1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2012, 2018 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 */ 36 37 /* Portions Copyright 2010 Robert Milkowski */ 38 39 #include <sys/cred.h> 40 #include <sys/zfs_context.h> 41 #include <sys/dmu_objset.h> 42 #include <sys/dsl_dir.h> 43 #include <sys/dsl_dataset.h> 44 #include <sys/dsl_prop.h> 45 #include <sys/dsl_pool.h> 46 #include <sys/dsl_synctask.h> 47 #include <sys/dsl_deleg.h> 48 #include <sys/dnode.h> 49 #include <sys/dbuf.h> 50 #include <sys/zvol.h> 51 #include <sys/dmu_tx.h> 52 #include <sys/zap.h> 53 #include <sys/zil.h> 54 #include <sys/dmu_impl.h> 55 #include <sys/zfs_ioctl.h> 56 #include <sys/sa.h> 57 #include <sys/zfs_onexit.h> 58 #include <sys/dsl_destroy.h> 59 #include <sys/vdev.h> 60 #include <sys/zfeature.h> 61 #include <sys/policy.h> 62 #include <sys/spa_impl.h> 63 #include <sys/dmu_recv.h> 64 #include <sys/zfs_project.h> 65 #include "zfs_namecheck.h" 66 67 /* 68 * Needed to close a window in dnode_move() that allows the objset to be freed 69 * before it can be safely accessed. 70 */ 71 krwlock_t os_lock; 72 73 /* 74 * Tunable to overwrite the maximum number of threads for the parallelization 75 * of dmu_objset_find_dp, needed to speed up the import of pools with many 76 * datasets. 77 * Default is 4 times the number of leaf vdevs. 78 */ 79 int dmu_find_threads = 0; 80 81 /* 82 * Backfill lower metadnode objects after this many have been freed. 83 * Backfilling negatively impacts object creation rates, so only do it 84 * if there are enough holes to fill. 85 */ 86 int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT; 87 88 static char *upgrade_tag = "upgrade_tag"; 89 90 static void dmu_objset_find_dp_cb(void *arg); 91 92 static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb); 93 static void dmu_objset_upgrade_stop(objset_t *os); 94 95 void 96 dmu_objset_init(void) 97 { 98 rw_init(&os_lock, NULL, RW_DEFAULT, NULL); 99 } 100 101 void 102 dmu_objset_fini(void) 103 { 104 rw_destroy(&os_lock); 105 } 106 107 spa_t * 108 dmu_objset_spa(objset_t *os) 109 { 110 return (os->os_spa); 111 } 112 113 zilog_t * 114 dmu_objset_zil(objset_t *os) 115 { 116 return (os->os_zil); 117 } 118 119 dsl_pool_t * 120 dmu_objset_pool(objset_t *os) 121 { 122 dsl_dataset_t *ds; 123 124 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) 125 return (ds->ds_dir->dd_pool); 126 else 127 return (spa_get_dsl(os->os_spa)); 128 } 129 130 dsl_dataset_t * 131 dmu_objset_ds(objset_t *os) 132 { 133 return (os->os_dsl_dataset); 134 } 135 136 dmu_objset_type_t 137 dmu_objset_type(objset_t *os) 138 { 139 return (os->os_phys->os_type); 140 } 141 142 void 143 dmu_objset_name(objset_t *os, char *buf) 144 { 145 dsl_dataset_name(os->os_dsl_dataset, buf); 146 } 147 148 uint64_t 149 dmu_objset_id(objset_t *os) 150 { 151 dsl_dataset_t *ds = os->os_dsl_dataset; 152 153 return (ds ? ds->ds_object : 0); 154 } 155 156 uint64_t 157 dmu_objset_dnodesize(objset_t *os) 158 { 159 return (os->os_dnodesize); 160 } 161 162 zfs_sync_type_t 163 dmu_objset_syncprop(objset_t *os) 164 { 165 return (os->os_sync); 166 } 167 168 zfs_logbias_op_t 169 dmu_objset_logbias(objset_t *os) 170 { 171 return (os->os_logbias); 172 } 173 174 static void 175 checksum_changed_cb(void *arg, uint64_t newval) 176 { 177 objset_t *os = arg; 178 179 /* 180 * Inheritance should have been done by now. 181 */ 182 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 183 184 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); 185 } 186 187 static void 188 compression_changed_cb(void *arg, uint64_t newval) 189 { 190 objset_t *os = arg; 191 192 /* 193 * Inheritance and range checking should have been done by now. 194 */ 195 ASSERT(newval != ZIO_COMPRESS_INHERIT); 196 197 os->os_compress = zio_compress_select(os->os_spa, 198 ZIO_COMPRESS_ALGO(newval), ZIO_COMPRESS_ON); 199 os->os_complevel = zio_complevel_select(os->os_spa, os->os_compress, 200 ZIO_COMPRESS_LEVEL(newval), ZIO_COMPLEVEL_DEFAULT); 201 } 202 203 static void 204 copies_changed_cb(void *arg, uint64_t newval) 205 { 206 objset_t *os = arg; 207 208 /* 209 * Inheritance and range checking should have been done by now. 210 */ 211 ASSERT(newval > 0); 212 ASSERT(newval <= spa_max_replication(os->os_spa)); 213 214 os->os_copies = newval; 215 } 216 217 static void 218 dedup_changed_cb(void *arg, uint64_t newval) 219 { 220 objset_t *os = arg; 221 spa_t *spa = os->os_spa; 222 enum zio_checksum checksum; 223 224 /* 225 * Inheritance should have been done by now. 226 */ 227 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 228 229 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); 230 231 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; 232 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); 233 } 234 235 static void 236 primary_cache_changed_cb(void *arg, uint64_t newval) 237 { 238 objset_t *os = arg; 239 240 /* 241 * Inheritance and range checking should have been done by now. 242 */ 243 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 244 newval == ZFS_CACHE_METADATA); 245 246 os->os_primary_cache = newval; 247 } 248 249 static void 250 secondary_cache_changed_cb(void *arg, uint64_t newval) 251 { 252 objset_t *os = arg; 253 254 /* 255 * Inheritance and range checking should have been done by now. 256 */ 257 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 258 newval == ZFS_CACHE_METADATA); 259 260 os->os_secondary_cache = newval; 261 } 262 263 static void 264 sync_changed_cb(void *arg, uint64_t newval) 265 { 266 objset_t *os = arg; 267 268 /* 269 * Inheritance and range checking should have been done by now. 270 */ 271 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || 272 newval == ZFS_SYNC_DISABLED); 273 274 os->os_sync = newval; 275 if (os->os_zil) 276 zil_set_sync(os->os_zil, newval); 277 } 278 279 static void 280 redundant_metadata_changed_cb(void *arg, uint64_t newval) 281 { 282 objset_t *os = arg; 283 284 /* 285 * Inheritance and range checking should have been done by now. 286 */ 287 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || 288 newval == ZFS_REDUNDANT_METADATA_MOST); 289 290 os->os_redundant_metadata = newval; 291 } 292 293 static void 294 dnodesize_changed_cb(void *arg, uint64_t newval) 295 { 296 objset_t *os = arg; 297 298 switch (newval) { 299 case ZFS_DNSIZE_LEGACY: 300 os->os_dnodesize = DNODE_MIN_SIZE; 301 break; 302 case ZFS_DNSIZE_AUTO: 303 /* 304 * Choose a dnode size that will work well for most 305 * workloads if the user specified "auto". Future code 306 * improvements could dynamically select a dnode size 307 * based on observed workload patterns. 308 */ 309 os->os_dnodesize = DNODE_MIN_SIZE * 2; 310 break; 311 case ZFS_DNSIZE_1K: 312 case ZFS_DNSIZE_2K: 313 case ZFS_DNSIZE_4K: 314 case ZFS_DNSIZE_8K: 315 case ZFS_DNSIZE_16K: 316 os->os_dnodesize = newval; 317 break; 318 } 319 } 320 321 static void 322 smallblk_changed_cb(void *arg, uint64_t newval) 323 { 324 objset_t *os = arg; 325 326 /* 327 * Inheritance and range checking should have been done by now. 328 */ 329 ASSERT(newval <= SPA_OLD_MAXBLOCKSIZE); 330 ASSERT(ISP2(newval)); 331 332 os->os_zpl_special_smallblock = newval; 333 } 334 335 static void 336 logbias_changed_cb(void *arg, uint64_t newval) 337 { 338 objset_t *os = arg; 339 340 ASSERT(newval == ZFS_LOGBIAS_LATENCY || 341 newval == ZFS_LOGBIAS_THROUGHPUT); 342 os->os_logbias = newval; 343 if (os->os_zil) 344 zil_set_logbias(os->os_zil, newval); 345 } 346 347 static void 348 recordsize_changed_cb(void *arg, uint64_t newval) 349 { 350 objset_t *os = arg; 351 352 os->os_recordsize = newval; 353 } 354 355 void 356 dmu_objset_byteswap(void *buf, size_t size) 357 { 358 objset_phys_t *osp = buf; 359 360 ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 || 361 size == sizeof (objset_phys_t)); 362 dnode_byteswap(&osp->os_meta_dnode); 363 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); 364 osp->os_type = BSWAP_64(osp->os_type); 365 osp->os_flags = BSWAP_64(osp->os_flags); 366 if (size >= OBJSET_PHYS_SIZE_V2) { 367 dnode_byteswap(&osp->os_userused_dnode); 368 dnode_byteswap(&osp->os_groupused_dnode); 369 if (size >= sizeof (objset_phys_t)) 370 dnode_byteswap(&osp->os_projectused_dnode); 371 } 372 } 373 374 /* 375 * The hash is a CRC-based hash of the objset_t pointer and the object number. 376 */ 377 static uint64_t 378 dnode_hash(const objset_t *os, uint64_t obj) 379 { 380 uintptr_t osv = (uintptr_t)os; 381 uint64_t crc = -1ULL; 382 383 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); 384 /* 385 * The low 6 bits of the pointer don't have much entropy, because 386 * the objset_t is larger than 2^6 bytes long. 387 */ 388 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; 389 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; 390 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; 391 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF]; 392 393 crc ^= (osv>>14) ^ (obj>>24); 394 395 return (crc); 396 } 397 398 static unsigned int 399 dnode_multilist_index_func(multilist_t *ml, void *obj) 400 { 401 dnode_t *dn = obj; 402 return (dnode_hash(dn->dn_objset, dn->dn_object) % 403 multilist_get_num_sublists(ml)); 404 } 405 406 /* 407 * Instantiates the objset_t in-memory structure corresponding to the 408 * objset_phys_t that's pointed to by the specified blkptr_t. 409 */ 410 int 411 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 412 objset_t **osp) 413 { 414 objset_t *os; 415 int i, err; 416 417 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); 418 ASSERT(!BP_IS_REDACTED(bp)); 419 420 /* 421 * We need the pool config lock to get properties. 422 */ 423 ASSERT(ds == NULL || dsl_pool_config_held(ds->ds_dir->dd_pool)); 424 425 /* 426 * The $ORIGIN dataset (if it exists) doesn't have an associated 427 * objset, so there's no reason to open it. The $ORIGIN dataset 428 * will not exist on pools older than SPA_VERSION_ORIGIN. 429 */ 430 if (ds != NULL && spa_get_dsl(spa) != NULL && 431 spa_get_dsl(spa)->dp_origin_snap != NULL) { 432 ASSERT3P(ds->ds_dir, !