1 // SPDX-License-Identifier: CDDL-1.0 2 /* 3 * This file and its contents are supplied under the terms of the 4 * Common Development and Distribution License ("CDDL"), version 1.0. 5 * You may only use this file in accordance with the terms of version 6 * 1.0 of the CDDL. 7 * 8 * A full copy of the text of the CDDL should have accompanied this 9 * source. A copy of the CDDL is also available via the Internet at 10 * https://opensource.org/license/CDDL-1.0. 11 */ 12 13 /* 14 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 15 * Copyright (c) 2012, 2020 by Delphix. All rights reserved. 16 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. 17 * Copyright (c) 2013, Joyent, Inc. All rights reserved. 18 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 19 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved. 20 * Copyright (c) 2016 Actifio, Inc. All rights reserved. 21 * Copyright 2017 Nexenta Systems, Inc. 22 * Copyright (c) 2017 Open-E, Inc. All Rights Reserved. 23 * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved. 24 * Copyright (c) 2019, Klara Inc. 25 * Copyright (c) 2019, Allan Jude 26 * Copyright (c) 2022 Hewlett Packard Enterprise Development LP. 27 * Copyright (c) 2025, Rob Norris <robn@despairlabs.com> 28 */ 29 30 /* Portions Copyright 2010 Robert Milkowski */ 31 32 #include <sys/cred.h> 33 #include <sys/zfs_context.h> 34 #include <sys/dmu_objset.h> 35 #include <sys/dsl_dir.h> 36 #include <sys/dsl_dataset.h> 37 #include <sys/dsl_prop.h> 38 #include <sys/dsl_pool.h> 39 #include <sys/dsl_synctask.h> 40 #include <sys/dsl_deleg.h> 41 #include <sys/dnode.h> 42 #include <sys/dbuf.h> 43 #include <sys/zvol.h> 44 #include <sys/dmu_tx.h> 45 #include <sys/zap.h> 46 #include <sys/zil.h> 47 #include <sys/dmu_impl.h> 48 #include <sys/zfs_ioctl.h> 49 #include <sys/sa.h> 50 #include <sys/zfs_onexit.h> 51 #include <sys/dsl_destroy.h> 52 #include <sys/vdev.h> 53 #include <sys/zfeature.h> 54 #include <sys/policy.h> 55 #include <sys/spa_impl.h> 56 #include <sys/dmu_recv.h> 57 #include <sys/zfs_project.h> 58 #include "zfs_namecheck.h" 59 #include <sys/vdev_impl.h> 60 #include <sys/arc.h> 61 #include <cityhash.h> 62 #include <sys/cred.h> 63 64 /* 65 * Needed to close a window in dnode_move() that allows the objset to be freed 66 * before it can be safely accessed. 67 */ 68 krwlock_t os_lock; 69 70 /* 71 * Tunable to overwrite the maximum number of threads for the parallelization 72 * of dmu_objset_find_dp, needed to speed up the import of pools with many 73 * datasets. 74 * Default is 4 times the number of leaf vdevs. 75 */ 76 static const int dmu_find_threads = 0; 77 78 /* 79 * Backfill lower metadnode objects after this many have been freed. 80 * Backfilling negatively impacts object creation rates, so only do it 81 * if there are enough holes to fill. 82 */ 83 static const int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT; 84 85 static const char *upgrade_tag = "upgrade_tag"; 86 87 static void dmu_objset_find_dp_cb(void *arg); 88 89 static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb); 90 static void dmu_objset_upgrade_stop(objset_t *os); 91 92 void 93 dmu_objset_init(void) 94 { 95 rw_init(&os_lock, NULL, RW_DEFAULT, NULL); 96 } 97 98 void 99 dmu_objset_fini(void) 100 { 101 rw_destroy(&os_lock); 102 } 103 104 spa_t * 105 dmu_objset_spa(objset_t *os) 106 { 107 return (os->os_spa); 108 } 109 110 zilog_t * 111 dmu_objset_zil(objset_t *os) 112 { 113 return (os->os_zil); 114 } 115 116 dsl_pool_t * 117 dmu_objset_pool(objset_t *os) 118 { 119 dsl_dataset_t *ds; 120 121 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) 122 return (ds->ds_dir->dd_pool); 123 else 124 return (spa_get_dsl(os->os_spa)); 125 } 126 127 dsl_dataset_t * 128 dmu_objset_ds(objset_t *os) 129 { 130 return (os->os_dsl_dataset); 131 } 132 133 dmu_objset_type_t 134 dmu_objset_type(objset_t *os) 135 { 136 return (os->os_phys->os_type); 137 } 138 139 void 140 dmu_objset_name(objset_t *os, char *buf) 141 { 142 dsl_dataset_name(os->os_dsl_dataset, buf); 143 } 144 145 uint64_t 146 dmu_objset_id(objset_t *os) 147 { 148 dsl_dataset_t *ds = os->os_dsl_dataset; 149 150 return (ds ? ds->ds_object : 0); 151 } 152 153 uint64_t 154 dmu_objset_dnodesize(objset_t *os) 155 { 156 return (os->os_dnodesize); 157 } 158 159 zfs_sync_type_t 160 dmu_objset_syncprop(objset_t *os) 161 { 162 return (os->os_sync); 163 } 164 165 zfs_logbias_op_t 166 dmu_objset_logbias(objset_t *os) 167 { 168 return (os->os_logbias); 169 } 170 171 static void 172 checksum_changed_cb(void *arg, uint64_t newval) 173 { 174 objset_t *os = arg; 175 176 /* 177 * Inheritance should have been done by now. 178 */ 179 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 180 181 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); 182 } 183 184 static void 185 compression_changed_cb(void *arg, uint64_t newval) 186 { 187 objset_t *os = arg; 188 189 /* 190 * Inheritance and range checking should have been done by now. 191 */ 192 ASSERT(newval != ZIO_COMPRESS_INHERIT); 193 194 os->os_compress = zio_compress_select(os->os_spa, 195 ZIO_COMPRESS_ALGO(newval), ZIO_COMPRESS_ON); 196 os->os_complevel = zio_complevel_select(os->os_spa, os->os_compress, 197 ZIO_COMPRESS_LEVEL(newval), ZIO_COMPLEVEL_DEFAULT); 198 } 199 200 static void 201 copies_changed_cb(void *arg, uint64_t newval) 202 { 203 objset_t *os = arg; 204 205 /* 206 * Inheritance and range checking should have been done by now. 207 */ 208 ASSERT(newval > 0); 209 ASSERT(newval <= spa_max_replication(os->os_spa)); 210 211 os->os_copies = newval; 212 } 213 214 static void 215 dedup_changed_cb(void *arg, uint64_t newval) 216 { 217 objset_t *os = arg; 218 spa_t *spa = os->os_spa; 219 enum zio_checksum checksum; 220 221 /* 222 * Inheritance should have been done by now. 223 */ 224 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 225 226 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); 227 228 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; 229 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); 230 } 231 232 static void 233 primary_cache_changed_cb(void *arg, uint64_t newval) 234 { 235 objset_t *os = arg; 236 237 /* 238 * Inheritance and range checking should have been done by now. 239 */ 240 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 241 newval == ZFS_CACHE_METADATA); 242 243 os->os_primary_cache = newval; 244 } 245 246 static void 247 secondary_cache_changed_cb(void *arg, uint64_t newval) 248 { 249 objset_t *os = arg; 250 251 /* 252 * Inheritance and range checking should have been done by now. 253 */ 254 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 255 newval == ZFS_CACHE_METADATA); 256 257 os->os_secondary_cache = newval; 258 } 259 260 static void 261 prefetch_changed_cb(void *arg, uint64_t newval) 262 { 263 objset_t *os = arg; 264 265 /* 266 * Inheritance should have been done by now. 267 */ 268 ASSERT(newval == ZFS_PREFETCH_ALL || newval == ZFS_PREFETCH_NONE || 269 newval == ZFS_PREFETCH_METADATA); 270 os->os_prefetch = newval; 271 } 272 273 static void 274 sync_changed_cb(void *arg, uint64_t newval) 275 { 276 objset_t *os = arg; 277 278 /* 279 * Inheritance and range checking should have been done by now. 280 */ 281 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || 282 newval == ZFS_SYNC_DISABLED); 283 284 os->os_sync = newval; 285 if (os->os_zil) 286 zil_set_sync(os->os_zil, newval); 287 } 288 289 static void 290 redundant_metadata_changed_cb(void *arg, uint64_t newval) 291 { 292 objset_t *os = arg; 293 294 /* 295 * Inheritance and range checking should have been done by now. 296 */ 297 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || 298 newval == ZFS_REDUNDANT_METADATA_MOST || 299 newval == ZFS_REDUNDANT_METADATA_SOME || 300 newval == ZFS_REDUNDANT_METADATA_NONE); 301 302 os->os_redundant_metadata = newval; 303 } 304 305 static void 306 dnodesize_changed_cb(void *arg, uint64_t newval) 307 { 308 objset_t *os = arg; 309 310 switch (newval) { 311 case ZFS_DNSIZE_LEGACY: 312 os->os_dnodesize = DNODE_MIN_SIZE; 313 break; 314 case ZFS_DNSIZE_AUTO: 315 /* 316 * Choose a dnode size that will work well for most 317 * workloads if the user specified "auto". Future code 318 * improvements could dynamically select a dnode size 319 * based on observed workload patterns. 320 */ 321 os->os_dnodesize = DNODE_MIN_SIZE * 2; 322 break; 323 case ZFS_DNSIZE_1K: 324 case ZFS_DNSIZE_2K: 325 case ZFS_DNSIZE_4K: 326 case ZFS_DNSIZE_8K: 327 case ZFS_DNSIZE_16K: 328 os->os_dnodesize = newval; 329 break; 330 } 331 } 332 333 static void 334 smallblk_changed_cb(void *arg, uint64_t newval) 335 { 336 objset_t *os = arg; 337 338 os->os_zpl_special_smallblock = newval; 339 } 340 341 static void 342 direct_changed_cb(void *arg, uint64_t newval) 343 { 344 objset_t *os = arg; 345 346 /* 347 * Inheritance and range checking should have been done by now. 348 */ 349 ASSERT(newval == ZFS_DIRECT_DISABLED || newval == ZFS_DIRECT_STANDARD || 350 newval == ZFS_DIRECT_ALWAYS); 351 352 os->os_direct = newval; 353 } 354 355 static void 356 logbias_changed_cb(void *arg, uint64_t newval) 357 { 358 objset_t *os = arg; 359 360 ASSERT(newval == ZFS_LOGBIAS_LATENCY || 361 newval == ZFS_LOGBIAS_THROUGHPUT); 362 os->os_logbias = newval; 363 if (os->os_zil) 364 zil_set_logbias(os->os_zil, newval); 365 } 366 367 static void 368 recordsize_changed_cb(void *arg, uint64_t newval) 369 { 370 objset_t *os = arg; 371 372 os->os_recordsize = newval; 373 } 374 375 void 376 dmu_objset_byteswap(void *buf, size_t size) 377 { 378 objset_phys_t *osp = buf; 379 380 ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 || 381 size == sizeof (objset_phys_t)); 382 dnode_byteswap(&osp->os_meta_dnode); 383 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); 384 osp->os_type = BSWAP_64(osp->os_type); 385 osp->os_flags = BSWAP_64(osp->os_flags); 386 if (size >= OBJSET_PHYS_SIZE_V2) { 387 dnode_byteswap(&osp->os_userused_dnode); 388 dnode_byteswap(&osp->os_groupused_dnode); 389 if (size >= sizeof (objset_phys_t)) 390 dnode_byteswap(&osp->os_projectused_dnode); 391 } 392 } 393 394 /* 395 * Runs cityhash on the objset_t pointer and the object number. 396 */ 397 static uint64_t 398 dnode_hash(const objset_t *os, uint64_t obj) 399 { 400 uintptr_t osv = (uintptr_t)os; 401 return (cityhash2((uint64_t)osv, obj)); 402 } 403 404 static unsigned int 405 dnode_multilist_index_func(multilist_t *ml, void *obj) 406 { 407 dnode_t *dn = obj; 408 409 /* 410 * The low order bits of the hash value are thought to be 411 * distributed evenly. Otherwise, in the case that the multilist 412 * has a power of two number of sublists, each sublists' usage 413 * would not be evenly distributed. In this context full 64bit 414 * division would be a waste of time, so limit it to 32 bits. 415 */ 416 return ((unsigned int)dnode_hash(dn->dn_objset, dn->dn_object) % 417 multilist_get_num_sublists(ml)); 418 } 419 420 static inline boolean_t 421 dmu_os_is_l2cacheable(objset_t *os) 422 { 423 if (os->os_secondary_cache == ZFS_CACHE_ALL || 424 os->os_secondary_cache == ZFS_CACHE_METADATA) { 425 if (l2arc_exclude_special == 0) 426 return (B_TRUE); 427 428 blkptr_t *bp = os->os_rootbp; 429 if (bp == NULL || BP_IS_HOLE(bp)) 430 return (B_FALSE); 431 uint64_t vdev = DVA_GET_VDEV(bp->blk_dva); 432 vdev_t *rvd = os->os_spa->spa_root_vdev; 433 vdev_t *vd = NULL; 434 435 if (vdev < rvd->vdev_children) 436 vd = rvd->vdev_child[vdev]; 437 438 if (vd == NULL) 439 return (B_TRUE); 440 441 if (vd->vdev_alloc_bias != VDEV_BIAS_SPECIAL && 442 vd->vdev_alloc_bias != VDEV_BIAS_DEDUP) 443 return (B_TRUE); 444 } 445 return (B_FALSE); 446 } 447 448 /* 449 * Instantiates the objset_t in-memory structure corresponding to the 450 * objset_phys_t that's pointed to by the specified blkptr_t. 451 */ 452 int 453 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 454 objset_t **osp) 455 { 456 objset_t *os; 457 int i, err; 458 459 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); 460 ASSERT(!BP_IS_REDACTED(bp)); 461 462 /* 463 * We need the pool config lock to get properties. 