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 */ 38 39 /* Portions Copyright 2010 Robert Milkowski */ 40 41 #include <sys/cred.h> 42 #include <sys/zfs_context.h> 43 #include <sys/dmu_objset.h> 44 #include <sys/dsl_dir.h> 45 #include <sys/dsl_dataset.h> 46 #include <sys/dsl_prop.h> 47 #include <sys/dsl_pool.h> 48 #include <sys/dsl_synctask.h> 49 #include <sys/dsl_deleg.h> 50 #include <sys/dnode.h> 51 #include <sys/dbuf.h> 52 #include <sys/zvol.h> 53 #include <sys/dmu_tx.h> 54 #include <sys/zap.h> 55 #include <sys/zil.h> 56 #include <sys/dmu_impl.h> 57 #include <sys/zfs_ioctl.h> 58 #include <sys/sa.h> 59 #include <sys/zfs_onexit.h> 60 #include <sys/dsl_destroy.h> 61 #include <sys/vdev.h> 62 #include <sys/zfeature.h> 63 #include <sys/policy.h> 64 #include <sys/spa_impl.h> 65 #include <sys/dmu_recv.h> 66 #include <sys/zfs_project.h> 67 #include "zfs_namecheck.h" 68 #include <sys/vdev_impl.h> 69 #include <sys/arc.h> 70 #include <cityhash.h> 71 72 /* 73 * Needed to close a window in dnode_move() that allows the objset to be freed 74 * before it can be safely accessed. 75 */ 76 krwlock_t os_lock; 77 78 /* 79 * Tunable to overwrite the maximum number of threads for the parallelization 80 * of dmu_objset_find_dp, needed to speed up the import of pools with many 81 * datasets. 82 * Default is 4 times the number of leaf vdevs. 83 */ 84 static const int dmu_find_threads = 0; 85 86 /* 87 * Backfill lower metadnode objects after this many have been freed. 88 * Backfilling negatively impacts object creation rates, so only do it 89 * if there are enough holes to fill. 90 */ 91 static const int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT; 92 93 static const char *upgrade_tag = "upgrade_tag"; 94 95 static void dmu_objset_find_dp_cb(void *arg); 96 97 static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb); 98 static void dmu_objset_upgrade_stop(objset_t *os); 99 100 void 101 dmu_objset_init(void) 102 { 103 rw_init(&os_lock, NULL, RW_DEFAULT, NULL); 104 } 105 106 void 107 dmu_objset_fini(void) 108 { 109 rw_destroy(&os_lock); 110 } 111 112 spa_t * 113 dmu_objset_spa(objset_t *os) 114 { 115 return (os->os_spa); 116 } 117 118 zilog_t * 119 dmu_objset_zil(objset_t *os) 120 { 121 return (os->os_zil); 122 } 123 124 dsl_pool_t * 125 dmu_objset_pool(objset_t *os) 126 { 127 dsl_dataset_t *ds; 128 129 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) 130 return (ds->ds_dir->dd_pool); 131 else 132 return (spa_get_dsl(os->os_spa)); 133 } 134 135 dsl_dataset_t * 136 dmu_objset_ds(objset_t *os) 137 { 138 return (os->os_dsl_dataset); 139 } 140 141 dmu_objset_type_t 142 dmu_objset_type(objset_t *os) 143 { 144 return (os->os_phys->os_type); 145 } 146 147 void 148 dmu_objset_name(objset_t *os, char *buf) 149 { 150 dsl_dataset_name(os->os_dsl_dataset, buf); 151 } 152 153 uint64_t 154 dmu_objset_id(objset_t *os) 155 { 156 dsl_dataset_t *ds = os->os_dsl_dataset; 157 158 return (ds ? ds->ds_object : 0); 159 } 160 161 uint64_t 162 dmu_objset_dnodesize(objset_t *os) 163 { 164 return (os->os_dnodesize); 165 } 166 167 zfs_sync_type_t 168 dmu_objset_syncprop(objset_t *os) 169 { 170 return (os->os_sync); 171 } 172 173 zfs_logbias_op_t 174 dmu_objset_logbias(objset_t *os) 175 { 176 return (os->os_logbias); 177 } 178 179 static void 180 checksum_changed_cb(void *arg, uint64_t newval) 181 { 182 objset_t *os = arg; 183 184 /* 185 * Inheritance should have been done by now. 186 */ 187 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 188 189 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); 190 } 191 192 static void 193 compression_changed_cb(void *arg, uint64_t newval) 194 { 195 objset_t *os = arg; 196 197 /* 198 * Inheritance and range checking should have been done by now. 199 */ 200 ASSERT(newval != ZIO_COMPRESS_INHERIT); 201 202 os->os_compress = zio_compress_select(os->os_spa, 203 ZIO_COMPRESS_ALGO(newval), ZIO_COMPRESS_ON); 204 os->os_complevel = zio_complevel_select(os->os_spa, os->os_compress, 205 ZIO_COMPRESS_LEVEL(newval), ZIO_COMPLEVEL_DEFAULT); 206 } 207 208 static void 209 copies_changed_cb(void *arg, uint64_t newval) 210 { 211 objset_t *os = arg; 212 213 /* 214 * Inheritance and range checking should have been done by now. 215 */ 216 ASSERT(newval > 0); 217 ASSERT(newval <= spa_max_replication(os->os_spa)); 218 219 os->os_copies = newval; 220 } 221 222 static void 223 dedup_changed_cb(void *arg, uint64_t newval) 224 { 225 objset_t *os = arg; 226 spa_t *spa = os->os_spa; 227 enum zio_checksum checksum; 228 229 /* 230 * Inheritance should have been done by now. 231 */ 232 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 233 234 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); 235 236 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; 237 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); 238 } 239 240 static void 241 primary_cache_changed_cb(void *arg, uint64_t newval) 242 { 243 objset_t *os = arg; 244 245 /* 246 * Inheritance and range checking should have been done by now. 247 */ 248 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 249 newval == ZFS_CACHE_METADATA); 250 251 os->os_primary_cache = newval; 252 } 253 254 static void 255 secondary_cache_changed_cb(void *arg, uint64_t newval) 256 { 257 objset_t *os = arg; 258 259 /* 260 * Inheritance and range checking should have been done by now. 261 */ 262 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 263 newval == ZFS_CACHE_METADATA); 264 265 os->os_secondary_cache = newval; 266 } 267 268 static void 269 prefetch_changed_cb(void *arg, uint64_t newval) 270 { 271 objset_t *os = arg; 272 273 /* 274 * Inheritance should have been done by now. 275 */ 276 ASSERT(newval == ZFS_PREFETCH_ALL || newval == ZFS_PREFETCH_NONE || 277 newval == ZFS_PREFETCH_METADATA); 278 os->os_prefetch = newval; 279 } 280 281 static void 282 sync_changed_cb(void *arg, uint64_t newval) 283 { 284 objset_t *os = arg; 285 286 /* 287 * Inheritance and range checking should have been done by now. 288 */ 289 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || 290 newval == ZFS_SYNC_DISABLED); 291 292 os->os_sync = newval; 293 if (os->os_zil) 294 zil_set_sync(os->os_zil, newval); 295 } 296 297 static void 298 redundant_metadata_changed_cb(void *arg, uint64_t newval) 299 { 300 objset_t *os = arg; 301 302 /* 303 * Inheritance and range checking should have been done by now. 304 */ 305 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || 306 newval == ZFS_REDUNDANT_METADATA_MOST || 307 newval == ZFS_REDUNDANT_METADATA_SOME || 308 newval == ZFS_REDUNDANT_METADATA_NONE); 309 310 os->os_redundant_metadata = newval; 311 } 312 313 static void 314 dnodesize_changed_cb(void *arg, uint64_t newval) 315 { 316 objset_t *os = arg; 317 318 switch (newval) { 319 case ZFS_DNSIZE_LEGACY: 320 os->os_dnodesize = DNODE_MIN_SIZE; 321 break; 322 case ZFS_DNSIZE_AUTO: 323 /* 324 * Choose a dnode size that will work well for most 325 * workloads if the user specified "auto". Future code 326 * improvements could dynamically select a dnode size 327 * based on observed workload patterns. 328 */ 329 os->os_dnodesize = DNODE_MIN_SIZE * 2; 330 break; 331 case ZFS_DNSIZE_1K: 332 case ZFS_DNSIZE_2K: 333 case ZFS_DNSIZE_4K: 334 case ZFS_DNSIZE_8K: 335 case ZFS_DNSIZE_16K: 336 os->os_dnodesize = newval; 337 break; 338 } 339 } 340 341 static void 342 smallblk_changed_cb(void *arg, uint64_t newval) 343 { 344 objset_t *os = arg; 345 346 /* 347 * Inheritance and range checking should have been done by now. 348 */ 349 ASSERT(newval <= SPA_MAXBLOCKSIZE); 350 ASSERT(ISP2(newval)); 351 352 os->os_zpl_special_smallblock = newval; 353 } 354 355 static void 356 direct_changed_cb(void *arg, uint64_t newval) 357 { 358 objset_t *os = arg; 359 360 /* 361 * Inheritance and range checking should have been done by now. 362 */ 363 ASSERT(newval == ZFS_DIRECT_DISABLED || newval == ZFS_DIRECT_STANDARD || 364 newval == ZFS_DIRECT_ALWAYS); 365 366 os->os_direct = newval; 367 } 368 369 static void 370 logbias_changed_cb(void *arg, uint64_t newval) 371 { 372 objset_t *os = arg; 373 374 ASSERT(newval == ZFS_LOGBIAS_LATENCY || 375 newval == ZFS_LOGBIAS_THROUGHPUT); 376 os->os_logbias = newval; 377 if (os->os_zil) 378 zil_set_logbias(os->os_zil, newval); 379 } 380 381 static void 382 recordsize_changed_cb(void *arg, uint64_t newval) 383 { 384 objset_t *os = arg; 385 386 os->os_recordsize = newval; 387 } 388 389 void 390 dmu_objset_byteswap(void *buf, size_t size) 391 { 392 objset_phys_t *osp = buf; 393 394 ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 || 395 size == sizeof (objset_phys_t)); 396 dnode_byteswap(&osp->os_meta_dnode); 397 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); 398 osp->os_type = BSWAP_64(osp->os_type); 399 osp->os_flags = BSWAP_64(osp->os_flags); 400 if (size >= OBJSET_PHYS_SIZE_V2) { 401 dnode_byteswap(&osp->os_userused_dnode); 402 dnode_byteswap(&osp->os_groupused_dnode); 403 if (size >= sizeof (objset_phys_t)) 404 dnode_byteswap(&osp->os_projectused_dnode); 405 } 406 } 407 408 /* 409 * Runs cityhash on the objset_t pointer and the object number. 410 */ 411 static uint64_t 412 dnode_hash(const objset_t *os, uint64_t obj) 413 { 414 uintptr_t osv = (uintptr_t)os; 415 return (cityhash2((uint64_t)osv, obj)); 416 } 417 418 static unsigned int 419 dnode_multilist_index_func(multilist_t *ml, void *obj) 420 { 421 dnode_t *dn = obj; 422 423 /* 424 * The low order bits of the hash value are thought to be 425 * distributed evenly. Otherwise, in the case that the multilist 426 * has a power of two number of sublists, each sublists' usage 427 * would not be evenly distributed. In this context full 64bit 428 * division would be a waste of time, so limit it to 32 bits. 429 */ 430 return ((unsigned int)dnode_hash(dn->dn_objset, dn->dn_object) % 431 multilist_get_num_sublists(ml)); 432 } 433 434 static inline boolean_t 435 dmu_os_is_l2cacheable(objset_t *os) 436 { 437 if (os->os_secondary_cache == ZFS_CACHE_ALL || 438 os->os_secondary_cache == ZFS_CACHE_METADATA) { 439 if (l2arc_exclude_special == 0) 440 return (B_TRUE); 441 442 blkptr_t *bp = os->os_rootbp; 443 if (bp == NULL || BP_IS_HOLE(bp)) 444 return (B_FALSE); 445 uint64_t vdev = DVA_GET_VDEV(bp->blk_dva); 446 vdev_t *rvd = os->os_spa->spa_root_vdev; 447 vdev_t *vd = NULL; 448 449 if (vdev < rvd->vdev_children) 450 vd = rvd->vdev_child[vdev]; 451 452 if (vd == NULL) 453 return (B_TRUE); 454 455 if (vd->vdev_alloc_bias != VDEV_BIAS_SPECIAL && 456 vd->vdev_alloc_bias != VDEV_BIAS_DEDUP) 457 return (B_TRUE); 458 } 459 return (B_FALSE); 460 } 461 462 /* 463 * Instantiates the objset_t in-memory structure corresponding to the 464 * objset_phys_t that's pointed to by the specified blkptr_t. 465 */ 466 int 467 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 468 objset_t **osp) 469 { 470 objset_t *os; 471 int i, err; 472 473 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); 474 ASSERT(!BP_IS_REDACTED(bp)); 475 476 /* 477 * We need the pool config lock to get properties. 478 */ 479 ASSERT(ds == NULL || dsl_pool_config_held(ds->ds_dir->dd_pool)); 480 481 /* 482 * The $ORIGIN dataset (if it exists) doesn't have an associated 483 * objset, so there's no reason to open it. The $ORIGIN dataset 484 * will not exist on pools older than SPA_VERSION_ORIGIN. 485 */ 486 if (ds != NULL && spa_get_dsl(spa) != NULL && 487 spa_get_dsl(spa)->dp_origin_snap != NULL) { 488 ASSERT3P(ds->ds_dir, !