=, 433 spa_get_dsl(spa)->dp_origin_snap->ds_dir); 434 } 435 436 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); 437 os->os_dsl_dataset = ds; 438 os->os_spa = spa; 439 os->os_rootbp = bp; 440 if (!BP_IS_HOLE(os->os_rootbp)) { 441 arc_flags_t aflags = ARC_FLAG_WAIT; 442 zbookmark_phys_t zb; 443 int size; 444 enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; 445 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, 446 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 447 448 if (DMU_OS_IS_L2CACHEABLE(os)) 449 aflags |= ARC_FLAG_L2CACHE; 450 451 if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) { 452 ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); 453 ASSERT(BP_IS_AUTHENTICATED(bp)); 454 zio_flags |= ZIO_FLAG_RAW; 455 } 456 457 dprintf_bp(os->os_rootbp, "reading %s", ""); 458 err = arc_read(NULL, spa, os->os_rootbp, 459 arc_getbuf_func, &os->os_phys_buf, 460 ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); 461 if (err != 0) { 462 kmem_free(os, sizeof (objset_t)); 463 /* convert checksum errors into IO errors */ 464 if (err == ECKSUM) 465 err = SET_ERROR(EIO); 466 return (err); 467 } 468 469 if (spa_version(spa) < SPA_VERSION_USERSPACE) 470 size = OBJSET_PHYS_SIZE_V1; 471 else if (!spa_feature_is_enabled(spa, 472 SPA_FEATURE_PROJECT_QUOTA)) 473 size = OBJSET_PHYS_SIZE_V2; 474 else 475 size = sizeof (objset_phys_t); 476 477 /* Increase the blocksize if we are permitted. */ 478 if (arc_buf_size(os->os_phys_buf) < size) { 479 arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, 480 ARC_BUFC_METADATA, size); 481 bzero(buf->b_data, size); 482 bcopy(os->os_phys_buf->b_data, buf->b_data, 483 arc_buf_size(os->os_phys_buf)); 484 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 485 os->os_phys_buf = buf; 486 } 487 488 os->os_phys = os->os_phys_buf->b_data; 489 os->os_flags = os->os_phys->os_flags; 490 } else { 491 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? 492 sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1; 493 os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, 494 ARC_BUFC_METADATA, size); 495 os->os_phys = os->os_phys_buf->b_data; 496 bzero(os->os_phys, size); 497 } 498 /* 499 * These properties will be filled in by the logic in zfs_get_zplprop() 500 * when they are queried for the first time. 501 */ 502 os->os_version = OBJSET_PROP_UNINITIALIZED; 503 os->os_normalization = OBJSET_PROP_UNINITIALIZED; 504 os->os_utf8only = OBJSET_PROP_UNINITIALIZED; 505 os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED; 506 507 /* 508 * Note: the changed_cb will be called once before the register 509 * func returns, thus changing the checksum/compression from the 510 * default (fletcher2/off). Snapshots don't need to know about 511 * checksum/compression/copies. 512 */ 513 if (ds != NULL) { 514 os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0); 515 516 err = dsl_prop_register(ds, 517 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), 518 primary_cache_changed_cb, os); 519 if (err == 0) { 520 err = dsl_prop_register(ds, 521 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), 522 secondary_cache_changed_cb, os); 523 } 524 if (!ds->ds_is_snapshot) { 525 if (err == 0) { 526 err = dsl_prop_register(ds, 527 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 528 checksum_changed_cb, os); 529 } 530 if (err == 0) { 531 err = dsl_prop_register(ds, 532 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 533 compression_changed_cb, os); 534 } 535 if (err == 0) { 536 err = dsl_prop_register(ds, 537 zfs_prop_to_name(ZFS_PROP_COPIES), 538 copies_changed_cb, os); 539 } 540 if (err == 0) { 541 err = dsl_prop_register(ds, 542 zfs_prop_to_name(ZFS_PROP_DEDUP), 543 dedup_changed_cb, os); 544 } 545 if (err == 0) { 546 err = dsl_prop_register(ds, 547 zfs_prop_to_name(ZFS_PROP_LOGBIAS), 548 logbias_changed_cb, os); 549 } 550 if (err == 0) { 551 err = dsl_prop_register(ds, 552 zfs_prop_to_name(ZFS_PROP_SYNC), 553 sync_changed_cb, os); 554 } 555 if (err == 0) { 556 err = dsl_prop_register(ds, 557 zfs_prop_to_name( 558 ZFS_PROP_REDUNDANT_METADATA), 559 redundant_metadata_changed_cb, os); 560 } 561 if (err == 0) { 562 err = dsl_prop_register(ds, 563 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 564 recordsize_changed_cb, os); 565 } 566 if (err == 0) { 567 err = dsl_prop_register(ds, 568 zfs_prop_to_name(ZFS_PROP_DNODESIZE), 569 dnodesize_changed_cb, os); 570 } 571 if (err == 0) { 572 err = dsl_prop_register(ds, 573 zfs_prop_to_name( 574 ZFS_PROP_SPECIAL_SMALL_BLOCKS), 575 smallblk_changed_cb, os); 576 } 577 } 578 if (err != 0) { 579 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 580 kmem_free(os, sizeof (objset_t)); 581 return (err); 582 } 583 } else { 584 /* It's the meta-objset. */ 585 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; 586 os->os_compress = ZIO_COMPRESS_ON; 587 os->os_complevel = ZIO_COMPLEVEL_DEFAULT; 588 os->os_encrypted = B_FALSE; 589 os->os_copies = spa_max_replication(spa); 590 os->os_dedup_checksum = ZIO_CHECKSUM_OFF; 591 os->os_dedup_verify = B_FALSE; 592 os->os_logbias = ZFS_LOGBIAS_LATENCY; 593 os->os_sync = ZFS_SYNC_STANDARD; 594 os->os_primary_cache = ZFS_CACHE_ALL; 595 os->os_secondary_cache = ZFS_CACHE_ALL; 596 os->os_dnodesize = DNODE_MIN_SIZE; 597 } 598 599 if (ds == NULL || !ds->ds_is_snapshot) 600 os->os_zil_header = os->os_phys->os_zil_header; 601 os->os_zil = zil_alloc(os, &os->os_zil_header); 602 603 for (i = 0; i < TXG_SIZE; i++) { 604 os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t), 605 offsetof(dnode_t, dn_dirty_link[i]), 606 dnode_multilist_index_func); 607 } 608 list_create(&os->os_dnodes, sizeof (dnode_t), 609 offsetof(dnode_t, dn_link)); 610 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), 611 offsetof(dmu_buf_impl_t, db_link)); 612 613 list_link_init(&os->os_evicting_node); 614 615 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); 616 mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); 617 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); 618 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); 619 os->os_obj_next_percpu_len = boot_ncpus; 620 os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * 621 sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); 622 623 dnode_special_open(os, &os->os_phys->os_meta_dnode, 624 DMU_META_DNODE_OBJECT, &os->os_meta_dnode); 625 if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) { 626 dnode_special_open(os, &os->os_phys->os_userused_dnode, 627 DMU_USERUSED_OBJECT, &os->os_userused_dnode); 628 dnode_special_open(os, &os->os_phys->os_groupused_dnode, 629 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); 630 if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf)) 631 dnode_special_open(os, 632 &os->os_phys->os_projectused_dnode, 633 DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode); 634 } 635 636 mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL); 637 638 *osp = os; 639 return (0); 640 } 641 642 int 643 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) 644 { 645 int err = 0; 646 647 /* 648 * We need the pool_config lock to manipulate the dsl_dataset_t. 649 * Even if the dataset is long-held, we need the pool_config lock 650 * to open the objset, as it needs to get properties. 651 */ 652 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool)); 653 654 mutex_enter(&ds->ds_opening_lock); 655 if (ds->ds_objset == NULL) { 656 objset_t *os; 657 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 658 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), 659 ds, dsl_dataset_get_blkptr(ds), &os); 660 rrw_exit(&ds->ds_bp_rwlock, FTAG); 661 662 if (err == 0) { 663 mutex_enter(&ds->ds_lock); 664 ASSERT(ds->ds_objset == NULL); 665 ds->ds_objset = os; 666 mutex_exit(&ds->ds_lock); 667 } 668 } 669 *osp = ds->ds_objset; 670 mutex_exit(&ds->ds_opening_lock); 671 return (err); 672 } 673 674 /* 675 * Holds the pool while the objset is held. Therefore only one objset 676 * can be held at a time. 677 */ 678 int 679 dmu_objset_hold_flags(const char *name, boolean_t decrypt, void *tag, 680 objset_t **osp) 681 { 682 dsl_pool_t *dp; 683 dsl_dataset_t *ds; 684 int err; 685 ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; 686 687 err = dsl_pool_hold(name, tag, &dp); 688 if (err != 0) 689 return (err); 690 err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds); 691 if (err != 0) { 692 dsl_pool_rele(dp, tag); 693 return (err); 694 } 695 696 err = dmu_objset_from_ds(ds, osp); 697 if (err != 0) { 698 dsl_dataset_rele(ds, tag); 699 dsl_pool_rele(dp, tag); 700 } 701 702 return (err); 703 } 704 705 int 706 dmu_objset_hold(const char *name, void *tag, objset_t **osp) 707 { 708 return (dmu_objset_hold_flags(name, B_FALSE, tag, osp)); 709 } 710 711 static int 712 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, 713 boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) 714 { 715 int err; 716 717 err = dmu_objset_from_ds(ds, osp); 718 if (err != 0) { 719 return (err); 720 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { 721 return (SET_ERROR(EINVAL)); 722 } else if (!readonly && dsl_dataset_is_snapshot(ds)) { 723 return (SET_ERROR(EROFS)); 724 } else if (!readonly && decrypt && 725 dsl_dir_incompatible_encryption_version(ds->ds_dir)) { 726 return (SET_ERROR(EROFS)); 727 } 728 729 /* if we are decrypting, we can now check MACs in os->os_phys_buf */ 730 if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) { 731 zbookmark_phys_t zb; 732 733 SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT, 734 ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 735 err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa, 736 &zb, B_FALSE); 737 if (err != 0) 738 return (err); 739 740 ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf)); 741 } 742 743 return (0); 744 } 745 746 /* 747 * dsl_pool must not be held when this is called. 748 * Upon successful return, there will be a longhold on the dataset, 749 * and the dsl_pool will not be held. 750 */ 751 int 752 dmu_objset_own(const char *name, dmu_objset_type_t type, 753 boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) 754 { 755 dsl_pool_t *dp; 756 dsl_dataset_t *ds; 757 int err; 758 ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; 759 760 err = dsl_pool_hold(name, FTAG, &dp); 761 if (err != 0) 762 return (err); 763 err = dsl_dataset_own(dp, name, flags, tag, &ds); 764 if (err != 0) { 765 dsl_pool_rele(dp, FTAG); 766 return (err); 767 } 768 err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); 769 if (err != 0) { 770 dsl_dataset_disown(ds, flags, tag); 771 dsl_pool_rele(dp, FTAG); 772 return (err); 773 } 774 775 /* 776 * User accounting requires the dataset to be decrypted and rw. 777 * We also don't begin user accounting during claiming to help 778 * speed up pool import times and to keep this txg reserved 779 * completely for recovery work. 780 */ 781 if ((dmu_objset_userobjspace_upgradable(*osp) || 782 dmu_objset_projectquota_upgradable(*osp)) && 783 !readonly && !dp->dp_spa->spa_claiming && 784 (ds->ds_dir->dd_crypto_obj == 0 || decrypt)) 785 dmu_objset_id_quota_upgrade(*osp); 786 787 dsl_pool_rele(dp, FTAG); 788 return (0); 789 } 790 791 int 792 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, 793 boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) 794 { 795 dsl_dataset_t *ds; 796 int err; 797 ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; 798 799 err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds); 800 if (err != 0) 801 return (err); 802 803 err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); 804 if (err != 0) { 805 dsl_dataset_disown(ds, flags, tag); 806 return (err); 807 } 808 809 return (0); 810 } 811 812 void 813 dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, void *tag) 814 { 815 ds_hold_flags_t flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0; 816 817 dsl_pool_t *dp = dmu_objset_pool(os); 818 dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag); 819 dsl_pool_rele(dp, tag); 820 } 821 822 void 823 dmu_objset_rele(objset_t *os, void *tag) 824 { 825 dmu_objset_rele_flags(os, B_FALSE, tag); 826 } 827 828 /* 829 * When we are called, os MUST refer to an objset associated with a dataset 830 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner 831 * == tag. We will then release and reacquire ownership of the dataset while 832 * holding the pool config_rwlock to avoid intervening namespace or ownership 833 * changes may occur. 