1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or https://opensource.org/licenses/CDDL-1.0. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2012, 2020 by Delphix. All rights reserved. 25 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. 26 * Copyright (c) 2013, Joyent, Inc. All rights reserved. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved. 29 * Copyright (c) 2016 Actifio, Inc. All rights reserved. 30 * Copyright 2017 Nexenta Systems, Inc. 31 * Copyright (c) 2017 Open-E, Inc. All Rights Reserved. 32 * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved. 33 * Copyright (c) 2019, Klara Inc. 34 * Copyright (c) 2019, Allan Jude 35 * Copyright (c) 2022 Hewlett Packard Enterprise Development LP. 36 */ 37 38 /* Portions Copyright 2010 Robert Milkowski */ 39 40 #include <sys/cred.h> 41 #include <sys/zfs_context.h> 42 #include <sys/dmu_objset.h> 43 #include <sys/dsl_dir.h> 44 #include <sys/dsl_dataset.h> 45 #include <sys/dsl_prop.h> 46 #include <sys/dsl_pool.h> 47 #include <sys/dsl_synctask.h> 48 #include <sys/dsl_deleg.h> 49 #include <sys/dnode.h> 50 #include <sys/dbuf.h> 51 #include <sys/zvol.h> 52 #include <sys/dmu_tx.h> 53 #include <sys/zap.h> 54 #include <sys/zil.h> 55 #include <sys/dmu_impl.h> 56 #include <sys/zfs_ioctl.h> 57 #include <sys/sa.h> 58 #include <sys/zfs_onexit.h> 59 #include <sys/dsl_destroy.h> 60 #include <sys/vdev.h> 61 #include <sys/zfeature.h> 62 #include <sys/policy.h> 63 #include <sys/spa_impl.h> 64 #include <sys/dmu_recv.h> 65 #include <sys/zfs_project.h> 66 #include "zfs_namecheck.h" 67 #include <sys/vdev_impl.h> 68 #include <sys/arc.h> 69 #include <cityhash.h> 70 71 /* 72 * Needed to close a window in dnode_move() that allows the objset to be freed 73 * before it can be safely accessed. 74 */ 75 krwlock_t os_lock; 76 77 /* 78 * Tunable to overwrite the maximum number of threads for the parallelization 79 * of dmu_objset_find_dp, needed to speed up the import of pools with many 80 * datasets. 81 * Default is 4 times the number of leaf vdevs. 82 */ 83 static const int dmu_find_threads = 0; 84 85 /* 86 * Backfill lower metadnode objects after this many have been freed. 87 * Backfilling negatively impacts object creation rates, so only do it 88 * if there are enough holes to fill. 89 */ 90 static const int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT; 91 92 static const char *upgrade_tag = "upgrade_tag"; 93 94 static void dmu_objset_find_dp_cb(void *arg); 95 96 static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb); 97 static void dmu_objset_upgrade_stop(objset_t *os); 98 99 void 100 dmu_objset_init(void) 101 { 102 rw_init(&os_lock, NULL, RW_DEFAULT, NULL); 103 } 104 105 void 106 dmu_objset_fini(void) 107 { 108 rw_destroy(&os_lock); 109 } 110 111 spa_t * 112 dmu_objset_spa(objset_t *os) 113 { 114 return (os->os_spa); 115 } 116 117 zilog_t * 118 dmu_objset_zil(objset_t *os) 119 { 120 return (os->os_zil); 121 } 122 123 dsl_pool_t * 124 dmu_objset_pool(objset_t *os) 125 { 126 dsl_dataset_t *ds; 127 128 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) 129 return (ds->ds_dir->dd_pool); 130 else 131 return (spa_get_dsl(os->os_spa)); 132 } 133 134 dsl_dataset_t * 135 dmu_objset_ds(objset_t *os) 136 { 137 return (os->os_dsl_dataset); 138 } 139 140 dmu_objset_type_t 141 dmu_objset_type(objset_t *os) 142 { 143 return (os->os_phys->os_type); 144 } 145 146 void 147 dmu_objset_name(objset_t *os, char *buf) 148 { 149 dsl_dataset_name(os->os_dsl_dataset, buf); 150 } 151 152 uint64_t 153 dmu_objset_id(objset_t *os) 154 { 155 dsl_dataset_t *ds = os->os_dsl_dataset; 156 157 return (ds ? ds->ds_object : 0); 158 } 159 160 uint64_t 161 dmu_objset_dnodesize(objset_t *os) 162 { 163 return (os->os_dnodesize); 164 } 165 166 zfs_sync_type_t 167 dmu_objset_syncprop(objset_t *os) 168 { 169 return (os->os_sync); 170 } 171 172 zfs_logbias_op_t 173 dmu_objset_logbias(objset_t *os) 174 { 175 return (os->os_logbias); 176 } 177 178 static void 179 checksum_changed_cb(void *arg, uint64_t newval) 180 { 181 objset_t *os = arg; 182 183 /* 184 * Inheritance should have been done by now. 185 */ 186 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 187 188 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); 189 } 190 191 static void 192 compression_changed_cb(void *arg, uint64_t newval) 193 { 194 objset_t *os = arg; 195 196 /* 197 * Inheritance and range checking should have been done by now. 198 */ 199 ASSERT(newval != ZIO_COMPRESS_INHERIT); 200 201 os->os_compress = zio_compress_select(os->os_spa, 202 ZIO_COMPRESS_ALGO(newval), ZIO_COMPRESS_ON); 203 os->os_complevel = zio_complevel_select(os->os_spa, os->os_compress, 204 ZIO_COMPRESS_LEVEL(newval), ZIO_COMPLEVEL_DEFAULT); 205 } 206 207 static void 208 copies_changed_cb(void *arg, uint64_t newval) 209 { 210 objset_t *os = arg; 211 212 /* 213 * Inheritance and range checking should have been done by now. 214 */ 215 ASSERT(newval > 0); 216 ASSERT(newval <= spa_max_replication(os->os_spa)); 217 218 os->os_copies = newval; 219 } 220 221 static void 222 dedup_changed_cb(void *arg, uint64_t newval) 223 { 224 objset_t *os = arg; 225 spa_t *spa = os->os_spa; 226 enum zio_checksum checksum; 227 228 /* 229 * Inheritance should have been done by now. 230 */ 231 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 232 233 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); 234 235 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; 236 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); 237 } 238 239 static void 240 primary_cache_changed_cb(void *arg, uint64_t newval) 241 { 242 objset_t *os = arg; 243 244 /* 245 * Inheritance and range checking should have been done by now. 246 */ 247 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 248 newval == ZFS_CACHE_METADATA); 249 250 os->os_primary_cache = newval; 251 } 252 253 static void 254 secondary_cache_changed_cb(void *arg, uint64_t newval) 255 { 256 objset_t *os = arg; 257 258 /* 259 * Inheritance and range checking should have been done by now. 260 */ 261 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 262 newval == ZFS_CACHE_METADATA); 263 264 os->os_secondary_cache = newval; 265 } 266 267 static void 268 prefetch_changed_cb(void *arg, uint64_t newval) 269 { 270 objset_t *os = arg; 271 272 /* 273 * Inheritance should have been done by now. 274 */ 275 ASSERT(newval == ZFS_PREFETCH_ALL || newval == ZFS_PREFETCH_NONE || 276 newval == ZFS_PREFETCH_METADATA); 277 os->os_prefetch = newval; 278 } 279 280 static void 281 sync_changed_cb(void *arg, uint64_t newval) 282 { 283 objset_t *os = arg; 284 285 /* 286 * Inheritance and range checking should have been done by now. 287 */ 288 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || 289 newval == ZFS_SYNC_DISABLED); 290 291 os->os_sync = newval; 292 if (os->os_zil) 293 zil_set_sync(os->os_zil, newval); 294 } 295 296 static void 297 redundant_metadata_changed_cb(void *arg, uint64_t newval) 298 { 299 objset_t *os = arg; 300 301 /* 302 * Inheritance and range checking should have been done by now. 303 */ 304 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || 305 newval == ZFS_REDUNDANT_METADATA_MOST || 306 newval == ZFS_REDUNDANT_METADATA_SOME || 307 newval == ZFS_REDUNDANT_METADATA_NONE); 308 309 os->os_redundant_metadata = newval; 310 } 311 312 static void 313 dnodesize_changed_cb(void *arg, uint64_t newval) 314 { 315 objset_t *os = arg; 316 317 switch (newval) { 318 case ZFS_DNSIZE_LEGACY: 319 os->os_dnodesize = DNODE_MIN_SIZE; 320 break; 321 case ZFS_DNSIZE_AUTO: 322 /* 323 * Choose a dnode size that will work well for most 324 * workloads if the user specified "auto". Future code 325 * improvements could dynamically select a dnode size 326 * based on observed workload patterns. 327 */ 328 os->os_dnodesize = DNODE_MIN_SIZE * 2; 329 break; 330 case ZFS_DNSIZE_1K: 331 case ZFS_DNSIZE_2K: 332 case ZFS_DNSIZE_4K: 333 case ZFS_DNSIZE_8K: 334 case ZFS_DNSIZE_16K: 335 os->os_dnodesize = newval; 336 break; 337 } 338 } 339 340 static void 341 smallblk_changed_cb(void *arg, uint64_t newval) 342 { 343 objset_t *os = arg; 344 345 /* 346 * Inheritance and range checking should have been done by now. 347 */ 348 ASSERT(newval <= SPA_MAXBLOCKSIZE); 349 ASSERT(ISP2(newval)); 350 351 os->os_zpl_special_smallblock = newval; 352 } 353 354 static void 355 direct_changed_cb(void *arg, uint64_t newval) 356 { 357 objset_t *os = arg; 358 359 /* 360 * Inheritance and range checking should have been done by now. 361 */ 362 ASSERT(newval == ZFS_DIRECT_DISABLED || newval == ZFS_DIRECT_STANDARD || 363 newval == ZFS_DIRECT_ALWAYS); 364 365 os->os_direct = newval; 366 } 367 368 static void 369 logbias_changed_cb(void *arg, uint64_t newval) 370 { 371 objset_t *os = arg; 372 373 ASSERT(newval == ZFS_LOGBIAS_LATENCY || 374 newval == ZFS_LOGBIAS_THROUGHPUT); 375 os->os_logbias = newval; 376 if (os->os_zil) 377 zil_set_logbias(os->os_zil, newval); 378 } 379 380 static void 381 recordsize_changed_cb(void *arg, uint64_t newval) 382 { 383 objset_t *os = arg; 384 385 os->os_recordsize = newval; 386 } 387 388 void 389 dmu_objset_byteswap(void *buf, size_t size) 390 { 391 objset_phys_t *osp = buf; 392 393 ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 || 394 size == sizeof (objset_phys_t)); 395 dnode_byteswap(&osp->os_meta_dnode); 396 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); 397 osp->os_type = BSWAP_64(osp->os_type); 398 osp->os_flags = BSWAP_64(osp->os_flags); 399 if (size >= OBJSET_PHYS_SIZE_V2) { 400 dnode_byteswap(&osp->os_userused_dnode); 401 dnode_byteswap(&osp->os_groupused_dnode); 402 if (size >= sizeof (objset_phys_t)) 403 dnode_byteswap(&osp->os_projectused_dnode); 404 } 405 } 406 407 /* 408 * Runs cityhash on the objset_t pointer and the object number. 409 */ 410 static uint64_t 411 dnode_hash(const objset_t *os, uint64_t obj) 412 { 413 uintptr_t osv = (uintptr_t)os; 414 return (cityhash2((uint64_t)osv, obj)); 415 } 416 417 static unsigned int 418 dnode_multilist_index_func(multilist_t *ml, void *obj) 419 { 420 dnode_t *dn = obj; 421 422 /* 423 * The low order bits of the hash value are thought to be 424 * distributed evenly. Otherwise, in the case that the multilist 425 * has a power of two number of sublists, each sublists' usage 426 * would not be evenly distributed. In this context full 64bit 427 * division would be a waste of time, so limit it to 32 bits. 428 */ 429 return ((unsigned int)dnode_hash(dn->dn_objset, dn->dn_object) % 430 multilist_get_num_sublists(ml)); 431 } 432 433 static inline boolean_t 434 dmu_os_is_l2cacheable(objset_t *os) 435 { 436 if (os->os_secondary_cache == ZFS_CACHE_ALL || 437 os->os_secondary_cache == ZFS_CACHE_METADATA) { 438 if (l2arc_exclude_special == 0) 439 return (B_TRUE); 440 441 blkptr_t *bp = os->os_rootbp; 442 if (bp == NULL || BP_IS_HOLE(bp)) 443 return (B_FALSE); 444 uint64_t vdev = DVA_GET_VDEV(bp->blk_dva); 445 vdev_t *rvd = os->os_spa->spa_root_vdev; 446 vdev_t *vd = NULL; 447 448 if (vdev < rvd->vdev_children) 449 vd = rvd->vdev_child[vdev]; 450 451 if (vd == NULL) 452 return (B_TRUE); 453 454 if (vd->vdev_alloc_bias != VDEV_BIAS_SPECIAL && 455 vd->vdev_alloc_bias != VDEV_BIAS_DEDUP) 456 return (B_TRUE); 457 } 458 return (B_FALSE); 459 } 460 461 /* 462 * Instantiates the objset_t in-memory structure corresponding to the 463 * objset_phys_t that's pointed to by the specified blkptr_t. 464 */ 465 int 466 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 467 objset_t **osp) 468 { 469 objset_t *os; 470 int i, err; 471 472 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); 473 ASSERT(!BP_IS_REDACTED(bp)); 474 475 /* 476 * We need the pool config lock to get properties. 477 */ 478 ASSERT(ds == NULL || dsl_pool_config_held(ds->ds_dir->dd_pool)); 479 480 /* 481 * The $ORIGIN dataset (if it exists) doesn't have an associated 482 * objset, so there's no reason to open it. The $ORIGIN dataset 483 * will not exist on pools older than SPA_VERSION_ORIGIN. 