* they should both be heads */ 2629 if (dsl_dataset_is_snapshot(clone) || 2630 dsl_dataset_is_snapshot(origin_head)) 2631 return (SET_ERROR(EINVAL)); 2632 2633 /* if we are not forcing, the branch point should be just before them */ 2634 if (!force && clone->ds_prev != origin_head->ds_prev) 2635 return (SET_ERROR(EINVAL)); 2636 2637 /* clone should be the clone (unless they are unrelated) */ 2638 if (clone->ds_prev != NULL && 2639 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap && 2640 origin_head->ds_dir != clone->ds_prev->ds_dir) 2641 return (SET_ERROR(EINVAL)); 2642 2643 /* the clone should be a child of the origin */ 2644 if (clone->ds_dir->dd_parent != origin_head->ds_dir) 2645 return (SET_ERROR(EINVAL)); 2646 2647 /* origin_head shouldn't be modified unless 'force' */ 2648 if (!force && 2649 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev)) 2650 return (SET_ERROR(ETXTBSY)); 2651 2652 /* origin_head should have no long holds (e.g. is not mounted) */ 2653 if (dsl_dataset_handoff_check(origin_head, owner, tx)) 2654 return (SET_ERROR(EBUSY)); 2655 2656 /* check amount of any unconsumed refreservation */ 2657 unused_refres_delta = 2658 (int64_t)MIN(origin_head->ds_reserved, 2659 origin_head->ds_phys->ds_unique_bytes) - 2660 (int64_t)MIN(origin_head->ds_reserved, 2661 clone->ds_phys->ds_unique_bytes); 2662 2663 if (unused_refres_delta > 0 && 2664 unused_refres_delta > 2665 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE)) 2666 return (SET_ERROR(ENOSPC)); 2667 2668 /* clone can't be over the head's refquota */ 2669 if (origin_head->ds_quota != 0 && 2670 clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota) 2671 return (SET_ERROR(EDQUOT)); 2672 2673 return (0); 2674} 2675 2676void 2677dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone, 2678 dsl_dataset_t *origin_head, dmu_tx_t *tx) 2679{ 2680 dsl_pool_t *dp = dmu_tx_pool(tx); 2681 int64_t unused_refres_delta; 2682 2683 ASSERT(clone->ds_reserved == 0); 2684 ASSERT(origin_head->ds_quota == 0 || 2685 clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota); 2686 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev); 2687 2688 dmu_buf_will_dirty(clone->ds_dbuf, tx); 2689 dmu_buf_will_dirty(origin_head->ds_dbuf, tx); 2690 2691 if (clone->ds_objset != NULL) { 2692 dmu_objset_evict(clone->ds_objset); 2693 clone->ds_objset = NULL; 2694 } 2695 2696 if (origin_head->ds_objset != NULL) { 2697 dmu_objset_evict(origin_head->ds_objset); 2698 origin_head->ds_objset = NULL; 2699 } 2700 2701 unused_refres_delta = 2702 (int64_t)MIN(origin_head->ds_reserved, 2703 origin_head->ds_phys->ds_unique_bytes) - 2704 (int64_t)MIN(origin_head->ds_reserved, 2705 clone->ds_phys->ds_unique_bytes); 2706 2707 /* 2708 * Reset origin's unique bytes, if it exists. 2709 */ 2710 if (clone->ds_prev) { 2711 dsl_dataset_t *origin = clone->ds_prev; 2712 uint64_t comp, uncomp; 2713 2714 dmu_buf_will_dirty(origin->ds_dbuf, tx); 2715 dsl_deadlist_space_range(&clone->ds_deadlist, 2716 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX, 2717 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp); 2718 } 2719 2720 /* swap blkptrs */ 2721 { 2722 blkptr_t tmp; 2723 tmp = origin_head->ds_phys->ds_bp; 2724 origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp; 2725 clone->ds_phys->ds_bp = tmp; 2726 } 2727 2728 /* set dd_*_bytes */ 2729 { 2730 int64_t dused, dcomp, duncomp; 2731 uint64_t cdl_used, cdl_comp, cdl_uncomp; 2732 uint64_t odl_used, odl_comp, odl_uncomp; 