834 * 835 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to 836 * release the hold on its dataset and acquire a new one on the dataset of the 837 * same name so that it can be partially torn down and reconstructed. 838 */ 839 void 840 dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds, 841 boolean_t decrypt, void *tag) 842 { 843 dsl_pool_t *dp; 844 char name[ZFS_MAX_DATASET_NAME_LEN]; 845 846 VERIFY3P(ds, !=, NULL); 847 VERIFY3P(ds->ds_owner, ==, tag); 848 VERIFY(dsl_dataset_long_held(ds)); 849 850 dsl_dataset_name(ds, name); 851 dp = ds->ds_dir->dd_pool; 852 dsl_pool_config_enter(dp, FTAG); 853 dsl_dataset_disown(ds, decrypt, tag); 854 VERIFY0(dsl_dataset_own(dp, name, 855 (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0, tag, newds)); 856 dsl_pool_config_exit(dp, FTAG); 857 } 858 859 void 860 dmu_objset_disown(objset_t *os, boolean_t decrypt, void *tag) 861 { 862 /* 863 * Stop upgrading thread 864 */ 865 dmu_objset_upgrade_stop(os); 866 dsl_dataset_disown(os->os_dsl_dataset, 867 (decrypt) ? DS_HOLD_FLAG_DECRYPT : 0, tag); 868 } 869 870 void 871 dmu_objset_evict_dbufs(objset_t *os) 872 { 873 dnode_t *dn_marker; 874 dnode_t *dn; 875 876 dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP); 877 878 mutex_enter(&os->os_lock); 879 dn = list_head(&os->os_dnodes); 880 while (dn != NULL) { 881 /* 882 * Skip dnodes without holds. We have to do this dance 883 * because dnode_add_ref() only works if there is already a 884 * hold. If the dnode has no holds, then it has no dbufs. 885 */ 886 if (dnode_add_ref(dn, FTAG)) { 887 list_insert_after(&os->os_dnodes, dn, dn_marker); 888 mutex_exit(&os->os_lock); 889 890 dnode_evict_dbufs(dn); 891 dnode_rele(dn, FTAG); 892 893 mutex_enter(&os->os_lock); 894 dn = list_next(&os->os_dnodes, dn_marker); 895 list_remove(&os->os_dnodes, dn_marker); 896 } else { 897 dn = list_next(&os->os_dnodes, dn); 898 } 899 } 900 mutex_exit(&os->os_lock); 901 902 kmem_free(dn_marker, sizeof (dnode_t)); 903 904 if (DMU_USERUSED_DNODE(os) != NULL) { 905 if (DMU_PROJECTUSED_DNODE(os) != NULL) 906 dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os)); 907 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); 908 dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); 909 } 910 dnode_evict_dbufs(DMU_META_DNODE(os)); 911 } 912 913 /* 914 * Objset eviction processing is split into into two pieces. 915 * The first marks the objset as evicting, evicts any dbufs that 916 * have a refcount of zero, and then queues up the objset for the 917 * second phase of eviction. Once os->os_dnodes has been cleared by 918 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. 919 * The second phase closes the special dnodes, dequeues the objset from 920 * the list of those undergoing eviction, and finally frees the objset. 921 * 922 * NOTE: Due to asynchronous eviction processing (invocation of 923 * dnode_buf_pageout()), it is possible for the meta dnode for the 924 * objset to have no holds even though os->os_dnodes is not empty. 925 */ 926 void 927 dmu_objset_evict(objset_t *os) 928 { 929 dsl_dataset_t *ds = os->os_dsl_dataset; 930 931 for (int t = 0; t < TXG_SIZE; t++) 932 ASSERT(!dmu_objset_is_dirty(os, t)); 933 934 if (ds) 935 dsl_prop_unregister_all(ds, os); 936 937 if (os->os_sa) 938 sa_tear_down(os); 939 940 dmu_objset_evict_dbufs(os); 941 942 mutex_enter(&os->os_lock); 943 spa_evicting_os_register(os->os_spa, os); 944 if (list_is_empty(&os->os_dnodes)) { 945 mutex_exit(&os->os_lock); 946 dmu_objset_evict_done(os); 947 } else { 948 mutex_exit(&os->os_lock); 949 } 950 951 952 } 953 954 void 955 dmu_objset_evict_done(objset_t *os) 956 { 957 ASSERT3P(list_head(&os->os_dnodes), ==, NULL); 958 959 dnode_special_close(&os->os_meta_dnode); 960 if (DMU_USERUSED_DNODE(os)) { 961 if (DMU_PROJECTUSED_DNODE(os)) 962 dnode_special_close(&os->os_projectused_dnode); 963 dnode_special_close(&os->os_userused_dnode); 964 dnode_special_close(&os->os_groupused_dnode); 965 } 966 zil_free(os->os_zil); 967 968 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 969 970 /* 971 * This is a barrier to prevent the objset from going away in 972 * dnode_move() until we can safely ensure that the objset is still in 973 * use. We consider the objset valid before the barrier and invalid 974 * after the barrier. 975 */ 976 rw_enter(&os_lock, RW_READER); 977 rw_exit(&os_lock); 978 979 kmem_free(os->os_obj_next_percpu, 980 os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); 981 982 mutex_destroy(&os->os_lock); 983 mutex_destroy(&os->os_userused_lock); 984 mutex_destroy(&os->os_obj_lock); 985 mutex_destroy(&os->os_user_ptr_lock); 986 mutex_destroy(&os->os_upgrade_lock); 987 for (int i = 0; i < TXG_SIZE; i++) { 988 multilist_destroy(os->os_dirty_dnodes[i]); 989 } 990 spa_evicting_os_deregister(os->os_spa, os); 991 kmem_free(os, sizeof (objset_t)); 992 } 993 994 inode_timespec_t 995 dmu_objset_snap_cmtime(objset_t *os) 996 { 997 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); 998 } 999 1000 objset_t * 1001 dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 1002 dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx) 1003 { 1004 objset_t *os; 1005 dnode_t *mdn; 1006 1007 ASSERT(dmu_tx_is_syncing(tx)); 1008 1009 if (blksz == 0) 1010 blksz = DNODE_BLOCK_SIZE; 1011 if (ibs == 0) 1012 ibs = DN_MAX_INDBLKSHIFT; 1013 1014 if (ds != NULL) 1015 VERIFY0(dmu_objset_from_ds(ds, &os)); 1016 else 1017 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); 1018 1019 mdn = DMU_META_DNODE(os); 1020 1021 dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0, 1022 DNODE_MIN_SLOTS, tx); 1023 1024 /* 1025 * We don't want to have to increase the meta-dnode's nlevels 1026 * later, because then we could do it in quiescing context while 1027 * we are also accessing it in open context. 1028 * 1029 * This precaution is not necessary for the MOS (ds == NULL), 1030 * because the MOS is only updated in syncing context. 1031 * This is most fortunate: the MOS is the only objset that 1032 * needs to be synced multiple times as spa_sync() iterates 1033 * to convergence, so minimizing its dn_nlevels matters. 1034 */ 1035 if (ds != NULL) { 1036 if (levels == 0) { 1037 levels = 1; 1038 1039 /* 1040 * Determine the number of levels necessary for the 1041 * meta-dnode to contain DN_MAX_OBJECT dnodes. Note 1042 * that in order to ensure that we do not overflow 1043 * 64 bits, there has to be a nlevels that gives us a 1044 * number of blocks > DN_MAX_OBJECT but < 2^64. 1045 * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) 1046 * (10) must be less than (64 - log2(DN_MAX_OBJECT)) 1047 * (16). 1048 */ 1049 while ((uint64_t)mdn->dn_nblkptr << 1050 (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) * 1051 (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < 1052 DN_MAX_OBJECT) 1053 levels++; 1054 } 1055 1056 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = 1057 mdn->dn_nlevels = levels; 1058 } 1059 1060 ASSERT(type != DMU_OST_NONE); 1061 ASSERT(type != DMU_OST_ANY); 1062 ASSERT(type < DMU_OST_NUMTYPES); 1063 os->os_phys->os_type = type; 1064 1065 /* 1066 * Enable user accounting if it is enabled and this is not an 1067 * encrypted receive. 1068 */ 1069 if (dmu_objset_userused_enabled(os) && 1070 (!os->os_encrypted || !dmu_objset_is_receiving(os))) { 1071 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 1072 if (dmu_objset_userobjused_enabled(os)) { 1073 ds->ds_feature_activation[ 1074 SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; 1075 os->os_phys->os_flags |= 1076 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; 1077 } 1078 if (dmu_objset_projectquota_enabled(os)) { 1079 ds->ds_feature_activation[ 1080 SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; 1081 os->os_phys->os_flags |= 1082 OBJSET_FLAG_PROJECTQUOTA_COMPLETE; 1083 } 1084 os->os_flags = os->os_phys->os_flags; 1085 } 1086 1087 dsl_dataset_dirty(ds, tx); 1088 1089 return (os); 1090 } 1091 1092 /* called from dsl for meta-objset */ 1093 objset_t * 1094 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 1095 dmu_objset_type_t type, dmu_tx_t *tx) 1096 { 1097 return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx)); 1098 } 1099 1100 typedef struct dmu_objset_create_arg { 1101 const char *doca_name; 1102 cred_t *doca_cred; 1103 proc_t *doca_proc; 1104 void (*doca_userfunc)(objset_t *os, void *arg, 1105 cred_t *cr, dmu_tx_t *tx); 1106 void *doca_userarg; 1107 dmu_objset_type_t doca_type; 1108 uint64_t doca_flags; 1109 dsl_crypto_params_t *doca_dcp; 1110 } dmu_objset_create_arg_t; 1111 1112 /*ARGSUSED*/ 1113 static int 1114 dmu_objset_create_check(void *arg, dmu_tx_t *tx) 1115 { 1116 dmu_objset_create_arg_t *doca = arg; 1117 dsl_pool_t *dp = dmu_tx_pool(tx); 1118 dsl_dir_t *pdd; 1119 dsl_dataset_t *parentds; 1120 objset_t *parentos; 1121 const char *tail; 1122 int error; 1123 1124 if (strchr(doca->doca_name, '@') != NULL) 1125 return (SET_ERROR(EINVAL)); 1126 1127 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) 1128 return (SET_ERROR(ENAMETOOLONG)); 1129 1130 if (dataset_nestcheck(doca->doca_name) != 0) 1131 return (SET_ERROR(ENAMETOOLONG)); 1132 1133 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); 1134 if (error != 0) 1135 return (error); 1136 if (tail == NULL) { 1137 dsl_dir_rele(pdd, FTAG); 1138 return (SET_ERROR(EEXIST)); 1139 } 1140 1141 error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp, NULL); 1142 if (error != 0) { 1143 dsl_dir_rele(pdd, FTAG); 1144 return (error); 1145 } 1146 1147 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 1148 doca->doca_cred, doca->doca_proc); 1149 if (error != 0) { 1150 dsl_dir_rele(pdd, FTAG); 1151 return (error); 1152 } 1153 1154 /* can't create below anything but filesystems (eg. no ZVOLs) */ 1155 error = dsl_dataset_hold_obj(pdd->dd_pool, 1156 dsl_dir_phys(pdd)->dd_head_dataset_obj, FTAG, &parentds); 1157 if (error != 0) { 1158 dsl_dir_rele(pdd, FTAG); 1159 return (error); 1160 } 1161 error = dmu_objset_from_ds(parentds, &parentos); 1162 if (error != 0) { 1163 dsl_dataset_rele(parentds, FTAG); 1164 dsl_dir_rele(pdd, FTAG); 1165 return (error); 1166 } 1167 if (dmu_objset_type(parentos) != DMU_OST_ZFS) { 1168 dsl_dataset_rele(parentds, FTAG); 1169 dsl_dir_rele(pdd, FTAG); 1170 return (SET_ERROR(ZFS_ERR_WRONG_PARENT)); 1171 } 1172 dsl_dataset_rele(parentds, FTAG); 1173 dsl_dir_rele(pdd, FTAG); 1174 1175 return (error); 1176 } 1177 1178 static void 1179 dmu_objset_create_sync(void *arg, dmu_tx_t *tx) 1180 { 1181 dmu_objset_create_arg_t *doca = arg; 1182 dsl_pool_t *dp = dmu_tx_pool(tx); 1183 spa_t *spa = dp->dp_spa; 1184 dsl_dir_t *pdd; 1185 const char *tail; 1186 dsl_dataset_t *ds; 1187 uint64_t obj; 1188 blkptr_t *bp; 1189 objset_t *os; 1190 zio_t *rzio; 1191 1192 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); 1193 1194 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, 1195 doca->doca_cred, doca->doca_dcp, tx); 1196 1197 VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj, 1198 DS_HOLD_FLAG_DECRYPT, FTAG, &ds)); 1199 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1200 bp = dsl_dataset_get_blkptr(ds); 1201 os = dmu_objset_create_impl(spa, ds, bp, doca->doca_type, tx); 1202 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1203 1204 if (doca->doca_userfunc != NULL) { 1205 doca->doca_userfunc(os, doca->doca_userarg, 1206 doca->doca_cred, tx); 1207 } 1208 1209 /* 1210 * The doca_userfunc() may write out some data that needs to be 1211 * encrypted if the dataset is encrypted (specifically the root 1212 * directory). This data must be written out before the encryption 1213 * key mapping is removed by dsl_dataset_rele_flags(). Force the 1214 * I/O to occur immediately by invoking the relevant sections of 1215 * dsl_pool_sync(). 