464 */ 465 ASSERT(ds == NULL || dsl_pool_config_held(ds->ds_dir->dd_pool)); 466 467 /* 468 * The $ORIGIN dataset (if it exists) doesn't have an associated 469 * objset, so there's no reason to open it. The $ORIGIN dataset 470 * will not exist on pools older than SPA_VERSION_ORIGIN. 471 */ 472 if (ds != NULL && spa_get_dsl(spa) != NULL && 473 spa_get_dsl(spa)->dp_origin_snap != NULL) { 474 ASSERT3P(ds->ds_dir, !=, 475 spa_get_dsl(spa)->dp_origin_snap->ds_dir); 476 } 477 478 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); 479 os->os_dsl_dataset = ds; 480 os->os_spa = spa; 481 os->os_rootbp = bp; 482 if (!BP_IS_HOLE(os->os_rootbp)) { 483 arc_flags_t aflags = ARC_FLAG_WAIT; 484 zbookmark_phys_t zb; 485 int size; 486 zio_flag_t zio_flags = ZIO_FLAG_CANFAIL; 487 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, 488 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 489 490 if (dmu_os_is_l2cacheable(os)) 491 aflags |= ARC_FLAG_L2CACHE; 492 493 if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) { 494 ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); 495 ASSERT(BP_IS_AUTHENTICATED(bp)); 496 zio_flags |= ZIO_FLAG_RAW; 497 } 498 499 dprintf_bp(os->os_rootbp, "reading %s", ""); 500 err = arc_read(NULL, spa, os->os_rootbp, 501 arc_getbuf_func, &os->os_phys_buf, 502 ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); 503 if (err != 0) { 504 kmem_free(os, sizeof (objset_t)); 505 /* convert checksum errors into IO errors */ 506 if (err == ECKSUM) 507 err = SET_ERROR(EIO); 508 return (err); 509 } 510 511 if (spa_version(spa) < SPA_VERSION_USERSPACE) 512 size = OBJSET_PHYS_SIZE_V1; 513 else if (!spa_feature_is_enabled(spa, 514 SPA_FEATURE_PROJECT_QUOTA)) 515 size = OBJSET_PHYS_SIZE_V2; 516 else 517 size = sizeof (objset_phys_t); 518 519 /* Increase the blocksize if we are permitted. */ 520 if (arc_buf_size(os->os_phys_buf) < size) { 521 arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, 522 ARC_BUFC_METADATA, size); 523 memset(buf->b_data, 0, size); 524 memcpy(buf->b_data, os->os_phys_buf->b_data, 525 arc_buf_size(os->os_phys_buf)); 526 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 527 os->os_phys_buf = buf; 528 } 529 530 os->os_phys = os->os_phys_buf->b_data; 531 os->os_flags = os->os_phys->os_flags; 532 } else { 533 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? 534 sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1; 535 os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, 536 ARC_BUFC_METADATA, size); 537 os->os_phys = os->os_phys_buf->b_data; 538 memset(os->os_phys, 0, size); 539 } 540 /* 541 * These properties will be filled in by the logic in zfs_get_zplprop() 542 * when they are queried for the first time. 543 */ 544 os->os_version = OBJSET_PROP_UNINITIALIZED; 545 os->os_normalization = OBJSET_PROP_UNINITIALIZED; 546 os->os_utf8only = OBJSET_PROP_UNINITIALIZED; 547 os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED; 548 549 /* 550 * Note: the changed_cb will be called once before the register 551 * func returns, thus changing the checksum/compression from the 552 * default (fletcher2/off). Snapshots don't need to know about 553 * checksum/compression/copies. 554 */ 555 if (ds != NULL) { 556 os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0); 557 558 err = dsl_prop_register(ds, 559 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), 560 primary_cache_changed_cb, os); 561 if (err == 0) { 562 err = dsl_prop_register(ds, 563 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), 564 secondary_cache_changed_cb, os); 565 } 566 if (err == 0) { 567 err = dsl_prop_register(ds, 568 zfs_prop_to_name(ZFS_PROP_PREFETCH), 569 prefetch_changed_cb, os); 570 } 571 if (!ds->ds_is_snapshot) { 572 if (err == 0) { 573 err = dsl_prop_register(ds, 574 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 575 checksum_changed_cb, os); 576 } 577 if (err == 0) { 578 err = dsl_prop_register(ds, 579 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 580 compression_changed_cb, os); 581 } 582 if (err == 0) { 583 err = dsl_prop_register(ds, 584 zfs_prop_to_name(ZFS_PROP_COPIES), 585 copies_changed_cb, os); 586 } 587 if (err == 0) { 588 err = dsl_prop_register(ds, 589 zfs_prop_to_name(ZFS_PROP_DEDUP), 590 dedup_changed_cb, os); 591 } 592 if (err == 0) { 593 err = dsl_prop_register(ds, 594 zfs_prop_to_name(ZFS_PROP_LOGBIAS), 595 logbias_changed_cb, os); 596 } 597 if (err == 0) { 598 err = dsl_prop_register(ds, 599 zfs_prop_to_name(ZFS_PROP_SYNC), 600 sync_changed_cb, os); 601 } 602 if (err == 0) { 603 err = dsl_prop_register(ds, 604 zfs_prop_to_name( 605 ZFS_PROP_REDUNDANT_METADATA), 606 redundant_metadata_changed_cb, os); 607 } 608 if (err == 0) { 609 err = dsl_prop_register(ds, 610 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 611 recordsize_changed_cb, os); 612 } 613 if (err == 0) { 614 err = dsl_prop_register(ds, 615 zfs_prop_to_name(ZFS_PROP_DNODESIZE), 616 dnodesize_changed_cb, os); 617 } 618 if (err == 0) { 619 err = dsl_prop_register(ds, 620 zfs_prop_to_name( 621 ZFS_PROP_SPECIAL_SMALL_BLOCKS), 622 smallblk_changed_cb, os); 623 } 624 if (err == 0) { 625 err = dsl_prop_register(ds, 626 zfs_prop_to_name(ZFS_PROP_DIRECT), 627 direct_changed_cb, os); 628 } 629 } 630 if (err != 0) { 631 dsl_prop_unregister_all(ds, os); 632 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 633 kmem_free(os, sizeof (objset_t)); 634 return (err); 635 } 636 } else { 637 /* It's the meta-objset. */ 638 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; 639 os->os_compress = ZIO_COMPRESS_ON; 640 os->os_complevel = ZIO_COMPLEVEL_DEFAULT; 641 os->os_encrypted = B_FALSE; 642 os->os_copies = spa_max_replication(spa); 643 os->os_dedup_checksum = ZIO_CHECKSUM_OFF; 644 os->os_dedup_verify = B_FALSE; 645 os->os_logbias = ZFS_LOGBIAS_LATENCY; 646 os->os_sync = ZFS_SYNC_STANDARD; 647 os->os_primary_cache = ZFS_CACHE_ALL; 648 os->os_secondary_cache = ZFS_CACHE_ALL; 649 os->os_dnodesize = DNODE_MIN_SIZE; 650 os->os_prefetch = ZFS_PREFETCH_ALL; 651 } 652 653 if (ds == NULL || !ds->ds_is_snapshot) 654 os->os_zil_header = os->os_phys->os_zil_header; 655 os->os_zil = zil_alloc(os, &os->os_zil_header); 656 657 for (i = 0; i < TXG_SIZE; i++) { 658 multilist_create(&os->os_dirty_dnodes[i], sizeof (dnode_t), 659 offsetof(dnode_t, dn_dirty_link[i]), 660 dnode_multilist_index_func); 661 } 662 list_create(&os->os_dnodes, sizeof (dnode_t), 663 offsetof(dnode_t, dn_link)); 664 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), 665 offsetof(dmu_buf_impl_t, db_link)); 666 667 list_link_init(&os->os_evicting_node); 668 669 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); 670 mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); 671 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); 672 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); 673 os->os_obj_next_percpu_len = boot_ncpus; 674 os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * 675 sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); 676 677 dnode_special_open(os, &os->os_phys->os_meta_dnode, 678 DMU_META_DNODE_OBJECT, &os->os_meta_dnode); 679 if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) { 680 dnode_special_open(os, &os->os_phys->os_userused_dnode, 681 DMU_USERUSED_OBJECT, &os->os_userused_dnode); 682 dnode_special_open(os, &os->os_phys->os_groupused_dnode, 683 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); 684 if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf)) 685 dnode_special_open(os, 686 &os->os_phys->os_projectused_dnode, 687 DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode); 688 } 689 690 mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL); 691 692 *osp = os; 693 return (0); 694 } 695 696 int 697 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) 698 { 699 int err = 0; 700 701 /* 702 * We need the pool_config lock to manipulate the dsl_dataset_t. 703 * Even if the dataset is long-held, we need the pool_config lock 704 * to open the objset, as it needs to get properties. 705 */ 706 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 707 708 mutex_enter(&ds->ds_opening_lock); 709 if (ds->ds_objset == NULL) { 710 objset_t *os; 711 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 712 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), 713 ds, dsl_dataset_get_blkptr(ds), &os); 714 rrw_exit(&ds->ds_bp_rwlock, FTAG); 715 716 if (err == 0) { 717 mutex_enter(&ds->ds_lock); 718 ASSERT0P(ds->ds_objset); 719 ds->ds_objset = os; 720 mutex_exit(&ds->ds_lock); 721 } 722 } 723 *osp = ds->ds_objset; 724 mutex_exit(&ds->ds_opening_lock); 725 return (err); 726 } 727 728 /* 729 * Holds the pool while the objset is held. Therefore only one objset 730 * can be held at a time. 731 */ 732 int 733 dmu_objset_hold_flags(const char *name, boolean_t decrypt, const void *tag, 734 objset_t **osp) 735 { 736 dsl_pool_t *dp; 737 dsl_dataset_t *ds; 738 int err; 739 ds_hold_flags_t flags; 740 741 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 742 err = dsl_pool_hold(name, tag, &dp); 743 if (err != 0) 744 return (err); 745 err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds); 746 if (err != 0) { 747 dsl_pool_rele(dp, tag); 748 return (err); 749 } 750 751 err = dmu_objset_from_ds(ds, osp); 752 if (err != 0) { 753 dsl_dataset_rele_flags(ds, flags, tag); 754 dsl_pool_rele(dp, tag); 755 } 756 757 return (err); 758 } 759 760 int 761 dmu_objset_hold(const char *name, const void *tag, objset_t **osp) 762 { 763 return (dmu_objset_hold_flags(name, B_FALSE, tag, osp)); 764 } 765 766 static int 767 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, 768 boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp) 769 { 770 (void) tag; 771 772 int err = dmu_objset_from_ds(ds, osp); 773 if (err != 0) { 774 return (err); 775 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { 776 return (SET_ERROR(EINVAL)); 777 } else if (!readonly && dsl_dataset_is_snapshot(ds)) { 778 return (SET_ERROR(EROFS)); 779 } else if (!readonly && decrypt && 780 dsl_dir_incompatible_encryption_version(ds->ds_dir)) { 781 return (SET_ERROR(EROFS)); 782 } 783 784 /* if we are decrypting, we can now check MACs in os->os_phys_buf */ 785 if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) { 786 zbookmark_phys_t zb; 787 788 SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT, 789 ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 790 err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa, 791 &zb, B_FALSE); 792 if (err != 0) 793 return (err); 794 795 ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf)); 796 } 797 798 return (0); 799 } 800 801 /* 802 * dsl_pool must not be held when this is called. 