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 2019 Joyent, Inc. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved. 29 * Copyright (c) 2014 Integros [integros.com] 30 * Copyright 2017 Nexenta Systems, Inc. 31 * Copyright (c) 2017 Open-E, Inc. All Rights Reserved. 32 */ 33 34 /* Portions Copyright 2010 Robert Milkowski */ 35 36 #include <sys/zfeature.h> 37 #include <sys/cred.h> 38 #include <sys/zfs_context.h> 39 #include <sys/dmu_objset.h> 40 #include <sys/dsl_dir.h> 41 #include <sys/dsl_dataset.h> 42 #include <sys/dsl_prop.h> 43 #include <sys/dsl_pool.h> 44 #include <sys/dsl_synctask.h> 45 #include <sys/dsl_deleg.h> 46 #include <sys/dnode.h> 47 #include <sys/dbuf.h> 48 #include <sys/zvol.h> 49 #include <sys/dmu_tx.h> 50 #include <sys/zap.h> 51 #include <sys/zil.h> 52 #include <sys/dmu_impl.h> 53 #include <sys/zfs_ioctl.h> 54 #include <sys/sa.h> 55 #include <sys/zfs_onexit.h> 56 #include <sys/dsl_destroy.h> 57 #include <sys/vdev.h> 58 #include <sys/zfeature.h> 59 #include <sys/spa_impl.h> 60 #include <sys/dmu_recv.h> 61 #include <sys/zfs_project.h> 62 #include "zfs_namecheck.h" 63 64 /* 65 * Needed to close a window in dnode_move() that allows the objset to be freed 66 * before it can be safely accessed. 67 */ 68 krwlock_t os_lock; 69 70 /* 71 * Tunable to overwrite the maximum number of threads for the parallization 72 * of dmu_objset_find_dp, needed to speed up the import of pools with many 73 * datasets. 74 * Default is 4 times the number of leaf vdevs. 75 */ 76 int dmu_find_threads = 0; 77 78 /* 79 * Backfill lower metadnode objects after this many have been freed. 80 * Backfilling negatively impacts object creation rates, so only do it 81 * if there are enough holes to fill. 82 */ 83 int dmu_rescan_dnode_threshold = 131072; 84 85 static char *upgrade_tag = "upgrade_tag"; 86 87 static void dmu_objset_find_dp_cb(void *arg); 88 89 static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb); 90 static void dmu_objset_upgrade_stop(objset_t *os); 91 92 void 93 dmu_objset_init(void) 94 { 95 rw_init(&os_lock, NULL, RW_DEFAULT, NULL); 96 } 97 98 void 99 dmu_objset_fini(void) 100 { 101 rw_destroy(&os_lock); 102 } 103 104 spa_t * 105 dmu_objset_spa(objset_t *os) 106 { 107 return (os->os_spa); 108 } 109 110 zilog_t * 111 dmu_objset_zil(objset_t *os) 112 { 113 return (os->os_zil); 114 } 115 116 dsl_pool_t * 117 dmu_objset_pool(objset_t *os) 118 { 119 dsl_dataset_t *ds; 120 121 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) 122 return (ds->ds_dir->dd_pool); 123 else 124 return (spa_get_dsl(os->os_spa)); 125 } 126 127 dsl_dataset_t * 128 dmu_objset_ds(objset_t *os) 129 { 130 return (os->os_dsl_dataset); 131 } 132 133 dmu_objset_type_t 134 dmu_objset_type(objset_t *os) 135 { 136 return (os->os_phys->os_type); 137 } 138 139 void 140 dmu_objset_name(objset_t *os, char *buf) 141 { 142 dsl_dataset_name(os->os_dsl_dataset, buf); 143 } 144 145 uint64_t 146 dmu_objset_id(objset_t *os) 147 { 148 dsl_dataset_t *ds = os->os_dsl_dataset; 149 150 return (ds ? ds->ds_object : 0); 151 } 152 153 uint64_t 154 dmu_objset_dnodesize(objset_t *os) 155 { 156 return (os->os_dnodesize); 157 } 158 159 zfs_sync_type_t 160 dmu_objset_syncprop(objset_t *os) 161 { 162 return (os->os_sync); 163 } 164 165 zfs_logbias_op_t 166 dmu_objset_logbias(objset_t *os) 167 { 168 return (os->os_logbias); 169 } 170 171 static void 172 checksum_changed_cb(void *arg, uint64_t newval) 173 { 174 objset_t *os = arg; 175 176 /* 177 * Inheritance should have been done by now. 178 */ 179 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 180 181 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); 182 } 183 184 static void 185 compression_changed_cb(void *arg, uint64_t newval) 186 { 187 objset_t *os = arg; 188 189 /* 190 * Inheritance and range checking should have been done by now. 191 */ 192 ASSERT(newval != ZIO_COMPRESS_INHERIT); 193 194 os->os_compress = zio_compress_select(os->os_spa, newval, 195 ZIO_COMPRESS_ON); 196 } 197 198 static void 199 copies_changed_cb(void *arg, uint64_t newval) 200 { 201 objset_t *os = arg; 202 203 /* 204 * Inheritance and range checking should have been done by now. 205 */ 206 ASSERT(newval > 0); 207 ASSERT(newval <= spa_max_replication(os->os_spa)); 208 209 os->os_copies = newval; 210 } 211 212 static void 213 dedup_changed_cb(void *arg, uint64_t newval) 214 { 215 objset_t *os = arg; 216 spa_t *spa = os->os_spa; 217 enum zio_checksum checksum; 218 219 /* 220 * Inheritance