1216 */ 1217 if (os->os_encrypted) { 1218 dsl_dataset_t *tmpds = NULL; 1219 boolean_t need_sync_done = B_FALSE; 1220 1221 mutex_enter(&ds->ds_lock); 1222 ds->ds_owner = FTAG; 1223 mutex_exit(&ds->ds_lock); 1224 1225 rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1226 tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, 1227 tx->tx_txg); 1228 if (tmpds != NULL) { 1229 dsl_dataset_sync(ds, rzio, tx); 1230 need_sync_done = B_TRUE; 1231 } 1232 VERIFY0(zio_wait(rzio)); 1233 1234 dmu_objset_do_userquota_updates(os, tx); 1235 taskq_wait(dp->dp_sync_taskq); 1236 if (txg_list_member(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1237 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1238 key_mapping_rele(spa, ds->ds_key_mapping, ds); 1239 } 1240 1241 rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1242 tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, 1243 tx->tx_txg); 1244 if (tmpds != NULL) { 1245 dmu_buf_rele(ds->ds_dbuf, ds); 1246 dsl_dataset_sync(ds, rzio, tx); 1247 } 1248 VERIFY0(zio_wait(rzio)); 1249 1250 if (need_sync_done) { 1251 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1252 key_mapping_rele(spa, ds->ds_key_mapping, ds); 1253 dsl_dataset_sync_done(ds, tx); 1254 } 1255 1256 mutex_enter(&ds->ds_lock); 1257 ds->ds_owner = NULL; 1258 mutex_exit(&ds->ds_lock); 1259 } 1260 1261 spa_history_log_internal_ds(ds, "create", tx, " "); 1262 1263 dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG); 1264 dsl_dir_rele(pdd, FTAG); 1265 } 1266 1267 int 1268 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 1269 dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg) 1270 { 1271 dmu_objset_create_arg_t doca; 1272 dsl_crypto_params_t tmp_dcp = { 0 }; 1273 1274 doca.doca_name = name; 1275 doca.doca_cred = CRED(); 1276 doca.doca_proc = curproc; 1277 doca.doca_flags = flags; 1278 doca.doca_userfunc = func; 1279 doca.doca_userarg = arg; 1280 doca.doca_type = type; 1281 1282 /* 1283 * Some callers (mostly for testing) do not provide a dcp on their 1284 * own but various code inside the sync task will require it to be 1285 * allocated. Rather than adding NULL checks throughout this code 1286 * or adding dummy dcp's to all of the callers we simply create a 1287 * dummy one here and use that. This zero dcp will have the same 1288 * effect as asking for inheritance of all encryption params. 1289 */ 1290 doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp; 1291 1292 int rv = dsl_sync_task(name, 1293 dmu_objset_create_check, dmu_objset_create_sync, &doca, 1294 6, ZFS_SPACE_CHECK_NORMAL); 1295 1296 if (rv == 0) 1297 zvol_create_minor(name); 1298 return (rv); 1299 } 1300 1301 typedef struct dmu_objset_clone_arg { 1302 const char *doca_clone; 1303 const char *doca_origin; 1304 cred_t *doca_cred; 1305 proc_t *doca_proc; 1306 } dmu_objset_clone_arg_t; 1307 1308 /*ARGSUSED*/ 1309 static int 1310 dmu_objset_clone_check(void *arg, dmu_tx_t *tx) 1311 { 1312 dmu_objset_clone_arg_t *doca = arg; 1313 dsl_dir_t *pdd; 1314 const char *tail; 1315 int error; 1316 dsl_dataset_t *origin; 1317 dsl_pool_t *dp = dmu_tx_pool(tx); 1318 1319 if (strchr(doca->doca_clone, '@') != NULL) 1320 return (SET_ERROR(EINVAL)); 1321 1322 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) 1323 return (SET_ERROR(ENAMETOOLONG)); 1324 1325 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); 1326 if (error != 0) 1327 return (error); 1328 if (tail == NULL) { 1329 dsl_dir_rele(pdd, FTAG); 1330 return (SET_ERROR(EEXIST)); 1331 } 1332 1333 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 1334 doca->doca_cred, doca->doca_proc); 1335 if (error != 0) { 1336 dsl_dir_rele(pdd, FTAG); 1337 return (SET_ERROR(EDQUOT)); 1338 } 1339 1340 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); 1341 if (error != 0) { 1342 dsl_dir_rele(pdd, FTAG); 1343 return (error); 1344 } 1345 1346 /* You can only clone snapshots, not the head datasets. */ 1347 if (!origin->ds_is_snapshot) { 1348 dsl_dataset_rele(origin, FTAG); 1349 dsl_dir_rele(pdd, FTAG); 1350 return (SET_ERROR(EINVAL)); 1351 } 1352 1353 dsl_dataset_rele(origin, FTAG); 1354 dsl_dir_rele(pdd, FTAG); 1355 1356 return (0); 1357 } 1358 1359 static void 1360 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) 1361 { 1362 dmu_objset_clone_arg_t *doca = arg; 1363 dsl_pool_t *dp = dmu_tx_pool(tx); 1364 dsl_dir_t *pdd; 1365 const char *tail; 1366 dsl_dataset_t *origin, *ds; 1367 uint64_t obj; 1368 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 1369 1370 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); 1371 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); 1372 1373 obj = dsl_dataset_create_sync(pdd, tail, origin, 0, 1374 doca->doca_cred, NULL, tx); 1375 1376 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 1377 dsl_dataset_name(origin, namebuf); 1378 spa_history_log_internal_ds(ds, "clone", tx, 1379 "origin=%s (%llu)", namebuf, (u_longlong_t)origin->ds_object); 1380 dsl_dataset_rele(ds, FTAG); 1381 dsl_dataset_rele(origin, FTAG); 1382 dsl_dir_rele(pdd, FTAG); 1383 } 1384 1385 int 1386 dmu_objset_clone(const char *clone, const char *origin) 1387 { 1388 dmu_objset_clone_arg_t doca; 1389 1390 doca.doca_clone = clone; 1391 doca.doca_origin = origin; 1392 doca.doca_cred = CRED(); 1393 doca.doca_proc = curproc; 1394 1395 int rv = dsl_sync_task(clone, 1396 dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 1397 6, ZFS_SPACE_CHECK_NORMAL); 1398 1399 if (rv == 0) 1400 zvol_create_minor(clone); 1401 1402 return (rv); 1403 } 1404 1405 int 1406 dmu_objset_snapshot_one(const char *fsname, const char *snapname) 1407 { 1408 int err; 1409 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); 1410 nvlist_t *snaps = fnvlist_alloc(); 1411 1412 fnvlist_add_boolean(snaps, longsnap); 1413 kmem_strfree(longsnap); 1414 err = dsl_dataset_snapshot(snaps, NULL, NULL); 1415 fnvlist_free(snaps); 1416 return (err); 1417 } 1418 1419 static void 1420 dmu_objset_upgrade_task_cb(void *data) 1421 { 1422 objset_t *os = data; 1423 1424 mutex_enter(&os->os_upgrade_lock); 1425 os->os_upgrade_status = EINTR; 1426 if (!os->os_upgrade_exit) { 1427 mutex_exit(&os->os_upgrade_lock); 1428 1429 os->os_upgrade_status = os->os_upgrade_cb(os); 1430 mutex_enter(&os->os_upgrade_lock); 1431 } 1432 os->os_upgrade_exit = B_TRUE; 1433 os->os_upgrade_id = 0; 1434 mutex_exit(&os->os_upgrade_lock); 1435 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1436 } 1437 1438 static void 1439 dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb) 1440 { 1441 if (os->os_upgrade_id != 0) 1442 return; 1443 1444 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 1445 dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag); 1446 1447 mutex_enter(&os->os_upgrade_lock); 1448 if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) { 1449 os->os_upgrade_exit = B_FALSE; 1450 os->os_upgrade_cb = cb; 1451 os->os_upgrade_id = taskq_dispatch( 1452 os->os_spa->spa_upgrade_taskq, 1453 dmu_objset_upgrade_task_cb, os, TQ_SLEEP); 1454 if (os->os_upgrade_id == TASKQID_INVALID) { 1455 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1456 os->os_upgrade_status = ENOMEM; 1457 } 1458 } 1459 mutex_exit(&os->os_upgrade_lock); 1460 } 1461 1462 static void 1463 dmu_objset_upgrade_stop(objset_t *os) 1464 { 1465 mutex_enter(&os->os_upgrade_lock); 1466 os->os_upgrade_exit = B_TRUE; 1467 if (os->os_upgrade_id != 0) { 1468 taskqid_t id = os->os_upgrade_id; 1469 1470 os->os_upgrade_id = 0; 1471 mutex_exit(&os->os_upgrade_lock); 1472 1473 if ((taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id)) == 0) { 1474 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1475 } 1476 txg_wait_synced(os->os_spa->spa_dsl_pool, 0); 1477 } else { 1478 mutex_exit(&os->os_upgrade_lock); 1479 } 1480 } 1481 1482 static void 1483 dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) 1484 { 1485 dnode_t *dn; 1486 1487 while ((dn = multilist_sublist_head(list)) != NULL) { 1488 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 1489 ASSERT(dn->dn_dbuf->db_data_pending); 1490 /* 1491 * Initialize dn_zio outside dnode_sync() because the 1492 * meta-dnode needs to set it outside dnode_sync(). 1493 */ 1494 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; 1495 ASSERT(dn->dn_zio); 1496 1497 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); 1498 multilist_sublist_remove(list, dn); 1499 1500 /* 1501 * If we are not doing useraccounting (os_synced_dnodes == NULL) 1502 * we are done with this dnode for this txg. Unset dn_dirty_txg 1503 * if later txgs aren't dirtying it so that future holders do 1504 * not get a stale value. Otherwise, we will do this in 1505 * userquota_updates_task() when processing has completely 1506 * finished for this txg. 1507 */ 1508 multilist_t *newlist = dn->dn_objset->os_synced_dnodes; 1509 if (newlist != NULL) { 1510 (void) dnode_add_ref(dn, newlist); 1511 multilist_insert(newlist, dn); 1512 } else { 1513 mutex_enter(&dn->dn_mtx); 1514 if (dn->dn_dirty_txg == tx->tx_txg) 1515 dn->dn_dirty_txg = 0; 1516 mutex_exit(&dn->dn_mtx); 1517 } 1518 1519 dnode_sync(dn, tx); 1520 } 1521 } 1522 1523 /* ARGSUSED */ 1524 static void 1525 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) 1526 { 1527 blkptr_t *bp = zio->io_bp; 1528 objset_t *os = arg; 1529 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; 1530 uint64_t fill = 0; 1531 1532 ASSERT(!BP_IS_EMBEDDED(bp)); 1533 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); 1534 ASSERT0(BP_GET_LEVEL(bp)); 1535 1536 /* 1537 * Update rootbp fill count: it should be the number of objects 1538 * allocated in the object set (not counting the "special" 1539 * objects that are stored in the objset_phys_t -- the meta 1540 * dnode and user/group/project accounting objects). 