=, 489 spa_get_dsl(spa)->dp_origin_snap->ds_dir); 490 } 491 492 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); 493 os->os_dsl_dataset = ds; 494 os->os_spa = spa; 495 os->os_rootbp = bp; 496 if (!BP_IS_HOLE(os->os_rootbp)) { 497 arc_flags_t aflags = ARC_FLAG_WAIT; 498 zbookmark_phys_t zb; 499 int size; 500 zio_flag_t zio_flags = ZIO_FLAG_CANFAIL; 501 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, 502 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 503 504 if (dmu_os_is_l2cacheable(os)) 505 aflags |= ARC_FLAG_L2CACHE; 506 507 if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) { 508 ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); 509 ASSERT(BP_IS_AUTHENTICATED(bp)); 510 zio_flags |= ZIO_FLAG_RAW; 511 } 512 513 dprintf_bp(os->os_rootbp, "reading %s", ""); 514 err = arc_read(NULL, spa, os->os_rootbp, 515 arc_getbuf_func, &os->os_phys_buf, 516 ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); 517 if (err != 0) { 518 kmem_free(os, sizeof (objset_t)); 519 /* convert checksum errors into IO errors */ 520 if (err == ECKSUM) 521 err = SET_ERROR(EIO); 522 return (err); 523 } 524 525 if (spa_version(spa) < SPA_VERSION_USERSPACE) 526 size = OBJSET_PHYS_SIZE_V1; 527 else if (!spa_feature_is_enabled(spa, 528 SPA_FEATURE_PROJECT_QUOTA)) 529 size = OBJSET_PHYS_SIZE_V2; 530 else 531 size = sizeof (objset_phys_t); 532 533 /* Increase the blocksize if we are permitted. */ 534 if (arc_buf_size(os->os_phys_buf) < size) { 535 arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, 536 ARC_BUFC_METADATA, size); 537 memset(buf->b_data, 0, size); 538 memcpy(buf->b_data, os->os_phys_buf->b_data, 539 arc_buf_size(os->os_phys_buf)); 540 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 541 os->os_phys_buf = buf; 542 } 543 544 os->os_phys = os->os_phys_buf->b_data; 545 os->os_flags = os->os_phys->os_flags; 546 } else { 547 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? 548 sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1; 549 os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, 550 ARC_BUFC_METADATA, size); 551 os->os_phys = os->os_phys_buf->b_data; 552 memset(os->os_phys, 0, size); 553 } 554 /* 555 * These properties will be filled in by the logic in zfs_get_zplprop() 556 * when they are queried for the first time. 557 */ 558 os->os_version = OBJSET_PROP_UNINITIALIZED; 559 os->os_normalization = OBJSET_PROP_UNINITIALIZED; 560 os->os_utf8only = OBJSET_PROP_UNINITIALIZED; 561 os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED; 562 563 /* 564 * Note: the changed_cb will be called once before the register 565 * func returns, thus changing the checksum/compression from the 566 * default (fletcher2/off). Snapshots don't need to know about 567 * checksum/compression/copies. 568 */ 569 if (ds != NULL) { 570 os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0); 571 572 err = dsl_prop_register(ds, 573 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), 574 primary_cache_changed_cb, os); 575 if (err == 0) { 576 err = dsl_prop_register(ds, 577 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), 578 secondary_cache_changed_cb, os); 579 } 580 if (err == 0) { 581 err = dsl_prop_register(ds, 582 zfs_prop_to_name(ZFS_PROP_PREFETCH), 583 prefetch_changed_cb, os); 584 } 585 if (!ds->ds_is_snapshot) { 586 if (err == 0) { 587 err = dsl_prop_register(ds, 588 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 589 checksum_changed_cb, os); 590 } 591 if (err == 0) { 592 err = dsl_prop_register(ds, 593 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 594 compression_changed_cb, os); 595 } 596 if (err == 0) { 597 err = dsl_prop_register(ds, 598 zfs_prop_to_name(ZFS_PROP_COPIES), 599 copies_changed_cb, os); 600 } 601 if (err == 0) { 602 err = dsl_prop_register(ds, 603 zfs_prop_to_name(ZFS_PROP_DEDUP), 604 dedup_changed_cb, os); 605 } 606 if (err == 0) { 607 err = dsl_prop_register(ds, 608 zfs_prop_to_name(ZFS_PROP_LOGBIAS), 609 logbias_changed_cb, os); 610 } 611 if (err == 0) { 612 err = dsl_prop_register(ds, 613 zfs_prop_to_name(ZFS_PROP_SYNC), 614 sync_changed_cb, os); 615 } 616 if (err == 0) { 617 err = dsl_prop_register(ds, 618 zfs_prop_to_name( 619 ZFS_PROP_REDUNDANT_METADATA), 620 redundant_metadata_changed_cb, os); 621 } 622 if (err == 0) { 623 err = dsl_prop_register(ds, 624 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 625 recordsize_changed_cb, os); 626 } 627 if (err == 0) { 628 err = dsl_prop_register(ds, 629 zfs_prop_to_name(ZFS_PROP_DNODESIZE), 630 dnodesize_changed_cb, os); 631 } 632 if (err == 0) { 633 err = dsl_prop_register(ds, 634 zfs_prop_to_name( 635 ZFS_PROP_SPECIAL_SMALL_BLOCKS), 636 smallblk_changed_cb, os); 637 } 638 if (err == 0) { 639 err = dsl_prop_register(ds, 640 zfs_prop_to_name(ZFS_PROP_DIRECT), 641 direct_changed_cb, os); 642 } 643 } 644 if (err != 0) { 645 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 646 kmem_free(os, sizeof (objset_t)); 647 return (err); 648 } 649 } else { 650 /* It's the meta-objset. */ 651 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; 652 os->os_compress = ZIO_COMPRESS_ON; 653 os->os_complevel = ZIO_COMPLEVEL_DEFAULT; 654 os->os_encrypted = B_FALSE; 655 os->os_copies = spa_max_replication(spa); 656 os->os_dedup_checksum = ZIO_CHECKSUM_OFF; 657 os->os_dedup_verify = B_FALSE; 658 os->os_logbias = ZFS_LOGBIAS_LATENCY; 659 os->os_sync = ZFS_SYNC_STANDARD; 660 os->os_primary_cache = ZFS_CACHE_ALL; 661 os->os_secondary_cache = ZFS_CACHE_ALL; 662 os->os_dnodesize = DNODE_MIN_SIZE; 663 os->os_prefetch = ZFS_PREFETCH_ALL; 664 } 665 666 if (ds == NULL || !ds->ds_is_snapshot) 667 os->os_zil_header = os->os_phys->os_zil_header; 668 os->os_zil = zil_alloc(os, &os->os_zil_header); 669 670 for (i = 0; i < TXG_SIZE; i++) { 671 multilist_create(&os->os_dirty_dnodes[i], sizeof (dnode_t), 672 offsetof(dnode_t, dn_dirty_link[i]), 673 dnode_multilist_index_func); 674 } 675 list_create(&os->os_dnodes, sizeof (dnode_t), 676 offsetof(dnode_t, dn_link)); 677 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), 678 offsetof(dmu_buf_impl_t, db_link)); 679 680 list_link_init(&os->os_evicting_node); 681 682 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); 683 mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); 684 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); 685 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); 686 os->os_obj_next_percpu_len = boot_ncpus; 687 os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * 688 sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); 689 690 dnode_special_open(os, &os->os_phys->os_meta_dnode, 691 DMU_META_DNODE_OBJECT, &os->os_meta_dnode); 692 if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) { 693 dnode_special_open(os, &os->os_phys->os_userused_dnode, 694 DMU_USERUSED_OBJECT, &os->os_userused_dnode); 695 dnode_special_open(os, &os->os_phys->os_groupused_dnode, 696 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); 697 if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf)) 698 dnode_special_open(os, 699 &os->os_phys->os_projectused_dnode, 700 DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode); 701 } 702 703 mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL); 704 705 *osp = os; 706 return (0); 707 } 708 709 int 710 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) 711 { 712 int err = 0; 713 714 /* 715 * We need the pool_config lock to manipulate the dsl_dataset_t. 716 * Even if the dataset is long-held, we need the pool_config lock 717 * to open the objset, as it needs to get properties. 718 */ 719 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 720 721 mutex_enter(&ds->ds_opening_lock); 722 if (ds->ds_objset == NULL) { 723 objset_t *os; 724 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 725 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), 726 ds, dsl_dataset_get_blkptr(ds), &os); 727 rrw_exit(&ds->ds_bp_rwlock, FTAG); 728 729 if (err == 0) { 730 mutex_enter(&ds->ds_lock); 731 ASSERT(ds->ds_objset == NULL); 732 ds->ds_objset = os; 733 mutex_exit(&ds->ds_lock); 734 } 735 } 736 *osp = ds->ds_objset; 737 mutex_exit(&ds->ds_opening_lock); 738 return (err); 739 } 740 741 /* 742 * Holds the pool while the objset is held. Therefore only one objset 743 * can be held at a time. 744 */ 745 int 746 dmu_objset_hold_flags(const char *name, boolean_t decrypt, const void *tag, 747 objset_t **osp) 748 { 749 dsl_pool_t *dp; 750 dsl_dataset_t *ds; 751 int err; 752 ds_hold_flags_t flags; 753 754 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 755 err = dsl_pool_hold(name, tag, &dp); 756 if (err != 0) 757 return (err); 758 err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds); 759 if (err != 0) { 760 dsl_pool_rele(dp, tag); 761 return (err); 762 } 763 764 err = dmu_objset_from_ds(ds, osp); 765 if (err != 0) { 766 dsl_dataset_rele(ds, tag); 767 dsl_pool_rele(dp, tag); 768 } 769 770 return (err); 771 } 772 773 int 774 dmu_objset_hold(const char *name, const void *tag, objset_t **osp) 775 { 776 return (dmu_objset_hold_flags(name, B_FALSE, tag, osp)); 777 } 778 779 static int 780 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, 781 boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp) 782 { 783 (void) tag; 784 785 int err = dmu_objset_from_ds(ds, osp); 786 if (err != 0) { 787 return (err); 788 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { 789 return (SET_ERROR(EINVAL)); 790 } else if (!readonly && dsl_dataset_is_snapshot(ds)) { 791 return (SET_ERROR(EROFS)); 792 } else if (!readonly && decrypt && 793 dsl_dir_incompatible_encryption_version(ds->ds_dir)) { 794 return (SET_ERROR(EROFS)); 795 } 796 797 /* if we are decrypting, we can now check MACs in os->os_phys_buf */ 798 if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) { 799 zbookmark_phys_t zb; 800 801 SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT, 802 ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 803 err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa, 804 &zb, B_FALSE); 805 if (err != 0) 806 return (err); 807 808 ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf)); 809 } 810 811 return (0); 812 } 813 814 /* 815 * dsl_pool must not be held when this is called. 816 * Upon successful return, there will be a longhold on the dataset, 817 * and the dsl_pool will not be held. 818 */ 819 int 820 dmu_objset_own(const char *name, dmu_objset_type_t type, 821 boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp) 822 { 823 dsl_pool_t *dp; 824 dsl_dataset_t *ds; 825 int err; 826 ds_hold_flags_t flags; 827 828 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 829 err = dsl_pool_hold(name, FTAG, &dp); 830 if (err != 0) 831 return (err); 832 err = dsl_dataset_own(dp, name, flags, tag, &ds); 833 if (err != 0) { 834 dsl_pool_rele(dp, FTAG); 835 return (err); 836 } 837 err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); 838 if (err != 0) { 839 dsl_dataset_disown(ds, flags, tag); 840 dsl_pool_rele(dp, FTAG); 841 return (err); 842 } 843 844 /* 845 * User accounting requires the dataset to be decrypted and rw. 846 * We also don't begin user accounting during claiming to help 847 * speed up pool import times and to keep this txg reserved 848 * completely for recovery work. 849 */ 850 if (!readonly && !dp->dp_spa->spa_claiming && 851 (ds->ds_dir->dd_crypto_obj == 0 || decrypt)) { 852 if (dmu_objset_userobjspace_upgradable(*osp) || 853 dmu_objset_projectquota_upgradable(*osp)) { 854 dmu_objset_id_quota_upgrade(*osp); 855 } else if (dmu_objset_userused_enabled(*osp)) { 856 dmu_objset_userspace_upgrade(*osp); 857 } 858 } 859 860 dsl_pool_rele(dp, FTAG); 861 return (0); 862 } 863 864 int 865 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, 866 boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp) 867 { 868 dsl_dataset_t *ds; 869 int err; 870 ds_hold_flags_t flags; 871 872 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 873 err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds); 874 if (err != 0) 875 return (err); 876 877 err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); 878 if (err != 0) { 879 dsl_dataset_disown(ds, flags, tag); 880 return (err); 881 } 882 883 return (0); 884 } 885 886 void 887 dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, const void *tag) 888 { 889 ds_hold_flags_t flags; 890 dsl_pool_t *dp = dmu_objset_pool(os); 891 892 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 893 dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag); 894 dsl_pool_rele(dp, tag); 895 } 896 897 void 898 dmu_objset_rele(objset_t *os, const void *tag) 899 { 900 dmu_objset_rele_flags(os, B_FALSE, tag); 901 } 902 903 /* 904 * When we are called, os MUST refer to an objset associated with a dataset 905 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner 906 * == tag. We will then release and reacquire ownership of the dataset while 907 * holding the pool config_rwlock to avoid intervening namespace or ownership 908 * changes may occur. 