803 * Upon successful return, there will be a longhold on the dataset, 804 * and the dsl_pool will not be held. 805 */ 806 int 807 dmu_objset_own(const char *name, dmu_objset_type_t type, 808 boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp) 809 { 810 dsl_pool_t *dp; 811 dsl_dataset_t *ds; 812 int err; 813 ds_hold_flags_t flags; 814 815 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 816 err = dsl_pool_hold(name, FTAG, &dp); 817 if (err != 0) 818 return (err); 819 err = dsl_dataset_own(dp, name, flags, tag, &ds); 820 if (err != 0) { 821 dsl_pool_rele(dp, FTAG); 822 return (err); 823 } 824 err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); 825 if (err != 0) { 826 dsl_dataset_disown(ds, flags, tag); 827 dsl_pool_rele(dp, FTAG); 828 return (err); 829 } 830 831 /* 832 * User accounting requires the dataset to be decrypted and rw. 833 * We also don't begin user accounting during claiming to help 834 * speed up pool import times and to keep this txg reserved 835 * completely for recovery work. 836 */ 837 if (!readonly && !dp->dp_spa->spa_claiming && 838 (ds->ds_dir->dd_crypto_obj == 0 || decrypt)) { 839 if (dmu_objset_userobjspace_upgradable(*osp) || 840 dmu_objset_projectquota_upgradable(*osp)) { 841 dmu_objset_id_quota_upgrade(*osp); 842 } else if (dmu_objset_userused_enabled(*osp)) { 843 dmu_objset_userspace_upgrade(*osp); 844 } 845 } 846 847 dsl_pool_rele(dp, FTAG); 848 return (0); 849 } 850 851 int 852 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, 853 boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp) 854 { 855 dsl_dataset_t *ds; 856 int err; 857 ds_hold_flags_t flags; 858 859 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 860 err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds); 861 if (err != 0) 862 return (err); 863 864 err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); 865 if (err != 0) { 866 dsl_dataset_disown(ds, flags, tag); 867 return (err); 868 } 869 870 return (0); 871 } 872 873 void 874 dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, const void *tag) 875 { 876 ds_hold_flags_t flags; 877 dsl_pool_t *dp = dmu_objset_pool(os); 878 879 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 880 dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag); 881 dsl_pool_rele(dp, tag); 882 } 883 884 void 885 dmu_objset_rele(objset_t *os, const void *tag) 886 { 887 dmu_objset_rele_flags(os, B_FALSE, tag); 888 } 889 890 /* 891 * When we are called, os MUST refer to an objset associated with a dataset 892 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner 893 * == tag. We will then release and reacquire ownership of the dataset while 894 * holding the pool config_rwlock to avoid intervening namespace or ownership 895 * changes may occur. 896 * 897 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to 898 * release the hold on its dataset and acquire a new one on the dataset of the 899 * same name so that it can be partially torn down and reconstructed. 900 */ 901 void 902 dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds, 903 boolean_t decrypt, const void *tag) 904 { 905 dsl_pool_t *dp; 906 char name[ZFS_MAX_DATASET_NAME_LEN]; 907 ds_hold_flags_t flags; 908 909 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 910 VERIFY3P(ds, !=, NULL); 911 VERIFY3P(ds->ds_owner, ==, tag); 912 VERIFY(dsl_dataset_long_held(ds)); 913 914 dsl_dataset_name(ds, name); 915 dp = ds->ds_dir->dd_pool; 916 dsl_pool_config_enter(dp, FTAG); 917 dsl_dataset_disown(ds, flags, tag); 918 VERIFY0(dsl_dataset_own(dp, name, flags, tag, newds)); 919 dsl_pool_config_exit(dp, FTAG); 920 } 921 922 void 923 dmu_objset_disown(objset_t *os, boolean_t decrypt, const void *tag) 924 { 925 ds_hold_flags_t flags; 926 927 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 928 /* 929 * Stop upgrading thread 930 */ 931 dmu_objset_upgrade_stop(os); 932 dsl_dataset_disown(os->os_dsl_dataset, flags, tag); 933 } 934 935 void 936 dmu_objset_evict_dbufs(objset_t *os) 937 { 938 dnode_t *dn_marker; 939 dnode_t *dn; 940 941 dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP); 942 943 mutex_enter(&os->os_lock); 944 dn = list_head(&os->os_dnodes); 945 while (dn != NULL) { 946 /* 947 * Skip dnodes without holds. We have to do this dance 948 * because dnode_add_ref() only works if there is already a 949 * hold. If the dnode has no holds, then it has no dbufs. 950 */ 951 if (dnode_add_ref(dn, FTAG)) { 952 list_insert_after(&os->os_dnodes, dn, dn_marker); 953 mutex_exit(&os->os_lock); 954 955 dnode_evict_dbufs(dn); 956 dnode_rele(dn, FTAG); 957 958 mutex_enter(&os->os_lock); 959 dn = list_next(&os->os_dnodes, dn_marker); 960 list_remove(&os->os_dnodes, dn_marker); 961 } else { 962 dn = list_next(&os->os_dnodes, dn); 963 } 964 } 965 mutex_exit(&os->os_lock); 966 967 kmem_free(dn_marker, sizeof (dnode_t)); 968 969 if (DMU_USERUSED_DNODE(os) != NULL) { 970 if (DMU_PROJECTUSED_DNODE(os) != NULL) 971 dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os)); 972 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); 973 dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); 974 } 975 dnode_evict_dbufs(DMU_META_DNODE(os)); 976 } 977 978 /* 979 * Objset eviction processing is split into into two pieces. 980 * The first marks the objset as evicting, evicts any dbufs that 981 * have a refcount of zero, and then queues up the objset for the 982 * second phase of eviction. Once os->os_dnodes has been cleared by 983 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. 984 * The second phase closes the special dnodes, dequeues the objset from 985 * the list of those undergoing eviction, and finally frees the objset. 986 * 987 * NOTE: Due to asynchronous eviction processing (invocation of 988 * dnode_buf_pageout()), it is possible for the meta dnode for the 989 * objset to have no holds even though os->os_dnodes is not empty. 990 */ 991 void 992 dmu_objset_evict(objset_t *os) 993 { 994 dsl_dataset_t *ds = os->os_dsl_dataset; 995 996 for (int t = 0; t < TXG_SIZE; t++) 997 ASSERT(!dmu_objset_is_dirty(os, t)); 998 999 if (ds) 1000 dsl_prop_unregister_all(ds, os); 1001 1002 if (os->os_sa) 1003 sa_tear_down(os); 1004 1005 dmu_objset_evict_dbufs(os); 1006 1007 mutex_enter(&os->os_lock); 1008 spa_evicting_os_register(os->os_spa, os); 1009 if (list_is_empty(&os->os_dnodes)) { 1010 mutex_exit(&os->os_lock); 1011 dmu_objset_evict_done(os); 1012 } else { 1013 mutex_exit(&os->os_lock); 1014 } 1015 1016 1017 } 1018 1019 void 1020 dmu_objset_evict_done(objset_t *os) 1021 { 1022 ASSERT3P(list_head(&os->os_dnodes), ==, NULL); 1023 1024 dnode_special_close(&os->os_meta_dnode); 1025 if (DMU_USERUSED_DNODE(os)) { 1026 if (DMU_PROJECTUSED_DNODE(os)) 1027 dnode_special_close(&os->os_projectused_dnode); 1028 dnode_special_close(&os->os_userused_dnode); 1029 dnode_special_close(&os->os_groupused_dnode); 1030 } 1031 zil_free(os->os_zil); 1032 1033 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 1034 1035 /* 1036 * This is a barrier to prevent the objset from going away in 1037 * dnode_move() until we can safely ensure that the objset is still in 1038 * use. We consider the objset valid before the barrier and invalid 1039 * after the barrier. 1040 */ 1041 rw_enter(&os_lock, RW_READER); 1042 rw_exit(&os_lock); 1043 1044 kmem_free(os->os_obj_next_percpu, 1045 os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); 1046 1047 mutex_destroy(&os->os_lock); 1048 mutex_destroy(&os->os_userused_lock); 1049 mutex_destroy(&os->os_obj_lock); 1050 mutex_destroy(&os->os_user_ptr_lock); 1051 mutex_destroy(&os->os_upgrade_lock); 1052 for (int i = 0; i < TXG_SIZE; i++) 1053 multilist_destroy(&os->os_dirty_dnodes[i]); 1054 spa_evicting_os_deregister(os->os_spa, os); 1055 kmem_free(os, sizeof (objset_t)); 1056 } 1057 1058 inode_timespec_t 1059 dmu_objset_snap_cmtime(objset_t *os) 1060 { 1061 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); 1062 } 1063 1064 objset_t * 1065 dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 1066 dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx) 1067 { 1068 objset_t *os; 1069 dnode_t *mdn; 1070 1071 ASSERT(dmu_tx_is_syncing(tx)); 1072 1073 if (blksz == 0) 1074 blksz = DNODE_BLOCK_SIZE; 1075 if (ibs == 0) 1076 ibs = DN_MAX_INDBLKSHIFT; 1077 1078 if (ds != NULL) 1079 VERIFY0(dmu_objset_from_ds(ds, &os)); 1080 else 1081 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); 1082 1083 mdn = DMU_META_DNODE(os); 1084 1085 dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0, 1086 DNODE_MIN_SLOTS, tx); 1087 1088 /* 1089 * We don't want to have to increase the meta-dnode's nlevels 1090 * later, because then we could do it in quiescing context while 1091 * we are also accessing it in open context. 1092 * 1093 * This precaution is not necessary for the MOS (ds == NULL), 1094 * because the MOS is only updated in syncing context. 1095 * This is most fortunate: the MOS is the only objset that 1096 * needs to be synced multiple times as spa_sync() iterates 1097 * to convergence, so minimizing its dn_nlevels matters. 1098 */ 1099 if (ds != NULL) { 1100 if (levels == 0) { 1101 levels = 1; 1102 1103 /* 1104 * Determine the number of levels necessary for the 1105 * meta-dnode to contain DN_MAX_OBJECT dnodes. Note 1106 * that in order to ensure that we do not overflow 1107 * 64 bits, there has to be a nlevels that gives us a 1108 * number of blocks > DN_MAX_OBJECT but < 2^64. 1109 * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) 1110 * (10) must be less than (64 - log2(DN_MAX_OBJECT)) 1111 * (16). 1112 */ 1113 while ((uint64_t)mdn->dn_nblkptr << 1114 (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) * 1115 (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < 1116 DN_MAX_OBJECT) 1117 levels++; 1118 } 1119 1120 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = 1121 mdn->dn_nlevels = levels; 1122 } 1123 1124 ASSERT(type != DMU_OST_NONE); 1125 ASSERT(type != DMU_OST_ANY); 1126 ASSERT(type < DMU_OST_NUMTYPES); 1127 os->os_phys->os_type = type; 1128 1129 /* 1130 * Enable user accounting if it is enabled and this is not an 1131 * encrypted receive. 1132 */ 1133 if (dmu_objset_userused_enabled(os) && 1134 (!os->os_encrypted || !dmu_objset_is_receiving(os))) { 1135 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 1136 if (dmu_objset_userobjused_enabled(os)) { 1137 ASSERT3P(ds, !=, NULL); 1138 ds->ds_feature_activation[ 1139 SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; 1140 os->os_phys->os_flags |= 1141 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; 1142 } 1143 if (dmu_objset_projectquota_enabled(os)) { 1144 ASSERT3P(ds, !