2733 2734 ASSERT3U(clone->ds_dir->dd_phys-> 2735 dd_used_breakdown[DD_USED_SNAP], ==, 0); 2736 2737 dsl_deadlist_space(&clone->ds_deadlist, 2738 &cdl_used, &cdl_comp, &cdl_uncomp); 2739 dsl_deadlist_space(&origin_head->ds_deadlist, 2740 &odl_used, &odl_comp, &odl_uncomp); 2741 2742 dused = clone->ds_phys->ds_referenced_bytes + cdl_used - 2743 (origin_head->ds_phys->ds_referenced_bytes + odl_used); 2744 dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp - 2745 (origin_head->ds_phys->ds_compressed_bytes + odl_comp); 2746 duncomp = clone->ds_phys->ds_uncompressed_bytes + 2747 cdl_uncomp - 2748 (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp); 2749 2750 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD, 2751 dused, dcomp, duncomp, tx); 2752 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD, 2753 -dused, -dcomp, -duncomp, tx); 2754 2755 /* 2756 * The difference in the space used by snapshots is the 2757 * difference in snapshot space due to the head's 2758 * deadlist (since that's the only thing that's 2759 * changing that affects the snapused). 2760 */ 2761 dsl_deadlist_space_range(&clone->ds_deadlist, 2762 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2763 &cdl_used, &cdl_comp, &cdl_uncomp); 2764 dsl_deadlist_space_range(&origin_head->ds_deadlist, 2765 origin_head->ds_dir->dd_origin_txg, UINT64_MAX, 2766 &odl_used, &odl_comp, &odl_uncomp); 2767 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used, 2768 DD_USED_HEAD, DD_USED_SNAP, tx); 2769 } 2770 2771 /* swap ds_*_bytes */ 2772 SWITCH64(origin_head->ds_phys->ds_referenced_bytes, 2773 clone->ds_phys->ds_referenced_bytes); 2774 SWITCH64(origin_head->ds_phys->ds_compressed_bytes, 2775 clone->ds_phys->ds_compressed_bytes); 2776 SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes, 2777 clone->ds_phys->ds_uncompressed_bytes); 2778 SWITCH64(origin_head->ds_phys->ds_unique_bytes, 2779 clone->ds_phys->ds_unique_bytes); 2780 2781 /* apply any parent delta for change in unconsumed refreservation */ 2782 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV, 2783 unused_refres_delta, 0, 0, tx); 2784 2785 /* 2786 * Swap deadlists. 2787 */ 2788 dsl_deadlist_close(&clone->ds_deadlist); 2789 dsl_deadlist_close(&origin_head->ds_deadlist); 2790 SWITCH64(origin_head->ds_phys->ds_deadlist_obj, 2791 clone->ds_phys->ds_deadlist_obj); 2792 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset, 2793 clone->ds_phys->ds_deadlist_obj); 2794 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset, 2795 origin_head->ds_phys->ds_deadlist_obj); 2796 2797 dsl_scan_ds_clone_swapped(origin_head, clone, tx); 2798 2799 spa_history_log_internal_ds(clone, "clone swap", tx, 2800 "parent=%s", origin_head->ds_dir->dd_myname); 2801} 2802 2803/* 2804 * Given a pool name and a dataset object number in that pool, 2805 * return the name of that dataset. 2806 */ 2807int 2808dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) 2809{ 2810 dsl_pool_t *dp; 2811 dsl_dataset_t *ds; 2812 int error; 2813 2814 error = dsl_pool_hold(pname, FTAG, &dp); 2815 if (error != 0) 2816 return (error); 2817 2818 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds); 2819 if (error == 0) { 2820 dsl_dataset_name(ds, buf); 2821 dsl_dataset_rele(ds, FTAG); 2822 } 2823 dsl_pool_rele(dp, FTAG); 2824 2825 return (error); 2826} 2827 2828int 2829dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, 2830 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) 2831{ 2832 int error = 0; 2833 2834 ASSERT3S(asize, >, 0); 2835 2836 /* 2837 * *ref_rsrv is the portion of asize that will come from any 2838 * unconsumed refreservation space. 