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