=, NULL); 1145 ds->ds_feature_activation[ 1146 SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; 1147 os->os_phys->os_flags |= 1148 OBJSET_FLAG_PROJECTQUOTA_COMPLETE; 1149 } 1150 os->os_flags = os->os_phys->os_flags; 1151 } 1152 1153 dsl_dataset_dirty(ds, tx); 1154 1155 return (os); 1156 } 1157 1158 /* called from dsl for meta-objset */ 1159 objset_t * 1160 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 1161 dmu_objset_type_t type, dmu_tx_t *tx) 1162 { 1163 return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx)); 1164 } 1165 1166 typedef struct dmu_objset_create_arg { 1167 const char *doca_name; 1168 cred_t *doca_cred; 1169 void (*doca_userfunc)(objset_t *os, void *arg, 1170 cred_t *cr, dmu_tx_t *tx); 1171 void *doca_userarg; 1172 dmu_objset_type_t doca_type; 1173 uint64_t doca_flags; 1174 dsl_crypto_params_t *doca_dcp; 1175 } dmu_objset_create_arg_t; 1176 1177 static int 1178 dmu_objset_create_check(void *arg, dmu_tx_t *tx) 1179 { 1180 dmu_objset_create_arg_t *doca = arg; 1181 dsl_pool_t *dp = dmu_tx_pool(tx); 1182 dsl_dir_t *pdd; 1183 dsl_dataset_t *parentds; 1184 objset_t *parentos; 1185 const char *tail; 1186 int error; 1187 1188 if (strchr(doca->doca_name, '@') != NULL) 1189 return (SET_ERROR(EINVAL)); 1190 1191 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) 1192 return (SET_ERROR(ENAMETOOLONG)); 1193 1194 if (dataset_nestcheck(doca->doca_name) != 0) 1195 return (SET_ERROR(ENAMETOOLONG)); 1196 1197 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); 1198 if (error != 0) 1199 return (error); 1200 if (tail == NULL) { 1201 dsl_dir_rele(pdd, FTAG); 1202 return (SET_ERROR(EEXIST)); 1203 } 1204 1205 error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp, NULL); 1206 if (error != 0) { 1207 dsl_dir_rele(pdd, FTAG); 1208 return (error); 1209 } 1210 1211 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 1212 doca->doca_cred); 1213 if (error != 0) { 1214 dsl_dir_rele(pdd, FTAG); 1215 return (error); 1216 } 1217 1218 /* can't create below anything but filesystems (eg. no ZVOLs) */ 1219 error = dsl_dataset_hold_obj(pdd->dd_pool, 1220 dsl_dir_phys(pdd)->dd_head_dataset_obj, FTAG, &parentds); 1221 if (error != 0) { 1222 dsl_dir_rele(pdd, FTAG); 1223 return (error); 1224 } 1225 error = dmu_objset_from_ds(parentds, &parentos); 1226 if (error != 0) { 1227 dsl_dataset_rele(parentds, FTAG); 1228 dsl_dir_rele(pdd, FTAG); 1229 return (error); 1230 } 1231 if (dmu_objset_type(parentos) != DMU_OST_ZFS) { 1232 dsl_dataset_rele(parentds, FTAG); 1233 dsl_dir_rele(pdd, FTAG); 1234 return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); 1235 } 1236 dsl_dataset_rele(parentds, FTAG); 1237 dsl_dir_rele(pdd, FTAG); 1238 1239 return (error); 1240 } 1241 1242 static void 1243 dmu_objset_create_sync(void *arg, dmu_tx_t *tx) 1244 { 1245 dmu_objset_create_arg_t *doca = arg; 1246 dsl_pool_t *dp = dmu_tx_pool(tx); 1247 spa_t *spa = dp->dp_spa; 1248 dsl_dir_t *pdd; 1249 const char *tail; 1250 dsl_dataset_t *ds; 1251 uint64_t obj; 1252 blkptr_t *bp; 1253 objset_t *os; 1254 zio_t *rzio; 1255 1256 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); 1257 1258 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, 1259 doca->doca_cred, doca->doca_dcp, tx); 1260 1261 VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj, 1262 DS_HOLD_FLAG_DECRYPT, FTAG, &ds)); 1263 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1264 bp = dsl_dataset_get_blkptr(ds); 1265 os = dmu_objset_create_impl(spa, ds, bp, doca->doca_type, tx); 1266 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1267 1268 if (doca->doca_userfunc != NULL) { 1269 doca->doca_userfunc(os, doca->doca_userarg, 1270 doca->doca_cred, tx); 1271 } 1272 1273 /* 1274 * The doca_userfunc() may write out some data that needs to be 1275 * encrypted if the dataset is encrypted (specifically the root 1276 * directory). This data must be written out before the encryption 1277 * key mapping is removed by dsl_dataset_rele_flags(). Force the 1278 * I/O to occur immediately by invoking the relevant sections of 1279 * dsl_pool_sync(). 1280 */ 1281 if (os->os_encrypted) { 1282 dsl_dataset_t *tmpds = NULL; 1283 boolean_t need_sync_done = B_FALSE; 1284 1285 mutex_enter(&ds->ds_lock); 1286 ds->ds_owner = FTAG; 1287 mutex_exit(&ds->ds_lock); 1288 1289 rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1290 tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, 1291 tx->tx_txg); 1292 if (tmpds != NULL) { 1293 dsl_dataset_sync(ds, rzio, tx); 1294 need_sync_done = B_TRUE; 1295 } 1296 VERIFY0(zio_wait(rzio)); 1297 1298 dmu_objset_sync_done(os, tx); 1299 taskq_wait(dp->dp_sync_taskq); 1300 if (txg_list_member(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1301 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1302 key_mapping_rele(spa, ds->ds_key_mapping, ds); 1303 } 1304 1305 rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1306 tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, 1307 tx->tx_txg); 1308 if (tmpds != NULL) { 1309 dmu_buf_rele(ds->ds_dbuf, ds); 1310 dsl_dataset_sync(ds, rzio, tx); 1311 } 1312 VERIFY0(zio_wait(rzio)); 1313 1314 if (need_sync_done) { 1315 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1316 key_mapping_rele(spa, ds->ds_key_mapping, ds); 1317 dsl_dataset_sync_done(ds, tx); 1318 dmu_buf_rele(ds->ds_dbuf, ds); 1319 } 1320 1321 mutex_enter(&ds->ds_lock); 1322 ds->ds_owner = NULL; 1323 mutex_exit(&ds->ds_lock); 1324 } 1325 1326 spa_history_log_internal_ds(ds, "create", tx, " "); 1327 1328 dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG); 1329 dsl_dir_rele(pdd, FTAG); 1330 } 1331 1332 int 1333 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 1334 dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg) 1335 { 1336 dmu_objset_create_arg_t doca; 1337 dsl_crypto_params_t tmp_dcp = { 0 }; 1338 1339 cred_t *cr = CRED(); 1340 crhold(cr); 1341 1342 doca.doca_name = name; 1343 doca.doca_cred = cr; 1344 doca.doca_flags = flags; 1345 doca.doca_userfunc = func; 1346 doca.doca_userarg = arg; 1347 doca.doca_type = type; 1348 1349 /* 1350 * Some callers (mostly for testing) do not provide a dcp on their 1351 * own but various code inside the sync task will require it to be 1352 * allocated. Rather than adding NULL checks throughout this code 1353 * or adding dummy dcp's to all of the callers we simply create a 1354 * dummy one here and use that. This zero dcp will have the same 1355 * effect as asking for inheritance of all encryption params. 1356 */ 1357 doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp; 1358 1359 int rv = dsl_sync_task(name, 1360 dmu_objset_create_check, dmu_objset_create_sync, &doca, 1361 6, ZFS_SPACE_CHECK_NORMAL); 1362 1363 if (rv == 0) 1364 zvol_create_minors(name); 1365 1366 crfree(cr); 1367 1368 return (rv); 1369 } 1370 1371 int 1372 dmu_objset_snapshot_one(const char *fsname, const char *snapname) 1373 { 1374 int err; 1375 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); 1376 nvlist_t *snaps = fnvlist_alloc(); 1377 1378 fnvlist_add_boolean(snaps, longsnap); 1379 kmem_strfree(longsnap); 1380 err = dsl_dataset_snapshot(snaps, NULL, NULL); 1381 fnvlist_free(snaps); 1382 return (err); 1383 } 1384 1385 static void 1386 dmu_objset_upgrade_task_cb(void *data) 1387 { 1388 objset_t *os = data; 1389 1390 mutex_enter(&os->os_upgrade_lock); 1391 os->os_upgrade_status = EINTR; 1392 if (!os->os_upgrade_exit) { 1393 int status; 1394 1395 mutex_exit(&os->os_upgrade_lock); 1396 1397 status = os->os_upgrade_cb(os); 1398 1399 mutex_enter(&os->os_upgrade_lock); 1400 1401 os->os_upgrade_status = status; 1402 } 1403 os->os_upgrade_exit = B_TRUE; 1404 os->os_upgrade_id = 0; 1405 mutex_exit(&os->os_upgrade_lock); 1406 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1407 } 1408 1409 static void 1410 dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb) 1411 { 1412 if (os->os_upgrade_id != 0) 1413 return; 1414 1415 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 1416 dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag); 1417 1418 mutex_enter(&os->os_upgrade_lock); 1419 if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) { 1420 os->os_upgrade_exit = B_FALSE; 1421 os->os_upgrade_cb = cb; 1422 os->os_upgrade_id = taskq_dispatch( 1423 os->os_spa->spa_upgrade_taskq, 1424 dmu_objset_upgrade_task_cb, os, TQ_SLEEP); 1425 if (os->os_upgrade_id == TASKQID_INVALID) { 1426 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1427 os->os_upgrade_status = ENOMEM; 1428 } 1429 } else { 1430 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1431 } 1432 mutex_exit(&os->os_upgrade_lock); 1433 } 1434 1435 static void 1436 dmu_objset_upgrade_stop(objset_t *os) 1437 { 1438 mutex_enter(&os->os_upgrade_lock); 1439 os->os_upgrade_exit = B_TRUE; 1440 if (os->os_upgrade_id != 0) { 1441 taskqid_t id = os->os_upgrade_id; 1442 1443 os->os_upgrade_id = 0; 1444 mutex_exit(&os->os_upgrade_lock); 1445 1446 if ((taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id, 1447 B_TRUE)) == 0) { 1448 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1449 } 1450 txg_wait_synced(os->os_spa->spa_dsl_pool, 0); 1451 } else { 1452 mutex_exit(&os->os_upgrade_lock); 1453 } 1454 } 1455 1456 static void 1457 dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) 1458 { 1459 dnode_t *dn; 1460 1461 while ((dn = multilist_sublist_head(list)) != NULL) { 1462 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 1463 ASSERT(dn->dn_dbuf->db_data_pending); 1464 /* 1465 * Initialize dn_zio outside dnode_sync() because the 1466 * meta-dnode needs to set it outside dnode_sync(). 1467 */ 1468 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; 1469 ASSERT(dn->dn_zio); 1470 1471 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); 1472 multilist_sublist_remove(list, dn); 1473 1474 /* 1475 * See the comment above dnode_rele_task() for an explanation 1476 * of why this dnode hold is always needed (even when not 1477 * doing user accounting). 1478 */ 1479 multilist_t *newlist = &dn->dn_objset->os_synced_dnodes; 1480 (void) dnode_add_ref(dn, newlist); 1481 multilist_insert(newlist, dn); 1482 1483 dnode_sync(dn, tx); 1484 } 1485 } 1486 1487 static void 1488 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) 1489 { 1490 (void) abuf; 1491 blkptr_t *bp = zio->io_bp; 1492 objset_t *os = arg; 1493 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; 1494 uint64_t fill = 0; 1495 1496 ASSERT(!BP_IS_EMBEDDED(bp)); 1497 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); 1498 ASSERT0(BP_GET_LEVEL(bp)); 1499 1500 /* 1501 * Update rootbp fill count: it should be the number of objects 1502 * allocated in the object set (not counting the "special" 1503 * objects that are stored in the objset_phys_t -- the meta 1504 * dnode and user/group/project accounting objects). 1505 */ 1506 for (int i = 0; i < dnp->dn_nblkptr; i++) 1507 fill += BP_GET_FILL(&dnp->dn_blkptr[i]); 1508 1509 BP_SET_FILL(bp, fill); 1510 1511 if (os->os_dsl_dataset != NULL) 1512 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); 1513 *os->os_rootbp = *bp; 1514 if (os->os_dsl_dataset != NULL) 1515 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); 1516 } 1517 1518 static void 1519 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) 1520 { 1521 (void) abuf; 1522 blkptr_t *bp = zio->io_bp; 1523 blkptr_t *bp_orig = &zio->io_bp_orig; 1524 objset_t *os = arg; 1525 1526 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { 1527 ASSERT(BP_EQUAL(bp, bp_orig)); 1528 } else { 1529 dsl_dataset_t *ds = os->os_dsl_dataset; 1530 dmu_tx_t *tx = os->os_synctx; 1531 1532 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); 1533 dsl_dataset_block_born(ds, bp, tx); 1534 } 1535 kmem_free(bp, sizeof (*bp)); 1536 } 1537 1538 typedef struct sync_objset_arg { 1539 zio_t *soa_zio; 1540 objset_t *soa_os; 1541 dmu_tx_t *soa_tx; 1542 kmutex_t soa_mutex; 1543 int soa_count; 1544 taskq_ent_t soa_tq_ent; 1545 } sync_objset_arg_t; 1546 1547 typedef struct sync_dnodes_arg { 1548 multilist_t *sda_list; 1549 int sda_sublist_idx; 1550 multilist_t *sda_newlist; 1551 sync_objset_arg_t *sda_soa; 1552 } sync_dnodes_arg_t; 1553 1554 static void sync_meta_dnode_task(void *arg); 1555 1556 static void 1557 sync_dnodes_task(void *arg) 1558 { 1559 sync_dnodes_arg_t *sda = arg; 1560 sync_objset_arg_t *soa = sda->sda_soa; 1561 objset_t *os = soa->soa_os; 1562 1563 uint_t allocator = spa_acq_allocator(os->os_spa); 1564 multilist_sublist_t *ms = 1565 multilist_sublist_lock_idx(sda->sda_list, sda->sda_sublist_idx); 1566 1567 dmu_objset_sync_dnodes(ms, soa->soa_tx); 1568 1569 multilist_sublist_unlock(ms); 1570 spa_rel_allocator(os->os_spa, allocator); 1571 1572 kmem_free(sda, sizeof (*sda)); 1573 1574 mutex_enter(&soa->soa_mutex); 1575 ASSERT(soa->soa_count != 0); 1576 if (--soa->soa_count != 0) { 1577 mutex_exit(&soa->soa_mutex); 1578 return; 1579 } 1580 mutex_exit(&soa->soa_mutex); 1581 1582 taskq_dispatch_ent(dmu_objset_pool(os)->dp_sync_taskq, 1583 sync_meta_dnode_task, soa, TQ_FRONT, &soa->soa_tq_ent); 1584 } 1585 1586 /* 1587 * Issue the zio_nowait() for all dirty record zios on the meta dnode, 1588 * then trigger the callback for the zil_sync. This runs once for each 1589 * objset, only after any/all sublists in the objset have been synced. 