2839 */ 2840 *ref_rsrv = 0; 2841 2842 mutex_enter(&ds->ds_lock); 2843 /* 2844 * Make a space adjustment for reserved bytes. 2845 */ 2846 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) { 2847 ASSERT3U(*used, >=, 2848 ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 2849 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes); 2850 *ref_rsrv = 2851 asize - MIN(asize, parent_delta(ds, asize + inflight)); 2852 } 2853 2854 if (!check_quota || ds->ds_quota == 0) { 2855 mutex_exit(&ds->ds_lock); 2856 return (0); 2857 } 2858 /* 2859 * If they are requesting more space, and our current estimate 2860 * is over quota, they get to try again unless the actual 2861 * on-disk is over quota and there are no pending changes (which 2862 * may free up space for us). 2863 */ 2864 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) { 2865 if (inflight > 0 || 2866 ds->ds_phys->ds_referenced_bytes < ds->ds_quota) 2867 error = SET_ERROR(ERESTART); 2868 else 2869 error = SET_ERROR(EDQUOT); 2870 } 2871 mutex_exit(&ds->ds_lock); 2872 2873 return (error); 2874} 2875 2876typedef struct dsl_dataset_set_qr_arg { 2877 const char *ddsqra_name; 2878 zprop_source_t ddsqra_source; 2879 uint64_t ddsqra_value; 2880} dsl_dataset_set_qr_arg_t; 2881 2882 2883/* ARGSUSED */ 2884static int 2885dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx) 2886{ 2887 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2888 dsl_pool_t *dp = dmu_tx_pool(tx); 2889 dsl_dataset_t *ds; 2890 int error; 2891 uint64_t newval; 2892 2893 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA) 2894 return (SET_ERROR(ENOTSUP)); 2895 2896 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 2897 if (error != 0) 2898 return (error); 2899 2900 if (dsl_dataset_is_snapshot(ds)) { 2901 dsl_dataset_rele(ds, FTAG); 2902 return (SET_ERROR(EINVAL)); 2903 } 2904 2905 error = dsl_prop_predict(ds->ds_dir, 2906 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 2907 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 2908 if (error != 0) { 2909 dsl_dataset_rele(ds, FTAG); 2910 return (error); 2911 } 2912 2913 if (newval == 0) { 2914 dsl_dataset_rele(ds, FTAG); 2915 return (0); 2916 } 2917 2918 if (newval < ds->ds_phys->ds_referenced_bytes || 2919 newval < ds->ds_reserved) { 2920 dsl_dataset_rele(ds, FTAG); 2921 return (SET_ERROR(ENOSPC)); 2922 } 2923 2924 dsl_dataset_rele(ds, FTAG); 2925 return (0); 2926} 2927 2928static void 2929dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx) 2930{ 2931 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2932 dsl_pool_t *dp = dmu_tx_pool(tx); 2933 dsl_dataset_t *ds; 2934 uint64_t newval; 2935 2936 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 2937 2938 dsl_prop_set_sync_impl(ds, 2939 zfs_prop_to_name(ZFS_PROP_REFQUOTA), 2940 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1, 2941 &ddsqra->ddsqra_value, tx); 2942 2943 VERIFY0(dsl_prop_get_int_ds(ds, 2944 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval)); 2945 2946 if (ds->ds_quota != newval) { 2947 dmu_buf_will_dirty(ds->ds_dbuf, tx); 2948 ds->ds_quota = newval; 2949 } 2950 dsl_dataset_rele(ds, FTAG); 2951} 2952 2953int 2954dsl_dataset_set_refquota(const char *dsname, zprop_source_t source, 2955 uint64_t refquota) 2956{ 2957 dsl_dataset_set_qr_arg_t ddsqra; 2958 