484 */ 485 if (ds != NULL && spa_get_dsl(spa) != NULL && 486 spa_get_dsl(spa)->dp_origin_snap != NULL) { 487 ASSERT3P(ds->ds_dir, !=, 488 spa_get_dsl(spa)->dp_origin_snap->ds_dir); 489 } 490 491 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); 492 os->os_dsl_dataset = ds; 493 os->os_spa = spa; 494 os->os_rootbp = bp; 495 if (!BP_IS_HOLE(os->os_rootbp)) { 496 arc_flags_t aflags = ARC_FLAG_WAIT; 497 zbookmark_phys_t zb; 498 int size; 499 zio_flag_t zio_flags = ZIO_FLAG_CANFAIL; 500 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, 501 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 502 503 if (dmu_os_is_l2cacheable(os)) 504 aflags |= ARC_FLAG_L2CACHE; 505 506 if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) { 507 ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); 508 ASSERT(BP_IS_AUTHENTICATED(bp)); 509 zio_flags |= ZIO_FLAG_RAW; 510 } 511 512 dprintf_bp(os->os_rootbp, "reading %s", ""); 513 err = arc_read(NULL, spa, os->os_rootbp, 514 arc_getbuf_func, &os->os_phys_buf, 515 ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); 516 if (err != 0) { 517 kmem_free(os, sizeof (objset_t)); 518 /* convert checksum errors into IO errors */ 519 if (err == ECKSUM) 520 err = SET_ERROR(EIO); 521 return (err); 522 } 523 524 if (spa_version(spa) < SPA_VERSION_USERSPACE) 525 size = OBJSET_PHYS_SIZE_V1; 526 else if (!spa_feature_is_enabled(spa, 527 SPA_FEATURE_PROJECT_QUOTA)) 528 size = OBJSET_PHYS_SIZE_V2; 529 else 530 size = sizeof (objset_phys_t); 531 532 /* Increase the blocksize if we are permitted. */ 533 if (arc_buf_size(os->os_phys_buf) < size) { 534 arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, 535 ARC_BUFC_METADATA, size); 536 memset(buf->b_data, 0, size); 537 memcpy(buf->b_data, os->os_phys_buf->b_data, 538 arc_buf_size(os->os_phys_buf)); 539 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 540 os->os_phys_buf = buf; 541 } 542 543 os->os_phys = os->os_phys_buf->b_data; 544 os->os_flags = os->os_phys->os_flags; 545 } else { 546 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? 547 sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1; 548 os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, 549 ARC_BUFC_METADATA, size); 550 os->os_phys = os->os_phys_buf->b_data; 551 memset(os->os_phys, 0, size); 552 } 553 /* 554 * These properties will be filled in by the logic in zfs_get_zplprop() 555 * when they are queried for the first time. 556 */ 557 os->os_version = OBJSET_PROP_UNINITIALIZED; 558 os->os_normalization = OBJSET_PROP_UNINITIALIZED; 559 os->os_utf8only = OBJSET_PROP_UNINITIALIZED; 560 os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED; 561 562 /* 563 * Note: the changed_cb will be called once before the register 564 * func returns, thus changing the checksum/compression from the 565 * default (fletcher2/off). Snapshots don't need to know about 566 * checksum/compression/copies. 567 */ 568 if (ds != NULL) { 569 os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0); 570 571 err = dsl_prop_register(ds, 572 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), 573 primary_cache_changed_cb, os); 574 if (err == 0) { 575 err = dsl_prop_register(ds, 576 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), 577 secondary_cache_changed_cb, os); 578 } 579 if (err == 0) { 580 err = dsl_prop_register(ds, 581 zfs_prop_to_name(ZFS_PROP_PREFETCH), 582 prefetch_changed_cb, os); 583 } 584 if (!ds->ds_is_snapshot) { 585 if (err == 0) { 586 err = dsl_prop_register(ds, 587 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 588 checksum_changed_cb, os); 589 } 590 if (err == 0) { 591 err = dsl_prop_register(ds, 592 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 593 compression_changed_cb, os); 594 } 595 if (err == 0) { 596 err = dsl_prop_register(ds, 597 zfs_prop_to_name(ZFS_PROP_COPIES), 598 copies_changed_cb, os); 599 } 600 if (err == 0) { 601 err = dsl_prop_register(ds, 602 zfs_prop_to_name(ZFS_PROP_DEDUP), 603 dedup_changed_cb, os); 604 } 605 if (err == 0) { 606 err = dsl_prop_register(ds, 607 zfs_prop_to_name(ZFS_PROP_LOGBIAS), 608 logbias_changed_cb, os); 609 } 610 if (err == 0) { 611 err = dsl_prop_register(ds, 612 zfs_prop_to_name(ZFS_PROP_SYNC), 613 sync_changed_cb, os); 614 } 615 if (err == 0) { 616 err = dsl_prop_register(ds, 617 zfs_prop_to_name( 618 ZFS_PROP_REDUNDANT_METADATA), 619 redundant_metadata_changed_cb, os); 620 } 621 if (err == 0) { 622 err = dsl_prop_register(ds, 623 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 624 recordsize_changed_cb, os); 625 } 626 if (err == 0) { 627 err = dsl_prop_register(ds, 628 zfs_prop_to_name(ZFS_PROP_DNODESIZE), 629 dnodesize_changed_cb, os); 630 } 631 if (err == 0) { 632 err = dsl_prop_register(ds, 633 zfs_prop_to_name( 634 ZFS_PROP_SPECIAL_SMALL_BLOCKS), 635 smallblk_changed_cb, os); 636 } 637 if (err == 0) { 638 err = dsl_prop_register(ds, 639 zfs_prop_to_name(ZFS_PROP_DIRECT), 640 direct_changed_cb, os); 641 } 642 } 643 if (err != 0) { 644 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 645 kmem_free(os, sizeof (objset_t)); 646 return (err); 647 } 648 } else { 649 /* It's the meta-objset. */ 650 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; 651 os->os_compress = ZIO_COMPRESS_ON; 652 os->os_complevel = ZIO_COMPLEVEL_DEFAULT; 653 os->os_encrypted = B_FALSE; 654 os->os_copies = spa_max_replication(spa); 655 os->os_dedup_checksum = ZIO_CHECKSUM_OFF; 656 os->os_dedup_verify = B_FALSE; 657 os->os_logbias = ZFS_LOGBIAS_LATENCY; 658 os->os_sync = ZFS_SYNC_STANDARD; 659 os->os_primary_cache = ZFS_CACHE_ALL; 660 os->os_secondary_cache = ZFS_CACHE_ALL; 661 os->os_dnodesize = DNODE_MIN_SIZE; 662 os->os_prefetch = ZFS_PREFETCH_ALL; 663 } 664 665 if (ds == NULL || !ds->ds_is_snapshot) 666 os->os_zil_header = os->os_phys->os_zil_header; 667 os->os_zil = zil_alloc(os, &os->os_zil_header); 668 669 for (i = 0; i < TXG_SIZE; i++) { 670 multilist_create(&os->os_dirty_dnodes[i], sizeof (dnode_t), 671 offsetof(dnode_t, dn_dirty_link[i]), 672 dnode_multilist_index_func); 673 } 674 list_create(&os->os_dnodes, sizeof (dnode_t), 675 offsetof(dnode_t, dn_link)); 676 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), 677 offsetof(dmu_buf_impl_t, db_link)); 678 679 list_link_init(&os->os_evicting_node); 680 681 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); 682 mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); 683 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); 684 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); 685 os->os_obj_next_percpu_len = boot_ncpus; 686 os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * 687 sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); 688 689 dnode_special_open(os, &os->os_phys->os_meta_dnode, 690 DMU_META_DNODE_OBJECT, &os->os_meta_dnode); 691 if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) { 692 dnode_special_open(os, &os->os_phys->os_userused_dnode, 693 DMU_USERUSED_OBJECT, &os->os_userused_dnode); 694 dnode_special_open(os, &os->os_phys->os_groupused_dnode, 695 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); 696 if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf)) 697 dnode_special_open(os, 698 &os->os_phys->os_projectused_dnode, 699 DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode); 700 } 701 702 mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL); 703 704 *osp = os; 705 return (0); 706 } 707 708 int 709 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) 710 { 711 int err = 0; 712 713 /* 714 * We need the pool_config lock to manipulate the dsl_dataset_t. 715 * Even if the dataset is long-held, we need the pool_config lock 716 * to open the objset, as it needs to get properties. 717 */ 718 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 719 720 mutex_enter(&ds->ds_opening_lock); 721 if (ds->ds_objset == NULL) { 722 objset_t *os; 723 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 724 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), 725 ds, dsl_dataset_get_blkptr(ds), &os); 726 rrw_exit(&ds->ds_bp_rwlock, FTAG); 727 728 if (err == 0) { 729 mutex_enter(&ds->ds_lock); 730 ASSERT(ds->ds_objset == NULL); 731 ds->ds_objset = os; 732 mutex_exit(&ds->ds_lock); 733 } 734 } 735 *osp = ds->ds_objset; 736 mutex_exit(&ds->ds_opening_lock); 737 return (err); 738 } 739 740 /* 741 * Holds the pool while the objset is held. Therefore only one objset 742 * can be held at a time. 743 */ 744 int 745 dmu_objset_hold_flags(const char *name, boolean_t decrypt, const void *tag, 746 objset_t **osp) 747 { 748 dsl_pool_t *dp; 749 dsl_dataset_t *ds; 750 int err; 751 ds_hold_flags_t flags; 752 753 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 754 err = dsl_pool_hold(name, tag, &dp); 755 if (err != 0) 756 return (err); 757 err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds); 758 if (err != 0) { 759 dsl_pool_rele(dp, tag); 760 return (err); 761 } 762 763 err = dmu_objset_from_ds(ds, osp); 764 if (err != 0) { 765 dsl_dataset_rele(ds, tag); 766 dsl_pool_rele(dp, tag); 767 } 768 769 return (err); 770 } 771 772 int 773 dmu_objset_hold(const char *name, const void *tag, objset_t **osp) 774 { 775 return (dmu_objset_hold_flags(name, B_FALSE, tag, osp)); 776 } 777 778 static int 779 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, 780 boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp) 781 { 782 (void) tag; 783 784 int err = dmu_objset_from_ds(ds, osp); 785 if (err != 0) { 786 return (err); 787 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { 788 return (SET_ERROR(EINVAL)); 789 } else if (!readonly && dsl_dataset_is_snapshot(ds)) { 790 return (SET_ERROR(EROFS)); 791 } else if (!readonly && decrypt && 792 dsl_dir_incompatible_encryption_version(ds->ds_dir)) { 793 return (SET_ERROR(EROFS)); 794 } 795 796 /* if we are decrypting, we can now check MACs in os->os_phys_buf */ 797 if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) { 798 zbookmark_phys_t zb; 799 800 SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT, 801 ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 802 err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa, 803 &zb, B_FALSE); 804 if (err != 0) 805 return (err); 806 807 ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf)); 808 } 809 810 return (0); 811 } 812 813 /* 814 * dsl_pool must not be held when this is called. 