1590 */ 1591 static void 1592 sync_meta_dnode_task(void *arg) 1593 { 1594 sync_objset_arg_t *soa = arg; 1595 objset_t *os = soa->soa_os; 1596 dmu_tx_t *tx = soa->soa_tx; 1597 int txgoff = tx->tx_txg & TXG_MASK; 1598 dbuf_dirty_record_t *dr; 1599 1600 ASSERT0(soa->soa_count); 1601 1602 list_t *list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; 1603 while ((dr = list_remove_head(list)) != NULL) { 1604 ASSERT0(dr->dr_dbuf->db_level); 1605 zio_nowait(dr->dr_zio); 1606 } 1607 1608 /* Enable dnode backfill if enough objects have been freed. */ 1609 if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { 1610 os->os_rescan_dnodes = B_TRUE; 1611 os->os_freed_dnodes = 0; 1612 } 1613 1614 /* 1615 * Free intent log blocks up to this tx. 1616 */ 1617 zil_sync(os->os_zil, tx); 1618 os->os_phys->os_zil_header = os->os_zil_header; 1619 zio_nowait(soa->soa_zio); 1620 1621 mutex_destroy(&soa->soa_mutex); 1622 kmem_free(soa, sizeof (*soa)); 1623 } 1624 1625 /* called from dsl */ 1626 void 1627 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) 1628 { 1629 int txgoff; 1630 zbookmark_phys_t zb; 1631 zio_prop_t zp; 1632 zio_t *zio; 1633 int num_sublists; 1634 multilist_t *ml; 1635 blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); 1636 *blkptr_copy = *os->os_rootbp; 1637 1638 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", (u_longlong_t)tx->tx_txg); 1639 1640 ASSERT(dmu_tx_is_syncing(tx)); 1641 /* XXX the write_done callback should really give us the tx... */ 1642 os->os_synctx = tx; 1643 1644 if (os->os_dsl_dataset == NULL) { 1645 /* 1646 * This is the MOS. If we have upgraded, 1647 * spa_max_replication() could change, so reset 1648 * os_copies here. 1649 */ 1650 os->os_copies = spa_max_replication(os->os_spa); 1651 } 1652 1653 /* 1654 * Create the root block IO 1655 */ 1656 SET_BOOKMARK(&zb, os->os_dsl_dataset ? 1657 os->os_dsl_dataset->ds_object : DMU_META_OBJSET, 1658 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 1659 arc_release(os->os_phys_buf, &os->os_phys_buf); 1660 1661 dmu_write_policy(os, NULL, 0, 0, &zp); 1662 1663 /* 1664 * If we are either claiming the ZIL or doing a raw receive, write 1665 * out the os_phys_buf raw. Neither of these actions will effect the 1666 * MAC at this point. 1667 */ 1668 if (os->os_raw_receive || 1669 os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { 1670 ASSERT(os->os_encrypted); 1671 arc_convert_to_raw(os->os_phys_buf, 1672 os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER, 1673 DMU_OT_OBJSET, NULL, NULL, NULL); 1674 } 1675 1676 zio = arc_write(pio, os->os_spa, tx->tx_txg, 1677 blkptr_copy, os->os_phys_buf, B_FALSE, dmu_os_is_l2cacheable(os), 1678 &zp, dmu_objset_write_ready, NULL, dmu_objset_write_done, 1679 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); 1680 1681 /* 1682 * Sync special dnodes - the parent IO for the sync is the root block 1683 */ 1684 DMU_META_DNODE(os)->dn_zio = zio; 1685 dnode_sync(DMU_META_DNODE(os), tx); 1686 1687 os->os_phys->os_flags = os->os_flags; 1688 1689 if (DMU_USERUSED_DNODE(os) && 1690 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1691 DMU_USERUSED_DNODE(os)->dn_zio = zio; 1692 dnode_sync(DMU_USERUSED_DNODE(os), tx); 1693 DMU_GROUPUSED_DNODE(os)->dn_zio = zio; 1694 dnode_sync(DMU_GROUPUSED_DNODE(os), tx); 1695 } 1696 1697 if (DMU_PROJECTUSED_DNODE(os) && 1698 DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1699 DMU_PROJECTUSED_DNODE(os)->dn_zio = zio; 1700 dnode_sync(DMU_PROJECTUSED_DNODE(os), tx); 1701 } 1702 1703 txgoff = tx->tx_txg & TXG_MASK; 1704 1705 /* 1706 * We must create the list here because it uses the 1707 * dn_dirty_link[] of this txg. But it may already 1708 * exist because we call dsl_dataset_sync() twice per txg. 1709 */ 1710 if (os->os_synced_dnodes.ml_sublists == NULL) { 1711 multilist_create(&os->os_synced_dnodes, sizeof (dnode_t), 1712 offsetof(dnode_t, dn_dirty_link[txgoff]), 1713 dnode_multilist_index_func); 1714 } else { 1715 ASSERT3U(os->os_synced_dnodes.ml_offset, ==, 1716 offsetof(dnode_t, dn_dirty_link[txgoff])); 1717 } 1718 1719 /* 1720 * zio_nowait(zio) is done after any/all sublist and meta dnode 1721 * zios have been nowaited, and the zil_sync() has been performed. 1722 * The soa is freed at the end of sync_meta_dnode_task. 1723 */ 1724 sync_objset_arg_t *soa = kmem_alloc(sizeof (*soa), KM_SLEEP); 1725 soa->soa_zio = zio; 1726 soa->soa_os = os; 1727 soa->soa_tx = tx; 1728 taskq_init_ent(&soa->soa_tq_ent); 1729 mutex_init(&soa->soa_mutex, NULL, MUTEX_DEFAULT, NULL); 1730 1731 ml = &os->os_dirty_dnodes[txgoff]; 1732 soa->soa_count = num_sublists = multilist_get_num_sublists(ml); 1733 1734 for (int i = 0; i < num_sublists; i++) { 1735 if (multilist_sublist_is_empty_idx(ml, i)) 1736 soa->soa_count--; 1737 } 1738 1739 if (soa->soa_count == 0) { 1740 taskq_dispatch_ent(dmu_objset_pool(os)->dp_sync_taskq, 1741 sync_meta_dnode_task, soa, TQ_FRONT, &soa->soa_tq_ent); 1742 } else { 1743 /* 1744 * Sync sublists in parallel. The last to finish 1745 * (i.e., when soa->soa_count reaches zero) must 1746 * dispatch sync_meta_dnode_task. 1747 */ 1748 for (int i = 0; i < num_sublists; i++) { 1749 if (multilist_sublist_is_empty_idx(ml, i)) 1750 continue; 1751 sync_dnodes_arg_t *sda = 1752 kmem_alloc(sizeof (*sda), KM_SLEEP); 1753 sda->sda_list = ml; 1754 sda->sda_sublist_idx = i; 1755 sda->sda_soa = soa; 1756 (void) taskq_dispatch( 1757 dmu_objset_pool(os)->dp_sync_taskq, 1758 sync_dnodes_task, sda, 0); 1759 /* sync_dnodes_task frees sda */ 1760 } 1761 } 1762 } 1763 1764 boolean_t 1765 dmu_objset_is_dirty(objset_t *os, uint64_t txg) 1766 { 1767 return (!multilist_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK])); 1768 } 1769 1770 static file_info_cb_t *file_cbs[DMU_OST_NUMTYPES]; 1771 1772 void 1773 dmu_objset_register_type(dmu_objset_type_t ost, file_info_cb_t *cb) 1774 { 1775 file_cbs[ost] = cb; 1776 } 1777 1778 int 1779 dmu_get_file_info(objset_t *os, dmu_object_type_t bonustype, const void *data, 1780 zfs_file_info_t *zfi) 1781 { 1782 file_info_cb_t *cb = file_cbs[os->os_phys->os_type]; 1783 if (cb == NULL) 1784 return (EINVAL); 1785 return (cb(bonustype, data, zfi)); 1786 } 1787 1788 boolean_t 1789 dmu_objset_userused_enabled(objset_t *os) 1790 { 1791 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && 1792 file_cbs[os->os_phys->os_type] != NULL && 1793 DMU_USERUSED_DNODE(os) != NULL); 1794 } 1795 1796 boolean_t 1797 dmu_objset_userobjused_enabled(objset_t *os) 1798 { 1799 return (dmu_objset_userused_enabled(os) && 1800 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING)); 1801 } 1802 1803 boolean_t 1804 dmu_objset_projectquota_enabled(objset_t *os) 1805 { 1806 return (file_cbs[os->os_phys->os_type] != NULL && 1807 DMU_PROJECTUSED_DNODE(os) != NULL && 1808 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA)); 1809 } 1810 1811 typedef struct userquota_node { 1812 /* must be in the first filed, see userquota_update_cache() */ 1813 char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN]; 1814 int64_t uqn_delta; 1815 avl_node_t uqn_node; 1816 } userquota_node_t; 1817 1818 typedef struct userquota_cache { 1819 avl_tree_t uqc_user_deltas; 1820 avl_tree_t uqc_group_deltas; 1821 avl_tree_t uqc_project_deltas; 1822 } userquota_cache_t; 1823 1824 static int 1825 userquota_compare(const void *l, const void *r) 1826 { 1827 const userquota_node_t *luqn = l; 1828 const userquota_node_t *ruqn = r; 1829 int rv; 1830 1831 /* 1832 * NB: can only access uqn_id because userquota_update_cache() doesn't 1833 * pass in an entire userquota_node_t. 1834 */ 1835 rv = strcmp(luqn->uqn_id, ruqn->uqn_id); 1836 1837 return (TREE_ISIGN(rv)); 1838 } 1839 1840 static void 1841 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) 1842 { 1843 void *cookie; 1844 userquota_node_t *uqn; 1845 1846 ASSERT(dmu_tx_is_syncing(tx)); 1847 1848 cookie = NULL; 1849 while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, 1850 &cookie)) != NULL) { 1851 /* 1852 * os_userused_lock protects against concurrent calls to 1853 * zap_increment(). It's needed because zap_increment() 1854 * is not thread-safe (i.e. not atomic). 1855 */ 1856 mutex_enter(&os->os_userused_lock); 1857 VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT, 1858 uqn->uqn_id, uqn->uqn_delta, tx)); 1859 mutex_exit(&os->os_userused_lock); 1860 kmem_free(uqn, sizeof (*uqn)); 1861 } 1862 avl_destroy(&cache->uqc_user_deltas); 1863 1864 cookie = NULL; 1865 while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, 1866 &cookie)) != NULL) { 1867 mutex_enter(&os->os_userused_lock); 1868 VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT, 1869 uqn->uqn_id, uqn->uqn_delta, tx)); 1870 mutex_exit(&os->os_userused_lock); 1871 kmem_free(uqn, sizeof (*uqn)); 1872 } 1873 avl_destroy(&cache->uqc_group_deltas); 1874 1875 if (dmu_objset_projectquota_enabled(os)) { 1876 cookie = NULL; 1877 while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas, 1878 &cookie)) != NULL) { 1879 mutex_enter(&os->os_userused_lock); 1880 VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT, 1881 uqn->uqn_id, uqn->uqn_delta, tx)); 1882 mutex_exit(&os->os_userused_lock); 1883 kmem_free(uqn, sizeof (*uqn)); 1884 } 1885 avl_destroy(&cache->uqc_project_deltas); 1886 } 1887 } 1888 1889 static void 1890 userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta) 1891 { 1892 userquota_node_t *uqn; 1893 avl_index_t idx; 1894 1895 ASSERT(strlen(id) < sizeof (uqn->uqn_id)); 1896 /* 1897 * Use id directly for searching because uqn_id is the first field of 1898 * userquota_node_t and fields after uqn_id won't be accessed in 1899 * avl_find(). 1900 */ 1901 uqn = avl_find(avl, (const void *)id, &idx); 1902 if (uqn == NULL) { 1903 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); 1904 strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id)); 1905 avl_insert(avl, uqn, idx); 1906 } 1907 uqn->uqn_delta += delta; 1908 } 1909 1910 static void 1911 do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used, 1912 uint64_t flags, uint64_t user, uint64_t group, uint64_t project, 1913 boolean_t subtract) 1914 { 1915 if (flags & DNODE_FLAG_USERUSED_ACCOUNTED) { 1916 int64_t delta = DNODE_MIN_SIZE + used; 1917 char name[20]; 1918 1919 if (subtract) 1920 delta = -delta; 1921 1922 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)user); 1923 userquota_update_cache(&cache->uqc_user_deltas, name, delta); 1924 1925 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)group); 1926 userquota_update_cache(&cache->uqc_group_deltas, name, delta); 1927 1928 if (dmu_objset_projectquota_enabled(os)) { 1929 (void) snprintf(name, sizeof (name), "%llx", 1930 (longlong_t)project); 1931 userquota_update_cache(&cache->uqc_project_deltas, 1932 name, delta); 1933 } 1934 } 1935 } 1936 1937 static void 1938 do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags, 1939 uint64_t user, uint64_t group, uint64_t project, boolean_t subtract) 1940 { 1941 if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) { 1942 char name[20 + DMU_OBJACCT_PREFIX_LEN]; 1943 int delta = subtract ? -1 : 1; 1944 1945 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", 1946 (longlong_t)user); 1947 userquota_update_cache(&cache->uqc_user_deltas, name, delta); 1948 1949 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", 1950 (longlong_t)group); 1951 userquota_update_cache(&cache->uqc_group_deltas, name, delta); 1952 1953 if (dmu_objset_projectquota_enabled(os)) { 1954 (void) snprintf(name, sizeof (name), 1955 DMU_OBJACCT_PREFIX "%llx", (longlong_t)project); 1956 