2959 ddsqra.ddsqra_name = dsname; 2960 ddsqra.ddsqra_source = source; 2961 ddsqra.ddsqra_value = refquota; 2962 2963 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check, 2964 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE)); 2965} 2966 2967static int 2968dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx) 2969{ 2970 dsl_dataset_set_qr_arg_t *ddsqra = arg; 2971 dsl_pool_t *dp = dmu_tx_pool(tx); 2972 dsl_dataset_t *ds; 2973 int error; 2974 uint64_t newval, unique; 2975 2976 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION) 2977 return (SET_ERROR(ENOTSUP)); 2978 2979 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds); 2980 if (error != 0) 2981 return (error); 2982 2983 if (dsl_dataset_is_snapshot(ds)) { 2984 dsl_dataset_rele(ds, FTAG); 2985 return (SET_ERROR(EINVAL)); 2986 } 2987 2988 error = dsl_prop_predict(ds->ds_dir, 2989 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 2990 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval); 2991 if (error != 0) { 2992 dsl_dataset_rele(ds, FTAG); 2993 return (error); 2994 } 2995 2996 /* 2997 * If we are doing the preliminary check in open context, the 2998 * space estimates may be inaccurate. 2999 */ 3000 if (!dmu_tx_is_syncing(tx)) { 3001 dsl_dataset_rele(ds, FTAG); 3002 return (0); 3003 } 3004 3005 mutex_enter(&ds->ds_lock); 3006 if (!DS_UNIQUE_IS_ACCURATE(ds)) 3007 dsl_dataset_recalc_head_uniq(ds); 3008 unique = ds->ds_phys->ds_unique_bytes; 3009 mutex_exit(&ds->ds_lock); 3010 3011 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) { 3012 uint64_t delta = MAX(unique, newval) - 3013 MAX(unique, ds->ds_reserved); 3014 3015 if (delta > 3016 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) || 3017 (ds->ds_quota > 0 && newval > ds->ds_quota)) { 3018 dsl_dataset_rele(ds, FTAG); 3019 return (SET_ERROR(ENOSPC)); 3020 } 3021 } 3022 3023 dsl_dataset_rele(ds, FTAG); 3024 return (0); 3025} 3026 3027void 3028dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds, 3029 zprop_source_t source, uint64_t value, dmu_tx_t *tx) 3030{ 3031 uint64_t newval; 3032 uint64_t unique; 3033 int64_t delta; 3034 3035 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 3036 source, sizeof (value), 1, &value, tx); 3037 3038 VERIFY0(dsl_prop_get_int_ds(ds, 3039 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval)); 3040 3041 dmu_buf_will_dirty(ds->ds_dbuf, tx); 3042 mutex_enter(&ds->ds_dir->dd_lock); 3043 mutex_enter(&ds->ds_lock); 3044 ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); 3045 unique = ds->ds_phys->ds_unique_bytes; 3046 delta = MAX(0, (int64_t)(newval - unique)) - 3047 MAX(0, (int64_t)(ds->ds_reserved - unique)); 3048 ds->ds_reserved = newval; 3049 mutex_exit(&ds->ds_lock); 3050 3051 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); 3052 mutex_exit(&ds->ds_dir->dd_lock); 3053} 3054 3055static void 3056dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx) 3057{ 3058 dsl_dataset_set_qr_arg_t *ddsqra = arg; 3059 dsl_pool_t *dp = dmu_tx_pool(tx); 3060 dsl_dataset_t *ds; 3061 3062 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds)); 3063 dsl_dataset_set_refreservation_sync_impl(ds, 3064 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx); 3065 dsl_dataset_rele(ds, FTAG); 3066} 3067 3068int 3069dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source, 3070 uint64_t refreservation) 3071{ 3072 dsl_dataset_set_qr_arg_t ddsqra; 3073 3074 ddsqra.ddsqra_name = dsname; 3075 ddsqra.ddsqra_source = source; 3076 ddsqra.ddsqra_value = refreservation; 3077 3078 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check, 3079 dsl_dataset_set_refreservation_sync, &ddsqra, 3080 0, ZFS_SPACE_CHECK_NONE)); 3081} 3082 3083/* 3084 * Return (in *usedp) the amount of space written in new that is not 3085 * present in oldsnap. New may be a snapshot or the head. Old must be 3086 * a snapshot before new, in new's filesystem (or its origin). If not then 3087 * fail and return EINVAL. 