815 * Upon successful return, there will be a longhold on the dataset, 816 * and the dsl_pool will not be held. 817 */ 818 int 819 dmu_objset_own(const char *name, dmu_objset_type_t type, 820 boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp) 821 { 822 dsl_pool_t *dp; 823 dsl_dataset_t *ds; 824 int err; 825 ds_hold_flags_t flags; 826 827 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 828 err = dsl_pool_hold(name, FTAG, &dp); 829 if (err != 0) 830 return (err); 831 err = dsl_dataset_own(dp, name, flags, tag, &ds); 832 if (err != 0) { 833 dsl_pool_rele(dp, FTAG); 834 return (err); 835 } 836 err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); 837 if (err != 0) { 838 dsl_dataset_disown(ds, flags, tag); 839 dsl_pool_rele(dp, FTAG); 840 return (err); 841 } 842 843 /* 844 * User accounting requires the dataset to be decrypted and rw. 845 * We also don't begin user accounting during claiming to help 846 * speed up pool import times and to keep this txg reserved 847 * completely for recovery work. 848 */ 849 if (!readonly && !dp->dp_spa->spa_claiming && 850 (ds->ds_dir->dd_crypto_obj == 0 || decrypt)) { 851 if (dmu_objset_userobjspace_upgradable(*osp) || 852 dmu_objset_projectquota_upgradable(*osp)) { 853 dmu_objset_id_quota_upgrade(*osp); 854 } else if (dmu_objset_userused_enabled(*osp)) { 855 dmu_objset_userspace_upgrade(*osp); 856 } 857 } 858 859 dsl_pool_rele(dp, FTAG); 860 return (0); 861 } 862 863 int 864 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, 865 boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp) 866 { 867 dsl_dataset_t *ds; 868 int err; 869 ds_hold_flags_t flags; 870 871 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 872 err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds); 873 if (err != 0) 874 return (err); 875 876 err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); 877 if (err != 0) { 878 dsl_dataset_disown(ds, flags, tag); 879 return (err); 880 } 881 882 return (0); 883 } 884 885 void 886 dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, const void *tag) 887 { 888 ds_hold_flags_t flags; 889 dsl_pool_t *dp = dmu_objset_pool(os); 890 891 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 892 dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag); 893 dsl_pool_rele(dp, tag); 894 } 895 896 void 897 dmu_objset_rele(objset_t *os, const void *tag) 898 { 899 dmu_objset_rele_flags(os, B_FALSE, tag); 900 } 901 902 /* 903 * When we are called, os MUST refer to an objset associated with a dataset 904 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner 905 * == tag. We will then release and reacquire ownership of the dataset while 906 * holding the pool config_rwlock to avoid intervening namespace or ownership 907 * changes may occur. 908 * 909 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to 910 * release the hold on its dataset and acquire a new one on the dataset of the 911 * same name so that it can be partially torn down and reconstructed. 912 */ 913 void 914 dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds, 915 boolean_t decrypt, const void *tag) 916 { 917 dsl_pool_t *dp; 918 char name[ZFS_MAX_DATASET_NAME_LEN]; 919 ds_hold_flags_t flags; 920 921 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 922 VERIFY3P(ds, !=, NULL); 923 VERIFY3P(ds->ds_owner, ==, tag); 924 VERIFY(dsl_dataset_long_held(ds)); 925 926 dsl_dataset_name(ds, name); 927 dp = ds->ds_dir->dd_pool; 928 dsl_pool_config_enter(dp, FTAG); 929 dsl_dataset_disown(ds, flags, tag); 930 VERIFY0(dsl_dataset_own(dp, name, flags, tag, newds)); 931 dsl_pool_config_exit(dp, FTAG); 932 } 933 934 void 935 dmu_objset_disown(objset_t *os, boolean_t decrypt, const void *tag) 936 { 937 ds_hold_flags_t flags; 938 939 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 940 /* 941 * Stop upgrading thread 942 */ 943 dmu_objset_upgrade_stop(os); 944 dsl_dataset_disown(os->os_dsl_dataset, flags, tag); 945 } 946 947 void 948 dmu_objset_evict_dbufs(objset_t *os) 949 { 950 dnode_t *dn_marker; 951 dnode_t *dn; 952 953 dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP); 954 955 mutex_enter(&os->os_lock); 956 dn = list_head(&os->os_dnodes); 957 while (dn != NULL) { 958 /* 959 * Skip dnodes without holds. We have to do this dance 960 * because dnode_add_ref() only works if there is already a 961 * hold. If the dnode has no holds, then it has no dbufs. 962 */ 963 if (dnode_add_ref(dn, FTAG)) { 964 list_insert_after(&os->os_dnodes, dn, dn_marker); 965 mutex_exit(&os->os_lock); 966 967 dnode_evict_dbufs(dn); 968 dnode_rele(dn, FTAG); 969 970 mutex_enter(&os->os_lock); 971 dn = list_next(&os->os_dnodes, dn_marker); 972 list_remove(&os->os_dnodes, dn_marker); 973 } else { 974 dn = list_next(&os->os_dnodes, dn); 975 } 976 } 977 mutex_exit(&os->os_lock); 978 979 kmem_free(dn_marker, sizeof (dnode_t)); 980 981 if (DMU_USERUSED_DNODE(os) != NULL) { 982 if (DMU_PROJECTUSED_DNODE(os) != NULL) 983 dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os)); 984 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); 985 dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); 986 } 987 dnode_evict_dbufs(DMU_META_DNODE(os)); 988 } 989 990 /* 991 * Objset eviction processing is split into into two pieces. 992 * The first marks the objset as evicting, evicts any dbufs that 993 * have a refcount of zero, and then queues up the objset for the 994 * second phase of eviction. Once os->os_dnodes has been cleared by 995 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. 996 * The second phase closes the special dnodes, dequeues the objset from 997 * the list of those undergoing eviction, and finally frees the objset. 998 * 999 * NOTE: Due to asynchronous eviction processing (invocation of 1000 * dnode_buf_pageout()), it is possible for the meta dnode for the 1001 * objset to have no holds even though os->os_dnodes is not empty. 1002 */ 1003 void 1004 dmu_objset_evict(objset_t *os) 1005 { 1006 dsl_dataset_t *ds = os->os_dsl_dataset; 1007 1008 for (int t = 0; t < TXG_SIZE; t++) 1009 ASSERT(!dmu_objset_is_dirty(os, t)); 1010 1011 if (ds) 1012 dsl_prop_unregister_all(ds, os); 1013 1014 if (os->os_sa) 1015 sa_tear_down(os); 1016 1017 dmu_objset_evict_dbufs(os); 1018 1019 mutex_enter(&os->os_lock); 1020 spa_evicting_os_register(os->os_spa, os); 1021 if (list_is_empty(&os->os_dnodes)) { 1022 mutex_exit(&os->os_lock); 1023 dmu_objset_evict_done(os); 1024 } else { 1025 mutex_exit(&os->os_lock); 1026 } 1027 1028 1029 } 1030 1031 void 1032 dmu_objset_evict_done(objset_t *os) 1033 { 1034 ASSERT3P(list_head(&os->os_dnodes), ==, NULL); 1035 1036 dnode_special_close(&os->os_meta_dnode); 1037 if (DMU_USERUSED_DNODE(os)) { 1038 if (DMU_PROJECTUSED_DNODE(os)) 1039 dnode_special_close(&os->os_projectused_dnode); 1040 dnode_special_close(&os->os_userused_dnode); 1041 dnode_special_close(&os->os_groupused_dnode); 1042 } 1043 zil_free(os->os_zil); 1044 1045 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 1046 1047 /* 1048 * This is a barrier to prevent the objset from going away in 1049 * dnode_move() until we can safely ensure that the objset is still in 1050 * use. We consider the objset valid before the barrier and invalid 1051 * after the barrier. 1052 */ 1053 rw_enter(&os_lock, RW_READER); 1054 rw_exit(&os_lock); 1055 1056 kmem_free(os->os_obj_next_percpu, 1057 os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); 1058 1059 mutex_destroy(&os->os_lock); 1060 mutex_destroy(&os->os_userused_lock); 1061 mutex_destroy(&os->os_obj_lock); 1062 mutex_destroy(&os->os_user_ptr_lock); 1063 mutex_destroy(&os->os_upgrade_lock); 1064 for (int i = 0; i < TXG_SIZE; i++) 1065 multilist_destroy(&os->os_dirty_dnodes[i]); 1066 spa_evicting_os_deregister(os->os_spa, os); 1067 kmem_free(os, sizeof (objset_t)); 1068 } 1069 1070 inode_timespec_t 1071 dmu_objset_snap_cmtime(objset_t *os) 1072 { 1073 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); 1074 } 1075 1076 objset_t * 1077 dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 1078 dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx) 1079 { 1080 objset_t *os; 1081 dnode_t *mdn; 1082 1083 ASSERT(dmu_tx_is_syncing(tx)); 1084 1085 if (blksz == 0) 1086 blksz = DNODE_BLOCK_SIZE; 1087 if (ibs == 0) 1088 ibs = DN_MAX_INDBLKSHIFT; 1089 1090 if (ds != NULL) 1091 VERIFY0(dmu_objset_from_ds(ds, &os)); 1092 else 1093 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); 1094 1095 mdn = DMU_META_DNODE(os); 1096 1097 dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0, 1098 DNODE_MIN_SLOTS, tx); 1099 1100 /* 1101 * We don't want to have to increase the meta-dnode's nlevels 1102 * later, because then we could do it in quiescing context while 1103 * we are also accessing it in open context. 1104 * 1105 * This precaution is not necessary for the MOS (ds == NULL), 1106 * because the MOS is only updated in syncing context. 1107 * This is most fortunate: the MOS is the only objset that 1108 * needs to be synced multiple times as spa_sync() iterates 1109 * to convergence, so minimizing its dn_nlevels matters. 1110 */ 1111 if (ds != NULL) { 1112 if (levels == 0) { 1113 levels = 1; 1114 1115 /* 1116 * Determine the number of levels necessary for the 1117 * meta-dnode to contain DN_MAX_OBJECT dnodes. Note 1118 * that in order to ensure that we do not overflow 1119 * 64 bits, there has to be a nlevels that gives us a 1120 * number of blocks > DN_MAX_OBJECT but < 2^64. 1121 * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) 1122 * (10) must be less than (64 - log2(DN_MAX_OBJECT)) 1123 * (16). 1124 */ 1125 while ((uint64_t)mdn->dn_nblkptr << 1126 (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) * 1127 (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < 1128 DN_MAX_OBJECT) 1129 levels++; 1130 } 1131 1132 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = 1133 mdn->dn_nlevels = levels; 1134 } 1135 1136 ASSERT(type != DMU_OST_NONE); 1137 ASSERT(type != DMU_OST_ANY); 1138 ASSERT(type < DMU_OST_NUMTYPES); 1139 os->os_phys->os_type = type; 1140 1141 /* 1142 * Enable user accounting if it is enabled and this is not an 1143 * encrypted receive. 1144 */ 1145 if (dmu_objset_userused_enabled(os) && 1146 (!os->os_encrypted || !dmu_objset_is_receiving(os))) { 1147 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 1148 if (dmu_objset_userobjused_enabled(os)) { 1149 ASSERT3P(ds, !=, NULL); 1150 ds->ds_feature_activation[ 1151 SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; 1152 os->os_phys->os_flags |= 1153 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; 1154 } 1155 if (dmu_objset_projectquota_enabled(os)) { 1156 ASSERT3P(ds, !=, NULL); 1157 ds->ds_feature_activation[ 1158 SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; 1159 os->os_phys->os_flags |= 1160 OBJSET_FLAG_PROJECTQUOTA_COMPLETE; 1161 } 1162 os->os_flags = os->os_phys->os_flags; 1163 } 1164 1165 dsl_dataset_dirty(ds, tx); 1166 1167 return (os); 1168 } 1169 1170 /* called from dsl for meta-objset */ 1171 objset_t * 1172 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 1173 dmu_objset_type_t type, dmu_tx_t *tx) 1174 { 1175 return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx)); 1176 } 1177 1178 typedef struct dmu_objset_create_arg { 1179 const char *doca_name; 1180 cred_t *doca_cred; 1181 proc_t *doca_proc; 1182 void (*doca_userfunc)(objset_t *os, void *arg, 1183 cred_t *cr, dmu_tx_t *tx); 1184 void *doca_userarg; 1185 dmu_objset_type_t doca_type; 1186 uint64_t doca_flags; 1187 dsl_crypto_params_t *doca_dcp; 1188 } dmu_objset_create_arg_t; 1189 1190 static int 1191 dmu_objset_create_check(void *arg, dmu_tx_t *tx) 1192 { 1193 dmu_objset_create_arg_t *doca = arg; 1194 dsl_pool_t *dp = dmu_tx_pool(tx); 1195 dsl_dir_t *pdd; 1196 dsl_dataset_t *parentds; 1197 objset_t *parentos; 1198 const char *tail; 1199 int error; 1200 1201 if (strchr(doca->doca_name, '@') != NULL) 1202 return (SET_ERROR(EINVAL)); 1203 1204 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) 1205 return (SET_ERROR(ENAMETOOLONG)); 1206 1207 if (dataset_nestcheck(doca->doca_name) != 0) 1208 return (SET_ERROR(ENAMETOOLONG)); 1209 1210 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); 1211 if (error != 0) 1212 return (error); 1213 if (tail == NULL) { 1214 dsl_dir_rele(pdd, FTAG); 1215 return (SET_ERROR(EEXIST)); 1216 } 1217 1218 error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp, NULL); 1219 if (error != 0) { 1220 dsl_dir_rele(pdd, FTAG); 1221 return (error); 1222 } 1223 1224 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 1225 doca->doca_cred, doca->doca_proc); 1226 if (error != 0) { 1227 dsl_dir_rele(pdd, FTAG); 1228 return (error); 1229 } 1230 1231 /* can't create below anything but filesystems (eg. no ZVOLs) */ 1232 error = dsl_dataset_hold_obj(pdd->dd_pool, 1233 dsl_dir_phys(pdd)->dd_head_dataset_obj, FTAG, &parentds); 1234 if (error != 0) { 1235 dsl_dir_rele(pdd, FTAG); 1236 return (error); 1237 } 1238 error = dmu_objset_from_ds(parentds, &parentos); 1239 if (error != 0) { 1240 dsl_dataset_rele(parentds, FTAG); 1241 dsl_dir_rele(pdd, FTAG); 1242 return (error); 1243 } 1244 if (dmu_objset_type(parentos) != DMU_OST_ZFS) { 1245 dsl_dataset_rele(parentds, FTAG); 1246 dsl_dir_rele(pdd, FTAG); 1247 return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); 1248 } 1249 dsl_dataset_rele(parentds, FTAG); 1250 dsl_dir_rele(pdd, FTAG); 1251 1252 return (error); 1253 } 1254 1255 static void 1256 dmu_objset_create_sync(void *arg, dmu_tx_t *tx) 1257 { 1258 dmu_objset_create_arg_t *doca = arg; 1259 dsl_pool_t *dp = dmu_tx_pool(tx); 1260 spa_t *spa = dp->dp_spa; 1261 dsl_dir_t *pdd; 1262 const char *tail; 1263 dsl_dataset_t *ds; 1264 uint64_t obj; 1265 blkptr_t *bp; 1266 objset_t *os; 1267 zio_t *rzio; 1268 1269 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); 1270 1271 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, 1272 doca->doca_cred, doca->doca_dcp, tx); 1273 1274 VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj, 1275 DS_HOLD_FLAG_DECRYPT, FTAG, &ds)); 1276 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1277 bp = dsl_dataset_get_blkptr(ds); 1278 os = dmu_objset_create_impl(spa, ds, bp, doca->doca_type, tx); 1279 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1280 1281 if (doca->doca_userfunc != NULL) { 1282 doca->doca_userfunc(os, doca->doca_userarg, 1283 doca->doca_cred, tx); 1284 } 1285 1286 /* 1287 * The doca_userfunc() may write out some data that needs to be 1288 * encrypted if the dataset is encrypted (specifically the root 1289 * directory). This data must be written out before the encryption 1290 * key mapping is removed by dsl_dataset_rele_flags(). Force the 1291 * I/O to occur immediately by invoking the relevant sections of 1292 * dsl_pool_sync(). 