userquota_update_cache(&cache->uqc_project_deltas, 1957 name, delta); 1958 } 1959 } 1960 } 1961 1962 typedef struct userquota_updates_arg { 1963 objset_t *uua_os; 1964 int uua_sublist_idx; 1965 dmu_tx_t *uua_tx; 1966 } userquota_updates_arg_t; 1967 1968 static void 1969 userquota_updates_task(void *arg) 1970 { 1971 userquota_updates_arg_t *uua = arg; 1972 objset_t *os = uua->uua_os; 1973 dmu_tx_t *tx = uua->uua_tx; 1974 dnode_t *dn; 1975 userquota_cache_t cache = { { 0 } }; 1976 1977 multilist_sublist_t *list = multilist_sublist_lock_idx( 1978 &os->os_synced_dnodes, uua->uua_sublist_idx); 1979 1980 ASSERT(multilist_sublist_head(list) == NULL || 1981 dmu_objset_userused_enabled(os)); 1982 avl_create(&cache.uqc_user_deltas, userquota_compare, 1983 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 1984 avl_create(&cache.uqc_group_deltas, userquota_compare, 1985 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 1986 if (dmu_objset_projectquota_enabled(os)) 1987 avl_create(&cache.uqc_project_deltas, userquota_compare, 1988 sizeof (userquota_node_t), offsetof(userquota_node_t, 1989 uqn_node)); 1990 1991 while ((dn = multilist_sublist_head(list)) != NULL) { 1992 int flags; 1993 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); 1994 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || 1995 dn->dn_phys->dn_flags & 1996 DNODE_FLAG_USERUSED_ACCOUNTED); 1997 1998 flags = dn->dn_id_flags; 1999 ASSERT(flags); 2000 if (flags & DN_ID_OLD_EXIST) { 2001 do_userquota_update(os, &cache, dn->dn_oldused, 2002 dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, 2003 dn->dn_oldprojid, B_TRUE); 2004 do_userobjquota_update(os, &cache, dn->dn_oldflags, 2005 dn->dn_olduid, dn->dn_oldgid, 2006 dn->dn_oldprojid, B_TRUE); 2007 } 2008 if (flags & DN_ID_NEW_EXIST) { 2009 do_userquota_update(os, &cache, 2010 DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, 2011 dn->dn_newuid, dn->dn_newgid, 2012 dn->dn_newprojid, B_FALSE); 2013 do_userobjquota_update(os, &cache, 2014 dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, 2015 dn->dn_newprojid, B_FALSE); 2016 } 2017 2018 mutex_enter(&dn->dn_mtx); 2019 dn->dn_oldused = 0; 2020 dn->dn_oldflags = 0; 2021 if (dn->dn_id_flags & DN_ID_NEW_EXIST) { 2022 dn->dn_olduid = dn->dn_newuid; 2023 dn->dn_oldgid = dn->dn_newgid; 2024 dn->dn_oldprojid = dn->dn_newprojid; 2025 dn->dn_id_flags |= DN_ID_OLD_EXIST; 2026 if (dn->dn_bonuslen == 0) 2027 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 2028 else 2029 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2030 } 2031 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); 2032 ASSERT3U(dn->dn_dirtycnt, >, 0); 2033 dn->dn_dirtycnt--; 2034 mutex_exit(&dn->dn_mtx); 2035 2036 multilist_sublist_remove(list, dn); 2037 dnode_rele(dn, &os->os_synced_dnodes); 2038 } 2039 do_userquota_cacheflush(os, &cache, tx); 2040 multilist_sublist_unlock(list); 2041 kmem_free(uua, sizeof (*uua)); 2042 } 2043 2044 /* 2045 * Release dnode holds from dmu_objset_sync_dnodes(). When the dnode is being 2046 * synced (i.e. we have issued the zio's for blocks in the dnode), it can't be 2047 * evicted because the block containing the dnode can't be evicted until it is 2048 * written out. However, this hold is necessary to prevent the dnode_t from 2049 * being moved (via dnode_move()) while it's still referenced by 2050 * dbuf_dirty_record_t:dr_dnode. And dr_dnode is needed for 2051 * dirty_lightweight_leaf-type dirty records. 2052 * 2053 * If we are doing user-object accounting, the dnode_rele() happens from 2054 * userquota_updates_task() instead. 2055 */ 2056 static void 2057 dnode_rele_task(void *arg) 2058 { 2059 userquota_updates_arg_t *uua = arg; 2060 objset_t *os = uua->uua_os; 2061 2062 multilist_sublist_t *list = multilist_sublist_lock_idx( 2063 &os->os_synced_dnodes, uua->uua_sublist_idx); 2064 2065 dnode_t *dn; 2066 while ((dn = multilist_sublist_head(list)) != NULL) { 2067 mutex_enter(&dn->dn_mtx); 2068 ASSERT3U(dn->dn_dirtycnt, >, 0); 2069 dn->dn_dirtycnt--; 2070 mutex_exit(&dn->dn_mtx); 2071 multilist_sublist_remove(list, dn); 2072 dnode_rele(dn, &os->os_synced_dnodes); 2073 } 2074 multilist_sublist_unlock(list); 2075 kmem_free(uua, sizeof (*uua)); 2076 } 2077 2078 /* 2079 * Return TRUE if userquota updates are needed. 2080 */ 2081 static boolean_t 2082 dmu_objset_do_userquota_updates_prep(objset_t *os, dmu_tx_t *tx) 2083 { 2084 if (!dmu_objset_userused_enabled(os)) 2085 return (B_FALSE); 2086 2087 /* 2088 * If this is a raw receive just return and handle accounting 2089 * later when we have the keys loaded. We also don't do user 2090 * accounting during claiming since the datasets are not owned 2091 * for the duration of claiming and this txg should only be 2092 * used for recovery. 2093 */ 2094 if (os->os_encrypted && dmu_objset_is_receiving(os)) 2095 return (B_FALSE); 2096 2097 if (tx->tx_txg <= os->os_spa->spa_claim_max_txg) 2098 return (B_FALSE); 2099 2100 /* Allocate the user/group/project used objects if necessary. */ 2101 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 2102 VERIFY0(zap_create_claim(os, 2103 DMU_USERUSED_OBJECT, 2104 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2105 VERIFY0(zap_create_claim(os, 2106 DMU_GROUPUSED_OBJECT, 2107 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2108 } 2109 2110 if (dmu_objset_projectquota_enabled(os) && 2111 DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 2112 VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT, 2113 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2114 } 2115 return (B_TRUE); 2116 } 2117 2118 /* 2119 * Dispatch taskq tasks to dp_sync_taskq to update the user accounting, and 2120 * also release the holds on the dnodes from dmu_objset_sync_dnodes(). 2121 * The caller must taskq_wait(dp_sync_taskq). 2122 */ 2123 void 2124 dmu_objset_sync_done(objset_t *os, dmu_tx_t *tx) 2125 { 2126 boolean_t need_userquota = dmu_objset_do_userquota_updates_prep(os, tx); 2127 2128 int num_sublists = multilist_get_num_sublists(&os->os_synced_dnodes); 2129 for (int i = 0; i < num_sublists; i++) { 2130 userquota_updates_arg_t *uua = 2131 kmem_alloc(sizeof (*uua), KM_SLEEP); 2132 uua->uua_os = os; 2133 uua->uua_sublist_idx = i; 2134 uua->uua_tx = tx; 2135 2136 /* 2137 * If we don't need to update userquotas, use 2138 * dnode_rele_task() to call dnode_rele() 2139 */ 2140 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, 2141 need_userquota ? userquota_updates_task : dnode_rele_task, 2142 uua, 0); 2143 /* callback frees uua */ 2144 } 2145 } 2146 2147 2148 /* 2149 * Returns a pointer to data to find uid/gid from 2150 * 2151 * If a dirty record for transaction group that is syncing can't 2152 * be found then NULL is returned. In the NULL case it is assumed 2153 * the uid/gid aren't changing. 2154 */ 2155 static void * 2156 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) 2157 { 2158 dbuf_dirty_record_t *dr; 2159 void *data; 2160 2161 if (db->db_dirtycnt == 0) { 2162 ASSERT(MUTEX_HELD(&db->db_mtx)); 2163 return (db->db.db_data); /* Nothing is changing */ 2164 } 2165 2166 dr = dbuf_find_dirty_eq(db, tx->tx_txg); 2167 2168 if (dr == NULL) { 2169 data = NULL; 2170 } else { 2171 if (dr->dr_dnode->dn_bonuslen == 0 && 2172 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) 2173 data = dr->dt.dl.dr_data->b_data; 2174 else 2175 data = dr->dt.dl.dr_data; 2176 } 2177 2178 return (data); 2179 } 2180 2181 void 2182 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) 2183 { 2184 objset_t *os = dn->dn_objset; 2185 void *data = NULL; 2186 dmu_buf_impl_t *db = NULL; 2187 int flags = dn->dn_id_flags; 2188 int error; 2189 boolean_t have_spill = B_FALSE; 2190 2191 if (!dmu_objset_userused_enabled(dn->dn_objset)) 2192 return; 2193 2194 /* 2195 * Raw receives introduce a problem with user accounting. Raw 2196 * receives cannot update the user accounting info because the 2197 * user ids and the sizes are encrypted. To guarantee that we 2198 * never end up with bad user accounting, we simply disable it 2199 * during raw receives. We also disable this for normal receives 2200 * so that an incremental raw receive may be done on top of an 2201 * existing non-raw receive. 2202 */ 2203 if (os->os_encrypted && dmu_objset_is_receiving(os)) 2204 return; 2205 2206 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| 2207 DN_ID_CHKED_SPILL))) 2208 return; 2209 2210 if (before && dn->dn_bonuslen != 0) 2211 data = DN_BONUS(dn->dn_phys); 2212 else if (!before && dn->dn_bonuslen != 0) { 2213 if (dn->dn_bonus) { 2214 db = dn->dn_bonus; 2215 mutex_enter(&db->db_mtx); 2216 data = dmu_objset_userquota_find_data(db, tx); 2217 } else { 2218 data = DN_BONUS(dn->dn_phys); 2219 } 2220 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { 2221 dmu_flags_t rf = DB_RF_MUST_SUCCEED; 2222 2223 if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) 2224 rf |= DB_RF_HAVESTRUCT; 2225 error = dmu_spill_hold_by_dnode(dn, rf, 2226 FTAG, (dmu_buf_t **)&db); 2227 ASSERT0(error); 2228 mutex_enter(&db->db_mtx); 2229 data = (before) ? db->db.db_data : 2230 dmu_objset_userquota_find_data(db, tx); 2231 have_spill = B_TRUE; 2232 } else { 2233 mutex_enter(&dn->dn_mtx); 2234 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2235 mutex_exit(&dn->dn_mtx); 2236 return; 2237 } 2238 2239 /* 2240 * Must always call the callback in case the object 2241 * type has changed and that type isn't an object type to track 2242 */ 2243 zfs_file_info_t zfi; 2244 error = file_cbs[os->os_phys->os_type](dn->dn_bonustype, data, &zfi); 2245 2246 if (before) { 2247 ASSERT(data); 2248 dn->dn_olduid = zfi.zfi_user; 2249 dn->dn_oldgid = zfi.zfi_group; 2250 dn->dn_oldprojid = zfi.zfi_project; 2251 } else if (data) { 2252 dn->dn_newuid = zfi.zfi_user; 2253 dn->dn_newgid = zfi.zfi_group; 2254 dn->dn_newprojid = zfi.zfi_project; 2255 } 2256 2257 /* 2258 * Preserve existing uid/gid when the callback can't determine 2259 * what the new uid/gid are and the callback returned EEXIST. 2260 * The EEXIST error tells us to just use the existing uid/gid. 2261 * If we don't know what the old values are then just assign 2262 * them to 0, since that is a new file being created. 2263 */ 2264 if (!before && data == NULL && error == EEXIST) { 2265 if (flags & DN_ID_OLD_EXIST) { 2266 dn->dn_newuid = dn->dn_olduid; 2267 dn->dn_newgid = dn->dn_oldgid; 2268 dn->dn_newprojid = dn->dn_oldprojid; 2269 } else { 2270 dn->dn_newuid = 0; 2271 dn->dn_newgid = 0; 2272 dn->dn_newprojid = ZFS_DEFAULT_PROJID; 2273 } 2274 error = 0; 2275 } 2276 2277 if (db) 2278 mutex_exit(&db->db_mtx); 2279 2280 mutex_enter(&dn->dn_mtx); 2281 if (error == 0 && before) 2282 dn->dn_id_flags |= DN_ID_OLD_EXIST; 2283 if (error == 0 && !before) 2284 dn->dn_id_flags |= DN_ID_NEW_EXIST; 2285 2286 if (have_spill) { 2287 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 2288 } else { 2289 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2290 } 2291 mutex_exit(&dn->dn_mtx); 2292 if (have_spill) 2293 dmu_buf_rele((dmu_buf_t *)db, FTAG); 2294 } 2295 2296 boolean_t 2297 dmu_objset_userspace_present(objset_t *os) 2298 { 2299 return (os->os_phys->os_flags & 2300 OBJSET_FLAG_USERACCOUNTING_COMPLETE); 2301 } 2302 2303 boolean_t 2304 dmu_objset_userobjspace_present(objset_t *os) 2305 { 2306 return (os->os_phys->os_flags & 2307 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE); 2308 } 2309 2310 boolean_t 2311 dmu_objset_projectquota_present(objset_t *os) 2312 { 2313 return (os->os_phys->os_flags & 2314 OBJSET_FLAG_PROJECTQUOTA_COMPLETE); 2315 } 2316 2317 static int 2318 dmu_objset_space_upgrade(objset_t *os) 2319 { 2320 uint64_t obj; 2321 int err = 0; 2322 2323 /* 2324 * We simply need to mark every object dirty, so that it will be 2325 * synced out and now accounted. If this is called 2326 * concurrently, or if we already did some work before crashing, 2327 * that's fine, since we track each object's accounted state 2328 * independently. 