1541 */ 1542 for (int i = 0; i < dnp->dn_nblkptr; i++) 1543 fill += BP_GET_FILL(&dnp->dn_blkptr[i]); 1544 1545 BP_SET_FILL(bp, fill); 1546 1547 if (os->os_dsl_dataset != NULL) 1548 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); 1549 *os->os_rootbp = *bp; 1550 if (os->os_dsl_dataset != NULL) 1551 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); 1552 } 1553 1554 /* ARGSUSED */ 1555 static void 1556 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) 1557 { 1558 blkptr_t *bp = zio->io_bp; 1559 blkptr_t *bp_orig = &zio->io_bp_orig; 1560 objset_t *os = arg; 1561 1562 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { 1563 ASSERT(BP_EQUAL(bp, bp_orig)); 1564 } else { 1565 dsl_dataset_t *ds = os->os_dsl_dataset; 1566 dmu_tx_t *tx = os->os_synctx; 1567 1568 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); 1569 dsl_dataset_block_born(ds, bp, tx); 1570 } 1571 kmem_free(bp, sizeof (*bp)); 1572 } 1573 1574 typedef struct sync_dnodes_arg { 1575 multilist_t *sda_list; 1576 int sda_sublist_idx; 1577 multilist_t *sda_newlist; 1578 dmu_tx_t *sda_tx; 1579 } sync_dnodes_arg_t; 1580 1581 static void 1582 sync_dnodes_task(void *arg) 1583 { 1584 sync_dnodes_arg_t *sda = arg; 1585 1586 multilist_sublist_t *ms = 1587 multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx); 1588 1589 dmu_objset_sync_dnodes(ms, sda->sda_tx); 1590 1591 multilist_sublist_unlock(ms); 1592 1593 kmem_free(sda, sizeof (*sda)); 1594 } 1595 1596 1597 /* called from dsl */ 1598 void 1599 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) 1600 { 1601 int txgoff; 1602 zbookmark_phys_t zb; 1603 zio_prop_t zp; 1604 zio_t *zio; 1605 list_t *list; 1606 dbuf_dirty_record_t *dr; 1607 int num_sublists; 1608 multilist_t *ml; 1609 blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); 1610 *blkptr_copy = *os->os_rootbp; 1611 1612 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); 1613 1614 ASSERT(dmu_tx_is_syncing(tx)); 1615 /* XXX the write_done callback should really give us the tx... */ 1616 os->os_synctx = tx; 1617 1618 if (os->os_dsl_dataset == NULL) { 1619 /* 1620 * This is the MOS. If we have upgraded, 1621 * spa_max_replication() could change, so reset 1622 * os_copies here. 1623 */ 1624 os->os_copies = spa_max_replication(os->os_spa); 1625 } 1626 1627 /* 1628 * Create the root block IO 1629 */ 1630 SET_BOOKMARK(&zb, os->os_dsl_dataset ? 1631 os->os_dsl_dataset->ds_object : DMU_META_OBJSET, 1632 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 1633 arc_release(os->os_phys_buf, &os->os_phys_buf); 1634 1635 dmu_write_policy(os, NULL, 0, 0, &zp); 1636 1637 /* 1638 * If we are either claiming the ZIL or doing a raw receive, write 1639 * out the os_phys_buf raw. Neither of these actions will effect the 1640 * MAC at this point. 1641 */ 1642 if (os->os_raw_receive || 1643 os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { 1644 ASSERT(os->os_encrypted); 1645 arc_convert_to_raw(os->os_phys_buf, 1646 os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER, 1647 DMU_OT_OBJSET, NULL, NULL, NULL); 1648 } 1649 1650 zio = arc_write(pio, os->os_spa, tx->tx_txg, 1651 blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), 1652 &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, 1653 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); 1654 1655 /* 1656 * Sync special dnodes - the parent IO for the sync is the root block 1657 */ 1658 DMU_META_DNODE(os)->dn_zio = zio; 1659 dnode_sync(DMU_META_DNODE(os), tx); 1660 1661 os->os_phys->os_flags = os->os_flags; 1662 1663 if (DMU_USERUSED_DNODE(os) && 1664 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1665 DMU_USERUSED_DNODE(os)->dn_zio = zio; 1666 dnode_sync(DMU_USERUSED_DNODE(os), tx); 1667 DMU_GROUPUSED_DNODE(os)->dn_zio = zio; 1668 dnode_sync(DMU_GROUPUSED_DNODE(os), tx); 1669 } 1670 1671 if (DMU_PROJECTUSED_DNODE(os) && 1672 DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1673 DMU_PROJECTUSED_DNODE(os)->dn_zio = zio; 1674 dnode_sync(DMU_PROJECTUSED_DNODE(os), tx); 1675 } 1676 1677 txgoff = tx->tx_txg & TXG_MASK; 1678 1679 if (dmu_objset_userused_enabled(os) && 1680 (!os->os_encrypted || !dmu_objset_is_receiving(os))) { 1681 /* 1682 * We must create the list here because it uses the 1683 * dn_dirty_link[] of this txg. But it may already 1684 * exist because we call dsl_dataset_sync() twice per txg. 1685 */ 1686 if (os->os_synced_dnodes == NULL) { 1687 os->os_synced_dnodes = 1688 multilist_create(sizeof (dnode_t), 1689 offsetof(dnode_t, dn_dirty_link[txgoff]), 1690 dnode_multilist_index_func); 1691 } else { 1692 ASSERT3U(os->os_synced_dnodes->ml_offset, ==, 1693 offsetof(dnode_t, dn_dirty_link[txgoff])); 1694 } 1695 } 1696 1697 ml = os->os_dirty_dnodes[txgoff]; 1698 num_sublists = multilist_get_num_sublists(ml); 1699 for (int i = 0; i < num_sublists; i++) { 1700 if (multilist_sublist_is_empty_idx(ml, i)) 1701 continue; 1702 sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP); 1703 sda->sda_list = ml; 1704 sda->sda_sublist_idx = i; 1705 sda->sda_tx = tx; 1706 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, 1707 sync_dnodes_task, sda, 0); 1708 /* callback frees sda */ 1709 } 1710 taskq_wait(dmu_objset_pool(os)->dp_sync_taskq); 1711 1712 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; 1713 while ((dr = list_head(list)) != NULL) { 1714 ASSERT0(dr->dr_dbuf->db_level); 1715 list_remove(list, dr); 1716 zio_nowait(dr->dr_zio); 1717 } 1718 1719 /* Enable dnode backfill if enough objects have been freed. */ 1720 if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { 1721 os->os_rescan_dnodes = B_TRUE; 1722 os->os_freed_dnodes = 0; 1723 } 1724 1725 /* 1726 * Free intent log blocks up to this tx. 1727 */ 1728 zil_sync(os->os_zil, tx); 1729 os->os_phys->os_zil_header = os->os_zil_header; 1730 zio_nowait(zio); 1731 } 1732 1733 boolean_t 1734 dmu_objset_is_dirty(objset_t *os, uint64_t txg) 1735 { 1736 return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK])); 1737 } 1738 1739 static file_info_cb_t *file_cbs[DMU_OST_NUMTYPES]; 1740 1741 void 1742 dmu_objset_register_type(dmu_objset_type_t ost, file_info_cb_t *cb) 1743 { 1744 file_cbs[ost] = cb; 1745 } 1746 1747 int 1748 dmu_get_file_info(objset_t *os, dmu_object_type_t bonustype, const void *data, 1749 zfs_file_info_t *zfi) 1750 { 1751 file_info_cb_t *cb = file_cbs[os->os_phys->os_type]; 1752 if (cb == NULL) 1753 return (EINVAL); 1754 return (cb(bonustype, data, zfi)); 1755 } 1756 1757 boolean_t 1758 dmu_objset_userused_enabled(objset_t *os) 1759 { 1760 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && 1761 file_cbs[os->os_phys->os_type] != NULL && 1762 DMU_USERUSED_DNODE(os) != NULL); 1763 } 1764 1765 boolean_t 1766 dmu_objset_userobjused_enabled(objset_t *os) 1767 { 1768 return (dmu_objset_userused_enabled(os) && 1769 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING)); 1770 } 1771 1772 boolean_t 1773 dmu_objset_projectquota_enabled(objset_t *os) 1774 { 1775 return (file_cbs[os->os_phys->os_type] != NULL && 1776 DMU_PROJECTUSED_DNODE(os) != NULL && 1777 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA)); 1778 } 1779 1780 typedef struct userquota_node { 1781 /* must be in the first filed, see userquota_update_cache() */ 1782 char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN]; 1783 int64_t uqn_delta; 1784 avl_node_t uqn_node; 1785 } userquota_node_t; 1786 1787 typedef struct userquota_cache { 1788 avl_tree_t uqc_user_deltas; 1789 avl_tree_t uqc_group_deltas; 1790 avl_tree_t uqc_project_deltas; 1791 } userquota_cache_t; 1792 1793 static int 1794 userquota_compare(const void *l, const void *r) 1795 { 1796 const userquota_node_t *luqn = l; 1797 const userquota_node_t *ruqn = r; 1798 int rv; 1799 1800 /* 1801 * NB: can only access uqn_id because userquota_update_cache() doesn't 1802 * pass in an entire userquota_node_t. 1803 */ 1804 rv = strcmp(luqn->uqn_id, ruqn->uqn_id); 1805 1806 return (TREE_ISIGN(rv)); 1807 } 1808 1809 static void 1810 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) 1811 { 1812 void *cookie; 1813 userquota_node_t *uqn; 1814 1815 ASSERT(dmu_tx_is_syncing(tx)); 1816 1817 cookie = NULL; 1818 while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, 1819 &cookie)) != NULL) { 1820 /* 1821 * os_userused_lock protects against concurrent calls to 1822 * zap_increment_int(). It's needed because zap_increment_int() 1823 * is not thread-safe (i.e. not atomic). 1824 */ 1825 mutex_enter(&os->os_userused_lock); 1826 VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT, 1827 uqn->uqn_id, uqn->uqn_delta, tx)); 1828 mutex_exit(&os->os_userused_lock); 1829 kmem_free(uqn, sizeof (*uqn)); 1830 } 1831 avl_destroy(&cache->uqc_user_deltas); 1832 1833 cookie = NULL; 1834 while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, 1835 &cookie)) != NULL) { 1836 mutex_enter(&os->os_userused_lock); 1837 VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT, 1838 uqn->uqn_id, uqn->uqn_delta, tx)); 1839 mutex_exit(&os->os_userused_lock); 1840 kmem_free(uqn, sizeof (*uqn)); 1841 } 1842 avl_destroy(&cache->uqc_group_deltas); 1843 1844 if (dmu_objset_projectquota_enabled(os)) { 1845 cookie = NULL; 1846 while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas, 1847 &cookie)) != NULL) { 1848 mutex_enter(&os->os_userused_lock); 1849 VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT, 1850 uqn->uqn_id, uqn->uqn_delta, tx)); 1851 mutex_exit(&os->os_userused_lock); 1852 kmem_free(uqn, sizeof (*uqn)); 1853 } 1854 avl_destroy(&cache->uqc_project_deltas); 1855 } 1856 } 1857 1858 static void 1859 userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta) 1860 { 1861 userquota_node_t *uqn; 1862 avl_index_t idx; 1863 1864 ASSERT(strlen(id) < sizeof (uqn->uqn_id)); 1865 /* 1866 * Use id directly for searching because uqn_id is the first field of 1867 * userquota_node_t and fields after uqn_id won't be accessed in 1868 * avl_find(). 