1293 */ 1294 if (os->os_encrypted) { 1295 dsl_dataset_t *tmpds = NULL; 1296 boolean_t need_sync_done = B_FALSE; 1297 1298 mutex_enter(&ds->ds_lock); 1299 ds->ds_owner = FTAG; 1300 mutex_exit(&ds->ds_lock); 1301 1302 rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1303 tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, 1304 tx->tx_txg); 1305 if (tmpds != NULL) { 1306 dsl_dataset_sync(ds, rzio, tx); 1307 need_sync_done = B_TRUE; 1308 } 1309 VERIFY0(zio_wait(rzio)); 1310 1311 dmu_objset_sync_done(os, tx); 1312 taskq_wait(dp->dp_sync_taskq); 1313 if (txg_list_member(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1314 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1315 key_mapping_rele(spa, ds->ds_key_mapping, ds); 1316 } 1317 1318 rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1319 tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, 1320 tx->tx_txg); 1321 if (tmpds != NULL) { 1322 dmu_buf_rele(ds->ds_dbuf, ds); 1323 dsl_dataset_sync(ds, rzio, tx); 1324 } 1325 VERIFY0(zio_wait(rzio)); 1326 1327 if (need_sync_done) { 1328 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1329 key_mapping_rele(spa, ds->ds_key_mapping, ds); 1330 dsl_dataset_sync_done(ds, tx); 1331 dmu_buf_rele(ds->ds_dbuf, ds); 1332 } 1333 1334 mutex_enter(&ds->ds_lock); 1335 ds->ds_owner = NULL; 1336 mutex_exit(&ds->ds_lock); 1337 } 1338 1339 spa_history_log_internal_ds(ds, "create", tx, " "); 1340 1341 dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG); 1342 dsl_dir_rele(pdd, FTAG); 1343 } 1344 1345 int 1346 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 1347 dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg) 1348 { 1349 dmu_objset_create_arg_t doca; 1350 dsl_crypto_params_t tmp_dcp = { 0 }; 1351 1352 doca.doca_name = name; 1353 doca.doca_cred = CRED(); 1354 doca.doca_proc = curproc; 1355 doca.doca_flags = flags; 1356 doca.doca_userfunc = func; 1357 doca.doca_userarg = arg; 1358 doca.doca_type = type; 1359 1360 /* 1361 * Some callers (mostly for testing) do not provide a dcp on their 1362 * own but various code inside the sync task will require it to be 1363 * allocated. Rather than adding NULL checks throughout this code 1364 * or adding dummy dcp's to all of the callers we simply create a 1365 * dummy one here and use that. This zero dcp will have the same 1366 * effect as asking for inheritance of all encryption params. 1367 */ 1368 doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp; 1369 1370 int rv = dsl_sync_task(name, 1371 dmu_objset_create_check, dmu_objset_create_sync, &doca, 1372 6, ZFS_SPACE_CHECK_NORMAL); 1373 1374 if (rv == 0) 1375 zvol_create_minor(name); 1376 return (rv); 1377 } 1378 1379 typedef struct dmu_objset_clone_arg { 1380 const char *doca_clone; 1381 const char *doca_origin; 1382 cred_t *doca_cred; 1383 proc_t *doca_proc; 1384 } dmu_objset_clone_arg_t; 1385 1386 static int 1387 dmu_objset_clone_check(void *arg, dmu_tx_t *tx) 1388 { 1389 dmu_objset_clone_arg_t *doca = arg; 1390 dsl_dir_t *pdd; 1391 const char *tail; 1392 int error; 1393 dsl_dataset_t *origin; 1394 dsl_pool_t *dp = dmu_tx_pool(tx); 1395 1396 if (strchr(doca->doca_clone, '@') != NULL) 1397 return (SET_ERROR(EINVAL)); 1398 1399 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) 1400 return (SET_ERROR(ENAMETOOLONG)); 1401 1402 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); 1403 if (error != 0) 1404 return (error); 1405 if (tail == NULL) { 1406 dsl_dir_rele(pdd, FTAG); 1407 return (SET_ERROR(EEXIST)); 1408 } 1409 1410 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 1411 doca->doca_cred, doca->doca_proc); 1412 if (error != 0) { 1413 dsl_dir_rele(pdd, FTAG); 1414 return (SET_ERROR(EDQUOT)); 1415 } 1416 1417 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); 1418 if (error != 0) { 1419 dsl_dir_rele(pdd, FTAG); 1420 return (error); 1421 } 1422 1423 /* You can only clone snapshots, not the head datasets. */ 1424 if (!origin->ds_is_snapshot) { 1425 dsl_dataset_rele(origin, FTAG); 1426 dsl_dir_rele(pdd, FTAG); 1427 return (SET_ERROR(EINVAL)); 1428 } 1429 1430 dsl_dataset_rele(origin, FTAG); 1431 dsl_dir_rele(pdd, FTAG); 1432 1433 return (0); 1434 } 1435 1436 static void 1437 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) 1438 { 1439 dmu_objset_clone_arg_t *doca = arg; 1440 dsl_pool_t *dp = dmu_tx_pool(tx); 1441 dsl_dir_t *pdd; 1442 const char *tail; 1443 dsl_dataset_t *origin, *ds; 1444 uint64_t obj; 1445 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 1446 1447 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); 1448 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); 1449 1450 obj = dsl_dataset_create_sync(pdd, tail, origin, 0, 1451 doca->doca_cred, NULL, tx); 1452 1453 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 1454 dsl_dataset_name(origin, namebuf); 1455 spa_history_log_internal_ds(ds, "clone", tx, 1456 "origin=%s (%llu)", namebuf, (u_longlong_t)origin->ds_object); 1457 dsl_dataset_rele(ds, FTAG); 1458 dsl_dataset_rele(origin, FTAG); 1459 dsl_dir_rele(pdd, FTAG); 1460 } 1461 1462 int 1463 dmu_objset_clone(const char *clone, const char *origin) 1464 { 1465 dmu_objset_clone_arg_t doca; 1466 1467 doca.doca_clone = clone; 1468 doca.doca_origin = origin; 1469 doca.doca_cred = CRED(); 1470 doca.doca_proc = curproc; 1471 1472 int rv = dsl_sync_task(clone, 1473 dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 1474 6, ZFS_SPACE_CHECK_NORMAL); 1475 1476 if (rv == 0) 1477 zvol_create_minor(clone); 1478 1479 return (rv); 1480 } 1481 1482 int 1483 dmu_objset_snapshot_one(const char *fsname, const char *snapname) 1484 { 1485 int err; 1486 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); 1487 nvlist_t *snaps = fnvlist_alloc(); 1488 1489 fnvlist_add_boolean(snaps, longsnap); 1490 kmem_strfree(longsnap); 1491 err = dsl_dataset_snapshot(snaps, NULL, NULL); 1492 fnvlist_free(snaps); 1493 return (err); 1494 } 1495 1496 static void 1497 dmu_objset_upgrade_task_cb(void *data) 1498 { 1499 objset_t *os = data; 1500 1501 mutex_enter(&os->os_upgrade_lock); 1502 os->os_upgrade_status = EINTR; 1503 if (!os->os_upgrade_exit) { 1504 int status; 1505 1506 mutex_exit(&os->os_upgrade_lock); 1507 1508 status = os->os_upgrade_cb(os); 1509 1510 mutex_enter(&os->os_upgrade_lock); 1511 1512 os->os_upgrade_status = status; 1513 } 1514 os->os_upgrade_exit = B_TRUE; 1515 os->os_upgrade_id = 0; 1516 mutex_exit(&os->os_upgrade_lock); 1517 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1518 } 1519 1520 static void 1521 dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb) 1522 { 1523 if (os->os_upgrade_id != 0) 1524 return; 1525 1526 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 1527 dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag); 1528 1529 mutex_enter(&os->os_upgrade_lock); 1530 if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) { 1531 os->os_upgrade_exit = B_FALSE; 1532 os->os_upgrade_cb = cb; 1533 os->os_upgrade_id = taskq_dispatch( 1534 os->os_spa->spa_upgrade_taskq, 1535 dmu_objset_upgrade_task_cb, os, TQ_SLEEP); 1536 if (os->os_upgrade_id == TASKQID_INVALID) { 1537 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1538 os->os_upgrade_status = ENOMEM; 1539 } 1540 } else { 1541 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1542 } 1543 mutex_exit(&os->os_upgrade_lock); 1544 } 1545 1546 static void 1547 dmu_objset_upgrade_stop(objset_t *os) 1548 { 1549 mutex_enter(&os->os_upgrade_lock); 1550 os->os_upgrade_exit = B_TRUE; 1551 if (os->os_upgrade_id != 0) { 1552 taskqid_t id = os->os_upgrade_id; 1553 1554 os->os_upgrade_id = 0; 1555 mutex_exit(&os->os_upgrade_lock); 1556 1557 if ((taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id)) == 0) { 1558 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1559 } 1560 txg_wait_synced(os->os_spa->spa_dsl_pool, 0); 1561 } else { 1562 mutex_exit(&os->os_upgrade_lock); 1563 } 1564 } 1565 1566 static void 1567 dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) 1568 { 1569 dnode_t *dn; 1570 1571 while ((dn = multilist_sublist_head(list)) != NULL) { 1572 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 1573 ASSERT(dn->dn_dbuf->db_data_pending); 1574 /* 1575 * Initialize dn_zio outside dnode_sync() because the 1576 * meta-dnode needs to set it outside dnode_sync(). 1577 */ 1578 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; 1579 ASSERT(dn->dn_zio); 1580 1581 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); 1582 multilist_sublist_remove(list, dn); 1583 1584 /* 1585 * See the comment above dnode_rele_task() for an explanation 1586 * of why this dnode hold is always needed (even when not 1587 * doing user accounting). 1588 */ 1589 multilist_t *newlist = &dn->dn_objset->os_synced_dnodes; 1590 (void) dnode_add_ref(dn, newlist); 1591 multilist_insert(newlist, dn); 1592 1593 dnode_sync(dn, tx); 1594 } 1595 } 1596 1597 static void 1598 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) 1599 { 1600 (void) abuf; 1601 blkptr_t *bp = zio->io_bp; 1602 objset_t *os = arg; 1603 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; 1604 uint64_t fill = 0; 1605 1606 ASSERT(!BP_IS_EMBEDDED(bp)); 1607 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); 1608 ASSERT0(BP_GET_LEVEL(bp)); 1609 1610 /* 1611 * Update rootbp fill count: it should be the number of objects 1612 * allocated in the object set (not counting the "special" 1613 * objects that are stored in the objset_phys_t -- the meta 1614 * dnode and user/group/project accounting objects). 