909 * 910 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to 911 * release the hold on its dataset and acquire a new one on the dataset of the 912 * same name so that it can be partially torn down and reconstructed. 913 */ 914 void 915 dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds, 916 boolean_t decrypt, const void *tag) 917 { 918 dsl_pool_t *dp; 919 char name[ZFS_MAX_DATASET_NAME_LEN]; 920 ds_hold_flags_t flags; 921 922 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 923 VERIFY3P(ds, !=, NULL); 924 VERIFY3P(ds->ds_owner, ==, tag); 925 VERIFY(dsl_dataset_long_held(ds)); 926 927 dsl_dataset_name(ds, name); 928 dp = ds->ds_dir->dd_pool; 929 dsl_pool_config_enter(dp, FTAG); 930 dsl_dataset_disown(ds, flags, tag); 931 VERIFY0(dsl_dataset_own(dp, name, flags, tag, newds)); 932 dsl_pool_config_exit(dp, FTAG); 933 } 934 935 void 936 dmu_objset_disown(objset_t *os, boolean_t decrypt, const void *tag) 937 { 938 ds_hold_flags_t flags; 939 940 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 941 /* 942 * Stop upgrading thread 943 */ 944 dmu_objset_upgrade_stop(os); 945 dsl_dataset_disown(os->os_dsl_dataset, flags, tag); 946 } 947 948 void 949 dmu_objset_evict_dbufs(objset_t *os) 950 { 951 dnode_t *dn_marker; 952 dnode_t *dn; 953 954 dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP); 955 956 mutex_enter(&os->os_lock); 957 dn = list_head(&os->os_dnodes); 958 while (dn != NULL) { 959 /* 960 * Skip dnodes without holds. We have to do this dance 961 * because dnode_add_ref() only works if there is already a 962 * hold. If the dnode has no holds, then it has no dbufs. 963 */ 964 if (dnode_add_ref(dn, FTAG)) { 965 list_insert_after(&os->os_dnodes, dn, dn_marker); 966 mutex_exit(&os->os_lock); 967 968 dnode_evict_dbufs(dn); 969 dnode_rele(dn, FTAG); 970 971 mutex_enter(&os->os_lock); 972 dn = list_next(&os->os_dnodes, dn_marker); 973 list_remove(&os->os_dnodes, dn_marker); 974 } else { 975 dn = list_next(&os->os_dnodes, dn); 976 } 977 } 978 mutex_exit(&os->os_lock); 979 980 kmem_free(dn_marker, sizeof (dnode_t)); 981 982 if (DMU_USERUSED_DNODE(os) != NULL) { 983 if (DMU_PROJECTUSED_DNODE(os) != NULL) 984 dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os)); 985 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); 986 dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); 987 } 988 dnode_evict_dbufs(DMU_META_DNODE(os)); 989 } 990 991 /* 992 * Objset eviction processing is split into into two pieces. 993 * The first marks the objset as evicting, evicts any dbufs that 994 * have a refcount of zero, and then queues up the objset for the 995 * second phase of eviction. Once os->os_dnodes has been cleared by 996 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. 997 * The second phase closes the special dnodes, dequeues the objset from 998 * the list of those undergoing eviction, and finally frees the objset. 999 * 1000 * NOTE: Due to asynchronous eviction processing (invocation of 1001 * dnode_buf_pageout()), it is possible for the meta dnode for the 1002 * objset to have no holds even though os->os_dnodes is not empty. 1003 */ 1004 void 1005 dmu_objset_evict(objset_t *os) 1006 { 1007 dsl_dataset_t *ds = os->os_dsl_dataset; 1008 1009 for (int t = 0; t < TXG_SIZE; t++) 1010 ASSERT(!dmu_objset_is_dirty(os, t)); 1011 1012 if (ds) 1013 dsl_prop_unregister_all(ds, os); 1014 1015 if (os->os_sa) 1016 sa_tear_down(os); 1017 1018 dmu_objset_evict_dbufs(os); 1019 1020 mutex_enter(&os->os_lock); 1021 spa_evicting_os_register(os->os_spa, os); 1022 if (list_is_empty(&os->os_dnodes)) { 1023 mutex_exit(&os->os_lock); 1024 dmu_objset_evict_done(os); 1025 } else { 1026 mutex_exit(&os->os_lock); 1027 } 1028 1029 1030 } 1031 1032 void 1033 dmu_objset_evict_done(objset_t *os) 1034 { 1035 ASSERT3P(list_head(&os->os_dnodes), ==, NULL); 1036 1037 dnode_special_close(&os->os_meta_dnode); 1038 if (DMU_USERUSED_DNODE(os)) { 1039 if (DMU_PROJECTUSED_DNODE(os)) 1040 dnode_special_close(&os->os_projectused_dnode); 1041 dnode_special_close(&os->os_userused_dnode); 1042 dnode_special_close(&os->os_groupused_dnode); 1043 } 1044 zil_free(os->os_zil); 1045 1046 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 1047 1048 /* 1049 * This is a barrier to prevent the objset from going away in 1050 * dnode_move() until we can safely ensure that the objset is still in 1051 * use. We consider the objset valid before the barrier and invalid 1052 * after the barrier. 1053 */ 1054 rw_enter(&os_lock, RW_READER); 1055 rw_exit(&os_lock); 1056 1057 kmem_free(os->os_obj_next_percpu, 1058 os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); 1059 1060 mutex_destroy(&os->os_lock); 1061 mutex_destroy(&os->os_userused_lock); 1062 mutex_destroy(&os->os_obj_lock); 1063 mutex_destroy(&os->os_user_ptr_lock); 1064 mutex_destroy(&os->os_upgrade_lock); 1065 for (int i = 0; i < TXG_SIZE; i++) 1066 multilist_destroy(&os->os_dirty_dnodes[i]); 1067 spa_evicting_os_deregister(os->os_spa, os); 1068 kmem_free(os, sizeof (objset_t)); 1069 } 1070 1071 inode_timespec_t 1072 dmu_objset_snap_cmtime(objset_t *os) 1073 { 1074 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); 1075 } 1076 1077 objset_t * 1078 dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 1079 dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx) 1080 { 1081 objset_t *os; 1082 dnode_t *mdn; 1083 1084 ASSERT(dmu_tx_is_syncing(tx)); 1085 1086 if (blksz == 0) 1087 blksz = DNODE_BLOCK_SIZE; 1088 if (ibs == 0) 1089 ibs = DN_MAX_INDBLKSHIFT; 1090 1091 if (ds != NULL) 1092 VERIFY0(dmu_objset_from_ds(ds, &os)); 1093 else 1094 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); 1095 1096 mdn = DMU_META_DNODE(os); 1097 1098 dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0, 1099 DNODE_MIN_SLOTS, tx); 1100 1101 /* 1102 * We don't want to have to increase the meta-dnode's nlevels 1103 * later, because then we could do it in quiescing context while 1104 * we are also accessing it in open context. 1105 * 1106 * This precaution is not necessary for the MOS (ds == NULL), 1107 * because the MOS is only updated in syncing context. 1108 * This is most fortunate: the MOS is the only objset that 1109 * needs to be synced multiple times as spa_sync() iterates 1110 * to convergence, so minimizing its dn_nlevels matters. 1111 */ 1112 if (ds != NULL) { 1113 if (levels == 0) { 1114 levels = 1; 1115 1116 /* 1117 * Determine the number of levels necessary for the 1118 * meta-dnode to contain DN_MAX_OBJECT dnodes. Note 1119 * that in order to ensure that we do not overflow 1120 * 64 bits, there has to be a nlevels that gives us a 1121 * number of blocks > DN_MAX_OBJECT but < 2^64. 1122 * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) 1123 * (10) must be less than (64 - log2(DN_MAX_OBJECT)) 1124 * (16). 1125 */ 1126 while ((uint64_t)mdn->dn_nblkptr << 1127 (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) * 1128 (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < 1129 DN_MAX_OBJECT) 1130 levels++; 1131 } 1132 1133 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = 1134 mdn->dn_nlevels = levels; 1135 } 1136 1137 ASSERT(type != DMU_OST_NONE); 1138 ASSERT(type != DMU_OST_ANY); 1139 ASSERT(type < DMU_OST_NUMTYPES); 1140 os->os_phys->os_type = type; 1141 1142 /* 1143 * Enable user accounting if it is enabled and this is not an 1144 * encrypted receive. 1145 */ 1146 if (dmu_objset_userused_enabled(os) && 1147 (!os->os_encrypted || !dmu_objset_is_receiving(os))) { 1148 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 1149 if (dmu_objset_userobjused_enabled(os)) { 1150 ASSERT3P(ds, !=, NULL); 1151 ds->ds_feature_activation[ 1152 SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; 1153 os->os_phys->os_flags |= 1154 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; 1155 } 1156 if (dmu_objset_projectquota_enabled(os)) { 1157 ASSERT3P(ds, !=, NULL); 1158 ds->ds_feature_activation[ 1159 SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; 1160 os->os_phys->os_flags |= 1161 OBJSET_FLAG_PROJECTQUOTA_COMPLETE; 1162 } 1163 os->os_flags = os->os_phys->os_flags; 1164 } 1165 1166 dsl_dataset_dirty(ds, tx); 1167 1168 return (os); 1169 } 1170 1171 /* called from dsl for meta-objset */ 1172 objset_t * 1173 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 1174 dmu_objset_type_t type, dmu_tx_t *tx) 1175 { 1176 return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx)); 1177 } 1178 1179 typedef struct dmu_objset_create_arg { 1180 const char *doca_name; 1181 cred_t *doca_cred; 1182 proc_t *doca_proc; 1183 void (*doca_userfunc)(objset_t *os, void *arg, 1184 cred_t *cr, dmu_tx_t *tx); 1185 void *doca_userarg; 1186 dmu_objset_type_t doca_type; 1187 uint64_t doca_flags; 1188 dsl_crypto_params_t *doca_dcp; 1189 } dmu_objset_create_arg_t; 1190 1191 static int 1192 dmu_objset_create_check(void *arg, dmu_tx_t *tx) 1193 { 1194 dmu_objset_create_arg_t *doca = arg; 1195 dsl_pool_t *dp = dmu_tx_pool(tx); 1196 dsl_dir_t *pdd; 1197 dsl_dataset_t *parentds; 1198 objset_t *parentos; 1199 const char *tail; 1200 int error; 1201 1202 if (strchr(doca->doca_name, '@') != NULL) 1203 return (SET_ERROR(EINVAL)); 1204 1205 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) 1206 return (SET_ERROR(ENAMETOOLONG)); 1207 1208 if (dataset_nestcheck(doca->doca_name) != 0) 1209 return (SET_ERROR(ENAMETOOLONG)); 1210 1211 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); 1212 if (error != 0) 1213 return (error); 1214 if (tail == NULL) { 1215 dsl_dir_rele(pdd, FTAG); 1216 return (SET_ERROR(EEXIST)); 1217 } 1218 1219 error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp, NULL); 1220 if (error != 0) { 1221 dsl_dir_rele(pdd, FTAG); 1222 return (error); 1223 } 1224 1225 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 1226 doca->doca_cred, doca->doca_proc); 1227 if (error != 0) { 1228 dsl_dir_rele(pdd, FTAG); 1229 return (error); 1230 } 1231 1232 /* can't create below anything but filesystems (eg. no ZVOLs) */ 1233 error = dsl_dataset_hold_obj(pdd->dd_pool, 1234 dsl_dir_phys(pdd)->dd_head_dataset_obj, FTAG, &parentds); 1235 if (error != 0) { 1236 dsl_dir_rele(pdd, FTAG); 1237 return (error); 1238 } 1239 error = dmu_objset_from_ds(parentds, &parentos); 1240 if (error != 0) { 1241 dsl_dataset_rele(parentds, FTAG); 1242 dsl_dir_rele(pdd, FTAG); 1243 return (error); 1244 } 1245 if (dmu_objset_type(parentos) != DMU_OST_ZFS) { 1246 dsl_dataset_rele(parentds, FTAG); 1247 dsl_dir_rele(pdd, FTAG); 1248 return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); 1249 } 1250 dsl_dataset_rele(parentds, FTAG); 1251 dsl_dir_rele(pdd, FTAG); 1252 1253 return (error); 1254 } 1255 1256 static void 1257 dmu_objset_create_sync(void *arg, dmu_tx_t *tx) 1258 { 1259 dmu_objset_create_arg_t *doca = arg; 1260 dsl_pool_t *dp = dmu_tx_pool(tx); 1261 spa_t *spa = dp->dp_spa; 1262 dsl_dir_t *pdd; 1263 const char *tail; 1264 dsl_dataset_t *ds; 1265 uint64_t obj; 1266 blkptr_t *bp; 1267 objset_t *os; 1268 zio_t *rzio; 1269 1270 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); 1271 1272 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, 1273 doca->doca_cred, doca->doca_dcp, tx); 1274 1275 VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj, 1276 DS_HOLD_FLAG_DECRYPT, FTAG, &ds)); 1277 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1278 bp = dsl_dataset_get_blkptr(ds); 1279 os = dmu_objset_create_impl(spa, ds, bp, doca->doca_type, tx); 1280 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1281 1282 if (doca->doca_userfunc != NULL) { 1283 doca->doca_userfunc(os, doca->doca_userarg, 1284 doca->doca_cred, tx); 1285 } 1286 1287 /* 1288 * The doca_userfunc() may write out some data that needs to be 1289 * encrypted if the dataset is encrypted (specifically the root 1290 * directory). This data must be written out before the encryption 1291 * key mapping is removed by dsl_dataset_rele_flags(). Force the 1292 * I/O to occur immediately by invoking the relevant sections of 1293 * dsl_pool_sync(). 