1869 */ 1870 uqn = avl_find(avl, (const void *)id, &idx); 1871 if (uqn == NULL) { 1872 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); 1873 strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id)); 1874 avl_insert(avl, uqn, idx); 1875 } 1876 uqn->uqn_delta += delta; 1877 } 1878 1879 static void 1880 do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used, 1881 uint64_t flags, uint64_t user, uint64_t group, uint64_t project, 1882 boolean_t subtract) 1883 { 1884 if (flags & DNODE_FLAG_USERUSED_ACCOUNTED) { 1885 int64_t delta = DNODE_MIN_SIZE + used; 1886 char name[20]; 1887 1888 if (subtract) 1889 delta = -delta; 1890 1891 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)user); 1892 userquota_update_cache(&cache->uqc_user_deltas, name, delta); 1893 1894 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)group); 1895 userquota_update_cache(&cache->uqc_group_deltas, name, delta); 1896 1897 if (dmu_objset_projectquota_enabled(os)) { 1898 (void) snprintf(name, sizeof (name), "%llx", 1899 (longlong_t)project); 1900 userquota_update_cache(&cache->uqc_project_deltas, 1901 name, delta); 1902 } 1903 } 1904 } 1905 1906 static void 1907 do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags, 1908 uint64_t user, uint64_t group, uint64_t project, boolean_t subtract) 1909 { 1910 if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) { 1911 char name[20 + DMU_OBJACCT_PREFIX_LEN]; 1912 int delta = subtract ? -1 : 1; 1913 1914 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", 1915 (longlong_t)user); 1916 userquota_update_cache(&cache->uqc_user_deltas, name, delta); 1917 1918 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", 1919 (longlong_t)group); 1920 userquota_update_cache(&cache->uqc_group_deltas, name, delta); 1921 1922 if (dmu_objset_projectquota_enabled(os)) { 1923 (void) snprintf(name, sizeof (name), 1924 DMU_OBJACCT_PREFIX "%llx", (longlong_t)project); 1925 userquota_update_cache(&cache->uqc_project_deltas, 1926 name, delta); 1927 } 1928 } 1929 } 1930 1931 typedef struct userquota_updates_arg { 1932 objset_t *uua_os; 1933 int uua_sublist_idx; 1934 dmu_tx_t *uua_tx; 1935 } userquota_updates_arg_t; 1936 1937 static void 1938 userquota_updates_task(void *arg) 1939 { 1940 userquota_updates_arg_t *uua = arg; 1941 objset_t *os = uua->uua_os; 1942 dmu_tx_t *tx = uua->uua_tx; 1943 dnode_t *dn; 1944 userquota_cache_t cache = { { 0 } }; 1945 1946 multilist_sublist_t *list = 1947 multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx); 1948 1949 ASSERT(multilist_sublist_head(list) == NULL || 1950 dmu_objset_userused_enabled(os)); 1951 avl_create(&cache.uqc_user_deltas, userquota_compare, 1952 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 1953 avl_create(&cache.uqc_group_deltas, userquota_compare, 1954 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 1955 if (dmu_objset_projectquota_enabled(os)) 1956 avl_create(&cache.uqc_project_deltas, userquota_compare, 1957 sizeof (userquota_node_t), offsetof(userquota_node_t, 1958 uqn_node)); 1959 1960 while ((dn = multilist_sublist_head(list)) != NULL) { 1961 int flags; 1962 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); 1963 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || 1964 dn->dn_phys->dn_flags & 1965 DNODE_FLAG_USERUSED_ACCOUNTED); 1966 1967 flags = dn->dn_id_flags; 1968 ASSERT(flags); 1969 if (flags & DN_ID_OLD_EXIST) { 1970 do_userquota_update(os, &cache, dn->dn_oldused, 1971 dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, 1972 dn->dn_oldprojid, B_TRUE); 1973 do_userobjquota_update(os, &cache, dn->dn_oldflags, 1974 dn->dn_olduid, dn->dn_oldgid, 1975 dn->dn_oldprojid, B_TRUE); 1976 } 1977 if (flags & DN_ID_NEW_EXIST) { 1978 do_userquota_update(os, &cache, 1979 DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, 1980 dn->dn_newuid, dn->dn_newgid, 1981 dn->dn_newprojid, B_FALSE); 1982 do_userobjquota_update(os, &cache, 1983 dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, 1984 dn->dn_newprojid, B_FALSE); 1985 } 1986 1987 mutex_enter(&dn->dn_mtx); 1988 dn->dn_oldused = 0; 1989 dn->dn_oldflags = 0; 1990 if (dn->dn_id_flags & DN_ID_NEW_EXIST) { 1991 dn->dn_olduid = dn->dn_newuid; 1992 dn->dn_oldgid = dn->dn_newgid; 1993 dn->dn_oldprojid = dn->dn_newprojid; 1994 dn->dn_id_flags |= DN_ID_OLD_EXIST; 1995 if (dn->dn_bonuslen == 0) 1996 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 1997 else 1998 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 1999 } 2000 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); 2001 if (dn->dn_dirty_txg == spa_syncing_txg(os->os_spa)) 2002 dn->dn_dirty_txg = 0; 2003 mutex_exit(&dn->dn_mtx); 2004 2005 multilist_sublist_remove(list, dn); 2006 dnode_rele(dn, os->os_synced_dnodes); 2007 } 2008 do_userquota_cacheflush(os, &cache, tx); 2009 multilist_sublist_unlock(list); 2010 kmem_free(uua, sizeof (*uua)); 2011 } 2012 2013 void 2014 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) 2015 { 2016 int num_sublists; 2017 2018 if (!dmu_objset_userused_enabled(os)) 2019 return; 2020 2021 /* 2022 * If this is a raw receive just return and handle accounting 2023 * later when we have the keys loaded. We also don't do user 2024 * accounting during claiming since the datasets are not owned 2025 * for the duration of claiming and this txg should only be 2026 * used for recovery. 2027 */ 2028 if (os->os_encrypted && dmu_objset_is_receiving(os)) 2029 return; 2030 2031 if (tx->tx_txg <= os->os_spa->spa_claim_max_txg) 2032 return; 2033 2034 /* Allocate the user/group/project used objects if necessary. */ 2035 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 2036 VERIFY0(zap_create_claim(os, 2037 DMU_USERUSED_OBJECT, 2038 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2039 VERIFY0(zap_create_claim(os, 2040 DMU_GROUPUSED_OBJECT, 2041 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2042 } 2043 2044 if (dmu_objset_projectquota_enabled(os) && 2045 DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 2046 VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT, 2047 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2048 } 2049 2050 num_sublists = multilist_get_num_sublists(os->os_synced_dnodes); 2051 for (int i = 0; i < num_sublists; i++) { 2052 if (multilist_sublist_is_empty_idx(os->os_synced_dnodes, i)) 2053 continue; 2054 userquota_updates_arg_t *uua = 2055 kmem_alloc(sizeof (*uua), KM_SLEEP); 2056 uua->uua_os = os; 2057 uua->uua_sublist_idx = i; 2058 uua->uua_tx = tx; 2059 /* note: caller does taskq_wait() */ 2060 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, 2061 userquota_updates_task, uua, 0); 2062 /* callback frees uua */ 2063 } 2064 } 2065 2066 /* 2067 * Returns a pointer to data to find uid/gid from 2068 * 2069 * If a dirty record for transaction group that is syncing can't 2070 * be found then NULL is returned. In the NULL case it is assumed 2071 * the uid/gid aren't changing. 2072 */ 2073 static void * 2074 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) 2075 { 2076 dbuf_dirty_record_t *dr; 2077 void *data; 2078 2079 if (db->db_dirtycnt == 0) 2080 return (db->db.db_data); /* Nothing is changing */ 2081 2082 dr = dbuf_find_dirty_eq(db, tx->tx_txg); 2083 2084 if (dr == NULL) { 2085 data = NULL; 2086 } else { 2087 dnode_t *dn; 2088 2089 DB_DNODE_ENTER(dr->dr_dbuf); 2090 dn = DB_DNODE(dr->dr_dbuf); 2091 2092 if (dn->dn_bonuslen == 0 && 2093 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) 2094 data = dr->dt.dl.dr_data->b_data; 2095 else 2096 data = dr->dt.dl.dr_data; 2097 2098 DB_DNODE_EXIT(dr->dr_dbuf); 2099 } 2100 2101 return (data); 2102 } 2103 2104 void 2105 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) 2106 { 2107 objset_t *os = dn->dn_objset; 2108 void *data = NULL; 2109 dmu_buf_impl_t *db = NULL; 2110 int flags = dn->dn_id_flags; 2111 int error; 2112 boolean_t have_spill = B_FALSE; 2113 2114 if (!dmu_objset_userused_enabled(dn->dn_objset)) 2115 return; 2116 2117 /* 2118 * Raw receives introduce a problem with user accounting. Raw 2119 * receives cannot update the user accounting info because the 2120 * user ids and the sizes are encrypted. To guarantee that we 2121 * never end up with bad user accounting, we simply disable it 2122 * during raw receives. We also disable this for normal receives 2123 * so that an incremental raw receive may be done on top of an 2124 * existing non-raw receive. 2125 */ 2126 if (os->os_encrypted && dmu_objset_is_receiving(os)) 2127 return; 2128 2129 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| 2130 DN_ID_CHKED_SPILL))) 2131 return; 2132 2133 if (before && dn->dn_bonuslen != 0) 2134 data = DN_BONUS(dn->dn_phys); 2135 else if (!before && dn->dn_bonuslen != 0) { 2136 if (dn->dn_bonus) { 2137 db = dn->dn_bonus; 2138 mutex_enter(&db->db_mtx); 2139 data = dmu_objset_userquota_find_data(db, tx); 2140 } else { 2141 data = DN_BONUS(dn->dn_phys); 2142 } 2143 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { 2144 int rf = 0; 2145 2146 if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) 2147 rf |= DB_RF_HAVESTRUCT; 2148 error = dmu_spill_hold_by_dnode(dn, 2149 rf | DB_RF_MUST_SUCCEED, 2150 FTAG, (dmu_buf_t **)&db); 2151 ASSERT(error == 0); 2152 mutex_enter(&db->db_mtx); 2153 data = (before) ? db->db.db_data : 2154 dmu_objset_userquota_find_data(db, tx); 2155 have_spill = B_TRUE; 2156 } else { 2157 mutex_enter(&dn->dn_mtx); 2158 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2159 mutex_exit(&dn->dn_mtx); 2160 return; 2161 } 2162 2163 /* 2164 * Must always call the callback in case the object 2165 * type has changed and that type isn't an object type to track 2166 */ 2167 zfs_file_info_t zfi; 2168 error = file_cbs[os->os_phys->os_type](dn->dn_bonustype, data, &zfi); 2169 2170 if (before) { 2171 ASSERT(data); 2172 dn->dn_olduid = zfi.zfi_user; 2173 dn->dn_oldgid = zfi.zfi_group; 2174 dn->dn_oldprojid = zfi.zfi_project; 2175 } else if (data) { 2176 dn->dn_newuid = zfi.zfi_user; 2177 dn->dn_newgid = zfi.zfi_group; 2178 dn->dn_newprojid = zfi.zfi_project; 2179 } 2180 2181 /* 2182 * Preserve existing uid/gid when the callback can't determine 2183 * what the new uid/gid are and the callback returned EEXIST. 2184 * The EEXIST error tells us to just use the existing uid/gid. 2185 * If we don't know what the old values are then just assign 2186 * them to 0, since that is a new file being created. 2187 */ 2188 if (!before && data == NULL && error == EEXIST) { 2189 if (flags & DN_ID_OLD_EXIST) { 2190 dn->dn_newuid = dn->dn_olduid; 2191 dn->dn_newgid = dn->dn_oldgid; 2192 dn->dn_newprojid = dn->dn_oldprojid; 2193 } else { 2194 dn->dn_newuid = 0; 2195 dn->dn_newgid = 0; 2196 dn->dn_newprojid = ZFS_DEFAULT_PROJID; 2197 } 2198 error = 0; 2199 } 2200 2201 if (db) 2202 mutex_exit(&db->db_mtx); 2203 2204 mutex_enter(&dn->dn_mtx); 2205 if (error == 0 && before) 2206 dn->dn_id_flags |= DN_ID_OLD_EXIST; 2207 if (error == 0 && !before) 2208 dn->dn_id_flags |= DN_ID_NEW_EXIST; 2209 2210 if (have_spill) { 2211 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 2212 } else { 2213 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2214 } 2215 mutex_exit(&dn->dn_mtx); 2216 if (have_spill) 2217 dmu_buf_rele((dmu_buf_t *)db, FTAG); 2218 } 2219 2220 boolean_t 2221 dmu_objset_userspace_present(objset_t *os) 2222 { 2223 return (os->os_phys->os_flags & 2224 OBJSET_FLAG_USERACCOUNTING_COMPLETE); 2225 } 2226 2227 boolean_t 2228 dmu_objset_userobjspace_present(objset_t *os) 2229 { 2230 return (os->os_phys->os_flags & 2231 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE); 2232 } 2233 2234 boolean_t 2235 dmu_objset_projectquota_present(objset_t *os) 2236 { 2237 return (os->os_phys->os_flags & 2238 OBJSET_FLAG_PROJECTQUOTA_COMPLETE); 2239 } 2240 2241 static int 2242 dmu_objset_space_upgrade(objset_t *os) 2243 { 2244 uint64_t obj; 2245 int err = 0; 2246 2247 /* 2248 * We simply need to mark every object dirty, so that it will be 2249 * synced out and now accounted. If this is called 2250 * concurrently, or if we already did some work before crashing, 2251 * that's fine, since we track each object's accounted state 2252 * independently. 