1615 */ 1616 for (int i = 0; i < dnp->dn_nblkptr; i++) 1617 fill += BP_GET_FILL(&dnp->dn_blkptr[i]); 1618 1619 BP_SET_FILL(bp, fill); 1620 1621 if (os->os_dsl_dataset != NULL) 1622 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); 1623 *os->os_rootbp = *bp; 1624 if (os->os_dsl_dataset != NULL) 1625 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); 1626 } 1627 1628 static void 1629 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) 1630 { 1631 (void) abuf; 1632 blkptr_t *bp = zio->io_bp; 1633 blkptr_t *bp_orig = &zio->io_bp_orig; 1634 objset_t *os = arg; 1635 1636 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { 1637 ASSERT(BP_EQUAL(bp, bp_orig)); 1638 } else { 1639 dsl_dataset_t *ds = os->os_dsl_dataset; 1640 dmu_tx_t *tx = os->os_synctx; 1641 1642 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); 1643 dsl_dataset_block_born(ds, bp, tx); 1644 } 1645 kmem_free(bp, sizeof (*bp)); 1646 } 1647 1648 typedef struct sync_objset_arg { 1649 zio_t *soa_zio; 1650 objset_t *soa_os; 1651 dmu_tx_t *soa_tx; 1652 kmutex_t soa_mutex; 1653 int soa_count; 1654 taskq_ent_t soa_tq_ent; 1655 } sync_objset_arg_t; 1656 1657 typedef struct sync_dnodes_arg { 1658 multilist_t *sda_list; 1659 int sda_sublist_idx; 1660 multilist_t *sda_newlist; 1661 sync_objset_arg_t *sda_soa; 1662 } sync_dnodes_arg_t; 1663 1664 static void sync_meta_dnode_task(void *arg); 1665 1666 static void 1667 sync_dnodes_task(void *arg) 1668 { 1669 sync_dnodes_arg_t *sda = arg; 1670 sync_objset_arg_t *soa = sda->sda_soa; 1671 objset_t *os = soa->soa_os; 1672 1673 uint_t allocator = spa_acq_allocator(os->os_spa); 1674 multilist_sublist_t *ms = 1675 multilist_sublist_lock_idx(sda->sda_list, sda->sda_sublist_idx); 1676 1677 dmu_objset_sync_dnodes(ms, soa->soa_tx); 1678 1679 multilist_sublist_unlock(ms); 1680 spa_rel_allocator(os->os_spa, allocator); 1681 1682 kmem_free(sda, sizeof (*sda)); 1683 1684 mutex_enter(&soa->soa_mutex); 1685 ASSERT(soa->soa_count != 0); 1686 if (--soa->soa_count != 0) { 1687 mutex_exit(&soa->soa_mutex); 1688 return; 1689 } 1690 mutex_exit(&soa->soa_mutex); 1691 1692 taskq_dispatch_ent(dmu_objset_pool(os)->dp_sync_taskq, 1693 sync_meta_dnode_task, soa, TQ_FRONT, &soa->soa_tq_ent); 1694 } 1695 1696 /* 1697 * Issue the zio_nowait() for all dirty record zios on the meta dnode, 1698 * then trigger the callback for the zil_sync. This runs once for each 1699 * objset, only after any/all sublists in the objset have been synced. 1700 */ 1701 static void 1702 sync_meta_dnode_task(void *arg) 1703 { 1704 sync_objset_arg_t *soa = arg; 1705 objset_t *os = soa->soa_os; 1706 dmu_tx_t *tx = soa->soa_tx; 1707 int txgoff = tx->tx_txg & TXG_MASK; 1708 dbuf_dirty_record_t *dr; 1709 1710 ASSERT0(soa->soa_count); 1711 1712 list_t *list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; 1713 while ((dr = list_remove_head(list)) != NULL) { 1714 ASSERT0(dr->dr_dbuf->db_level); 1715 zio_nowait(dr->dr_zio); 1716 } 1717 1718 /* Enable dnode backfill if enough objects have been freed. */ 1719 if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { 1720 os->os_rescan_dnodes = B_TRUE; 1721 os->os_freed_dnodes = 0; 1722 } 1723 1724 /* 1725 * Free intent log blocks up to this tx. 1726 */ 1727 zil_sync(os->os_zil, tx); 1728 os->os_phys->os_zil_header = os->os_zil_header; 1729 zio_nowait(soa->soa_zio); 1730 1731 mutex_destroy(&soa->soa_mutex); 1732 kmem_free(soa, sizeof (*soa)); 1733 } 1734 1735 /* called from dsl */ 1736 void 1737 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) 1738 { 1739 int txgoff; 1740 zbookmark_phys_t zb; 1741 zio_prop_t zp; 1742 zio_t *zio; 1743 int num_sublists; 1744 multilist_t *ml; 1745 blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); 1746 *blkptr_copy = *os->os_rootbp; 1747 1748 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", (u_longlong_t)tx->tx_txg); 1749 1750 ASSERT(dmu_tx_is_syncing(tx)); 1751 /* XXX the write_done callback should really give us the tx... */ 1752 os->os_synctx = tx; 1753 1754 if (os->os_dsl_dataset == NULL) { 1755 /* 1756 * This is the MOS. If we have upgraded, 1757 * spa_max_replication() could change, so reset 1758 * os_copies here. 1759 */ 1760 os->os_copies = spa_max_replication(os->os_spa); 1761 } 1762 1763 /* 1764 * Create the root block IO 1765 */ 1766 SET_BOOKMARK(&zb, os->os_dsl_dataset ? 1767 os->os_dsl_dataset->ds_object : DMU_META_OBJSET, 1768 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 1769 arc_release(os->os_phys_buf, &os->os_phys_buf); 1770 1771 dmu_write_policy(os, NULL, 0, 0, &zp); 1772 1773 /* 1774 * If we are either claiming the ZIL or doing a raw receive, write 1775 * out the os_phys_buf raw. Neither of these actions will effect the 1776 * MAC at this point. 1777 */ 1778 if (os->os_raw_receive || 1779 os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { 1780 ASSERT(os->os_encrypted); 1781 arc_convert_to_raw(os->os_phys_buf, 1782 os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER, 1783 DMU_OT_OBJSET, NULL, NULL, NULL); 1784 } 1785 1786 zio = arc_write(pio, os->os_spa, tx->tx_txg, 1787 blkptr_copy, os->os_phys_buf, B_FALSE, dmu_os_is_l2cacheable(os), 1788 &zp, dmu_objset_write_ready, NULL, dmu_objset_write_done, 1789 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); 1790 1791 /* 1792 * Sync special dnodes - the parent IO for the sync is the root block 1793 */ 1794 DMU_META_DNODE(os)->dn_zio = zio; 1795 dnode_sync(DMU_META_DNODE(os), tx); 1796 1797 os->os_phys->os_flags = os->os_flags; 1798 1799 if (DMU_USERUSED_DNODE(os) && 1800 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1801 DMU_USERUSED_DNODE(os)->dn_zio = zio; 1802 dnode_sync(DMU_USERUSED_DNODE(os), tx); 1803 DMU_GROUPUSED_DNODE(os)->dn_zio = zio; 1804 dnode_sync(DMU_GROUPUSED_DNODE(os), tx); 1805 } 1806 1807 if (DMU_PROJECTUSED_DNODE(os) && 1808 DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1809 DMU_PROJECTUSED_DNODE(os)->dn_zio = zio; 1810 dnode_sync(DMU_PROJECTUSED_DNODE(os), tx); 1811 } 1812 1813 txgoff = tx->tx_txg & TXG_MASK; 1814 1815 /* 1816 * We must create the list here because it uses the 1817 * dn_dirty_link[] of this txg. But it may already 1818 * exist because we call dsl_dataset_sync() twice per txg. 1819 */ 1820 if (os->os_synced_dnodes.ml_sublists == NULL) { 1821 multilist_create(&os->os_synced_dnodes, sizeof (dnode_t), 1822 offsetof(dnode_t, dn_dirty_link[txgoff]), 1823 dnode_multilist_index_func); 1824 } else { 1825 ASSERT3U(os->os_synced_dnodes.ml_offset, ==, 1826 offsetof(dnode_t, dn_dirty_link[txgoff])); 1827 } 1828 1829 /* 1830 * zio_nowait(zio) is done after any/all sublist and meta dnode 1831 * zios have been nowaited, and the zil_sync() has been performed. 1832 * The soa is freed at the end of sync_meta_dnode_task. 1833 */ 1834 sync_objset_arg_t *soa = kmem_alloc(sizeof (*soa), KM_SLEEP); 1835 soa->soa_zio = zio; 1836 soa->soa_os = os; 1837 soa->soa_tx = tx; 1838 taskq_init_ent(&soa->soa_tq_ent); 1839 mutex_init(&soa->soa_mutex, NULL, MUTEX_DEFAULT, NULL); 1840 1841 ml = &os->os_dirty_dnodes[txgoff]; 1842 soa->soa_count = num_sublists = multilist_get_num_sublists(ml); 1843 1844 for (int i = 0; i < num_sublists; i++) { 1845 if (multilist_sublist_is_empty_idx(ml, i)) 1846 soa->soa_count--; 1847 } 1848 1849 if (soa->soa_count == 0) { 1850 taskq_dispatch_ent(dmu_objset_pool(os)->dp_sync_taskq, 1851 sync_meta_dnode_task, soa, TQ_FRONT, &soa->soa_tq_ent); 1852 } else { 1853 /* 1854 * Sync sublists in parallel. The last to finish 1855 * (i.e., when soa->soa_count reaches zero) must 1856 * dispatch sync_meta_dnode_task. 1857 */ 1858 for (int i = 0; i < num_sublists; i++) { 1859 if (multilist_sublist_is_empty_idx(ml, i)) 1860 continue; 1861 sync_dnodes_arg_t *sda = 1862 kmem_alloc(sizeof (*sda), KM_SLEEP); 1863 sda->sda_list = ml; 1864 sda->sda_sublist_idx = i; 1865 sda->sda_soa = soa; 1866 (void) taskq_dispatch( 1867 dmu_objset_pool(os)->dp_sync_taskq, 1868 sync_dnodes_task, sda, 0); 1869 /* sync_dnodes_task frees sda */ 1870 } 1871 } 1872 } 1873 1874 boolean_t 1875 dmu_objset_is_dirty(objset_t *os, uint64_t txg) 1876 { 1877 return (!multilist_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK])); 1878 } 1879 1880 static file_info_cb_t *file_cbs[DMU_OST_NUMTYPES]; 1881 1882 void 1883 dmu_objset_register_type(dmu_objset_type_t ost, file_info_cb_t *cb) 1884 { 1885 file_cbs[ost] = cb; 1886 } 1887 1888 int 1889 dmu_get_file_info(objset_t *os, dmu_object_type_t bonustype, const void *data, 1890 zfs_file_info_t *zfi) 1891 { 1892 file_info_cb_t *cb = file_cbs[os->os_phys->os_type]; 1893 if (cb == NULL) 1894 return (EINVAL); 1895 return (cb(bonustype, data, zfi)); 1896 } 1897 1898 boolean_t 1899 dmu_objset_userused_enabled(objset_t *os) 1900 { 1901 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && 1902 file_cbs[os->os_phys->os_type] != NULL && 1903 DMU_USERUSED_DNODE(os) != NULL); 1904 } 1905 1906 boolean_t 1907 dmu_objset_userobjused_enabled(objset_t *os) 1908 { 1909 return (dmu_objset_userused_enabled(os) && 1910 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING)); 1911 } 1912 1913 boolean_t 1914 dmu_objset_projectquota_enabled(objset_t *os) 1915 { 1916 return (file_cbs[os->os_phys->os_type] != NULL && 1917 DMU_PROJECTUSED_DNODE(os) != NULL && 1918 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA)); 1919 } 1920 1921 typedef struct userquota_node { 1922 /* must be in the first filed, see userquota_update_cache() */ 1923 char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN]; 1924 int64_t uqn_delta; 1925 avl_node_t uqn_node; 1926 } userquota_node_t; 1927 1928 typedef struct userquota_cache { 1929 avl_tree_t uqc_user_deltas; 1930 avl_tree_t uqc_group_deltas; 1931 avl_tree_t uqc_project_deltas; 1932 } userquota_cache_t; 1933 1934 static int 1935 userquota_compare(const void *l, const void *r) 1936 { 1937 const userquota_node_t *luqn = l; 1938 const userquota_node_t *ruqn = r; 1939 int rv; 1940 1941 /* 1942 * NB: can only access uqn_id because userquota_update_cache() doesn't 1943 * pass in an entire userquota_node_t. 1944 */ 1945 rv = strcmp(luqn->uqn_id, ruqn->uqn_id); 1946 1947 return (TREE_ISIGN(rv)); 1948 } 1949 1950 static void 1951 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) 1952 { 1953 void *cookie; 1954 userquota_node_t *uqn; 1955 1956 ASSERT(dmu_tx_is_syncing(tx)); 1957 1958 cookie = NULL; 1959 while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, 1960 &cookie)) != NULL) { 1961 /* 1962 * os_userused_lock protects against concurrent calls to 1963 * zap_increment_int(). It's needed because zap_increment_int() 1964 * is not thread-safe (i.e. not atomic). 1965 */ 1966 mutex_enter(&os->os_userused_lock); 1967 VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT, 1968 uqn->uqn_id, uqn->uqn_delta, tx)); 1969 mutex_exit(&os->os_userused_lock); 1970 kmem_free(uqn, sizeof (*uqn)); 1971 } 1972 avl_destroy(&cache->uqc_user_deltas); 1973 1974 cookie = NULL; 1975 while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, 1976 &cookie)) != NULL) { 1977 mutex_enter(&os->os_userused_lock); 1978 VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT, 1979 uqn->uqn_id, uqn->uqn_delta, tx)); 1980 mutex_exit(&os->os_userused_lock); 1981 kmem_free(uqn, sizeof (*uqn)); 1982 } 1983 avl_destroy(&cache->uqc_group_deltas); 1984 1985 if (dmu_objset_projectquota_enabled(os)) { 1986 cookie = NULL; 1987 while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas, 1988 &cookie)) != NULL) { 1989 mutex_enter(&os->os_userused_lock); 1990 VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT, 1991 uqn->uqn_id, uqn->uqn_delta, tx)); 1992 mutex_exit(&os->os_userused_lock); 1993 kmem_free(uqn, sizeof (*uqn)); 1994 } 1995 avl_destroy(&cache->uqc_project_deltas); 1996 } 1997 } 1998 1999 static void 2000 userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta) 2001 { 2002 userquota_node_t *uqn; 2003 avl_index_t idx; 2004 2005 ASSERT(strlen(id) < sizeof (uqn->uqn_id)); 2006 /* 2007 * Use id directly for searching because uqn_id is the first field of 2008 * userquota_node_t and fields after uqn_id won't be accessed in 2009 * avl_find(). 