1294 */ 1295 if (os->os_encrypted) { 1296 dsl_dataset_t *tmpds = NULL; 1297 boolean_t need_sync_done = B_FALSE; 1298 1299 mutex_enter(&ds->ds_lock); 1300 ds->ds_owner = FTAG; 1301 mutex_exit(&ds->ds_lock); 1302 1303 rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1304 tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, 1305 tx->tx_txg); 1306 if (tmpds != NULL) { 1307 dsl_dataset_sync(ds, rzio, tx); 1308 need_sync_done = B_TRUE; 1309 } 1310 VERIFY0(zio_wait(rzio)); 1311 1312 dmu_objset_sync_done(os, tx); 1313 taskq_wait(dp->dp_sync_taskq); 1314 if (txg_list_member(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1315 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1316 key_mapping_rele(spa, ds->ds_key_mapping, ds); 1317 } 1318 1319 rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1320 tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, 1321 tx->tx_txg); 1322 if (tmpds != NULL) { 1323 dmu_buf_rele(ds->ds_dbuf, ds); 1324 dsl_dataset_sync(ds, rzio, tx); 1325 } 1326 VERIFY0(zio_wait(rzio)); 1327 1328 if (need_sync_done) { 1329 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1330 key_mapping_rele(spa, ds->ds_key_mapping, ds); 1331 dsl_dataset_sync_done(ds, tx); 1332 dmu_buf_rele(ds->ds_dbuf, ds); 1333 } 1334 1335 mutex_enter(&ds->ds_lock); 1336 ds->ds_owner = NULL; 1337 mutex_exit(&ds->ds_lock); 1338 } 1339 1340 spa_history_log_internal_ds(ds, "create", tx, " "); 1341 1342 dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG); 1343 dsl_dir_rele(pdd, FTAG); 1344 } 1345 1346 int 1347 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 1348 dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg) 1349 { 1350 dmu_objset_create_arg_t doca; 1351 dsl_crypto_params_t tmp_dcp = { 0 }; 1352 1353 doca.doca_name = name; 1354 doca.doca_cred = CRED(); 1355 doca.doca_proc = curproc; 1356 doca.doca_flags = flags; 1357 doca.doca_userfunc = func; 1358 doca.doca_userarg = arg; 1359 doca.doca_type = type; 1360 1361 /* 1362 * Some callers (mostly for testing) do not provide a dcp on their 1363 * own but various code inside the sync task will require it to be 1364 * allocated. Rather than adding NULL checks throughout this code 1365 * or adding dummy dcp's to all of the callers we simply create a 1366 * dummy one here and use that. This zero dcp will have the same 1367 * effect as asking for inheritance of all encryption params. 1368 */ 1369 doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp; 1370 1371 int rv = dsl_sync_task(name, 1372 dmu_objset_create_check, dmu_objset_create_sync, &doca, 1373 6, ZFS_SPACE_CHECK_NORMAL); 1374 1375 if (rv == 0) 1376 zvol_create_minor(name); 1377 return (rv); 1378 } 1379 1380 typedef struct dmu_objset_clone_arg { 1381 const char *doca_clone; 1382 const char *doca_origin; 1383 cred_t *doca_cred; 1384 proc_t *doca_proc; 1385 } dmu_objset_clone_arg_t; 1386 1387 static int 1388 dmu_objset_clone_check(void *arg, dmu_tx_t *tx) 1389 { 1390 dmu_objset_clone_arg_t *doca = arg; 1391 dsl_dir_t *pdd; 1392 const char *tail; 1393 int error; 1394 dsl_dataset_t *origin; 1395 dsl_pool_t *dp = dmu_tx_pool(tx); 1396 1397 if (strchr(doca->doca_clone, '@') != NULL) 1398 return (SET_ERROR(EINVAL)); 1399 1400 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) 1401 return (SET_ERROR(ENAMETOOLONG)); 1402 1403 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); 1404 if (error != 0) 1405 return (error); 1406 if (tail == NULL) { 1407 dsl_dir_rele(pdd, FTAG); 1408 return (SET_ERROR(EEXIST)); 1409 } 1410 1411 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 1412 doca->doca_cred, doca->doca_proc); 1413 if (error != 0) { 1414 dsl_dir_rele(pdd, FTAG); 1415 return (SET_ERROR(EDQUOT)); 1416 } 1417 1418 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); 1419 if (error != 0) { 1420 dsl_dir_rele(pdd, FTAG); 1421 return (error); 1422 } 1423 1424 /* You can only clone snapshots, not the head datasets. */ 1425 if (!origin->ds_is_snapshot) { 1426 dsl_dataset_rele(origin, FTAG); 1427 dsl_dir_rele(pdd, FTAG); 1428 return (SET_ERROR(EINVAL)); 1429 } 1430 1431 dsl_dataset_rele(origin, FTAG); 1432 dsl_dir_rele(pdd, FTAG); 1433 1434 return (0); 1435 } 1436 1437 static void 1438 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) 1439 { 1440 dmu_objset_clone_arg_t *doca = arg; 1441 dsl_pool_t *dp = dmu_tx_pool(tx); 1442 dsl_dir_t *pdd; 1443 const char *tail; 1444 dsl_dataset_t *origin, *ds; 1445 uint64_t obj; 1446 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 1447 1448 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); 1449 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); 1450 1451 obj = dsl_dataset_create_sync(pdd, tail, origin, 0, 1452 doca->doca_cred, NULL, tx); 1453 1454 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 1455 dsl_dataset_name(origin, namebuf); 1456 spa_history_log_internal_ds(ds, "clone", tx, 1457 "origin=%s (%llu)", namebuf, (u_longlong_t)origin->ds_object); 1458 dsl_dataset_rele(ds, FTAG); 1459 dsl_dataset_rele(origin, FTAG); 1460 dsl_dir_rele(pdd, FTAG); 1461 } 1462 1463 int 1464 dmu_objset_clone(const char *clone, const char *origin) 1465 { 1466 dmu_objset_clone_arg_t doca; 1467 1468 doca.doca_clone = clone; 1469 doca.doca_origin = origin; 1470 doca.doca_cred = CRED(); 1471 doca.doca_proc = curproc; 1472 1473 int rv = dsl_sync_task(clone, 1474 dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 1475 6, ZFS_SPACE_CHECK_NORMAL); 1476 1477 if (rv == 0) 1478 zvol_create_minor(clone); 1479 1480 return (rv); 1481 } 1482 1483 int 1484 dmu_objset_snapshot_one(const char *fsname, const char *snapname) 1485 { 1486 int err; 1487 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); 1488 nvlist_t *snaps = fnvlist_alloc(); 1489 1490 fnvlist_add_boolean(snaps, longsnap); 1491 kmem_strfree(longsnap); 1492 err = dsl_dataset_snapshot(snaps, NULL, NULL); 1493 fnvlist_free(snaps); 1494 return (err); 1495 } 1496 1497 static void 1498 dmu_objset_upgrade_task_cb(void *data) 1499 { 1500 objset_t *os = data; 1501 1502 mutex_enter(&os->os_upgrade_lock); 1503 os->os_upgrade_status = EINTR; 1504 if (!os->os_upgrade_exit) { 1505 int status; 1506 1507 mutex_exit(&os->os_upgrade_lock); 1508 1509 status = os->os_upgrade_cb(os); 1510 1511 mutex_enter(&os->os_upgrade_lock); 1512 1513 os->os_upgrade_status = status; 1514 } 1515 os->os_upgrade_exit = B_TRUE; 1516 os->os_upgrade_id = 0; 1517 mutex_exit(&os->os_upgrade_lock); 1518 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1519 } 1520 1521 static void 1522 dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb) 1523 { 1524 if (os->os_upgrade_id != 0) 1525 return; 1526 1527 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 1528 dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag); 1529 1530 mutex_enter(&os->os_upgrade_lock); 1531 if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) { 1532 os->os_upgrade_exit = B_FALSE; 1533 os->os_upgrade_cb = cb; 1534 os->os_upgrade_id = taskq_dispatch( 1535 os->os_spa->spa_upgrade_taskq, 1536 dmu_objset_upgrade_task_cb, os, TQ_SLEEP); 1537 if (os->os_upgrade_id == TASKQID_INVALID) { 1538 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1539 os->os_upgrade_status = ENOMEM; 1540 } 1541 } else { 1542 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1543 } 1544 mutex_exit(&os->os_upgrade_lock); 1545 } 1546 1547 static void 1548 dmu_objset_upgrade_stop(objset_t *os) 1549 { 1550 mutex_enter(&os->os_upgrade_lock); 1551 os->os_upgrade_exit = B_TRUE; 1552 if (os->os_upgrade_id != 0) { 1553 taskqid_t id = os->os_upgrade_id; 1554 1555 os->os_upgrade_id = 0; 1556 mutex_exit(&os->os_upgrade_lock); 1557 1558 if ((taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id)) == 0) { 1559 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1560 } 1561 txg_wait_synced(os->os_spa->spa_dsl_pool, 0); 1562 } else { 1563 mutex_exit(&os->os_upgrade_lock); 1564 } 1565 } 1566 1567 static void 1568 dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) 1569 { 1570 dnode_t *dn; 1571 1572 while ((dn = multilist_sublist_head(list)) != NULL) { 1573 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 1574 ASSERT(dn->dn_dbuf->db_data_pending); 1575 /* 1576 * Initialize dn_zio outside dnode_sync() because the 1577 * meta-dnode needs to set it outside dnode_sync(). 1578 */ 1579 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; 1580 ASSERT(dn->dn_zio); 1581 1582 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); 1583 multilist_sublist_remove(list, dn); 1584 1585 /* 1586 * See the comment above dnode_rele_task() for an explanation 1587 * of why this dnode hold is always needed (even when not 1588 * doing user accounting). 1589 */ 1590 multilist_t *newlist = &dn->dn_objset->os_synced_dnodes; 1591 (void) dnode_add_ref(dn, newlist); 1592 multilist_insert(newlist, dn); 1593 1594 dnode_sync(dn, tx); 1595 } 1596 } 1597 1598 static void 1599 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) 1600 { 1601 (void) abuf; 1602 blkptr_t *bp = zio->io_bp; 1603 objset_t *os = arg; 1604 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; 1605 uint64_t fill = 0; 1606 1607 ASSERT(!BP_IS_EMBEDDED(bp)); 1608 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); 1609 ASSERT0(BP_GET_LEVEL(bp)); 1610 1611 /* 1612 * Update rootbp fill count: it should be the number of objects 1613 * allocated in the object set (not counting the "special" 1614 * objects that are stored in the objset_phys_t -- the meta 1615 * dnode and user/group/project accounting objects). 