3088 * 3089 * The written space is calculated by considering two components: First, we 3090 * ignore any freed space, and calculate the written as new's used space 3091 * minus old's used space. Next, we add in the amount of space that was freed 3092 * between the two snapshots, thus reducing new's used space relative to old's. 3093 * Specifically, this is the space that was born before old->ds_creation_txg, 3094 * and freed before new (ie. on new's deadlist or a previous deadlist). 3095 * 3096 * space freed [---------------------] 3097 * snapshots ---O-------O--------O-------O------ 3098 * oldsnap new 3099 */ 3100int 3101dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new, 3102 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3103{ 3104 int err = 0; 3105 uint64_t snapobj; 3106 dsl_pool_t *dp = new->ds_dir->dd_pool; 3107 3108 ASSERT(dsl_pool_config_held(dp)); 3109 3110 *usedp = 0; 3111 *usedp += new->ds_phys->ds_referenced_bytes; 3112 *usedp -= oldsnap->ds_phys->ds_referenced_bytes; 3113 3114 *compp = 0; 3115 *compp += new->ds_phys->ds_compressed_bytes; 3116 *compp -= oldsnap->ds_phys->ds_compressed_bytes; 3117 3118 *uncompp = 0; 3119 *uncompp += new->ds_phys->ds_uncompressed_bytes; 3120 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes; 3121 3122 snapobj = new->ds_object; 3123 while (snapobj != oldsnap->ds_object) { 3124 dsl_dataset_t *snap; 3125 uint64_t used, comp, uncomp; 3126 3127 if (snapobj == new->ds_object) { 3128 snap = new; 3129 } else { 3130 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap); 3131 if (err != 0) 3132 break; 3133 } 3134 3135 if (snap->ds_phys->ds_prev_snap_txg == 3136 oldsnap->ds_phys->ds_creation_txg) { 3137 /* 3138 * The blocks in the deadlist can not be born after 3139 * ds_prev_snap_txg, so get the whole deadlist space, 3140 * which is more efficient (especially for old-format 3141 * deadlists). Unfortunately the deadlist code 3142 * doesn't have enough information to make this 3143 * optimization itself. 3144 */ 3145 dsl_deadlist_space(&snap->ds_deadlist, 3146 &used, &comp, &uncomp); 3147 } else { 3148 dsl_deadlist_space_range(&snap->ds_deadlist, 3149 0, oldsnap->ds_phys->ds_creation_txg, 3150 &used, &comp, &uncomp); 3151 } 3152 *usedp += used; 3153 *compp += comp; 3154 *uncompp += uncomp; 3155 3156 /* 3157 * If we get to the beginning of the chain of snapshots 3158 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap 3159 * was not a snapshot of/before new. 3160 */ 3161 snapobj = snap->ds_phys->ds_prev_snap_obj; 3162 if (snap != new) 3163 dsl_dataset_rele(snap, FTAG); 3164 if (snapobj == 0) { 3165 err = SET_ERROR(EINVAL); 3166 break; 3167 } 3168 3169 } 3170 return (err); 3171} 3172 3173/* 3174 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap, 3175 * lastsnap, and all snapshots in between are deleted. 