2010 */ 2011 uqn = avl_find(avl, (const void *)id, &idx); 2012 if (uqn == NULL) { 2013 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); 2014 strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id)); 2015 avl_insert(avl, uqn, idx); 2016 } 2017 uqn->uqn_delta += delta; 2018 } 2019 2020 static void 2021 do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used, 2022 uint64_t flags, uint64_t user, uint64_t group, uint64_t project, 2023 boolean_t subtract) 2024 { 2025 if (flags & DNODE_FLAG_USERUSED_ACCOUNTED) { 2026 int64_t delta = DNODE_MIN_SIZE + used; 2027 char name[20]; 2028 2029 if (subtract) 2030 delta = -delta; 2031 2032 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)user); 2033 userquota_update_cache(&cache->uqc_user_deltas, name, delta); 2034 2035 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)group); 2036 userquota_update_cache(&cache->uqc_group_deltas, name, delta); 2037 2038 if (dmu_objset_projectquota_enabled(os)) { 2039 (void) snprintf(name, sizeof (name), "%llx", 2040 (longlong_t)project); 2041 userquota_update_cache(&cache->uqc_project_deltas, 2042 name, delta); 2043 } 2044 } 2045 } 2046 2047 static void 2048 do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags, 2049 uint64_t user, uint64_t group, uint64_t project, boolean_t subtract) 2050 { 2051 if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) { 2052 char name[20 + DMU_OBJACCT_PREFIX_LEN]; 2053 int delta = subtract ? -1 : 1; 2054 2055 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", 2056 (longlong_t)user); 2057 userquota_update_cache(&cache->uqc_user_deltas, name, delta); 2058 2059 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", 2060 (longlong_t)group); 2061 userquota_update_cache(&cache->uqc_group_deltas, name, delta); 2062 2063 if (dmu_objset_projectquota_enabled(os)) { 2064 (void) snprintf(name, sizeof (name), 2065 DMU_OBJACCT_PREFIX "%llx", (longlong_t)project); 2066 userquota_update_cache(&cache->uqc_project_deltas, 2067 name, delta); 2068 } 2069 } 2070 } 2071 2072 typedef struct userquota_updates_arg { 2073 objset_t *uua_os; 2074 int uua_sublist_idx; 2075 dmu_tx_t *uua_tx; 2076 } userquota_updates_arg_t; 2077 2078 static void 2079 userquota_updates_task(void *arg) 2080 { 2081 userquota_updates_arg_t *uua = arg; 2082 objset_t *os = uua->uua_os; 2083 dmu_tx_t *tx = uua->uua_tx; 2084 dnode_t *dn; 2085 userquota_cache_t cache = { { 0 } }; 2086 2087 multilist_sublist_t *list = multilist_sublist_lock_idx( 2088 &os->os_synced_dnodes, uua->uua_sublist_idx); 2089 2090 ASSERT(multilist_sublist_head(list) == NULL || 2091 dmu_objset_userused_enabled(os)); 2092 avl_create(&cache.uqc_user_deltas, userquota_compare, 2093 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 2094 avl_create(&cache.uqc_group_deltas, userquota_compare, 2095 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 2096 if (dmu_objset_projectquota_enabled(os)) 2097 avl_create(&cache.uqc_project_deltas, userquota_compare, 2098 sizeof (userquota_node_t), offsetof(userquota_node_t, 2099 uqn_node)); 2100 2101 while ((dn = multilist_sublist_head(list)) != NULL) { 2102 int flags; 2103 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); 2104 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || 2105 dn->dn_phys->dn_flags & 2106 DNODE_FLAG_USERUSED_ACCOUNTED); 2107 2108 flags = dn->dn_id_flags; 2109 ASSERT(flags); 2110 if (flags & DN_ID_OLD_EXIST) { 2111 do_userquota_update(os, &cache, dn->dn_oldused, 2112 dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, 2113 dn->dn_oldprojid, B_TRUE); 2114 do_userobjquota_update(os, &cache, dn->dn_oldflags, 2115 dn->dn_olduid, dn->dn_oldgid, 2116 dn->dn_oldprojid, B_TRUE); 2117 } 2118 if (flags & DN_ID_NEW_EXIST) { 2119 do_userquota_update(os, &cache, 2120 DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, 2121 dn->dn_newuid, dn->dn_newgid, 2122 dn->dn_newprojid, B_FALSE); 2123 do_userobjquota_update(os, &cache, 2124 dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, 2125 dn->dn_newprojid, B_FALSE); 2126 } 2127 2128 mutex_enter(&dn->dn_mtx); 2129 dn->dn_oldused = 0; 2130 dn->dn_oldflags = 0; 2131 if (dn->dn_id_flags & DN_ID_NEW_EXIST) { 2132 dn->dn_olduid = dn->dn_newuid; 2133 dn->dn_oldgid = dn->dn_newgid; 2134 dn->dn_oldprojid = dn->dn_newprojid; 2135 dn->dn_id_flags |= DN_ID_OLD_EXIST; 2136 if (dn->dn_bonuslen == 0) 2137 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 2138 else 2139 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2140 } 2141 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); 2142 mutex_exit(&dn->dn_mtx); 2143 2144 multilist_sublist_remove(list, dn); 2145 dnode_rele(dn, &os->os_synced_dnodes); 2146 } 2147 do_userquota_cacheflush(os, &cache, tx); 2148 multilist_sublist_unlock(list); 2149 kmem_free(uua, sizeof (*uua)); 2150 } 2151 2152 /* 2153 * Release dnode holds from dmu_objset_sync_dnodes(). When the dnode is being 2154 * synced (i.e. we have issued the zio's for blocks in the dnode), it can't be 2155 * evicted because the block containing the dnode can't be evicted until it is 2156 * written out. However, this hold is necessary to prevent the dnode_t from 2157 * being moved (via dnode_move()) while it's still referenced by 2158 * dbuf_dirty_record_t:dr_dnode. And dr_dnode is needed for 2159 * dirty_lightweight_leaf-type dirty records. 2160 * 2161 * If we are doing user-object accounting, the dnode_rele() happens from 2162 * userquota_updates_task() instead. 2163 */ 2164 static void 2165 dnode_rele_task(void *arg) 2166 { 2167 userquota_updates_arg_t *uua = arg; 2168 objset_t *os = uua->uua_os; 2169 2170 multilist_sublist_t *list = multilist_sublist_lock_idx( 2171 &os->os_synced_dnodes, uua->uua_sublist_idx); 2172 2173 dnode_t *dn; 2174 while ((dn = multilist_sublist_head(list)) != NULL) { 2175 multilist_sublist_remove(list, dn); 2176 dnode_rele(dn, &os->os_synced_dnodes); 2177 } 2178 multilist_sublist_unlock(list); 2179 kmem_free(uua, sizeof (*uua)); 2180 } 2181 2182 /* 2183 * Return TRUE if userquota updates are needed. 2184 */ 2185 static boolean_t 2186 dmu_objset_do_userquota_updates_prep(objset_t *os, dmu_tx_t *tx) 2187 { 2188 if (!dmu_objset_userused_enabled(os)) 2189 return (B_FALSE); 2190 2191 /* 2192 * If this is a raw receive just return and handle accounting 2193 * later when we have the keys loaded. We also don't do user 2194 * accounting during claiming since the datasets are not owned 2195 * for the duration of claiming and this txg should only be 2196 * used for recovery. 2197 */ 2198 if (os->os_encrypted && dmu_objset_is_receiving(os)) 2199 return (B_FALSE); 2200 2201 if (tx->tx_txg <= os->os_spa->spa_claim_max_txg) 2202 return (B_FALSE); 2203 2204 /* Allocate the user/group/project used objects if necessary. */ 2205 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 2206 VERIFY0(zap_create_claim(os, 2207 DMU_USERUSED_OBJECT, 2208 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2209 VERIFY0(zap_create_claim(os, 2210 DMU_GROUPUSED_OBJECT, 2211 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2212 } 2213 2214 if (dmu_objset_projectquota_enabled(os) && 2215 DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 2216 VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT, 2217 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2218 } 2219 return (B_TRUE); 2220 } 2221 2222 /* 2223 * Dispatch taskq tasks to dp_sync_taskq to update the user accounting, and 2224 * also release the holds on the dnodes from dmu_objset_sync_dnodes(). 2225 * The caller must taskq_wait(dp_sync_taskq). 2226 */ 2227 void 2228 dmu_objset_sync_done(objset_t *os, dmu_tx_t *tx) 2229 { 2230 boolean_t need_userquota = dmu_objset_do_userquota_updates_prep(os, tx); 2231 2232 int num_sublists = multilist_get_num_sublists(&os->os_synced_dnodes); 2233 for (int i = 0; i < num_sublists; i++) { 2234 userquota_updates_arg_t *uua = 2235 kmem_alloc(sizeof (*uua), KM_SLEEP); 2236 uua->uua_os = os; 2237 uua->uua_sublist_idx = i; 2238 uua->uua_tx = tx; 2239 2240 /* 2241 * If we don't need to update userquotas, use 2242 * dnode_rele_task() to call dnode_rele() 2243 */ 2244 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, 2245 need_userquota ? userquota_updates_task : dnode_rele_task, 2246 uua, 0); 2247 /* callback frees uua */ 2248 } 2249 } 2250 2251 2252 /* 2253 * Returns a pointer to data to find uid/gid from 2254 * 2255 * If a dirty record for transaction group that is syncing can't 2256 * be found then NULL is returned. In the NULL case it is assumed 2257 * the uid/gid aren't changing. 2258 */ 2259 static void * 2260 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) 2261 { 2262 dbuf_dirty_record_t *dr; 2263 void *data; 2264 2265 if (db->db_dirtycnt == 0) 2266 return (db->db.db_data); /* Nothing is changing */ 2267 2268 dr = dbuf_find_dirty_eq(db, tx->tx_txg); 2269 2270 if (dr == NULL) { 2271 data = NULL; 2272 } else { 2273 if (dr->dr_dnode->dn_bonuslen == 0 && 2274 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) 2275 data = dr->dt.dl.dr_data->b_data; 2276 else 2277 data = dr->dt.dl.dr_data; 2278 } 2279 2280 return (data); 2281 } 2282 2283 void 2284 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) 2285 { 2286 objset_t *os = dn->dn_objset; 2287 void *data = NULL; 2288 dmu_buf_impl_t *db = NULL; 2289 int flags = dn->dn_id_flags; 2290 int error; 2291 boolean_t have_spill = B_FALSE; 2292 2293 if (!dmu_objset_userused_enabled(dn->dn_objset)) 2294 return; 2295 2296 /* 2297 * Raw receives introduce a problem with user accounting. Raw 2298 * receives cannot update the user accounting info because the 2299 * user ids and the sizes are encrypted. To guarantee that we 2300 * never end up with bad user accounting, we simply disable it 2301 * during raw receives. We also disable this for normal receives 2302 * so that an incremental raw receive may be done on top of an 2303 * existing non-raw receive. 2304 */ 2305 if (os->os_encrypted && dmu_objset_is_receiving(os)) 2306 return; 2307 2308 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| 2309 DN_ID_CHKED_SPILL))) 2310 return; 2311 2312 if (before && dn->dn_bonuslen != 0) 2313 data = DN_BONUS(dn->dn_phys); 2314 else if (!before && dn->dn_bonuslen != 0) { 2315 if (dn->dn_bonus) { 2316 db = dn->dn_bonus; 2317 mutex_enter(&db->db_mtx); 2318 data = dmu_objset_userquota_find_data(db, tx); 2319 } else { 2320 data = DN_BONUS(dn->dn_phys); 2321 } 2322 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { 2323 int rf = 0; 2324 2325 if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) 2326 rf |= DB_RF_HAVESTRUCT; 2327 error = dmu_spill_hold_by_dnode(dn, 2328 rf | DB_RF_MUST_SUCCEED, 2329 FTAG, (dmu_buf_t **)&db); 2330 ASSERT(error == 0); 2331 mutex_enter(&db->db_mtx); 2332 data = (before) ? db->db.db_data : 2333 dmu_objset_userquota_find_data(db, tx); 2334 have_spill = B_TRUE; 2335 } else { 2336 mutex_enter(&dn->dn_mtx); 2337 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2338 mutex_exit(&dn->dn_mtx); 2339 return; 2340 } 2341 2342 /* 2343 * Must always call the callback in case the object 2344 * type has changed and that type isn't an object type to track 2345 */ 2346 zfs_file_info_t zfi; 2347 error = file_cbs[os->os_phys->os_type](dn->dn_bonustype, data, &zfi); 2348 2349 if (before) { 2350 ASSERT(data); 2351 dn->dn_olduid = zfi.zfi_user; 2352 dn->dn_oldgid = zfi.zfi_group; 2353 dn->dn_oldprojid = zfi.zfi_project; 2354 } else if (data) { 2355 dn->dn_newuid = zfi.zfi_user; 2356 dn->dn_newgid = zfi.zfi_group; 2357 dn->dn_newprojid = zfi.zfi_project; 2358 } 2359 2360 /* 2361 * Preserve existing uid/gid when the callback can't determine 2362 * what the new uid/gid are and the callback returned EEXIST. 2363 * The EEXIST error tells us to just use the existing uid/gid. 2364 * If we don't know what the old values are then just assign 2365 * them to 0, since that is a new file being created. 