1616 */ 1617 for (int i = 0; i < dnp->dn_nblkptr; i++) 1618 fill += BP_GET_FILL(&dnp->dn_blkptr[i]); 1619 1620 BP_SET_FILL(bp, fill); 1621 1622 if (os->os_dsl_dataset != NULL) 1623 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); 1624 *os->os_rootbp = *bp; 1625 if (os->os_dsl_dataset != NULL) 1626 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); 1627 } 1628 1629 static void 1630 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) 1631 { 1632 (void) abuf; 1633 blkptr_t *bp = zio->io_bp; 1634 blkptr_t *bp_orig = &zio->io_bp_orig; 1635 objset_t *os = arg; 1636 1637 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { 1638 ASSERT(BP_EQUAL(bp, bp_orig)); 1639 } else { 1640 dsl_dataset_t *ds = os->os_dsl_dataset; 1641 dmu_tx_t *tx = os->os_synctx; 1642 1643 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); 1644 dsl_dataset_block_born(ds, bp, tx); 1645 } 1646 kmem_free(bp, sizeof (*bp)); 1647 } 1648 1649 typedef struct sync_objset_arg { 1650 zio_t *soa_zio; 1651 objset_t *soa_os; 1652 dmu_tx_t *soa_tx; 1653 kmutex_t soa_mutex; 1654 int soa_count; 1655 taskq_ent_t soa_tq_ent; 1656 } sync_objset_arg_t; 1657 1658 typedef struct sync_dnodes_arg { 1659 multilist_t *sda_list; 1660 int sda_sublist_idx; 1661 multilist_t *sda_newlist; 1662 sync_objset_arg_t *sda_soa; 1663 } sync_dnodes_arg_t; 1664 1665 static void sync_meta_dnode_task(void *arg); 1666 1667 static void 1668 sync_dnodes_task(void *arg) 1669 { 1670 sync_dnodes_arg_t *sda = arg; 1671 sync_objset_arg_t *soa = sda->sda_soa; 1672 objset_t *os = soa->soa_os; 1673 1674 uint_t allocator = spa_acq_allocator(os->os_spa); 1675 multilist_sublist_t *ms = 1676 multilist_sublist_lock_idx(sda->sda_list, sda->sda_sublist_idx); 1677 1678 dmu_objset_sync_dnodes(ms, soa->soa_tx); 1679 1680 multilist_sublist_unlock(ms); 1681 spa_rel_allocator(os->os_spa, allocator); 1682 1683 kmem_free(sda, sizeof (*sda)); 1684 1685 mutex_enter(&soa->soa_mutex); 1686 ASSERT(soa->soa_count != 0); 1687 if (--soa->soa_count != 0) { 1688 mutex_exit(&soa->soa_mutex); 1689 return; 1690 } 1691 mutex_exit(&soa->soa_mutex); 1692 1693 taskq_dispatch_ent(dmu_objset_pool(os)->dp_sync_taskq, 1694 sync_meta_dnode_task, soa, TQ_FRONT, &soa->soa_tq_ent); 1695 } 1696 1697 /* 1698 * Issue the zio_nowait() for all dirty record zios on the meta dnode, 1699 * then trigger the callback for the zil_sync. This runs once for each 1700 * objset, only after any/all sublists in the objset have been synced. 1701 */ 1702 static void 1703 sync_meta_dnode_task(void *arg) 1704 { 1705 sync_objset_arg_t *soa = arg; 1706 objset_t *os = soa->soa_os; 1707 dmu_tx_t *tx = soa->soa_tx; 1708 int txgoff = tx->tx_txg & TXG_MASK; 1709 dbuf_dirty_record_t *dr; 1710 1711 ASSERT0(soa->soa_count); 1712 1713 list_t *list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; 1714 while ((dr = list_remove_head(list)) != NULL) { 1715 ASSERT0(dr->dr_dbuf->db_level); 1716 zio_nowait(dr->dr_zio); 1717 } 1718 1719 /* Enable dnode backfill if enough objects have been freed. */ 1720 if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { 1721 os->os_rescan_dnodes = B_TRUE; 1722 os->os_freed_dnodes = 0; 1723 } 1724 1725 /* 1726 * Free intent log blocks up to this tx. 1727 */ 1728 zil_sync(os->os_zil, tx); 1729 os->os_phys->os_zil_header = os->os_zil_header; 1730 zio_nowait(soa->soa_zio); 1731 1732 mutex_destroy(&soa->soa_mutex); 1733 kmem_free(soa, sizeof (*soa)); 1734 } 1735 1736 /* called from dsl */ 1737 void 1738 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) 1739 { 1740 int txgoff; 1741 zbookmark_phys_t zb; 1742 zio_prop_t zp; 1743 zio_t *zio; 1744 int num_sublists; 1745 multilist_t *ml; 1746 blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); 1747 *blkptr_copy = *os->os_rootbp; 1748 1749 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", (u_longlong_t)tx->tx_txg); 1750 1751 ASSERT(dmu_tx_is_syncing(tx)); 1752 /* XXX the write_done callback should really give us the tx... */ 1753 os->os_synctx = tx; 1754 1755 if (os->os_dsl_dataset == NULL) { 1756 /* 1757 * This is the MOS. If we have upgraded, 1758 * spa_max_replication() could change, so reset 1759 * os_copies here. 1760 */ 1761 os->os_copies = spa_max_replication(os->os_spa); 1762 } 1763 1764 /* 1765 * Create the root block IO 1766 */ 1767 SET_BOOKMARK(&zb, os->os_dsl_dataset ? 1768 os->os_dsl_dataset->ds_object : DMU_META_OBJSET, 1769 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 1770 arc_release(os->os_phys_buf, &os->os_phys_buf); 1771 1772 dmu_write_policy(os, NULL, 0, 0, &zp); 1773 1774 /* 1775 * If we are either claiming the ZIL or doing a raw receive, write 1776 * out the os_phys_buf raw. Neither of these actions will effect the 1777 * MAC at this point. 1778 */ 1779 if (os->os_raw_receive || 1780 os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { 1781 ASSERT(os->os_encrypted); 1782 arc_convert_to_raw(os->os_phys_buf, 1783 os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER, 1784 DMU_OT_OBJSET, NULL, NULL, NULL); 1785 } 1786 1787 zio = arc_write(pio, os->os_spa, tx->tx_txg, 1788 blkptr_copy, os->os_phys_buf, B_FALSE, dmu_os_is_l2cacheable(os), 1789 &zp, dmu_objset_write_ready, NULL, dmu_objset_write_done, 1790 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); 1791 1792 /* 1793 * Sync special dnodes - the parent IO for the sync is the root block 1794 */ 1795 DMU_META_DNODE(os)->dn_zio = zio; 1796 dnode_sync(DMU_META_DNODE(os), tx); 1797 1798 os->os_phys->os_flags = os->os_flags; 1799 1800 if (DMU_USERUSED_DNODE(os) && 1801 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1802 DMU_USERUSED_DNODE(os)->dn_zio = zio; 1803 dnode_sync(DMU_USERUSED_DNODE(os), tx); 1804 DMU_GROUPUSED_DNODE(os)->dn_zio = zio; 1805 dnode_sync(DMU_GROUPUSED_DNODE(os), tx); 1806 } 1807 1808 if (DMU_PROJECTUSED_DNODE(os) && 1809 DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1810 DMU_PROJECTUSED_DNODE(os)->dn_zio = zio; 1811 dnode_sync(DMU_PROJECTUSED_DNODE(os), tx); 1812 } 1813 1814 txgoff = tx->tx_txg & TXG_MASK; 1815 1816 /* 1817 * We must create the list here because it uses the 1818 * dn_dirty_link[] of this txg. But it may already 1819 * exist because we call dsl_dataset_sync() twice per txg. 1820 */ 1821 if (os->os_synced_dnodes.ml_sublists == NULL) { 1822 multilist_create(&os->os_synced_dnodes, sizeof (dnode_t), 1823 offsetof(dnode_t, dn_dirty_link[txgoff]), 1824 dnode_multilist_index_func); 1825 } else { 1826 ASSERT3U(os->os_synced_dnodes.ml_offset, ==, 1827 offsetof(dnode_t, dn_dirty_link[txgoff])); 1828 } 1829 1830 /* 1831 * zio_nowait(zio) is done after any/all sublist and meta dnode 1832 * zios have been nowaited, and the zil_sync() has been performed. 1833 * The soa is freed at the end of sync_meta_dnode_task. 1834 */ 1835 sync_objset_arg_t *soa = kmem_alloc(sizeof (*soa), KM_SLEEP); 1836 soa->soa_zio = zio; 1837 soa->soa_os = os; 1838 soa->soa_tx = tx; 1839 taskq_init_ent(&soa->soa_tq_ent); 1840 mutex_init(&soa->soa_mutex, NULL, MUTEX_DEFAULT, NULL); 1841 1842 ml = &os->os_dirty_dnodes[txgoff]; 1843 soa->soa_count = num_sublists = multilist_get_num_sublists(ml); 1844 1845 for (int i = 0; i < num_sublists; i++) { 1846 if (multilist_sublist_is_empty_idx(ml, i)) 1847 soa->soa_count--; 1848 } 1849 1850 if (soa->soa_count == 0) { 1851 taskq_dispatch_ent(dmu_objset_pool(os)->dp_sync_taskq, 1852 sync_meta_dnode_task, soa, TQ_FRONT, &soa->soa_tq_ent); 1853 } else { 1854 /* 1855 * Sync sublists in parallel. The last to finish 1856 * (i.e., when soa->soa_count reaches zero) must 1857 * dispatch sync_meta_dnode_task. 1858 */ 1859 for (int i = 0; i < num_sublists; i++) { 1860 if (multilist_sublist_is_empty_idx(ml, i)) 1861 continue; 1862 sync_dnodes_arg_t *sda = 1863 kmem_alloc(sizeof (*sda), KM_SLEEP); 1864 sda->sda_list = ml; 1865 sda->sda_sublist_idx = i; 1866 sda->sda_soa = soa; 1867 (void) taskq_dispatch( 1868 dmu_objset_pool(os)->dp_sync_taskq, 1869 sync_dnodes_task, sda, 0); 1870 /* sync_dnodes_task frees sda */ 1871 } 1872 } 1873 } 1874 1875 boolean_t 1876 dmu_objset_is_dirty(objset_t *os, uint64_t txg) 1877 { 1878 return (!multilist_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK])); 1879 } 1880 1881 static file_info_cb_t *file_cbs[DMU_OST_NUMTYPES]; 1882 1883 void 1884 dmu_objset_register_type(dmu_objset_type_t ost, file_info_cb_t *cb) 1885 { 1886 file_cbs[ost] = cb; 1887 } 1888 1889 int 1890 dmu_get_file_info(objset_t *os, dmu_object_type_t bonustype, const void *data, 1891 zfs_file_info_t *zfi) 1892 { 1893 file_info_cb_t *cb = file_cbs[os->os_phys->os_type]; 1894 if (cb == NULL) 1895 return (EINVAL); 1896 return (cb(bonustype, data, zfi)); 1897 } 1898 1899 boolean_t 1900 dmu_objset_userused_enabled(objset_t *os) 1901 { 1902 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && 1903 file_cbs[os->os_phys->os_type] != NULL && 1904 DMU_USERUSED_DNODE(os) != NULL); 1905 } 1906 1907 boolean_t 1908 dmu_objset_userobjused_enabled(objset_t *os) 1909 { 1910 return (dmu_objset_userused_enabled(os) && 1911 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING)); 1912 } 1913 1914 boolean_t 1915 dmu_objset_projectquota_enabled(objset_t *os) 1916 { 1917 return (file_cbs[os->os_phys->os_type] != NULL && 1918 DMU_PROJECTUSED_DNODE(os) != NULL && 1919 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA)); 1920 } 1921 1922 typedef struct userquota_node { 1923 /* must be in the first filed, see userquota_update_cache() */ 1924 char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN]; 1925 int64_t uqn_delta; 1926 avl_node_t uqn_node; 1927 } userquota_node_t; 1928 1929 typedef struct userquota_cache { 1930 avl_tree_t uqc_user_deltas; 1931 avl_tree_t uqc_group_deltas; 1932 avl_tree_t uqc_project_deltas; 1933 } userquota_cache_t; 1934 1935 static int 1936 userquota_compare(const void *l, const void *r) 1937 { 1938 const userquota_node_t *luqn = l; 1939 const userquota_node_t *ruqn = r; 1940 int rv; 1941 1942 /* 1943 * NB: can only access uqn_id because userquota_update_cache() doesn't 1944 * pass in an entire userquota_node_t. 1945 */ 1946 rv = strcmp(luqn->uqn_id, ruqn->uqn_id); 1947 1948 return (TREE_ISIGN(rv)); 1949 } 1950 1951 static void 1952 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) 1953 { 1954 void *cookie; 1955 userquota_node_t *uqn; 1956 1957 ASSERT(dmu_tx_is_syncing(tx)); 1958 1959 cookie = NULL; 1960 while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, 1961 &cookie)) != NULL) { 1962 /* 1963 * os_userused_lock protects against concurrent calls to 1964 * zap_increment_int(). It's needed because zap_increment_int() 1965 * is not thread-safe (i.e. not atomic). 1966 */ 1967 mutex_enter(&os->os_userused_lock); 1968 VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT, 1969 uqn->uqn_id, uqn->uqn_delta, tx)); 1970 mutex_exit(&os->os_userused_lock); 1971 kmem_free(uqn, sizeof (*uqn)); 1972 } 1973 avl_destroy(&cache->uqc_user_deltas); 1974 1975 cookie = NULL; 1976 while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, 1977 &cookie)) != NULL) { 1978 mutex_enter(&os->os_userused_lock); 1979 VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT, 1980 uqn->uqn_id, uqn->uqn_delta, tx)); 1981 mutex_exit(&os->os_userused_lock); 1982 kmem_free(uqn, sizeof (*uqn)); 1983 } 1984 avl_destroy(&cache->uqc_group_deltas); 1985 1986 if (dmu_objset_projectquota_enabled(os)) { 1987 cookie = NULL; 1988 while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas, 1989 &cookie)) != NULL) { 1990 mutex_enter(&os->os_userused_lock); 1991 VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT, 1992 uqn->uqn_id, uqn->uqn_delta, tx)); 1993 mutex_exit(&os->os_userused_lock); 1994 kmem_free(uqn, sizeof (*uqn)); 1995 } 1996 avl_destroy(&cache->uqc_project_deltas); 1997 } 1998 } 1999 2000 static void 2001 userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta) 2002 { 2003 userquota_node_t *uqn; 2004 avl_index_t idx; 2005 2006 ASSERT(strlen(id) < sizeof (uqn->uqn_id)); 2007 /* 2008 * Use id directly for searching because uqn_id is the first field of 2009 * userquota_node_t and fields after uqn_id won't be accessed in 2010 * avl_find(). 