2253 */ 2254 2255 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { 2256 dmu_tx_t *tx; 2257 dmu_buf_t *db; 2258 int objerr; 2259 2260 mutex_enter(&os->os_upgrade_lock); 2261 if (os->os_upgrade_exit) 2262 err = SET_ERROR(EINTR); 2263 mutex_exit(&os->os_upgrade_lock); 2264 if (err != 0) 2265 return (err); 2266 2267 if (issig(JUSTLOOKING) && issig(FORREAL)) 2268 return (SET_ERROR(EINTR)); 2269 2270 objerr = dmu_bonus_hold(os, obj, FTAG, &db); 2271 if (objerr != 0) 2272 continue; 2273 tx = dmu_tx_create(os); 2274 dmu_tx_hold_bonus(tx, obj); 2275 objerr = dmu_tx_assign(tx, TXG_WAIT); 2276 if (objerr != 0) { 2277 dmu_buf_rele(db, FTAG); 2278 dmu_tx_abort(tx); 2279 continue; 2280 } 2281 dmu_buf_will_dirty(db, tx); 2282 dmu_buf_rele(db, FTAG); 2283 dmu_tx_commit(tx); 2284 } 2285 return (0); 2286 } 2287 2288 int 2289 dmu_objset_userspace_upgrade(objset_t *os) 2290 { 2291 int err = 0; 2292 2293 if (dmu_objset_userspace_present(os)) 2294 return (0); 2295 if (dmu_objset_is_snapshot(os)) 2296 return (SET_ERROR(EINVAL)); 2297 if (!dmu_objset_userused_enabled(os)) 2298 return (SET_ERROR(ENOTSUP)); 2299 2300 err = dmu_objset_space_upgrade(os); 2301 if (err) 2302 return (err); 2303 2304 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 2305 txg_wait_synced(dmu_objset_pool(os), 0); 2306 return (0); 2307 } 2308 2309 static int 2310 dmu_objset_id_quota_upgrade_cb(objset_t *os) 2311 { 2312 int err = 0; 2313 2314 if (dmu_objset_userobjspace_present(os) && 2315 dmu_objset_projectquota_present(os)) 2316 return (0); 2317 if (dmu_objset_is_snapshot(os)) 2318 return (SET_ERROR(EINVAL)); 2319 if (!dmu_objset_userobjused_enabled(os)) 2320 return (SET_ERROR(ENOTSUP)); 2321 if (!dmu_objset_projectquota_enabled(os) && 2322 dmu_objset_userobjspace_present(os)) 2323 return (SET_ERROR(ENOTSUP)); 2324 2325 dmu_objset_ds(os)->ds_feature_activation[ 2326 SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE; 2327 if (dmu_objset_projectquota_enabled(os)) 2328 dmu_objset_ds(os)->ds_feature_activation[ 2329 SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE; 2330 2331 err = dmu_objset_space_upgrade(os); 2332 if (err) 2333 return (err); 2334 2335 os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; 2336 if (dmu_objset_projectquota_enabled(os)) 2337 os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE; 2338 2339 txg_wait_synced(dmu_objset_pool(os), 0); 2340 return (0); 2341 } 2342 2343 void 2344 dmu_objset_id_quota_upgrade(objset_t *os) 2345 { 2346 dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb); 2347 } 2348 2349 boolean_t 2350 dmu_objset_userobjspace_upgradable(objset_t *os) 2351 { 2352 return (dmu_objset_type(os) == DMU_OST_ZFS && 2353 !dmu_objset_is_snapshot(os) && 2354 dmu_objset_userobjused_enabled(os) && 2355 !dmu_objset_userobjspace_present(os) && 2356 spa_writeable(dmu_objset_spa(os))); 2357 } 2358 2359 boolean_t 2360 dmu_objset_projectquota_upgradable(objset_t *os) 2361 { 2362 return (dmu_objset_type(os) == DMU_OST_ZFS && 2363 !dmu_objset_is_snapshot(os) && 2364 dmu_objset_projectquota_enabled(os) && 2365 !dmu_objset_projectquota_present(os) && 2366 spa_writeable(dmu_objset_spa(os))); 2367 } 2368 2369 void 2370 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 2371 uint64_t *usedobjsp, uint64_t *availobjsp) 2372 { 2373 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, 2374 usedobjsp, availobjsp); 2375 } 2376 2377 uint64_t 2378 dmu_objset_fsid_guid(objset_t *os) 2379 { 2380 return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); 2381 } 2382 2383 void 2384 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) 2385 { 2386 stat->dds_type = os->os_phys->os_type; 2387 if (os->os_dsl_dataset) 2388 dsl_dataset_fast_stat(os->os_dsl_dataset, stat); 2389 } 2390 2391 void 2392 dmu_objset_stats(objset_t *os, nvlist_t *nv) 2393 { 2394 ASSERT(os->os_dsl_dataset || 2395 os->os_phys->os_type == DMU_OST_META); 2396 2397 if (os->os_dsl_dataset != NULL) 2398 dsl_dataset_stats(os->os_dsl_dataset, nv); 2399 2400 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, 2401 os->os_phys->os_type); 2402 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, 2403 dmu_objset_userspace_present(os)); 2404 } 2405 2406 int 2407 dmu_objset_is_snapshot(objset_t *os) 2408 { 2409 if (os->os_dsl_dataset != NULL) 2410 return (os->os_dsl_dataset->ds_is_snapshot); 2411 else 2412 return (B_FALSE); 2413 } 2414 2415 int 2416 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen, 2417 boolean_t *conflict) 2418 { 2419 dsl_dataset_t *ds = os->os_dsl_dataset; 2420 uint64_t ignored; 2421 2422 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 2423 return (SET_ERROR(ENOENT)); 2424 2425 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, 2426 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, 2427 MT_NORMALIZE, real, maxlen, conflict)); 2428 } 2429 2430 int 2431 dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 2432 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) 2433 { 2434 dsl_dataset_t *ds = os->os_dsl_dataset; 2435 zap_cursor_t cursor; 2436 zap_attribute_t attr; 2437 2438 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 2439 2440 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 2441 return (SET_ERROR(ENOENT)); 2442 2443 zap_cursor_init_serialized(&cursor, 2444 ds->ds_dir->dd_pool->dp_meta_objset, 2445 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); 2446 2447 if (zap_cursor_retrieve(&cursor, &attr) != 0) { 2448 zap_cursor_fini(&cursor); 2449 return (SET_ERROR(ENOENT)); 2450 } 2451 2452 if (strlen(attr.za_name) + 1 > namelen) { 2453 zap_cursor_fini(&cursor); 2454 return (SET_ERROR(ENAMETOOLONG)); 2455 } 2456 2457 (void) strlcpy(name, attr.za_name, namelen); 2458 if (idp) 2459 *idp = attr.za_first_integer; 2460 if (case_conflict) 2461 *case_conflict = attr.za_normalization_conflict; 2462 zap_cursor_advance(&cursor); 2463 *offp = zap_cursor_serialize(&cursor); 2464 zap_cursor_fini(&cursor); 2465 2466 return (0); 2467 } 2468 2469 int 2470 dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value) 2471 { 2472 return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value)); 2473 } 2474 2475 int 2476 dmu_dir_list_next(objset_t *os, int namelen, char *name, 2477 uint64_t *idp, uint64_t *offp) 2478 { 2479 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 2480 zap_cursor_t cursor; 2481 zap_attribute_t attr; 2482 2483 /* there is no next dir on a snapshot! */ 2484 if (os->os_dsl_dataset->ds_object != 2485 dsl_dir_phys(dd)->dd_head_dataset_obj) 2486 return (SET_ERROR(ENOENT)); 2487 2488 zap_cursor_init_serialized(&cursor, 2489 dd->dd_pool->dp_meta_objset, 2490 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); 2491 2492 if (zap_cursor_retrieve(&cursor, &attr) != 0) { 2493 zap_cursor_fini(&cursor); 2494 return (SET_ERROR(ENOENT)); 2495 } 2496 2497 if (strlen(attr.za_name) + 1 > namelen) { 2498 zap_cursor_fini(&cursor); 2499 return (SET_ERROR(ENAMETOOLONG)); 2500 } 2501 2502 (void) strlcpy(name, attr.za_name, namelen); 2503 if (idp) 2504 *idp = attr.za_first_integer; 2505 zap_cursor_advance(&cursor); 2506 *offp = zap_cursor_serialize(&cursor); 2507 zap_cursor_fini(&cursor); 2508 2509 return (0); 2510 } 2511 2512 typedef struct dmu_objset_find_ctx { 2513 taskq_t *dc_tq; 2514 dsl_pool_t *dc_dp; 2515 uint64_t dc_ddobj; 2516 char *dc_ddname; /* last component of ddobj's name */ 2517 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); 2518 void *dc_arg; 2519 int dc_flags; 2520 kmutex_t *dc_error_lock; 2521 int *dc_error; 2522 } dmu_objset_find_ctx_t; 2523 2524 static void 2525 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) 2526 { 2527 dsl_pool_t *dp = dcp->dc_dp; 2528 dsl_dir_t *dd; 2529 dsl_dataset_t *ds; 2530 zap_cursor_t zc; 2531 zap_attribute_t *attr; 2532 uint64_t thisobj; 2533 int err = 0; 2534 2535 /* don't process if there already was an error */ 2536 if (*dcp->dc_error != 0) 2537 goto out; 2538 2539 /* 2540 * Note: passing the name (dc_ddname) here is optional, but it 2541 * improves performance because we don't need to call 2542 * zap_value_search() to determine the name. 2543 */ 2544 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); 2545 if (err != 0) 2546 goto out; 2547 2548 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2549 if (dd->dd_myname[0] == '$') { 2550 dsl_dir_rele(dd, FTAG); 2551 goto out; 2552 } 2553 2554 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2555 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 2556 2557 /* 2558 * Iterate over all children. 2559 */ 2560 if (dcp->dc_flags & DS_FIND_CHILDREN) { 2561 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2562 dsl_dir_phys(dd)->dd_child_dir_zapobj); 2563 zap_cursor_retrieve(&zc, attr) == 0; 2564 (void) zap_cursor_advance(&zc)) { 2565 ASSERT3U(attr->za_integer_length, ==, 2566 sizeof (uint64_t)); 2567 ASSERT3U(attr->za_num_integers, ==, 1); 2568 2569 dmu_objset_find_ctx_t *child_dcp = 2570 kmem_alloc(sizeof (*child_dcp), KM_SLEEP); 2571 *child_dcp = *dcp; 2572 child_dcp->dc_ddobj = attr->za_first_integer; 2573 child_dcp->dc_ddname = spa_strdup(attr->za_name); 2574 if (dcp->dc_tq != NULL) 2575 (void) taskq_dispatch(dcp->dc_tq, 2576 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); 2577 else 2578 dmu_objset_find_dp_impl(child_dcp); 2579 } 2580 zap_cursor_fini(&zc); 2581 } 2582 2583 /* 2584 * Iterate over all snapshots. 2585 */ 2586 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { 2587 dsl_dataset_t *ds; 2588 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2589 2590 if (err == 0) { 2591 uint64_t snapobj; 2592 2593 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 2594 dsl_dataset_rele(ds, FTAG); 2595 2596 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 2597 zap_cursor_retrieve(&zc, attr) == 0; 2598 (void) zap_cursor_advance(&zc)) { 2599 ASSERT3U(attr->za_integer_length, ==, 2600 sizeof (uint64_t)); 2601 ASSERT3U(attr->za_num_integers, ==, 1); 2602 2603 err = dsl_dataset_hold_obj(dp, 2604 attr->za_first_integer, FTAG, &ds); 2605 if (err != 0) 2606 break; 2607 err = dcp->dc_func(dp, ds, dcp->dc_arg); 2608 dsl_dataset_rele(ds, FTAG); 2609 if (err != 0) 2610 break; 2611 } 2612 zap_cursor_fini(&zc); 2613 } 2614 } 2615 2616 kmem_free(attr, sizeof (zap_attribute_t)); 2617 2618 if (err != 0) { 2619 dsl_dir_rele(dd, FTAG); 2620 goto out; 2621 } 2622 2623 /* 2624 * Apply to self. 2625 */ 2626 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2627 2628 /* 2629 * Note: we hold the dir while calling dsl_dataset_hold_obj() so 2630 * that the dir will remain cached, and we won't have to re-instantiate 2631 * it (which could be expensive due to finding its name via 2632 * zap_value_search()). 