2366 */ 2367 if (!before && data == NULL && error == EEXIST) { 2368 if (flags & DN_ID_OLD_EXIST) { 2369 dn->dn_newuid = dn->dn_olduid; 2370 dn->dn_newgid = dn->dn_oldgid; 2371 dn->dn_newprojid = dn->dn_oldprojid; 2372 } else { 2373 dn->dn_newuid = 0; 2374 dn->dn_newgid = 0; 2375 dn->dn_newprojid = ZFS_DEFAULT_PROJID; 2376 } 2377 error = 0; 2378 } 2379 2380 if (db) 2381 mutex_exit(&db->db_mtx); 2382 2383 mutex_enter(&dn->dn_mtx); 2384 if (error == 0 && before) 2385 dn->dn_id_flags |= DN_ID_OLD_EXIST; 2386 if (error == 0 && !before) 2387 dn->dn_id_flags |= DN_ID_NEW_EXIST; 2388 2389 if (have_spill) { 2390 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 2391 } else { 2392 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2393 } 2394 mutex_exit(&dn->dn_mtx); 2395 if (have_spill) 2396 dmu_buf_rele((dmu_buf_t *)db, FTAG); 2397 } 2398 2399 boolean_t 2400 dmu_objset_userspace_present(objset_t *os) 2401 { 2402 return (os->os_phys->os_flags & 2403 OBJSET_FLAG_USERACCOUNTING_COMPLETE); 2404 } 2405 2406 boolean_t 2407 dmu_objset_userobjspace_present(objset_t *os) 2408 { 2409 return (os->os_phys->os_flags & 2410 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE); 2411 } 2412 2413 boolean_t 2414 dmu_objset_projectquota_present(objset_t *os) 2415 { 2416 return (os->os_phys->os_flags & 2417 OBJSET_FLAG_PROJECTQUOTA_COMPLETE); 2418 } 2419 2420 static int 2421 dmu_objset_space_upgrade(objset_t *os) 2422 { 2423 uint64_t obj; 2424 int err = 0; 2425 2426 /* 2427 * We simply need to mark every object dirty, so that it will be 2428 * synced out and now accounted. If this is called 2429 * concurrently, or if we already did some work before crashing, 2430 * that's fine, since we track each object's accounted state 2431 * independently. 2432 */ 2433 2434 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { 2435 dmu_tx_t *tx; 2436 dmu_buf_t *db; 2437 int objerr; 2438 2439 mutex_enter(&os->os_upgrade_lock); 2440 if (os->os_upgrade_exit) 2441 err = SET_ERROR(EINTR); 2442 mutex_exit(&os->os_upgrade_lock); 2443 if (err != 0) 2444 return (err); 2445 2446 if (issig()) 2447 return (SET_ERROR(EINTR)); 2448 2449 objerr = dmu_bonus_hold(os, obj, FTAG, &db); 2450 if (objerr != 0) 2451 continue; 2452 tx = dmu_tx_create(os); 2453 dmu_tx_hold_bonus(tx, obj); 2454 objerr = dmu_tx_assign(tx, TXG_WAIT); 2455 if (objerr != 0) { 2456 dmu_buf_rele(db, FTAG); 2457 dmu_tx_abort(tx); 2458 continue; 2459 } 2460 dmu_buf_will_dirty(db, tx); 2461 dmu_buf_rele(db, FTAG); 2462 dmu_tx_commit(tx); 2463 } 2464 return (0); 2465 } 2466 2467 static int 2468 dmu_objset_userspace_upgrade_cb(objset_t *os) 2469 { 2470 int err = 0; 2471 2472 if (dmu_objset_userspace_present(os)) 2473 return (0); 2474 if (dmu_objset_is_snapshot(os)) 2475 return (SET_ERROR(EINVAL)); 2476 if (!dmu_objset_userused_enabled(os)) 2477 return (SET_ERROR(ENOTSUP)); 2478 2479 err = dmu_objset_space_upgrade(os); 2480 if (err) 2481 return (err); 2482 2483 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 2484 txg_wait_synced(dmu_objset_pool(os), 0); 2485 return (0); 2486 } 2487 2488 void 2489 dmu_objset_userspace_upgrade(objset_t *os) 2490 { 2491 dmu_objset_upgrade(os, dmu_objset_userspace_upgrade_cb); 2492 } 2493 2494 static int 2495 dmu_objset_id_quota_upgrade_cb(objset_t *os) 2496 { 2497 int err = 0; 2498 2499 if (dmu_objset_userobjspace_present(os) && 2500 dmu_objset_projectquota_present(os)) 2501 return (0); 2502 if (dmu_objset_is_snapshot(os)) 2503 return (SET_ERROR(EINVAL)); 2504 if (!dmu_objset_userused_enabled(os)) 2505 return (SET_ERROR(ENOTSUP)); 2506 if (!dmu_objset_projectquota_enabled(os) && 2507 dmu_objset_userobjspace_present(os)) 2508 return (SET_ERROR(ENOTSUP)); 2509 2510 if (dmu_objset_userobjused_enabled(os)) 2511 dmu_objset_ds(os)->ds_feature_activation[ 2512 SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; 2513 if (dmu_objset_projectquota_enabled(os)) 2514 dmu_objset_ds(os)->ds_feature_activation[ 2515 SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; 2516 2517 err = dmu_objset_space_upgrade(os); 2518 if (err) 2519 return (err); 2520 2521 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 2522 if (dmu_objset_userobjused_enabled(os)) 2523 os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; 2524 if (dmu_objset_projectquota_enabled(os)) 2525 os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE; 2526 2527 txg_wait_synced(dmu_objset_pool(os), 0); 2528 return (0); 2529 } 2530 2531 void 2532 dmu_objset_id_quota_upgrade(objset_t *os) 2533 { 2534 dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb); 2535 } 2536 2537 boolean_t 2538 dmu_objset_userobjspace_upgradable(objset_t *os) 2539 { 2540 return (dmu_objset_type(os) == DMU_OST_ZFS && 2541 !dmu_objset_is_snapshot(os) && 2542 dmu_objset_userobjused_enabled(os) && 2543 !dmu_objset_userobjspace_present(os) && 2544 spa_writeable(dmu_objset_spa(os))); 2545 } 2546 2547 boolean_t 2548 dmu_objset_projectquota_upgradable(objset_t *os) 2549 { 2550 return (dmu_objset_type(os) == DMU_OST_ZFS && 2551 !dmu_objset_is_snapshot(os) && 2552 dmu_objset_projectquota_enabled(os) && 2553 !dmu_objset_projectquota_present(os) && 2554 spa_writeable(dmu_objset_spa(os))); 2555 } 2556 2557 void 2558 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 2559 uint64_t *usedobjsp, uint64_t *availobjsp) 2560 { 2561 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, 2562 usedobjsp, availobjsp); 2563 } 2564 2565 uint64_t 2566 dmu_objset_fsid_guid(objset_t *os) 2567 { 2568 return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); 2569 } 2570 2571 void 2572 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) 2573 { 2574 stat->dds_type = os->os_phys->os_type; 2575 if (os->os_dsl_dataset) 2576 dsl_dataset_fast_stat(os->os_dsl_dataset, stat); 2577 } 2578 2579 void 2580 dmu_objset_stats(objset_t *os, nvlist_t *nv) 2581 { 2582 ASSERT(os->os_dsl_dataset || 2583 os->os_phys->os_type == DMU_OST_META); 2584 2585 if (os->os_dsl_dataset != NULL) 2586 dsl_dataset_stats(os->os_dsl_dataset, nv); 2587 2588 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, 2589 os->os_phys->os_type); 2590 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, 2591 dmu_objset_userspace_present(os)); 2592 } 2593 2594 int 2595 dmu_objset_is_snapshot(objset_t *os) 2596 { 2597 if (os->os_dsl_dataset != NULL) 2598 return (os->os_dsl_dataset->ds_is_snapshot); 2599 else 2600 return (B_FALSE); 2601 } 2602 2603 int 2604 dmu_snapshot_realname(objset_t *os, const char *name, char *real, int maxlen, 2605 boolean_t *conflict) 2606 { 2607 dsl_dataset_t *ds = os->os_dsl_dataset; 2608 uint64_t ignored; 2609 2610 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 2611 return (SET_ERROR(ENOENT)); 2612 2613 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, 2614 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, 2615 MT_NORMALIZE, real, maxlen, conflict)); 2616 } 2617 2618 int 2619 dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 2620 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) 2621 { 2622 dsl_dataset_t *ds = os->os_dsl_dataset; 2623 zap_cursor_t cursor; 2624 zap_attribute_t *attr; 2625 2626 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 2627 2628 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 2629 return (SET_ERROR(ENOENT)); 2630 2631 attr = zap_attribute_alloc(); 2632 zap_cursor_init_serialized(&cursor, 2633 ds->ds_dir->dd_pool->dp_meta_objset, 2634 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); 2635 2636 if (zap_cursor_retrieve(&cursor, attr) != 0) { 2637 zap_cursor_fini(&cursor); 2638 zap_attribute_free(attr); 2639 return (SET_ERROR(ENOENT)); 2640 } 2641 2642 if (strlen(attr->za_name) + 1 > namelen) { 2643 zap_cursor_fini(&cursor); 2644 zap_attribute_free(attr); 2645 return (SET_ERROR(ENAMETOOLONG)); 2646 } 2647 2648 (void) strlcpy(name, attr->za_name, namelen); 2649 if (idp) 2650 *idp = attr->za_first_integer; 2651 if (case_conflict) 2652 *case_conflict = attr->za_normalization_conflict; 2653 zap_cursor_advance(&cursor); 2654 *offp = zap_cursor_serialize(&cursor); 2655 zap_cursor_fini(&cursor); 2656 zap_attribute_free(attr); 2657 2658 return (0); 2659 } 2660 2661 int 2662 dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value) 2663 { 2664 return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value)); 2665 } 2666 2667 int 2668 dmu_dir_list_next(objset_t *os, int namelen, char *name, 2669 uint64_t *idp, uint64_t *offp) 2670 { 2671 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 2672 zap_cursor_t cursor; 2673 zap_attribute_t *attr; 2674 2675 /* there is no next dir on a snapshot! */ 2676 if (os->os_dsl_dataset->ds_object != 2677 dsl_dir_phys(dd)->dd_head_dataset_obj) 2678 return (SET_ERROR(ENOENT)); 2679 2680 attr = zap_attribute_alloc(); 2681 zap_cursor_init_serialized(&cursor, 2682 dd->dd_pool->dp_meta_objset, 2683 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); 2684 2685 if (zap_cursor_retrieve(&cursor, attr) != 0) { 2686 zap_cursor_fini(&cursor); 2687 zap_attribute_free(attr); 2688 return (SET_ERROR(ENOENT)); 2689 } 2690 2691 if (strlen(attr->za_name) + 1 > namelen) { 2692 zap_cursor_fini(&cursor); 2693 zap_attribute_free(attr); 2694 return (SET_ERROR(ENAMETOOLONG)); 2695 } 2696 2697 (void) strlcpy(name, attr->za_name, namelen); 2698 if (idp) 2699 *idp = attr->za_first_integer; 2700 zap_cursor_advance(&cursor); 2701 *offp = zap_cursor_serialize(&cursor); 2702 zap_cursor_fini(&cursor); 2703 zap_attribute_free(attr); 2704 2705 return (0); 2706 } 2707 2708 typedef struct dmu_objset_find_ctx { 2709 taskq_t *dc_tq; 2710 dsl_pool_t *dc_dp; 2711 uint64_t dc_ddobj; 2712 char *dc_ddname; /* last component of ddobj's name */ 2713 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); 2714 void *dc_arg; 2715 int dc_flags; 2716 kmutex_t *dc_error_lock; 2717 int *dc_error; 2718 } dmu_objset_find_ctx_t; 2719 2720 static void 2721 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) 2722 { 2723 dsl_pool_t *dp = dcp->dc_dp; 2724 dsl_dir_t *dd; 2725 dsl_dataset_t *ds; 2726 zap_cursor_t zc; 2727 zap_attribute_t *attr; 2728 uint64_t thisobj; 2729 int err = 0; 2730 2731 /* don't process if there already was an error */ 2732 if (*dcp->dc_error != 0) 2733 goto out; 2734 2735 /* 2736 * Note: passing the name (dc_ddname) here is optional, but it 2737 * improves performance because we don't need to call 2738 * zap_value_search() to determine the name. 2739 */ 2740 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); 2741 if (err != 0) 2742 goto out; 2743 2744 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2745 if (dd->dd_myname[0] == '$') { 2746 dsl_dir_rele(dd, FTAG); 2747 goto out; 2748 } 2749 2750 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2751 attr = zap_attribute_alloc(); 2752 2753 /* 2754 * Iterate over all children. 2755 */ 2756 if (dcp->dc_flags & DS_FIND_CHILDREN) { 2757 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2758 dsl_dir_phys(dd)->dd_child_dir_zapobj); 2759 zap_cursor_retrieve(&zc, attr) == 0; 2760 (void) zap_cursor_advance(&zc)) { 2761 ASSERT3U(attr->za_integer_length, ==, 2762 sizeof (uint64_t)); 2763 ASSERT3U(attr->za_num_integers, ==, 1); 2764 2765 dmu_objset_find_ctx_t *child_dcp = 2766 kmem_alloc(sizeof (*child_dcp), KM_SLEEP); 2767 *child_dcp = *dcp; 2768 child_dcp->dc_ddobj = attr->za_first_integer; 2769 child_dcp->dc_ddname = spa_strdup(attr->za_name); 2770 if (dcp->dc_tq != NULL) 2771 (void) taskq_dispatch(dcp->dc_tq, 2772 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); 2773 else 2774 dmu_objset_find_dp_impl(child_dcp); 2775 } 2776 zap_cursor_fini(&zc); 2777 } 2778 2779 /* 2780 * Iterate over all snapshots. 2781 */ 2782 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { 2783 dsl_dataset_t *ds; 2784 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2785 2786 if (err == 0) { 2787 uint64_t snapobj; 2788 2789 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 2790 dsl_dataset_rele(ds, FTAG); 2791 2792 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 2793 zap_cursor_retrieve(&zc, attr) == 0; 2794 (void) zap_cursor_advance(&zc)) { 2795 ASSERT3U(attr->za_integer_length, ==, 2796 sizeof (uint64_t)); 2797 ASSERT3U(attr->za_num_integers, ==, 1); 2798 2799 err = dsl_dataset_hold_obj(dp, 2800 attr->za_first_integer, FTAG, &ds); 2801 if (err != 0) 2802 break; 2803 err = dcp->dc_func(dp, ds, dcp->dc_arg); 2804 dsl_dataset_rele(ds, FTAG); 2805 if (err != 0) 2806 break; 2807 } 2808 zap_cursor_fini(&zc); 2809 } 2810 } 2811 2812 zap_attribute_free(attr); 2813 2814 if (err != 0) { 2815 dsl_dir_rele(dd, FTAG); 2816 goto out; 2817 } 2818 2819 /* 2820 * Apply to self. 2821 */ 2822 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2823 2824 /* 2825 * Note: we hold the dir while calling dsl_dataset_hold_obj() so 2826 * that the dir will remain cached, and we won't have to re-instantiate 2827 * it (which could be expensive due to finding its name via 2828 * zap_value_search()). 