2011 */ 2012 uqn = avl_find(avl, (const void *)id, &idx); 2013 if (uqn == NULL) { 2014 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); 2015 strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id)); 2016 avl_insert(avl, uqn, idx); 2017 } 2018 uqn->uqn_delta += delta; 2019 } 2020 2021 static void 2022 do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used, 2023 uint64_t flags, uint64_t user, uint64_t group, uint64_t project, 2024 boolean_t subtract) 2025 { 2026 if (flags & DNODE_FLAG_USERUSED_ACCOUNTED) { 2027 int64_t delta = DNODE_MIN_SIZE + used; 2028 char name[20]; 2029 2030 if (subtract) 2031 delta = -delta; 2032 2033 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)user); 2034 userquota_update_cache(&cache->uqc_user_deltas, name, delta); 2035 2036 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)group); 2037 userquota_update_cache(&cache->uqc_group_deltas, name, delta); 2038 2039 if (dmu_objset_projectquota_enabled(os)) { 2040 (void) snprintf(name, sizeof (name), "%llx", 2041 (longlong_t)project); 2042 userquota_update_cache(&cache->uqc_project_deltas, 2043 name, delta); 2044 } 2045 } 2046 } 2047 2048 static void 2049 do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags, 2050 uint64_t user, uint64_t group, uint64_t project, boolean_t subtract) 2051 { 2052 if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) { 2053 char name[20 + DMU_OBJACCT_PREFIX_LEN]; 2054 int delta = subtract ? -1 : 1; 2055 2056 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", 2057 (longlong_t)user); 2058 userquota_update_cache(&cache->uqc_user_deltas, name, delta); 2059 2060 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", 2061 (longlong_t)group); 2062 userquota_update_cache(&cache->uqc_group_deltas, name, delta); 2063 2064 if (dmu_objset_projectquota_enabled(os)) { 2065 (void) snprintf(name, sizeof (name), 2066 DMU_OBJACCT_PREFIX "%llx", (longlong_t)project); 2067 userquota_update_cache(&cache->uqc_project_deltas, 2068 name, delta); 2069 } 2070 } 2071 } 2072 2073 typedef struct userquota_updates_arg { 2074 objset_t *uua_os; 2075 int uua_sublist_idx; 2076 dmu_tx_t *uua_tx; 2077 } userquota_updates_arg_t; 2078 2079 static void 2080 userquota_updates_task(void *arg) 2081 { 2082 userquota_updates_arg_t *uua = arg; 2083 objset_t *os = uua->uua_os; 2084 dmu_tx_t *tx = uua->uua_tx; 2085 dnode_t *dn; 2086 userquota_cache_t cache = { { 0 } }; 2087 2088 multilist_sublist_t *list = multilist_sublist_lock_idx( 2089 &os->os_synced_dnodes, uua->uua_sublist_idx); 2090 2091 ASSERT(multilist_sublist_head(list) == NULL || 2092 dmu_objset_userused_enabled(os)); 2093 avl_create(&cache.uqc_user_deltas, userquota_compare, 2094 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 2095 avl_create(&cache.uqc_group_deltas, userquota_compare, 2096 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 2097 if (dmu_objset_projectquota_enabled(os)) 2098 avl_create(&cache.uqc_project_deltas, userquota_compare, 2099 sizeof (userquota_node_t), offsetof(userquota_node_t, 2100 uqn_node)); 2101 2102 while ((dn = multilist_sublist_head(list)) != NULL) { 2103 int flags; 2104 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); 2105 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || 2106 dn->dn_phys->dn_flags & 2107 DNODE_FLAG_USERUSED_ACCOUNTED); 2108 2109 flags = dn->dn_id_flags; 2110 ASSERT(flags); 2111 if (flags & DN_ID_OLD_EXIST) { 2112 do_userquota_update(os, &cache, dn->dn_oldused, 2113 dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, 2114 dn->dn_oldprojid, B_TRUE); 2115 do_userobjquota_update(os, &cache, dn->dn_oldflags, 2116 dn->dn_olduid, dn->dn_oldgid, 2117 dn->dn_oldprojid, B_TRUE); 2118 } 2119 if (flags & DN_ID_NEW_EXIST) { 2120 do_userquota_update(os, &cache, 2121 DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, 2122 dn->dn_newuid, dn->dn_newgid, 2123 dn->dn_newprojid, B_FALSE); 2124 do_userobjquota_update(os, &cache, 2125 dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, 2126 dn->dn_newprojid, B_FALSE); 2127 } 2128 2129 mutex_enter(&dn->dn_mtx); 2130 dn->dn_oldused = 0; 2131 dn->dn_oldflags = 0; 2132 if (dn->dn_id_flags & DN_ID_NEW_EXIST) { 2133 dn->dn_olduid = dn->dn_newuid; 2134 dn->dn_oldgid = dn->dn_newgid; 2135 dn->dn_oldprojid = dn->dn_newprojid; 2136 dn->dn_id_flags |= DN_ID_OLD_EXIST; 2137 if (dn->dn_bonuslen == 0) 2138 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 2139 else 2140 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2141 } 2142 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); 2143 mutex_exit(&dn->dn_mtx); 2144 2145 multilist_sublist_remove(list, dn); 2146 dnode_rele(dn, &os->os_synced_dnodes); 2147 } 2148 do_userquota_cacheflush(os, &cache, tx); 2149 multilist_sublist_unlock(list); 2150 kmem_free(uua, sizeof (*uua)); 2151 } 2152 2153 /* 2154 * Release dnode holds from dmu_objset_sync_dnodes(). When the dnode is being 2155 * synced (i.e. we have issued the zio's for blocks in the dnode), it can't be 2156 * evicted because the block containing the dnode can't be evicted until it is 2157 * written out. However, this hold is necessary to prevent the dnode_t from 2158 * being moved (via dnode_move()) while it's still referenced by 2159 * dbuf_dirty_record_t:dr_dnode. And dr_dnode is needed for 2160 * dirty_lightweight_leaf-type dirty records. 2161 * 2162 * If we are doing user-object accounting, the dnode_rele() happens from 2163 * userquota_updates_task() instead. 2164 */ 2165 static void 2166 dnode_rele_task(void *arg) 2167 { 2168 userquota_updates_arg_t *uua = arg; 2169 objset_t *os = uua->uua_os; 2170 2171 multilist_sublist_t *list = multilist_sublist_lock_idx( 2172 &os->os_synced_dnodes, uua->uua_sublist_idx); 2173 2174 dnode_t *dn; 2175 while ((dn = multilist_sublist_head(list)) != NULL) { 2176 multilist_sublist_remove(list, dn); 2177 dnode_rele(dn, &os->os_synced_dnodes); 2178 } 2179 multilist_sublist_unlock(list); 2180 kmem_free(uua, sizeof (*uua)); 2181 } 2182 2183 /* 2184 * Return TRUE if userquota updates are needed. 2185 */ 2186 static boolean_t 2187 dmu_objset_do_userquota_updates_prep(objset_t *os, dmu_tx_t *tx) 2188 { 2189 if (!dmu_objset_userused_enabled(os)) 2190 return (B_FALSE); 2191 2192 /* 2193 * If this is a raw receive just return and handle accounting 2194 * later when we have the keys loaded. We also don't do user 2195 * accounting during claiming since the datasets are not owned 2196 * for the duration of claiming and this txg should only be 2197 * used for recovery. 2198 */ 2199 if (os->os_encrypted && dmu_objset_is_receiving(os)) 2200 return (B_FALSE); 2201 2202 if (tx->tx_txg <= os->os_spa->spa_claim_max_txg) 2203 return (B_FALSE); 2204 2205 /* Allocate the user/group/project used objects if necessary. */ 2206 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 2207 VERIFY0(zap_create_claim(os, 2208 DMU_USERUSED_OBJECT, 2209 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2210 VERIFY0(zap_create_claim(os, 2211 DMU_GROUPUSED_OBJECT, 2212 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2213 } 2214 2215 if (dmu_objset_projectquota_enabled(os) && 2216 DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 2217 VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT, 2218 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2219 } 2220 return (B_TRUE); 2221 } 2222 2223 /* 2224 * Dispatch taskq tasks to dp_sync_taskq to update the user accounting, and 2225 * also release the holds on the dnodes from dmu_objset_sync_dnodes(). 2226 * The caller must taskq_wait(dp_sync_taskq). 2227 */ 2228 void 2229 dmu_objset_sync_done(objset_t *os, dmu_tx_t *tx) 2230 { 2231 boolean_t need_userquota = dmu_objset_do_userquota_updates_prep(os, tx); 2232 2233 int num_sublists = multilist_get_num_sublists(&os->os_synced_dnodes); 2234 for (int i = 0; i < num_sublists; i++) { 2235 userquota_updates_arg_t *uua = 2236 kmem_alloc(sizeof (*uua), KM_SLEEP); 2237 uua->uua_os = os; 2238 uua->uua_sublist_idx = i; 2239 uua->uua_tx = tx; 2240 2241 /* 2242 * If we don't need to update userquotas, use 2243 * dnode_rele_task() to call dnode_rele() 2244 */ 2245 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, 2246 need_userquota ? userquota_updates_task : dnode_rele_task, 2247 uua, 0); 2248 /* callback frees uua */ 2249 } 2250 } 2251 2252 2253 /* 2254 * Returns a pointer to data to find uid/gid from 2255 * 2256 * If a dirty record for transaction group that is syncing can't 2257 * be found then NULL is returned. In the NULL case it is assumed 2258 * the uid/gid aren't changing. 2259 */ 2260 static void * 2261 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) 2262 { 2263 dbuf_dirty_record_t *dr; 2264 void *data; 2265 2266 if (db->db_dirtycnt == 0) 2267 return (db->db.db_data); /* Nothing is changing */ 2268 2269 dr = dbuf_find_dirty_eq(db, tx->tx_txg); 2270 2271 if (dr == NULL) { 2272 data = NULL; 2273 } else { 2274 if (dr->dr_dnode->dn_bonuslen == 0 && 2275 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) 2276 data = dr->dt.dl.dr_data->b_data; 2277 else 2278 data = dr->dt.dl.dr_data; 2279 } 2280 2281 return (data); 2282 } 2283 2284 void 2285 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) 2286 { 2287 objset_t *os = dn->dn_objset; 2288 void *data = NULL; 2289 dmu_buf_impl_t *db = NULL; 2290 int flags = dn->dn_id_flags; 2291 int error; 2292 boolean_t have_spill = B_FALSE; 2293 2294 if (!dmu_objset_userused_enabled(dn->dn_objset)) 2295 return; 2296 2297 /* 2298 * Raw receives introduce a problem with user accounting. Raw 2299 * receives cannot update the user accounting info because the 2300 * user ids and the sizes are encrypted. To guarantee that we 2301 * never end up with bad user accounting, we simply disable it 2302 * during raw receives. We also disable this for normal receives 2303 * so that an incremental raw receive may be done on top of an 2304 * existing non-raw receive. 2305 */ 2306 if (os->os_encrypted && dmu_objset_is_receiving(os)) 2307 return; 2308 2309 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| 2310 DN_ID_CHKED_SPILL))) 2311 return; 2312 2313 if (before && dn->dn_bonuslen != 0) 2314 data = DN_BONUS(dn->dn_phys); 2315 else if (!before && dn->dn_bonuslen != 0) { 2316 if (dn->dn_bonus) { 2317 db = dn->dn_bonus; 2318 mutex_enter(&db->db_mtx); 2319 data = dmu_objset_userquota_find_data(db, tx); 2320 } else { 2321 data = DN_BONUS(dn->dn_phys); 2322 } 2323 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { 2324 int rf = 0; 2325 2326 if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) 2327 rf |= DB_RF_HAVESTRUCT; 2328 error = dmu_spill_hold_by_dnode(dn, 2329 rf | DB_RF_MUST_SUCCEED, 2330 FTAG, (dmu_buf_t **)&db); 2331 ASSERT(error == 0); 2332 mutex_enter(&db->db_mtx); 2333 data = (before) ? db->db.db_data : 2334 dmu_objset_userquota_find_data(db, tx); 2335 have_spill = B_TRUE; 2336 } else { 2337 mutex_enter(&dn->dn_mtx); 2338 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2339 mutex_exit(&dn->dn_mtx); 2340 return; 2341 } 2342 2343 /* 2344 * Must always call the callback in case the object 2345 * type has changed and that type isn't an object type to track 2346 */ 2347 zfs_file_info_t zfi; 2348 error = file_cbs[os->os_phys->os_type](dn->dn_bonustype, data, &zfi); 2349 2350 if (before) { 2351 ASSERT(data); 2352 dn->dn_olduid = zfi.zfi_user; 2353 dn->dn_oldgid = zfi.zfi_group; 2354 dn->dn_oldprojid = zfi.zfi_project; 2355 } else if (data) { 2356 dn->dn_newuid = zfi.zfi_user; 2357 dn->dn_newgid = zfi.zfi_group; 2358 dn->dn_newprojid = zfi.zfi_project; 2359 } 2360 2361 /* 2362 * Preserve existing uid/gid when the callback can't determine 2363 * what the new uid/gid are and the callback returned EEXIST. 2364 * The EEXIST error tells us to just use the existing uid/gid. 2365 * If we don't know what the old values are then just assign 2366 * them to 0, since that is a new file being created. 