3176 * 3177 * blocks that would be freed [---------------------------] 3178 * snapshots ---O-------O--------O-------O--------O 3179 * firstsnap lastsnap 3180 * 3181 * This is the set of blocks that were born after the snap before firstsnap, 3182 * (birth > firstsnap->prev_snap_txg) and died before the snap after the 3183 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist). 3184 * We calculate this by iterating over the relevant deadlists (from the snap 3185 * after lastsnap, backward to the snap after firstsnap), summing up the 3186 * space on the deadlist that was born after the snap before firstsnap. 3187 */ 3188int 3189dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, 3190 dsl_dataset_t *lastsnap, 3191 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp) 3192{ 3193 int err = 0; 3194 uint64_t snapobj; 3195 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool; 3196 3197 ASSERT(dsl_dataset_is_snapshot(firstsnap)); 3198 ASSERT(dsl_dataset_is_snapshot(lastsnap)); 3199 3200 /* 3201 * Check that the snapshots are in the same dsl_dir, and firstsnap 3202 * is before lastsnap. 3203 */ 3204 if (firstsnap->ds_dir != lastsnap->ds_dir || 3205 firstsnap->ds_phys->ds_creation_txg > 3206 lastsnap->ds_phys->ds_creation_txg) 3207 return (SET_ERROR(EINVAL)); 3208 3209 *usedp = *compp = *uncompp = 0; 3210 3211 snapobj = lastsnap->ds_phys->ds_next_snap_obj; 3212 while (snapobj != firstsnap->ds_object) { 3213 dsl_dataset_t *ds; 3214 uint64_t used, comp, uncomp; 3215 3216 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds); 3217 if (err != 0) 3218 break; 3219 3220 dsl_deadlist_space_range(&ds->ds_deadlist, 3221 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX, 3222 &used, &comp, &uncomp); 3223 *usedp += used; 3224 *compp += comp; 3225 *uncompp += uncomp; 3226 3227 snapobj = ds->ds_phys->ds_prev_snap_obj; 3228 ASSERT3U(snapobj, !=, 0); 3229 dsl_dataset_rele(ds, FTAG); 3230 } 3231 return (err); 3232} 3233 3234/* 3235 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline. 3236 * For example, they could both be snapshots of the same filesystem, and 3237 * 'earlier' is before 'later'. Or 'earlier' could be the origin of 3238 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's 3239 * filesystem. Or 'earlier' could be the origin's origin. 3240 * 3241 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg. 3242 */ 3243boolean_t 3244dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier, 3245 uint64_t earlier_txg) 3246{ 3247 dsl_pool_t *dp = later->ds_dir->dd_pool; 3248 int error; 3249 boolean_t ret; 3250 3251 ASSERT(dsl_pool_config_held(dp)); 3252 ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0); 3253 3254 if (earlier_txg == 0) 3255 earlier_txg = earlier->ds_phys->ds_creation_txg; 3256 3257 if (dsl_dataset_is_snapshot(later) && 3258 earlier_txg >= later->ds_phys->ds_creation_txg) 3259 return (B_FALSE); 3260 3261 if (later->ds_dir == earlier->ds_dir) 3262 return (B_TRUE); 3263 if (!dsl_dir_is_clone(later->ds_dir)) 3264 return (B_FALSE); 3265 3266 if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object) 3267 return (B_TRUE); 3268 dsl_dataset_t *origin; 3269 error = dsl_dataset_hold_obj(dp, 3270 later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin); 3271 if (error != 0) 3272 return (B_FALSE); 3273 ret = dsl_dataset_is_before(origin, earlier, earlier_txg); 3274 dsl_dataset_rele(origin, FTAG); 3275 return (ret); 3276} 3277 3278 3279void 3280dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx) 3281{ 3282 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; 3283 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx); 3284} 3285