2829 */ 2830 dsl_dir_rele(dd, FTAG); 2831 if (err != 0) 2832 goto out; 2833 err = dcp->dc_func(dp, ds, dcp->dc_arg); 2834 dsl_dataset_rele(ds, FTAG); 2835 2836 out: 2837 if (err != 0) { 2838 mutex_enter(dcp->dc_error_lock); 2839 /* only keep first error */ 2840 if (*dcp->dc_error == 0) 2841 *dcp->dc_error = err; 2842 mutex_exit(dcp->dc_error_lock); 2843 } 2844 2845 if (dcp->dc_ddname != NULL) 2846 spa_strfree(dcp->dc_ddname); 2847 kmem_free(dcp, sizeof (*dcp)); 2848 } 2849 2850 static void 2851 dmu_objset_find_dp_cb(void *arg) 2852 { 2853 dmu_objset_find_ctx_t *dcp = arg; 2854 dsl_pool_t *dp = dcp->dc_dp; 2855 2856 /* 2857 * We need to get a pool_config_lock here, as there are several 2858 * assert(pool_config_held) down the stack. Getting a lock via 2859 * dsl_pool_config_enter is risky, as it might be stalled by a 2860 * pending writer. This would deadlock, as the write lock can 2861 * only be granted when our parent thread gives up the lock. 2862 * The _prio interface gives us priority over a pending writer. 2863 */ 2864 dsl_pool_config_enter_prio(dp, FTAG); 2865 2866 dmu_objset_find_dp_impl(dcp); 2867 2868 dsl_pool_config_exit(dp, FTAG); 2869 } 2870 2871 /* 2872 * Find objsets under and including ddobj, call func(ds) on each. 2873 * The order for the enumeration is completely undefined. 2874 * func is called with dsl_pool_config held. 2875 */ 2876 int 2877 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, 2878 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) 2879 { 2880 int error = 0; 2881 taskq_t *tq = NULL; 2882 int ntasks; 2883 dmu_objset_find_ctx_t *dcp; 2884 kmutex_t err_lock; 2885 2886 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); 2887 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); 2888 dcp->dc_tq = NULL; 2889 dcp->dc_dp = dp; 2890 dcp->dc_ddobj = ddobj; 2891 dcp->dc_ddname = NULL; 2892 dcp->dc_func = func; 2893 dcp->dc_arg = arg; 2894 dcp->dc_flags = flags; 2895 dcp->dc_error_lock = &err_lock; 2896 dcp->dc_error = &error; 2897 2898 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { 2899 /* 2900 * In case a write lock is held we can't make use of 2901 * parallelism, as down the stack of the worker threads 2902 * the lock is asserted via dsl_pool_config_held. 2903 * In case of a read lock this is solved by getting a read 2904 * lock in each worker thread, which isn't possible in case 2905 * of a writer lock. So we fall back to the synchronous path 2906 * here. 2907 * In the future it might be possible to get some magic into 2908 * dsl_pool_config_held in a way that it returns true for 2909 * the worker threads so that a single lock held from this 2910 * thread suffices. For now, stay single threaded. 2911 */ 2912 dmu_objset_find_dp_impl(dcp); 2913 mutex_destroy(&err_lock); 2914 2915 return (error); 2916 } 2917 2918 ntasks = dmu_find_threads; 2919 if (ntasks == 0) 2920 ntasks = vdev_count_leaves(dp->dp_spa) * 4; 2921 tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks, 2922 INT_MAX, 0); 2923 if (tq == NULL) { 2924 kmem_free(dcp, sizeof (*dcp)); 2925 mutex_destroy(&err_lock); 2926 2927 return (SET_ERROR(ENOMEM)); 2928 } 2929 dcp->dc_tq = tq; 2930 2931 /* dcp will be freed by task */ 2932 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); 2933 2934 /* 2935 * PORTING: this code relies on the property of taskq_wait to wait 2936 * until no more tasks are queued and no more tasks are active. As 2937 * we always queue new tasks from within other tasks, task_wait 2938 * reliably waits for the full recursion to finish, even though we 2939 * enqueue new tasks after taskq_wait has been called. 2940 * On platforms other than illumos, taskq_wait may not have this 2941 * property. 2942 */ 2943 taskq_wait(tq); 2944 taskq_destroy(tq); 2945 mutex_destroy(&err_lock); 2946 2947 return (error); 2948 } 2949 2950 /* 2951 * Find all objsets under name, and for each, call 'func(child_name, arg)'. 2952 * The dp_config_rwlock must not be held when this is called, and it 2953 * will not be held when the callback is called. 2954 * Therefore this function should only be used when the pool is not changing 2955 * (e.g. in syncing context), or the callback can deal with the possible races. 2956 */ 2957 static int 2958 dmu_objset_find_impl(spa_t *spa, const char *name, 2959 int func(const char *, void *), void *arg, int flags) 2960 { 2961 dsl_dir_t *dd; 2962 dsl_pool_t *dp = spa_get_dsl(spa); 2963 dsl_dataset_t *ds; 2964 zap_cursor_t zc; 2965 zap_attribute_t *attr; 2966 char *child; 2967 uint64_t thisobj; 2968 int err; 2969 2970 dsl_pool_config_enter(dp, FTAG); 2971 2972 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); 2973 if (err != 0) { 2974 dsl_pool_config_exit(dp, FTAG); 2975 return (err); 2976 } 2977 2978 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2979 if (dd->dd_myname[0] == '$') { 2980 dsl_dir_rele(dd, FTAG); 2981 dsl_pool_config_exit(dp, FTAG); 2982 return (0); 2983 } 2984 2985 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2986 attr = zap_attribute_alloc(); 2987 2988 /* 2989 * Iterate over all children. 2990 */ 2991 if (flags & DS_FIND_CHILDREN) { 2992 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2993 dsl_dir_phys(dd)->dd_child_dir_zapobj); 2994 zap_cursor_retrieve(&zc, attr) == 0; 2995 (void) zap_cursor_advance(&zc)) { 2996 ASSERT3U(attr->za_integer_length, ==, 2997 sizeof (uint64_t)); 2998 ASSERT3U(attr->za_num_integers, ==, 1); 2999 3000 child = kmem_asprintf("%s/%s", name, attr->za_name); 3001 dsl_pool_config_exit(dp, FTAG); 3002 err = dmu_objset_find_impl(spa, child, 3003 func, arg, flags); 3004 dsl_pool_config_enter(dp, FTAG); 3005 kmem_strfree(child); 3006 if (err != 0) 3007 break; 3008 } 3009 zap_cursor_fini(&zc); 3010 3011 if (err != 0) { 3012 dsl_dir_rele(dd, FTAG); 3013 dsl_pool_config_exit(dp, FTAG); 3014 zap_attribute_free(attr); 3015 return (err); 3016 } 3017 } 3018 3019 /* 3020 * Iterate over all snapshots. 3021 */ 3022 if (flags & DS_FIND_SNAPSHOTS) { 3023 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 3024 3025 if (err == 0) { 3026 uint64_t snapobj; 3027 3028 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 3029 dsl_dataset_rele(ds, FTAG); 3030 3031 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 3032 zap_cursor_retrieve(&zc, attr) == 0; 3033 (void) zap_cursor_advance(&zc)) { 3034 ASSERT3U(attr->za_integer_length, ==, 3035 sizeof (uint64_t)); 3036 ASSERT3U(attr->za_num_integers, ==, 1); 3037 3038 child = kmem_asprintf("%s@%s", 3039 name, attr->za_name); 3040 dsl_pool_config_exit(dp, FTAG); 3041 err = func(child, arg); 3042 dsl_pool_config_enter(dp, FTAG); 3043 kmem_strfree(child); 3044 if (err != 0) 3045 break; 3046 } 3047 zap_cursor_fini(&zc); 3048 } 3049 } 3050 3051 dsl_dir_rele(dd, FTAG); 3052 zap_attribute_free(attr); 3053 dsl_pool_config_exit(dp, FTAG); 3054 3055 if (err != 0) 3056 return (err); 3057 3058 /* Apply to self. */ 3059 return (func(name, arg)); 3060 } 3061 3062 /* 3063 * See comment above dmu_objset_find_impl(). 3064 */ 3065 int 3066 dmu_objset_find(const char *name, int func(const char *, void *), void *arg, 3067 int flags) 3068 { 3069 spa_t *spa; 3070 int error; 3071 3072 error = spa_open(name, &spa, FTAG); 3073 if (error != 0) 3074 return (error); 3075 error = dmu_objset_find_impl(spa, name, func, arg, flags); 3076 spa_close(spa, FTAG); 3077 return (error); 3078 } 3079 3080 boolean_t 3081 dmu_objset_incompatible_encryption_version(objset_t *os) 3082 { 3083 return (dsl_dir_incompatible_encryption_version( 3084 os->os_dsl_dataset->ds_dir)); 3085 } 3086 3087 void 3088 dmu_objset_set_user(objset_t *os, void *user_ptr) 3089 { 3090 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 3091 os->os_user_ptr = user_ptr; 3092 } 3093 3094 void * 3095 dmu_objset_get_user(objset_t *os) 3096 { 3097 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 3098 return (os->os_user_ptr); 3099 } 3100 3101 /* 3102 * Determine name of filesystem, given name of snapshot. 3103 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes 3104 */ 3105 int 3106 dmu_fsname(const char *snapname, char *buf) 3107 { 3108 char *atp = strchr(snapname, '@'); 3109 if (atp == NULL) 3110 return (SET_ERROR(EINVAL)); 3111 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) 3112 return (SET_ERROR(ENAMETOOLONG)); 3113 (void) strlcpy(buf, snapname, atp - snapname + 1); 3114 return (0); 3115 } 3116 3117 /* 3118 * Call when we think we're going to write/free space in open context 3119 * to track the amount of dirty data in the open txg, which is also the 3120 * amount of memory that can not be evicted until this txg syncs. 3121 * 3122 * Note that there are two conditions where this can be called from 3123 * syncing context: 3124 * 3125 * [1] When we just created the dataset, in which case we go on with 3126 * updating any accounting of dirty data as usual. 3127 * [2] When we are dirtying MOS data, in which case we only update the 3128 * pool's accounting of dirty data. 3129 */ 3130 void 3131 dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) 3132 { 3133 dsl_dataset_t *ds = os->os_dsl_dataset; 3134 int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); 3135 3136 if (ds != NULL) { 3137 dsl_dir_willuse_space(ds->ds_dir, aspace, tx); 3138 } 3139 3140 dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); 3141 } 3142 3143 #if defined(_KERNEL) 3144 EXPORT_SYMBOL(dmu_objset_zil); 3145 EXPORT_SYMBOL(dmu_objset_pool); 3146 EXPORT_SYMBOL(dmu_objset_ds); 3147 EXPORT_SYMBOL(dmu_objset_type); 3148 EXPORT_SYMBOL(dmu_objset_name); 3149 EXPORT_SYMBOL(dmu_objset_hold); 3150 EXPORT_SYMBOL(dmu_objset_hold_flags); 3151 EXPORT_SYMBOL(dmu_objset_own); 3152 EXPORT_SYMBOL(dmu_objset_rele); 3153 EXPORT_SYMBOL(dmu_objset_rele_flags); 3154 EXPORT_SYMBOL(dmu_objset_disown); 3155 EXPORT_SYMBOL(dmu_objset_from_ds); 3156 EXPORT_SYMBOL(dmu_objset_create); 3157 EXPORT_SYMBOL(dmu_objset_clone); 3158 EXPORT_SYMBOL(dmu_objset_stats); 3159 EXPORT_SYMBOL(dmu_objset_fast_stat); 3160 EXPORT_SYMBOL(dmu_objset_spa); 3161 EXPORT_SYMBOL(dmu_objset_space); 3162 EXPORT_SYMBOL(dmu_objset_fsid_guid); 3163 EXPORT_SYMBOL(dmu_objset_find); 3164 EXPORT_SYMBOL(dmu_objset_byteswap); 3165 EXPORT_SYMBOL(dmu_objset_evict_dbufs); 3166 EXPORT_SYMBOL(dmu_objset_snap_cmtime); 3167 EXPORT_SYMBOL(dmu_objset_dnodesize); 3168 3169 EXPORT_SYMBOL(dmu_objset_sync); 3170 EXPORT_SYMBOL(dmu_objset_is_dirty); 3171 EXPORT_SYMBOL(dmu_objset_create_impl_dnstats); 3172 EXPORT_SYMBOL(dmu_objset_create_impl); 3173 EXPORT_SYMBOL(dmu_objset_open_impl); 3174 EXPORT_SYMBOL(dmu_objset_evict); 3175 EXPORT_SYMBOL(dmu_objset_register_type); 3176 EXPORT_SYMBOL(dmu_objset_sync_done); 3177 EXPORT_SYMBOL(dmu_objset_userquota_get_ids); 3178 EXPORT_SYMBOL(dmu_objset_userused_enabled); 3179 EXPORT_SYMBOL(dmu_objset_userspace_upgrade); 3180 EXPORT_SYMBOL(dmu_objset_userspace_present); 3181 EXPORT_SYMBOL(dmu_objset_userobjused_enabled); 3182 EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable); 3183 EXPORT_SYMBOL(dmu_objset_userobjspace_present); 3184 EXPORT_SYMBOL(dmu_objset_projectquota_enabled); 3185 EXPORT_SYMBOL(dmu_objset_projectquota_present); 3186 EXPORT_SYMBOL(dmu_objset_projectquota_upgradable); 3187 EXPORT_SYMBOL(dmu_objset_id_quota_upgrade); 3188 #endif 3189