2367 */ 2368 if (!before && data == NULL && error == EEXIST) { 2369 if (flags & DN_ID_OLD_EXIST) { 2370 dn->dn_newuid = dn->dn_olduid; 2371 dn->dn_newgid = dn->dn_oldgid; 2372 dn->dn_newprojid = dn->dn_oldprojid; 2373 } else { 2374 dn->dn_newuid = 0; 2375 dn->dn_newgid = 0; 2376 dn->dn_newprojid = ZFS_DEFAULT_PROJID; 2377 } 2378 error = 0; 2379 } 2380 2381 if (db) 2382 mutex_exit(&db->db_mtx); 2383 2384 mutex_enter(&dn->dn_mtx); 2385 if (error == 0 && before) 2386 dn->dn_id_flags |= DN_ID_OLD_EXIST; 2387 if (error == 0 && !before) 2388 dn->dn_id_flags |= DN_ID_NEW_EXIST; 2389 2390 if (have_spill) { 2391 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 2392 } else { 2393 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2394 } 2395 mutex_exit(&dn->dn_mtx); 2396 if (have_spill) 2397 dmu_buf_rele((dmu_buf_t *)db, FTAG); 2398 } 2399 2400 boolean_t 2401 dmu_objset_userspace_present(objset_t *os) 2402 { 2403 return (os->os_phys->os_flags & 2404 OBJSET_FLAG_USERACCOUNTING_COMPLETE); 2405 } 2406 2407 boolean_t 2408 dmu_objset_userobjspace_present(objset_t *os) 2409 { 2410 return (os->os_phys->os_flags & 2411 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE); 2412 } 2413 2414 boolean_t 2415 dmu_objset_projectquota_present(objset_t *os) 2416 { 2417 return (os->os_phys->os_flags & 2418 OBJSET_FLAG_PROJECTQUOTA_COMPLETE); 2419 } 2420 2421 static int 2422 dmu_objset_space_upgrade(objset_t *os) 2423 { 2424 uint64_t obj; 2425 int err = 0; 2426 2427 /* 2428 * We simply need to mark every object dirty, so that it will be 2429 * synced out and now accounted. If this is called 2430 * concurrently, or if we already did some work before crashing, 2431 * that's fine, since we track each object's accounted state 2432 * independently. 2433 */ 2434 2435 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { 2436 dmu_tx_t *tx; 2437 dmu_buf_t *db; 2438 int objerr; 2439 2440 mutex_enter(&os->os_upgrade_lock); 2441 if (os->os_upgrade_exit) 2442 err = SET_ERROR(EINTR); 2443 mutex_exit(&os->os_upgrade_lock); 2444 if (err != 0) 2445 return (err); 2446 2447 if (issig()) 2448 return (SET_ERROR(EINTR)); 2449 2450 objerr = dmu_bonus_hold(os, obj, FTAG, &db); 2451 if (objerr != 0) 2452 continue; 2453 tx = dmu_tx_create(os); 2454 dmu_tx_hold_bonus(tx, obj); 2455 objerr = dmu_tx_assign(tx, DMU_TX_WAIT); 2456 if (objerr != 0) { 2457 dmu_buf_rele(db, FTAG); 2458 dmu_tx_abort(tx); 2459 continue; 2460 } 2461 dmu_buf_will_dirty(db, tx); 2462 dmu_buf_rele(db, FTAG); 2463 dmu_tx_commit(tx); 2464 } 2465 return (0); 2466 } 2467 2468 static int 2469 dmu_objset_userspace_upgrade_cb(objset_t *os) 2470 { 2471 int err = 0; 2472 2473 if (dmu_objset_userspace_present(os)) 2474 return (0); 2475 if (dmu_objset_is_snapshot(os)) 2476 return (SET_ERROR(EINVAL)); 2477 if (!dmu_objset_userused_enabled(os)) 2478 return (SET_ERROR(ENOTSUP)); 2479 2480 err = dmu_objset_space_upgrade(os); 2481 if (err) 2482 return (err); 2483 2484 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 2485 txg_wait_synced(dmu_objset_pool(os), 0); 2486 return (0); 2487 } 2488 2489 void 2490 dmu_objset_userspace_upgrade(objset_t *os) 2491 { 2492 dmu_objset_upgrade(os, dmu_objset_userspace_upgrade_cb); 2493 } 2494 2495 static int 2496 dmu_objset_id_quota_upgrade_cb(objset_t *os) 2497 { 2498 int err = 0; 2499 2500 if (dmu_objset_userobjspace_present(os) && 2501 dmu_objset_projectquota_present(os)) 2502 return (0); 2503 if (dmu_objset_is_snapshot(os)) 2504 return (SET_ERROR(EINVAL)); 2505 if (!dmu_objset_userused_enabled(os)) 2506 return (SET_ERROR(ENOTSUP)); 2507 if (!dmu_objset_projectquota_enabled(os) && 2508 dmu_objset_userobjspace_present(os)) 2509 return (SET_ERROR(ENOTSUP)); 2510 2511 if (dmu_objset_userobjused_enabled(os)) 2512 dmu_objset_ds(os)->ds_feature_activation[ 2513 SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; 2514 if (dmu_objset_projectquota_enabled(os)) 2515 dmu_objset_ds(os)->ds_feature_activation[ 2516 SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; 2517 2518 err = dmu_objset_space_upgrade(os); 2519 if (err) 2520 return (err); 2521 2522 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 2523 if (dmu_objset_userobjused_enabled(os)) 2524 os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; 2525 if (dmu_objset_projectquota_enabled(os)) 2526 os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE; 2527 2528 txg_wait_synced(dmu_objset_pool(os), 0); 2529 return (0); 2530 } 2531 2532 void 2533 dmu_objset_id_quota_upgrade(objset_t *os) 2534 { 2535 dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb); 2536 } 2537 2538 boolean_t 2539 dmu_objset_userobjspace_upgradable(objset_t *os) 2540 { 2541 return (dmu_objset_type(os) == DMU_OST_ZFS && 2542 !dmu_objset_is_snapshot(os) && 2543 dmu_objset_userobjused_enabled(os) && 2544 !dmu_objset_userobjspace_present(os) && 2545 spa_writeable(dmu_objset_spa(os))); 2546 } 2547 2548 boolean_t 2549 dmu_objset_projectquota_upgradable(objset_t *os) 2550 { 2551 return (dmu_objset_type(os) == DMU_OST_ZFS && 2552 !dmu_objset_is_snapshot(os) && 2553 dmu_objset_projectquota_enabled(os) && 2554 !dmu_objset_projectquota_present(os) && 2555 spa_writeable(dmu_objset_spa(os))); 2556 } 2557 2558 void 2559 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 2560 uint64_t *usedobjsp, uint64_t *availobjsp) 2561 { 2562 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, 2563 usedobjsp, availobjsp); 2564 } 2565 2566 uint64_t 2567 dmu_objset_fsid_guid(objset_t *os) 2568 { 2569 return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); 2570 } 2571 2572 void 2573 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) 2574 { 2575 stat->dds_type = os->os_phys->os_type; 2576 if (os->os_dsl_dataset) 2577 dsl_dataset_fast_stat(os->os_dsl_dataset, stat); 2578 } 2579 2580 void 2581 dmu_objset_stats(objset_t *os, nvlist_t *nv) 2582 { 2583 ASSERT(os->os_dsl_dataset || 2584 os->os_phys->os_type == DMU_OST_META); 2585 2586 if (os->os_dsl_dataset != NULL) 2587 dsl_dataset_stats(os->os_dsl_dataset, nv); 2588 2589 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, 2590 os->os_phys->os_type); 2591 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, 2592 dmu_objset_userspace_present(os)); 2593 } 2594 2595 int 2596 dmu_objset_is_snapshot(objset_t *os) 2597 { 2598 if (os->os_dsl_dataset != NULL) 2599 return (os->os_dsl_dataset->ds_is_snapshot); 2600 else 2601 return (B_FALSE); 2602 } 2603 2604 int 2605 dmu_snapshot_realname(objset_t *os, const char *name, char *real, int maxlen, 2606 boolean_t *conflict) 2607 { 2608 dsl_dataset_t *ds = os->os_dsl_dataset; 2609 uint64_t ignored; 2610 2611 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 2612 return (SET_ERROR(ENOENT)); 2613 2614 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, 2615 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, 2616 MT_NORMALIZE, real, maxlen, conflict)); 2617 } 2618 2619 int 2620 dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 2621 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) 2622 { 2623 dsl_dataset_t *ds = os->os_dsl_dataset; 2624 zap_cursor_t cursor; 2625 zap_attribute_t *attr; 2626 2627 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 2628 2629 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 2630 return (SET_ERROR(ENOENT)); 2631 2632 attr = zap_attribute_alloc(); 2633 zap_cursor_init_serialized(&cursor, 2634 ds->ds_dir->dd_pool->dp_meta_objset, 2635 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); 2636 2637 if (zap_cursor_retrieve(&cursor, attr) != 0) { 2638 zap_cursor_fini(&cursor); 2639 zap_attribute_free(attr); 2640 return (SET_ERROR(ENOENT)); 2641 } 2642 2643 if (strlen(attr->za_name) + 1 > namelen) { 2644 zap_cursor_fini(&cursor); 2645 zap_attribute_free(attr); 2646 return (SET_ERROR(ENAMETOOLONG)); 2647 } 2648 2649 (void) strlcpy(name, attr->za_name, namelen); 2650 if (idp) 2651 *idp = attr->za_first_integer; 2652 if (case_conflict) 2653 *case_conflict = attr->za_normalization_conflict; 2654 zap_cursor_advance(&cursor); 2655 *offp = zap_cursor_serialize(&cursor); 2656 zap_cursor_fini(&cursor); 2657 zap_attribute_free(attr); 2658 2659 return (0); 2660 } 2661 2662 int 2663 dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value) 2664 { 2665 return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value)); 2666 } 2667 2668 int 2669 dmu_dir_list_next(objset_t *os, int namelen, char *name, 2670 uint64_t *idp, uint64_t *offp) 2671 { 2672 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 2673 zap_cursor_t cursor; 2674 zap_attribute_t *attr; 2675 2676 /* there is no next dir on a snapshot! */ 2677 if (os->os_dsl_dataset->ds_object != 2678 dsl_dir_phys(dd)->dd_head_dataset_obj) 2679 return (SET_ERROR(ENOENT)); 2680 2681 attr = zap_attribute_alloc(); 2682 zap_cursor_init_serialized(&cursor, 2683 dd->dd_pool->dp_meta_objset, 2684 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); 2685 2686 if (zap_cursor_retrieve(&cursor, attr) != 0) { 2687 zap_cursor_fini(&cursor); 2688 zap_attribute_free(attr); 2689 return (SET_ERROR(ENOENT)); 2690 } 2691 2692 if (strlen(attr->za_name) + 1 > namelen) { 2693 zap_cursor_fini(&cursor); 2694 zap_attribute_free(attr); 2695 return (SET_ERROR(ENAMETOOLONG)); 2696 } 2697 2698 (void) strlcpy(name, attr->za_name, namelen); 2699 if (idp) 2700 *idp = attr->za_first_integer; 2701 zap_cursor_advance(&cursor); 2702 *offp = zap_cursor_serialize(&cursor); 2703 zap_cursor_fini(&cursor); 2704 zap_attribute_free(attr); 2705 2706 return (0); 2707 } 2708 2709 typedef struct dmu_objset_find_ctx { 2710 taskq_t *dc_tq; 2711 dsl_pool_t *dc_dp; 2712 uint64_t dc_ddobj; 2713 char *dc_ddname; /* last component of ddobj's name */ 2714 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); 2715 void *dc_arg; 2716 int dc_flags; 2717 kmutex_t *dc_error_lock; 2718 int *dc_error; 2719 } dmu_objset_find_ctx_t; 2720 2721 static void 2722 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) 2723 { 2724 dsl_pool_t *dp = dcp->dc_dp; 2725 dsl_dir_t *dd; 2726 dsl_dataset_t *ds; 2727 zap_cursor_t zc; 2728 zap_attribute_t *attr; 2729 uint64_t thisobj; 2730 int err = 0; 2731 2732 /* don't process if there already was an error */ 2733 if (*dcp->dc_error != 0) 2734 goto out; 2735 2736 /* 2737 * Note: passing the name (dc_ddname) here is optional, but it 2738 * improves performance because we don't need to call 2739 * zap_value_search() to determine the name. 2740 */ 2741 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); 2742 if (err != 0) 2743 goto out; 2744 2745 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2746 if (dd->dd_myname[0] == '$') { 2747 dsl_dir_rele(dd, FTAG); 2748 goto out; 2749 } 2750 2751 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2752 attr = zap_attribute_alloc(); 2753 2754 /* 2755 * Iterate over all children. 2756 */ 2757 if (dcp->dc_flags & DS_FIND_CHILDREN) { 2758 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2759 dsl_dir_phys(dd)->dd_child_dir_zapobj); 2760 zap_cursor_retrieve(&zc, attr) == 0; 2761 (void) zap_cursor_advance(&zc)) { 2762 ASSERT3U(attr->za_integer_length, ==, 2763 sizeof (uint64_t)); 2764 ASSERT3U(attr->za_num_integers, ==, 1); 2765 2766 dmu_objset_find_ctx_t *child_dcp = 2767 kmem_alloc(sizeof (*child_dcp), KM_SLEEP); 2768 *child_dcp = *dcp; 2769 child_dcp->dc_ddobj = attr->za_first_integer; 2770 child_dcp->dc_ddname = spa_strdup(attr->za_name); 2771 if (dcp->dc_tq != NULL) 2772 (void) taskq_dispatch(dcp->dc_tq, 2773 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); 2774 else 2775 dmu_objset_find_dp_impl(child_dcp); 2776 } 2777 zap_cursor_fini(&zc); 2778 } 2779 2780 /* 2781 * Iterate over all snapshots. 2782 */ 2783 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { 2784 dsl_dataset_t *ds; 2785 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2786 2787 if (err == 0) { 2788 uint64_t snapobj; 2789 2790 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 2791 dsl_dataset_rele(ds, FTAG); 2792 2793 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 2794 zap_cursor_retrieve(&zc, attr) == 0; 2795 (void) zap_cursor_advance(&zc)) { 2796 ASSERT3U(attr->za_integer_length, ==, 2797 sizeof (uint64_t)); 2798 ASSERT3U(attr->za_num_integers, ==, 1); 2799 2800 err = dsl_dataset_hold_obj(dp, 2801 attr->za_first_integer, FTAG, &ds); 2802 if (err != 0) 2803 break; 2804 err = dcp->dc_func(dp, ds, dcp->dc_arg); 2805 dsl_dataset_rele(ds, FTAG); 2806 if (err != 0) 2807 break; 2808 } 2809 zap_cursor_fini(&zc); 2810 } 2811 } 2812 2813 zap_attribute_free(attr); 2814 2815 if (err != 0) { 2816 dsl_dir_rele(dd, FTAG); 2817 goto out; 2818 } 2819 2820 /* 2821 * Apply to self. 2822 */ 2823 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2824 2825 /* 2826 * Note: we hold the dir while calling dsl_dataset_hold_obj() so 2827 * that the dir will remain cached, and we won't have to re-instantiate 2828 * it (which could be expensive due to finding its name via 2829 * zap_value_search()). 