2329 */ 2330 2331 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { 2332 dmu_tx_t *tx; 2333 dmu_buf_t *db; 2334 int objerr; 2335 2336 mutex_enter(&os->os_upgrade_lock); 2337 if (os->os_upgrade_exit) 2338 err = SET_ERROR(EINTR); 2339 mutex_exit(&os->os_upgrade_lock); 2340 if (err != 0) 2341 return (err); 2342 2343 if (issig()) 2344 return (SET_ERROR(EINTR)); 2345 2346 objerr = dmu_bonus_hold(os, obj, FTAG, &db); 2347 if (objerr != 0) 2348 continue; 2349 tx = dmu_tx_create(os); 2350 dmu_tx_hold_bonus(tx, obj); 2351 objerr = dmu_tx_assign(tx, DMU_TX_WAIT); 2352 if (objerr != 0) { 2353 dmu_buf_rele(db, FTAG); 2354 dmu_tx_abort(tx); 2355 continue; 2356 } 2357 dmu_buf_will_dirty(db, tx); 2358 dmu_buf_rele(db, FTAG); 2359 dmu_tx_commit(tx); 2360 } 2361 return (0); 2362 } 2363 2364 static int 2365 dmu_objset_userspace_upgrade_cb(objset_t *os) 2366 { 2367 int err = 0; 2368 2369 if (dmu_objset_userspace_present(os)) 2370 return (0); 2371 if (dmu_objset_is_snapshot(os)) 2372 return (SET_ERROR(EINVAL)); 2373 if (!dmu_objset_userused_enabled(os)) 2374 return (SET_ERROR(ENOTSUP)); 2375 2376 err = dmu_objset_space_upgrade(os); 2377 if (err) 2378 return (err); 2379 2380 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 2381 txg_wait_synced(dmu_objset_pool(os), 0); 2382 return (0); 2383 } 2384 2385 void 2386 dmu_objset_userspace_upgrade(objset_t *os) 2387 { 2388 dmu_objset_upgrade(os, dmu_objset_userspace_upgrade_cb); 2389 } 2390 2391 static int 2392 dmu_objset_id_quota_upgrade_cb(objset_t *os) 2393 { 2394 int err = 0; 2395 2396 if (dmu_objset_userobjspace_present(os) && 2397 dmu_objset_projectquota_present(os)) 2398 return (0); 2399 if (dmu_objset_is_snapshot(os)) 2400 return (SET_ERROR(EINVAL)); 2401 if (!dmu_objset_userused_enabled(os)) 2402 return (SET_ERROR(ENOTSUP)); 2403 if (!dmu_objset_projectquota_enabled(os) && 2404 dmu_objset_userobjspace_present(os)) 2405 return (SET_ERROR(ENOTSUP)); 2406 2407 if (dmu_objset_userobjused_enabled(os)) 2408 dmu_objset_ds(os)->ds_feature_activation[ 2409 SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; 2410 if (dmu_objset_projectquota_enabled(os)) 2411 dmu_objset_ds(os)->ds_feature_activation[ 2412 SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; 2413 2414 err = dmu_objset_space_upgrade(os); 2415 if (err) 2416 return (err); 2417 2418 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 2419 if (dmu_objset_userobjused_enabled(os)) 2420 os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; 2421 if (dmu_objset_projectquota_enabled(os)) 2422 os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE; 2423 2424 txg_wait_synced(dmu_objset_pool(os), 0); 2425 return (0); 2426 } 2427 2428 void 2429 dmu_objset_id_quota_upgrade(objset_t *os) 2430 { 2431 dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb); 2432 } 2433 2434 boolean_t 2435 dmu_objset_userobjspace_upgradable(objset_t *os) 2436 { 2437 return (dmu_objset_type(os) == DMU_OST_ZFS && 2438 !dmu_objset_is_snapshot(os) && 2439 dmu_objset_userobjused_enabled(os) && 2440 !dmu_objset_userobjspace_present(os) && 2441 spa_writeable(dmu_objset_spa(os))); 2442 } 2443 2444 boolean_t 2445 dmu_objset_projectquota_upgradable(objset_t *os) 2446 { 2447 return (dmu_objset_type(os) == DMU_OST_ZFS && 2448 !dmu_objset_is_snapshot(os) && 2449 dmu_objset_projectquota_enabled(os) && 2450 !dmu_objset_projectquota_present(os) && 2451 spa_writeable(dmu_objset_spa(os))); 2452 } 2453 2454 void 2455 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 2456 uint64_t *usedobjsp, uint64_t *availobjsp) 2457 { 2458 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, 2459 usedobjsp, availobjsp); 2460 } 2461 2462 uint64_t 2463 dmu_objset_fsid_guid(objset_t *os) 2464 { 2465 return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); 2466 } 2467 2468 void 2469 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) 2470 { 2471 stat->dds_type = os->os_phys->os_type; 2472 if (os->os_dsl_dataset) 2473 dsl_dataset_fast_stat(os->os_dsl_dataset, stat); 2474 } 2475 2476 void 2477 dmu_objset_stats(objset_t *os, nvlist_t *nv) 2478 { 2479 ASSERT(os->os_dsl_dataset || 2480 os->os_phys->os_type == DMU_OST_META); 2481 2482 if (os->os_dsl_dataset != NULL) 2483 dsl_dataset_stats(os->os_dsl_dataset, nv); 2484 2485 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, 2486 os->os_phys->os_type); 2487 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, 2488 dmu_objset_userspace_present(os)); 2489 } 2490 2491 int 2492 dmu_objset_is_snapshot(objset_t *os) 2493 { 2494 if (os->os_dsl_dataset != NULL) 2495 return (os->os_dsl_dataset->ds_is_snapshot); 2496 else 2497 return (B_FALSE); 2498 } 2499 2500 int 2501 dmu_snapshot_realname(objset_t *os, const char *name, char *real, int maxlen, 2502 boolean_t *conflict) 2503 { 2504 dsl_dataset_t *ds = os->os_dsl_dataset; 2505 uint64_t ignored; 2506 2507 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 2508 return (SET_ERROR(ENOENT)); 2509 2510 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, 2511 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, 2512 MT_NORMALIZE, real, maxlen, conflict)); 2513 } 2514 2515 int 2516 dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 2517 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) 2518 { 2519 dsl_dataset_t *ds = os->os_dsl_dataset; 2520 zap_cursor_t cursor; 2521 zap_attribute_t *attr; 2522 2523 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 2524 2525 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 2526 return (SET_ERROR(ENOENT)); 2527 2528 attr = zap_attribute_alloc(); 2529 zap_cursor_init_serialized(&cursor, 2530 ds->ds_dir->dd_pool->dp_meta_objset, 2531 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); 2532 2533 if (zap_cursor_retrieve(&cursor, attr) != 0) { 2534 zap_cursor_fini(&cursor); 2535 zap_attribute_free(attr); 2536 return (SET_ERROR(ENOENT)); 2537 } 2538 2539 if (strlen(attr->za_name) + 1 > namelen) { 2540 zap_cursor_fini(&cursor); 2541 zap_attribute_free(attr); 2542 return (SET_ERROR(ENAMETOOLONG)); 2543 } 2544 2545 (void) strlcpy(name, attr->za_name, namelen); 2546 if (idp) 2547 *idp = attr->za_first_integer; 2548 if (case_conflict) 2549 *case_conflict = attr->za_normalization_conflict; 2550 zap_cursor_advance(&cursor); 2551 *offp = zap_cursor_serialize(&cursor); 2552 zap_cursor_fini(&cursor); 2553 zap_attribute_free(attr); 2554 2555 return (0); 2556 } 2557 2558 int 2559 dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value) 2560 { 2561 return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value)); 2562 } 2563 2564 int 2565 dmu_dir_list_next(objset_t *os, int namelen, char *name, 2566 uint64_t *idp, uint64_t *offp) 2567 { 2568 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 2569 zap_cursor_t cursor; 2570 zap_attribute_t *attr; 2571 2572 /* there is no next dir on a snapshot! */ 2573 if (os->os_dsl_dataset->ds_object != 2574 dsl_dir_phys(dd)->dd_head_dataset_obj) 2575 return (SET_ERROR(ENOENT)); 2576 2577 attr = zap_attribute_alloc(); 2578 zap_cursor_init_serialized(&cursor, 2579 dd->dd_pool->dp_meta_objset, 2580 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); 2581 2582 if (zap_cursor_retrieve(&cursor, attr) != 0) { 2583 zap_cursor_fini(&cursor); 2584 zap_attribute_free(attr); 2585 return (SET_ERROR(ENOENT)); 2586 } 2587 2588 if (strlen(attr->za_name) + 1 > namelen) { 2589 zap_cursor_fini(&cursor); 2590 zap_attribute_free(attr); 2591 return (SET_ERROR(ENAMETOOLONG)); 2592 } 2593 2594 (void) strlcpy(name, attr->za_name, namelen); 2595 if (idp) 2596 *idp = attr->za_first_integer; 2597 zap_cursor_advance(&cursor); 2598 *offp = zap_cursor_serialize(&cursor); 2599 zap_cursor_fini(&cursor); 2600 zap_attribute_free(attr); 2601 2602 return (0); 2603 } 2604 2605 typedef struct dmu_objset_find_ctx { 2606 taskq_t *dc_tq; 2607 dsl_pool_t *dc_dp; 2608 uint64_t dc_ddobj; 2609 char *dc_ddname; /* last component of ddobj's name */ 2610 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); 2611 void *dc_arg; 2612 int dc_flags; 2613 kmutex_t *dc_error_lock; 2614 int *dc_error; 2615 } dmu_objset_find_ctx_t; 2616 2617 static void 2618 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) 2619 { 2620 dsl_pool_t *dp = dcp->dc_dp; 2621 dsl_dir_t *dd; 2622 dsl_dataset_t *ds; 2623 zap_cursor_t zc; 2624 zap_attribute_t *attr; 2625 uint64_t thisobj; 2626 int err = 0; 2627 2628 /* don't process if there already was an error */ 2629 if (*dcp->dc_error != 0) 2630 goto out; 2631 2632 /* 2633 * Note: passing the name (dc_ddname) here is optional, but it 2634 * improves performance because we don't need to call 2635 * zap_value_search() to determine the name. 2636 */ 2637 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); 2638 if (err != 0) 2639 goto out; 2640 2641 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2642 if (dd->dd_myname[0] == '$') { 2643 dsl_dir_rele(dd, FTAG); 2644 goto out; 2645 } 2646 2647 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2648 attr = zap_attribute_alloc(); 2649 2650 /* 2651 * Iterate over all children. 2652 */ 2653 if (dcp->dc_flags & DS_FIND_CHILDREN) { 2654 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2655 dsl_dir_phys(dd)->dd_child_dir_zapobj); 2656 zap_cursor_retrieve(&zc, attr) == 0; 2657 (void) zap_cursor_advance(&zc)) { 2658 ASSERT3U(attr->za_integer_length, ==, 2659 sizeof (uint64_t)); 2660 ASSERT3U(attr->za_num_integers, ==, 1); 2661 2662 dmu_objset_find_ctx_t *child_dcp = 2663 kmem_alloc(sizeof (*child_dcp), KM_SLEEP); 2664 *child_dcp = *dcp; 2665 child_dcp->dc_ddobj = attr->za_first_integer; 2666 child_dcp->dc_ddname = spa_strdup(attr->za_name); 2667 if (dcp->dc_tq != NULL) 2668 (void) taskq_dispatch(dcp->dc_tq, 2669 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); 2670 else 2671 dmu_objset_find_dp_impl(child_dcp); 2672 } 2673 zap_cursor_fini(&zc); 2674 } 2675 2676 /* 2677 * Iterate over all snapshots. 2678 */ 2679 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { 2680 dsl_dataset_t *ds; 2681 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2682 2683 if (err == 0) { 2684 uint64_t snapobj; 2685 2686 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 2687 dsl_dataset_rele(ds, FTAG); 2688 2689 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 2690 zap_cursor_retrieve(&zc, attr) == 0; 2691 (void) zap_cursor_advance(&zc)) { 2692 ASSERT3U(attr->za_integer_length, ==, 2693 sizeof (uint64_t)); 2694 ASSERT3U(attr->za_num_integers, ==, 1); 2695 2696 err = dsl_dataset_hold_obj(dp, 2697 attr->za_first_integer, FTAG, &ds); 2698 if (err != 0) 2699 break; 2700 err = dcp->dc_func(dp, ds, dcp->dc_arg); 2701 dsl_dataset_rele(ds, FTAG); 2702 if (err != 0) 2703 break; 2704 } 2705 zap_cursor_fini(&zc); 2706 } 2707 } 2708 2709 zap_attribute_free(attr); 2710 2711 if (err != 0) { 2712 dsl_dir_rele(dd, FTAG); 2713 goto out; 2714 } 2715 2716 /* 2717 * Apply to self. 2718 */ 2719 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2720 2721 /* 2722 * Note: we hold the dir while calling dsl_dataset_hold_obj() so 2723 * that the dir will remain cached, and we won't have to re-instantiate 2724 * it (which could be expensive due to finding its name via 2725 * zap_value_search()). 