2633 */ 2634 dsl_dir_rele(dd, FTAG); 2635 if (err != 0) 2636 goto out; 2637 err = dcp->dc_func(dp, ds, dcp->dc_arg); 2638 dsl_dataset_rele(ds, FTAG); 2639 2640 out: 2641 if (err != 0) { 2642 mutex_enter(dcp->dc_error_lock); 2643 /* only keep first error */ 2644 if (*dcp->dc_error == 0) 2645 *dcp->dc_error = err; 2646 mutex_exit(dcp->dc_error_lock); 2647 } 2648 2649 if (dcp->dc_ddname != NULL) 2650 spa_strfree(dcp->dc_ddname); 2651 kmem_free(dcp, sizeof (*dcp)); 2652 } 2653 2654 static void 2655 dmu_objset_find_dp_cb(void *arg) 2656 { 2657 dmu_objset_find_ctx_t *dcp = arg; 2658 dsl_pool_t *dp = dcp->dc_dp; 2659 2660 /* 2661 * We need to get a pool_config_lock here, as there are several 2662 * assert(pool_config_held) down the stack. Getting a lock via 2663 * dsl_pool_config_enter is risky, as it might be stalled by a 2664 * pending writer. This would deadlock, as the write lock can 2665 * only be granted when our parent thread gives up the lock. 2666 * The _prio interface gives us priority over a pending writer. 2667 */ 2668 dsl_pool_config_enter_prio(dp, FTAG); 2669 2670 dmu_objset_find_dp_impl(dcp); 2671 2672 dsl_pool_config_exit(dp, FTAG); 2673 } 2674 2675 /* 2676 * Find objsets under and including ddobj, call func(ds) on each. 2677 * The order for the enumeration is completely undefined. 2678 * func is called with dsl_pool_config held. 2679 */ 2680 int 2681 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, 2682 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) 2683 { 2684 int error = 0; 2685 taskq_t *tq = NULL; 2686 int ntasks; 2687 dmu_objset_find_ctx_t *dcp; 2688 kmutex_t err_lock; 2689 2690 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); 2691 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); 2692 dcp->dc_tq = NULL; 2693 dcp->dc_dp = dp; 2694 dcp->dc_ddobj = ddobj; 2695 dcp->dc_ddname = NULL; 2696 dcp->dc_func = func; 2697 dcp->dc_arg = arg; 2698 dcp->dc_flags = flags; 2699 dcp->dc_error_lock = &err_lock; 2700 dcp->dc_error = &error; 2701 2702 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { 2703 /* 2704 * In case a write lock is held we can't make use of 2705 * parallelism, as down the stack of the worker threads 2706 * the lock is asserted via dsl_pool_config_held. 2707 * In case of a read lock this is solved by getting a read 2708 * lock in each worker thread, which isn't possible in case 2709 * of a writer lock. So we fall back to the synchronous path 2710 * here. 2711 * In the future it might be possible to get some magic into 2712 * dsl_pool_config_held in a way that it returns true for 2713 * the worker threads so that a single lock held from this 2714 * thread suffices. For now, stay single threaded. 2715 */ 2716 dmu_objset_find_dp_impl(dcp); 2717 mutex_destroy(&err_lock); 2718 2719 return (error); 2720 } 2721 2722 ntasks = dmu_find_threads; 2723 if (ntasks == 0) 2724 ntasks = vdev_count_leaves(dp->dp_spa) * 4; 2725 tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks, 2726 INT_MAX, 0); 2727 if (tq == NULL) { 2728 kmem_free(dcp, sizeof (*dcp)); 2729 mutex_destroy(&err_lock); 2730 2731 return (SET_ERROR(ENOMEM)); 2732 } 2733 dcp->dc_tq = tq; 2734 2735 /* dcp will be freed by task */ 2736 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); 2737 2738 /* 2739 * PORTING: this code relies on the property of taskq_wait to wait 2740 * until no more tasks are queued and no more tasks are active. As 2741 * we always queue new tasks from within other tasks, task_wait 2742 * reliably waits for the full recursion to finish, even though we 2743 * enqueue new tasks after taskq_wait has been called. 2744 * On platforms other than illumos, taskq_wait may not have this 2745 * property. 2746 */ 2747 taskq_wait(tq); 2748 taskq_destroy(tq); 2749 mutex_destroy(&err_lock); 2750 2751 return (error); 2752 } 2753 2754 /* 2755 * Find all objsets under name, and for each, call 'func(child_name, arg)'. 2756 * The dp_config_rwlock must not be held when this is called, and it 2757 * will not be held when the callback is called. 2758 * Therefore this function should only be used when the pool is not changing 2759 * (e.g. in syncing context), or the callback can deal with the possible races. 2760 */ 2761 static int 2762 dmu_objset_find_impl(spa_t *spa, const char *name, 2763 int func(const char *, void *), void *arg, int flags) 2764 { 2765 dsl_dir_t *dd; 2766 dsl_pool_t *dp = spa_get_dsl(spa); 2767 dsl_dataset_t *ds; 2768 zap_cursor_t zc; 2769 zap_attribute_t *attr; 2770 char *child; 2771 uint64_t thisobj; 2772 int err; 2773 2774 dsl_pool_config_enter(dp, FTAG); 2775 2776 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); 2777 if (err != 0) { 2778 dsl_pool_config_exit(dp, FTAG); 2779 return (err); 2780 } 2781 2782 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2783 if (dd->dd_myname[0] == '$') { 2784 dsl_dir_rele(dd, FTAG); 2785 dsl_pool_config_exit(dp, FTAG); 2786 return (0); 2787 } 2788 2789 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2790 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 2791 2792 /* 2793 * Iterate over all children. 2794 */ 2795 if (flags & DS_FIND_CHILDREN) { 2796 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2797 dsl_dir_phys(dd)->dd_child_dir_zapobj); 2798 zap_cursor_retrieve(&zc, attr) == 0; 2799 (void) zap_cursor_advance(&zc)) { 2800 ASSERT3U(attr->za_integer_length, ==, 2801 sizeof (uint64_t)); 2802 ASSERT3U(attr->za_num_integers, ==, 1); 2803 2804 child = kmem_asprintf("%s/%s", name, attr->za_name); 2805 dsl_pool_config_exit(dp, FTAG); 2806 err = dmu_objset_find_impl(spa, child, 2807 func, arg, flags); 2808 dsl_pool_config_enter(dp, FTAG); 2809 kmem_strfree(child); 2810 if (err != 0) 2811 break; 2812 } 2813 zap_cursor_fini(&zc); 2814 2815 if (err != 0) { 2816 dsl_dir_rele(dd, FTAG); 2817 dsl_pool_config_exit(dp, FTAG); 2818 kmem_free(attr, sizeof (zap_attribute_t)); 2819 return (err); 2820 } 2821 } 2822 2823 /* 2824 * Iterate over all snapshots. 2825 */ 2826 if (flags & DS_FIND_SNAPSHOTS) { 2827 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2828 2829 if (err == 0) { 2830 uint64_t snapobj; 2831 2832 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 2833 dsl_dataset_rele(ds, FTAG); 2834 2835 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 2836 zap_cursor_retrieve(&zc, attr) == 0; 2837 (void) zap_cursor_advance(&zc)) { 2838 ASSERT3U(attr->za_integer_length, ==, 2839 sizeof (uint64_t)); 2840 ASSERT3U(attr->za_num_integers, ==, 1); 2841 2842 child = kmem_asprintf("%s@%s", 2843 name, attr->za_name); 2844 dsl_pool_config_exit(dp, FTAG); 2845 err = func(child, arg); 2846 dsl_pool_config_enter(dp, FTAG); 2847 kmem_strfree(child); 2848 if (err != 0) 2849 break; 2850 } 2851 zap_cursor_fini(&zc); 2852 } 2853 } 2854 2855 dsl_dir_rele(dd, FTAG); 2856 kmem_free(attr, sizeof (zap_attribute_t)); 2857 dsl_pool_config_exit(dp, FTAG); 2858 2859 if (err != 0) 2860 return (err); 2861 2862 /* Apply to self. */ 2863 return (func(name, arg)); 2864 } 2865 2866 /* 2867 * See comment above dmu_objset_find_impl(). 2868 */ 2869 int 2870 dmu_objset_find(const char *name, int func(const char *, void *), void *arg, 2871 int flags) 2872 { 2873 spa_t *spa; 2874 int error; 2875 2876 error = spa_open(name, &spa, FTAG); 2877 if (error != 0) 2878 return (error); 2879 error = dmu_objset_find_impl(spa, name, func, arg, flags); 2880 spa_close(spa, FTAG); 2881 return (error); 2882 } 2883 2884 boolean_t 2885 dmu_objset_incompatible_encryption_version(objset_t *os) 2886 { 2887 return (dsl_dir_incompatible_encryption_version( 2888 os->os_dsl_dataset->ds_dir)); 2889 } 2890 2891 void 2892 dmu_objset_set_user(objset_t *os, void *user_ptr) 2893 { 2894 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 2895 os->os_user_ptr = user_ptr; 2896 } 2897 2898 void * 2899 dmu_objset_get_user(objset_t *os) 2900 { 2901 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 2902 return (os->os_user_ptr); 2903 } 2904 2905 /* 2906 * Determine name of filesystem, given name of snapshot. 2907 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes 2908 */ 2909 int 2910 dmu_fsname(const char *snapname, char *buf) 2911 { 2912 char *atp = strchr(snapname, '@'); 2913 if (atp == NULL) 2914 return (SET_ERROR(EINVAL)); 2915 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) 2916 return (SET_ERROR(ENAMETOOLONG)); 2917 (void) strlcpy(buf, snapname, atp - snapname + 1); 2918 return (0); 2919 } 2920 2921 /* 2922 * Call when we think we're going to write/free space in open context 2923 * to track the amount of dirty data in the open txg, which is also the 2924 * amount of memory that can not be evicted until this txg syncs. 2925 * 2926 * Note that there are two conditions where this can be called from 2927 * syncing context: 2928 * 2929 * [1] When we just created the dataset, in which case we go on with 2930 * updating any accounting of dirty data as usual. 2931 * [2] When we are dirtying MOS data, in which case we only update the 2932 * pool's accounting of dirty data. 2933 */ 2934 void 2935 dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) 2936 { 2937 dsl_dataset_t *ds = os->os_dsl_dataset; 2938 int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); 2939 2940 if (ds != NULL) { 2941 dsl_dir_willuse_space(ds->ds_dir, aspace, tx); 2942 } 2943 2944 dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); 2945 } 2946 2947 #if defined(_KERNEL) 2948 EXPORT_SYMBOL(dmu_objset_zil); 2949 EXPORT_SYMBOL(dmu_objset_pool); 2950 EXPORT_SYMBOL(dmu_objset_ds); 2951 EXPORT_SYMBOL(dmu_objset_type); 2952 EXPORT_SYMBOL(dmu_objset_name); 2953 EXPORT_SYMBOL(dmu_objset_hold); 2954 EXPORT_SYMBOL(dmu_objset_hold_flags); 2955 EXPORT_SYMBOL(dmu_objset_own); 2956 EXPORT_SYMBOL(dmu_objset_rele); 2957 EXPORT_SYMBOL(dmu_objset_rele_flags); 2958 EXPORT_SYMBOL(dmu_objset_disown); 2959 EXPORT_SYMBOL(dmu_objset_from_ds); 2960 EXPORT_SYMBOL(dmu_objset_create); 2961 EXPORT_SYMBOL(dmu_objset_clone); 2962 EXPORT_SYMBOL(dmu_objset_stats); 2963 EXPORT_SYMBOL(dmu_objset_fast_stat); 2964 EXPORT_SYMBOL(dmu_objset_spa); 2965 EXPORT_SYMBOL(dmu_objset_space); 2966 EXPORT_SYMBOL(dmu_objset_fsid_guid); 2967 EXPORT_SYMBOL(dmu_objset_find); 2968 EXPORT_SYMBOL(dmu_objset_byteswap); 2969 EXPORT_SYMBOL(dmu_objset_evict_dbufs); 2970 EXPORT_SYMBOL(dmu_objset_snap_cmtime); 2971 EXPORT_SYMBOL(dmu_objset_dnodesize); 2972 2973 EXPORT_SYMBOL(dmu_objset_sync); 2974 EXPORT_SYMBOL(dmu_objset_is_dirty); 2975 EXPORT_SYMBOL(dmu_objset_create_impl_dnstats); 2976 EXPORT_SYMBOL(dmu_objset_create_impl); 2977 EXPORT_SYMBOL(dmu_objset_open_impl); 2978 EXPORT_SYMBOL(dmu_objset_evict); 2979 EXPORT_SYMBOL(dmu_objset_register_type); 2980 EXPORT_SYMBOL(dmu_objset_do_userquota_updates); 2981 EXPORT_SYMBOL(dmu_objset_userquota_get_ids); 2982 EXPORT_SYMBOL(dmu_objset_userused_enabled); 2983 EXPORT_SYMBOL(dmu_objset_userspace_upgrade); 2984 EXPORT_SYMBOL(dmu_objset_userspace_present); 2985 EXPORT_SYMBOL(dmu_objset_userobjused_enabled); 2986 EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable); 2987 EXPORT_SYMBOL(dmu_objset_userobjspace_present); 2988 EXPORT_SYMBOL(dmu_objset_projectquota_enabled); 2989 EXPORT_SYMBOL(dmu_objset_projectquota_present); 2990 EXPORT_SYMBOL(dmu_objset_projectquota_upgradable); 2991 EXPORT_SYMBOL(dmu_objset_id_quota_upgrade); 2992 #endif 2993