2830 */ 2831 dsl_dir_rele(dd, FTAG); 2832 if (err != 0) 2833 goto out; 2834 err = dcp->dc_func(dp, ds, dcp->dc_arg); 2835 dsl_dataset_rele(ds, FTAG); 2836 2837 out: 2838 if (err != 0) { 2839 mutex_enter(dcp->dc_error_lock); 2840 /* only keep first error */ 2841 if (*dcp->dc_error == 0) 2842 *dcp->dc_error = err; 2843 mutex_exit(dcp->dc_error_lock); 2844 } 2845 2846 if (dcp->dc_ddname != NULL) 2847 spa_strfree(dcp->dc_ddname); 2848 kmem_free(dcp, sizeof (*dcp)); 2849 } 2850 2851 static void 2852 dmu_objset_find_dp_cb(void *arg) 2853 { 2854 dmu_objset_find_ctx_t *dcp = arg; 2855 dsl_pool_t *dp = dcp->dc_dp; 2856 2857 /* 2858 * We need to get a pool_config_lock here, as there are several 2859 * assert(pool_config_held) down the stack. Getting a lock via 2860 * dsl_pool_config_enter is risky, as it might be stalled by a 2861 * pending writer. This would deadlock, as the write lock can 2862 * only be granted when our parent thread gives up the lock. 2863 * The _prio interface gives us priority over a pending writer. 2864 */ 2865 dsl_pool_config_enter_prio(dp, FTAG); 2866 2867 dmu_objset_find_dp_impl(dcp); 2868 2869 dsl_pool_config_exit(dp, FTAG); 2870 } 2871 2872 /* 2873 * Find objsets under and including ddobj, call func(ds) on each. 2874 * The order for the enumeration is completely undefined. 2875 * func is called with dsl_pool_config held. 2876 */ 2877 int 2878 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, 2879 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) 2880 { 2881 int error = 0; 2882 taskq_t *tq = NULL; 2883 int ntasks; 2884 dmu_objset_find_ctx_t *dcp; 2885 kmutex_t err_lock; 2886 2887 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); 2888 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); 2889 dcp->dc_tq = NULL; 2890 dcp->dc_dp = dp; 2891 dcp->dc_ddobj = ddobj; 2892 dcp->dc_ddname = NULL; 2893 dcp->dc_func = func; 2894 dcp->dc_arg = arg; 2895 dcp->dc_flags = flags; 2896 dcp->dc_error_lock = &err_lock; 2897 dcp->dc_error = &error; 2898 2899 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { 2900 /* 2901 * In case a write lock is held we can't make use of 2902 * parallelism, as down the stack of the worker threads 2903 * the lock is asserted via dsl_pool_config_held. 2904 * In case of a read lock this is solved by getting a read 2905 * lock in each worker thread, which isn't possible in case 2906 * of a writer lock. So we fall back to the synchronous path 2907 * here. 2908 * In the future it might be possible to get some magic into 2909 * dsl_pool_config_held in a way that it returns true for 2910 * the worker threads so that a single lock held from this 2911 * thread suffices. For now, stay single threaded. 2912 */ 2913 dmu_objset_find_dp_impl(dcp); 2914 mutex_destroy(&err_lock); 2915 2916 return (error); 2917 } 2918 2919 ntasks = dmu_find_threads; 2920 if (ntasks == 0) 2921 ntasks = vdev_count_leaves(dp->dp_spa) * 4; 2922 tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks, 2923 INT_MAX, 0); 2924 if (tq == NULL) { 2925 kmem_free(dcp, sizeof (*dcp)); 2926 mutex_destroy(&err_lock); 2927 2928 return (SET_ERROR(ENOMEM)); 2929 } 2930 dcp->dc_tq = tq; 2931 2932 /* dcp will be freed by task */ 2933 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); 2934 2935 /* 2936 * PORTING: this code relies on the property of taskq_wait to wait 2937 * until no more tasks are queued and no more tasks are active. As 2938 * we always queue new tasks from within other tasks, task_wait 2939 * reliably waits for the full recursion to finish, even though we 2940 * enqueue new tasks after taskq_wait has been called. 2941 * On platforms other than illumos, taskq_wait may not have this 2942 * property. 2943 */ 2944 taskq_wait(tq); 2945 taskq_destroy(tq); 2946 mutex_destroy(&err_lock); 2947 2948 return (error); 2949 } 2950 2951 /* 2952 * Find all objsets under name, and for each, call 'func(child_name, arg)'. 2953 * The dp_config_rwlock must not be held when this is called, and it 2954 * will not be held when the callback is called. 2955 * Therefore this function should only be used when the pool is not changing 2956 * (e.g. in syncing context), or the callback can deal with the possible races. 2957 */ 2958 static int 2959 dmu_objset_find_impl(spa_t *spa, const char *name, 2960 int func(const char *, void *), void *arg, int flags) 2961 { 2962 dsl_dir_t *dd; 2963 dsl_pool_t *dp = spa_get_dsl(spa); 2964 dsl_dataset_t *ds; 2965 zap_cursor_t zc; 2966 zap_attribute_t *attr; 2967 char *child; 2968 uint64_t thisobj; 2969 int err; 2970 2971 dsl_pool_config_enter(dp, FTAG); 2972 2973 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); 2974 if (err != 0) { 2975 dsl_pool_config_exit(dp, FTAG); 2976 return (err); 2977 } 2978 2979 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2980 if (dd->dd_myname[0] == '$') { 2981 dsl_dir_rele(dd, FTAG); 2982 dsl_pool_config_exit(dp, FTAG); 2983 return (0); 2984 } 2985 2986 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2987 attr = zap_attribute_alloc(); 2988 2989 /* 2990 * Iterate over all children. 2991 */ 2992 if (flags & DS_FIND_CHILDREN) { 2993 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2994 dsl_dir_phys(dd)->dd_child_dir_zapobj); 2995 zap_cursor_retrieve(&zc, attr) == 0; 2996 (void) zap_cursor_advance(&zc)) { 2997 ASSERT3U(attr->za_integer_length, ==, 2998 sizeof (uint64_t)); 2999 ASSERT3U(attr->za_num_integers, ==, 1); 3000 3001 child = kmem_asprintf("%s/%s", name, attr->za_name); 3002 dsl_pool_config_exit(dp, FTAG); 3003 err = dmu_objset_find_impl(spa, child, 3004 func, arg, flags); 3005 dsl_pool_config_enter(dp, FTAG); 3006 kmem_strfree(child); 3007 if (err != 0) 3008 break; 3009 } 3010 zap_cursor_fini(&zc); 3011 3012 if (err != 0) { 3013 dsl_dir_rele(dd, FTAG); 3014 dsl_pool_config_exit(dp, FTAG); 3015 zap_attribute_free(attr); 3016 return (err); 3017 } 3018 } 3019 3020 /* 3021 * Iterate over all snapshots. 3022 */ 3023 if (flags & DS_FIND_SNAPSHOTS) { 3024 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 3025 3026 if (err == 0) { 3027 uint64_t snapobj; 3028 3029 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 3030 dsl_dataset_rele(ds, FTAG); 3031 3032 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 3033 zap_cursor_retrieve(&zc, attr) == 0; 3034 (void) zap_cursor_advance(&zc)) { 3035 ASSERT3U(attr->za_integer_length, ==, 3036 sizeof (uint64_t)); 3037 ASSERT3U(attr->za_num_integers, ==, 1); 3038 3039 child = kmem_asprintf("%s@%s", 3040 name, attr->za_name); 3041 dsl_pool_config_exit(dp, FTAG); 3042 err = func(child, arg); 3043 dsl_pool_config_enter(dp, FTAG); 3044 kmem_strfree(child); 3045 if (err != 0) 3046 break; 3047 } 3048 zap_cursor_fini(&zc); 3049 } 3050 } 3051 3052 dsl_dir_rele(dd, FTAG); 3053 zap_attribute_free(attr); 3054 dsl_pool_config_exit(dp, FTAG); 3055 3056 if (err != 0) 3057 return (err); 3058 3059 /* Apply to self. */ 3060 return (func(name, arg)); 3061 } 3062 3063 /* 3064 * See comment above dmu_objset_find_impl(). 3065 */ 3066 int 3067 dmu_objset_find(const char *name, int func(const char *, void *), void *arg, 3068 int flags) 3069 { 3070 spa_t *spa; 3071 int error; 3072 3073 error = spa_open(name, &spa, FTAG); 3074 if (error != 0) 3075 return (error); 3076 error = dmu_objset_find_impl(spa, name, func, arg, flags); 3077 spa_close(spa, FTAG); 3078 return (error); 3079 } 3080 3081 boolean_t 3082 dmu_objset_incompatible_encryption_version(objset_t *os) 3083 { 3084 return (dsl_dir_incompatible_encryption_version( 3085 os->os_dsl_dataset->ds_dir)); 3086 } 3087 3088 void 3089 dmu_objset_set_user(objset_t *os, void *user_ptr) 3090 { 3091 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 3092 os->os_user_ptr = user_ptr; 3093 } 3094 3095 void * 3096 dmu_objset_get_user(objset_t *os) 3097 { 3098 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 3099 return (os->os_user_ptr); 3100 } 3101 3102 /* 3103 * Determine name of filesystem, given name of snapshot. 3104 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes 3105 */ 3106 int 3107 dmu_fsname(const char *snapname, char *buf) 3108 { 3109 char *atp = strchr(snapname, '@'); 3110 if (atp == NULL) 3111 return (SET_ERROR(EINVAL)); 3112 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) 3113 return (SET_ERROR(ENAMETOOLONG)); 3114 (void) strlcpy(buf, snapname, atp - snapname + 1); 3115 return (0); 3116 } 3117 3118 /* 3119 * Call when we think we're going to write/free space in open context 3120 * to track the amount of dirty data in the open txg, which is also the 3121 * amount of memory that can not be evicted until this txg syncs. 3122 * 3123 * Note that there are two conditions where this can be called from 3124 * syncing context: 3125 * 3126 * [1] When we just created the dataset, in which case we go on with 3127 * updating any accounting of dirty data as usual. 3128 * [2] When we are dirtying MOS data, in which case we only update the 3129 * pool's accounting of dirty data. 3130 */ 3131 void 3132 dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) 3133 { 3134 dsl_dataset_t *ds = os->os_dsl_dataset; 3135 int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); 3136 3137 if (ds != NULL) { 3138 dsl_dir_willuse_space(ds->ds_dir, aspace, tx); 3139 } 3140 3141 dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); 3142 } 3143 3144 #if defined(_KERNEL) 3145 EXPORT_SYMBOL(dmu_objset_zil); 3146 EXPORT_SYMBOL(dmu_objset_pool); 3147 EXPORT_SYMBOL(dmu_objset_ds); 3148 EXPORT_SYMBOL(dmu_objset_type); 3149 EXPORT_SYMBOL(dmu_objset_name); 3150 EXPORT_SYMBOL(dmu_objset_hold); 3151 EXPORT_SYMBOL(dmu_objset_hold_flags); 3152 EXPORT_SYMBOL(dmu_objset_own); 3153 EXPORT_SYMBOL(dmu_objset_rele); 3154 EXPORT_SYMBOL(dmu_objset_rele_flags); 3155 EXPORT_SYMBOL(dmu_objset_disown); 3156 EXPORT_SYMBOL(dmu_objset_from_ds); 3157 EXPORT_SYMBOL(dmu_objset_create); 3158 EXPORT_SYMBOL(dmu_objset_clone); 3159 EXPORT_SYMBOL(dmu_objset_stats); 3160 EXPORT_SYMBOL(dmu_objset_fast_stat); 3161 EXPORT_SYMBOL(dmu_objset_spa); 3162 EXPORT_SYMBOL(dmu_objset_space); 3163 EXPORT_SYMBOL(dmu_objset_fsid_guid); 3164 EXPORT_SYMBOL(dmu_objset_find); 3165 EXPORT_SYMBOL(dmu_objset_byteswap); 3166 EXPORT_SYMBOL(dmu_objset_evict_dbufs); 3167 EXPORT_SYMBOL(dmu_objset_snap_cmtime); 3168 EXPORT_SYMBOL(dmu_objset_dnodesize); 3169 3170 EXPORT_SYMBOL(dmu_objset_sync); 3171 EXPORT_SYMBOL(dmu_objset_is_dirty); 3172 EXPORT_SYMBOL(dmu_objset_create_impl_dnstats); 3173 EXPORT_SYMBOL(dmu_objset_create_impl); 3174 EXPORT_SYMBOL(dmu_objset_open_impl); 3175 EXPORT_SYMBOL(dmu_objset_evict); 3176 EXPORT_SYMBOL(dmu_objset_register_type); 3177 EXPORT_SYMBOL(dmu_objset_sync_done); 3178 EXPORT_SYMBOL(dmu_objset_userquota_get_ids); 3179 EXPORT_SYMBOL(dmu_objset_userused_enabled); 3180 EXPORT_SYMBOL(dmu_objset_userspace_upgrade); 3181 EXPORT_SYMBOL(dmu_objset_userspace_present); 3182 EXPORT_SYMBOL(dmu_objset_userobjused_enabled); 3183 EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable); 3184 EXPORT_SYMBOL(dmu_objset_userobjspace_present); 3185 EXPORT_SYMBOL(dmu_objset_projectquota_enabled); 3186 EXPORT_SYMBOL(dmu_objset_projectquota_present); 3187 EXPORT_SYMBOL(dmu_objset_projectquota_upgradable); 3188 EXPORT_SYMBOL(dmu_objset_id_quota_upgrade); 3189 #endif 3190