2726 */ 2727 dsl_dir_rele(dd, FTAG); 2728 if (err != 0) 2729 goto out; 2730 err = dcp->dc_func(dp, ds, dcp->dc_arg); 2731 dsl_dataset_rele(ds, FTAG); 2732 2733 out: 2734 if (err != 0) { 2735 mutex_enter(dcp->dc_error_lock); 2736 /* only keep first error */ 2737 if (*dcp->dc_error == 0) 2738 *dcp->dc_error = err; 2739 mutex_exit(dcp->dc_error_lock); 2740 } 2741 2742 if (dcp->dc_ddname != NULL) 2743 spa_strfree(dcp->dc_ddname); 2744 kmem_free(dcp, sizeof (*dcp)); 2745 } 2746 2747 static void 2748 dmu_objset_find_dp_cb(void *arg) 2749 { 2750 dmu_objset_find_ctx_t *dcp = arg; 2751 dsl_pool_t *dp = dcp->dc_dp; 2752 2753 /* 2754 * We need to get a pool_config_lock here, as there are several 2755 * assert(pool_config_held) down the stack. Getting a lock via 2756 * dsl_pool_config_enter is risky, as it might be stalled by a 2757 * pending writer. This would deadlock, as the write lock can 2758 * only be granted when our parent thread gives up the lock. 2759 * The _prio interface gives us priority over a pending writer. 2760 */ 2761 dsl_pool_config_enter_prio(dp, FTAG); 2762 2763 dmu_objset_find_dp_impl(dcp); 2764 2765 dsl_pool_config_exit(dp, FTAG); 2766 } 2767 2768 /* 2769 * Find objsets under and including ddobj, call func(ds) on each. 2770 * The order for the enumeration is completely undefined. 2771 * func is called with dsl_pool_config held. 2772 */ 2773 int 2774 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, 2775 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) 2776 { 2777 int error = 0; 2778 taskq_t *tq = NULL; 2779 int ntasks; 2780 dmu_objset_find_ctx_t *dcp; 2781 kmutex_t err_lock; 2782 2783 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); 2784 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); 2785 dcp->dc_tq = NULL; 2786 dcp->dc_dp = dp; 2787 dcp->dc_ddobj = ddobj; 2788 dcp->dc_ddname = NULL; 2789 dcp->dc_func = func; 2790 dcp->dc_arg = arg; 2791 dcp->dc_flags = flags; 2792 dcp->dc_error_lock = &err_lock; 2793 dcp->dc_error = &error; 2794 2795 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { 2796 /* 2797 * In case a write lock is held we can't make use of 2798 * parallelism, as down the stack of the worker threads 2799 * the lock is asserted via dsl_pool_config_held. 2800 * In case of a read lock this is solved by getting a read 2801 * lock in each worker thread, which isn't possible in case 2802 * of a writer lock. So we fall back to the synchronous path 2803 * here. 2804 * In the future it might be possible to get some magic into 2805 * dsl_pool_config_held in a way that it returns true for 2806 * the worker threads so that a single lock held from this 2807 * thread suffices. For now, stay single threaded. 2808 */ 2809 dmu_objset_find_dp_impl(dcp); 2810 mutex_destroy(&err_lock); 2811 2812 return (error); 2813 } 2814 2815 ntasks = dmu_find_threads; 2816 if (ntasks == 0) 2817 ntasks = vdev_count_leaves(dp->dp_spa) * 4; 2818 tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks, 2819 INT_MAX, 0); 2820 if (tq == NULL) { 2821 kmem_free(dcp, sizeof (*dcp)); 2822 mutex_destroy(&err_lock); 2823 2824 return (SET_ERROR(ENOMEM)); 2825 } 2826 dcp->dc_tq = tq; 2827 2828 /* dcp will be freed by task */ 2829 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); 2830 2831 /* 2832 * PORTING: this code relies on the property of taskq_wait to wait 2833 * until no more tasks are queued and no more tasks are active. As 2834 * we always queue new tasks from within other tasks, task_wait 2835 * reliably waits for the full recursion to finish, even though we 2836 * enqueue new tasks after taskq_wait has been called. 2837 * On platforms other than illumos, taskq_wait may not have this 2838 * property. 2839 */ 2840 taskq_wait(tq); 2841 taskq_destroy(tq); 2842 mutex_destroy(&err_lock); 2843 2844 return (error); 2845 } 2846 2847 /* 2848 * Find all objsets under name, and for each, call 'func(child_name, arg)'. 2849 * The dp_config_rwlock must not be held when this is called, and it 2850 * will not be held when the callback is called. 2851 * Therefore this function should only be used when the pool is not changing 2852 * (e.g. in syncing context), or the callback can deal with the possible races. 2853 */ 2854 static int 2855 dmu_objset_find_impl(spa_t *spa, const char *name, 2856 int func(const char *, void *), void *arg, int flags) 2857 { 2858 dsl_dir_t *dd; 2859 dsl_pool_t *dp = spa_get_dsl(spa); 2860 dsl_dataset_t *ds; 2861 zap_cursor_t zc; 2862 zap_attribute_t *attr; 2863 char *child; 2864 uint64_t thisobj; 2865 int err; 2866 2867 dsl_pool_config_enter(dp, FTAG); 2868 2869 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); 2870 if (err != 0) { 2871 dsl_pool_config_exit(dp, FTAG); 2872 return (err); 2873 } 2874 2875 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2876 if (dd->dd_myname[0] == '$') { 2877 dsl_dir_rele(dd, FTAG); 2878 dsl_pool_config_exit(dp, FTAG); 2879 return (0); 2880 } 2881 2882 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2883 attr = zap_attribute_alloc(); 2884 2885 /* 2886 * Iterate over all children. 2887 */ 2888 if (flags & DS_FIND_CHILDREN) { 2889 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2890 dsl_dir_phys(dd)->dd_child_dir_zapobj); 2891 zap_cursor_retrieve(&zc, attr) == 0; 2892 (void) zap_cursor_advance(&zc)) { 2893 ASSERT3U(attr->za_integer_length, ==, 2894 sizeof (uint64_t)); 2895 ASSERT3U(attr->za_num_integers, ==, 1); 2896 2897 child = kmem_asprintf("%s/%s", name, attr->za_name); 2898 dsl_pool_config_exit(dp, FTAG); 2899 err = dmu_objset_find_impl(spa, child, 2900 func, arg, flags); 2901 dsl_pool_config_enter(dp, FTAG); 2902 kmem_strfree(child); 2903 if (err != 0) 2904 break; 2905 } 2906 zap_cursor_fini(&zc); 2907 2908 if (err != 0) { 2909 dsl_dir_rele(dd, FTAG); 2910 dsl_pool_config_exit(dp, FTAG); 2911 zap_attribute_free(attr); 2912 return (err); 2913 } 2914 } 2915 2916 /* 2917 * Iterate over all snapshots. 2918 */ 2919 if (flags & DS_FIND_SNAPSHOTS) { 2920 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2921 2922 if (err == 0) { 2923 uint64_t snapobj; 2924 2925 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 2926 dsl_dataset_rele(ds, FTAG); 2927 2928 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 2929 zap_cursor_retrieve(&zc, attr) == 0; 2930 (void) zap_cursor_advance(&zc)) { 2931 ASSERT3U(attr->za_integer_length, ==, 2932 sizeof (uint64_t)); 2933 ASSERT3U(attr->za_num_integers, ==, 1); 2934 2935 child = kmem_asprintf("%s@%s", 2936 name, attr->za_name); 2937 dsl_pool_config_exit(dp, FTAG); 2938 err = func(child, arg); 2939 dsl_pool_config_enter(dp, FTAG); 2940 kmem_strfree(child); 2941 if (err != 0) 2942 break; 2943 } 2944 zap_cursor_fini(&zc); 2945 } 2946 } 2947 2948 dsl_dir_rele(dd, FTAG); 2949 zap_attribute_free(attr); 2950 dsl_pool_config_exit(dp, FTAG); 2951 2952 if (err != 0) 2953 return (err); 2954 2955 /* Apply to self. */ 2956 return (func(name, arg)); 2957 } 2958 2959 /* 2960 * See comment above dmu_objset_find_impl(). 2961 */ 2962 int 2963 dmu_objset_find(const char *name, int func(const char *, void *), void *arg, 2964 int flags) 2965 { 2966 spa_t *spa; 2967 int error; 2968 2969 error = spa_open(name, &spa, FTAG); 2970 if (error != 0) 2971 return (error); 2972 error = dmu_objset_find_impl(spa, name, func, arg, flags); 2973 spa_close(spa, FTAG); 2974 return (error); 2975 } 2976 2977 boolean_t 2978 dmu_objset_incompatible_encryption_version(objset_t *os) 2979 { 2980 return (dsl_dir_incompatible_encryption_version( 2981 os->os_dsl_dataset->ds_dir)); 2982 } 2983 2984 void 2985 dmu_objset_set_user(objset_t *os, void *user_ptr) 2986 { 2987 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 2988 os->os_user_ptr = user_ptr; 2989 } 2990 2991 void * 2992 dmu_objset_get_user(objset_t *os) 2993 { 2994 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 2995 return (os->os_user_ptr); 2996 } 2997 2998 /* 2999 * Determine name of filesystem, given name of snapshot. 3000 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes 3001 */ 3002 int 3003 dmu_fsname(const char *snapname, char *buf) 3004 { 3005 const char *atp = strchr(snapname, '@'); 3006 if (atp == NULL) 3007 return (SET_ERROR(EINVAL)); 3008 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) 3009 return (SET_ERROR(ENAMETOOLONG)); 3010 (void) strlcpy(buf, snapname, atp - snapname + 1); 3011 return (0); 3012 } 3013 3014 /* 3015 * Call when we think we're going to write/free space in open context 3016 * to track the amount of dirty data in the open txg, which is also the 3017 * amount of memory that can not be evicted until this txg syncs. 3018 * 3019 * Note that there are two conditions where this can be called from 3020 * syncing context: 3021 * 3022 * [1] When we just created the dataset, in which case we go on with 3023 * updating any accounting of dirty data as usual. 3024 * [2] When we are dirtying MOS data, in which case we only update the 3025 * pool's accounting of dirty data. 3026 */ 3027 void 3028 dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) 3029 { 3030 dsl_dataset_t *ds = os->os_dsl_dataset; 3031 int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); 3032 3033 if (ds != NULL) { 3034 dsl_dir_willuse_space(ds->ds_dir, aspace, tx); 3035 dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); 3036 } else { 3037 dsl_pool_dirty_mos_space(dmu_tx_pool(tx), space, tx); 3038 } 3039 } 3040 3041 /* 3042 * Check if a block is shared with a snapshot in this objset. 3043 * Returns B_TRUE if block was created before or at the time of the 3044 * previous snapshot, B_FALSE otherwise. 3045 */ 3046 boolean_t 3047 dmu_objset_block_is_shared(objset_t *os, const blkptr_t *bp) 3048 { 3049 if (BP_IS_HOLE(bp)) 3050 return (B_FALSE); 3051 3052 dsl_dataset_t *ds = os->os_dsl_dataset; 3053 if (ds == NULL) 3054 return (B_FALSE); 3055 3056 return (BP_GET_BIRTH(bp) <= dsl_dataset_phys(ds)->ds_prev_snap_txg); 3057 } 3058 3059 EXPORT_SYMBOL(dmu_objset_zil); 3060 EXPORT_SYMBOL(dmu_objset_pool); 3061 EXPORT_SYMBOL(dmu_objset_ds); 3062 EXPORT_SYMBOL(dmu_objset_type); 3063 EXPORT_SYMBOL(dmu_objset_name); 3064 EXPORT_SYMBOL(dmu_objset_hold); 3065 EXPORT_SYMBOL(dmu_objset_hold_flags); 3066 EXPORT_SYMBOL(dmu_objset_own); 3067 EXPORT_SYMBOL(dmu_objset_rele); 3068 EXPORT_SYMBOL(dmu_objset_rele_flags); 3069 EXPORT_SYMBOL(dmu_objset_disown); 3070 EXPORT_SYMBOL(dmu_objset_from_ds); 3071 EXPORT_SYMBOL(dmu_objset_create); 3072 EXPORT_SYMBOL(dmu_objset_stats); 3073 EXPORT_SYMBOL(dmu_objset_fast_stat); 3074 EXPORT_SYMBOL(dmu_objset_spa); 3075 EXPORT_SYMBOL(dmu_objset_space); 3076 EXPORT_SYMBOL(dmu_objset_fsid_guid); 3077 EXPORT_SYMBOL(dmu_objset_find); 3078 EXPORT_SYMBOL(dmu_objset_byteswap); 3079 EXPORT_SYMBOL(dmu_objset_evict_dbufs); 3080 EXPORT_SYMBOL(dmu_objset_snap_cmtime); 3081 EXPORT_SYMBOL(dmu_objset_dnodesize); 3082 3083 EXPORT_SYMBOL(dmu_objset_sync); 3084 EXPORT_SYMBOL(dmu_objset_is_dirty); 3085 EXPORT_SYMBOL(dmu_objset_create_impl_dnstats); 3086 EXPORT_SYMBOL(dmu_objset_create_impl); 3087 EXPORT_SYMBOL(dmu_objset_open_impl); 3088 EXPORT_SYMBOL(dmu_objset_evict); 3089 EXPORT_SYMBOL(dmu_objset_register_type); 3090 EXPORT_SYMBOL(dmu_objset_sync_done); 3091 EXPORT_SYMBOL(dmu_objset_userquota_get_ids); 3092 EXPORT_SYMBOL(dmu_objset_userused_enabled); 3093 EXPORT_SYMBOL(dmu_objset_userspace_upgrade); 3094 EXPORT_SYMBOL(dmu_objset_userspace_present); 3095 EXPORT_SYMBOL(dmu_objset_userobjused_enabled); 3096 EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable); 3097 EXPORT_SYMBOL(dmu_objset_userobjspace_present); 3098 EXPORT_SYMBOL(dmu_objset_projectquota_enabled); 3099 EXPORT_SYMBOL(dmu_objset_projectquota_present); 3100 EXPORT_SYMBOL(dmu_objset_projectquota_upgradable); 3101 EXPORT_SYMBOL(dmu_objset_id_quota_upgrade); 3102 EXPORT_SYMBOL(dmu_objset_block_is_shared); 3103