should have been done by now. 221 */ 222 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 223 224 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); 225 226 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; 227 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); 228 } 229 230 static void 231 primary_cache_changed_cb(void *arg, uint64_t newval) 232 { 233 objset_t *os = arg; 234 235 /* 236 * Inheritance and range checking should have been done by now. 237 */ 238 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 239 newval == ZFS_CACHE_METADATA); 240 241 os->os_primary_cache = newval; 242 } 243 244 static void 245 secondary_cache_changed_cb(void *arg, uint64_t newval) 246 { 247 objset_t *os = arg; 248 249 /* 250 * Inheritance and range checking should have been done by now. 251 */ 252 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 253 newval == ZFS_CACHE_METADATA); 254 255 os->os_secondary_cache = newval; 256 } 257 258 static void 259 sync_changed_cb(void *arg, uint64_t newval) 260 { 261 objset_t *os = arg; 262 263 /* 264 * Inheritance and range checking should have been done by now. 265 */ 266 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || 267 newval == ZFS_SYNC_DISABLED); 268 269 os->os_sync = newval; 270 if (os->os_zil) 271 zil_set_sync(os->os_zil, newval); 272 } 273 274 static void 275 redundant_metadata_changed_cb(void *arg, uint64_t newval) 276 { 277 objset_t *os = arg; 278 279 /* 280 * Inheritance and range checking should have been done by now. 281 */ 282 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || 283 newval == ZFS_REDUNDANT_METADATA_MOST); 284 285 os->os_redundant_metadata = newval; 286 } 287 288 static void 289 dnodesize_changed_cb(void *arg, uint64_t newval) 290 { 291 objset_t *os = arg; 292 293 switch (newval) { 294 case ZFS_DNSIZE_LEGACY: 295 os->os_dnodesize = DNODE_MIN_SIZE; 296 break; 297 case ZFS_DNSIZE_AUTO: 298 /* 299 * Choose a dnode size that will work well for most 300 * workloads if the user specified "auto". Future code 301 * improvements could dynamically select a dnode size 302 * based on observed workload patterns. 303 */ 304 os->os_dnodesize = DNODE_MIN_SIZE * 2; 305 break; 306 case ZFS_DNSIZE_1K: 307 case ZFS_DNSIZE_2K: 308 case ZFS_DNSIZE_4K: 309 case ZFS_DNSIZE_8K: 310 case ZFS_DNSIZE_16K: 311 os->os_dnodesize = newval; 312 break; 313 } 314 } 315 316 static void 317 smallblk_changed_cb(void *arg, uint64_t newval) 318 { 319 objset_t *os = arg; 320 321 /* 322 * Inheritance and range checking should have been done by now. 323 */ 324 ASSERT(newval <= SPA_OLD_MAXBLOCKSIZE); 325 ASSERT(ISP2(newval)); 326 327 os->os_zpl_special_smallblock = newval; 328 } 329 330 static void 331 logbias_changed_cb(void *arg, uint64_t newval) 332 { 333 objset_t *os = arg; 334 335 ASSERT(newval == ZFS_LOGBIAS_LATENCY || 336 newval == ZFS_LOGBIAS_THROUGHPUT); 337 os->os_logbias = newval; 338 if (os->os_zil) 339 zil_set_logbias(os->os_zil, newval); 340 } 341 342 static void 343 recordsize_changed_cb(void *arg, uint64_t newval) 344 { 345 objset_t *os = arg; 346 347 os->os_recordsize = newval; 348 } 349 350 void 351 dmu_objset_byteswap(void *buf, size_t size) 352 { 353 objset_phys_t *osp = buf; 354 355 ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 || 356 size == sizeof (objset_phys_t)); 357 dnode_byteswap(&osp->os_meta_dnode); 358 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); 359 osp->os_type = BSWAP_64(osp->os_type); 360 osp->os_flags = BSWAP_64(osp->os_flags); 361 if (size >= OBJSET_PHYS_SIZE_V2) { 362 dnode_byteswap(&osp->os_userused_dnode); 363 dnode_byteswap(&osp->os_groupused_dnode); 364 if (size >= sizeof (objset_phys_t)) 365 dnode_byteswap(&osp->os_projectused_dnode); 366 } 367 } 368 369 /* 370 * The hash is a CRC-based hash of the objset_t pointer and the object number. 371 */ 372 static uint64_t 373 dnode_hash(const objset_t *os, uint64_t obj) 374 { 375 uintptr_t osv = (uintptr_t)os; 376 uint64_t crc = -1ULL; 377 378 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); 379 /* 380 * The low 6 bits of the pointer don't have much entropy, because 381 * the objset_t is larger than 2^6 bytes long. 382 */ 383 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; 384 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; 385 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; 386 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF]; 387 388 crc ^= (osv>>14) ^ (obj>>24); 389 390 return (crc); 391 } 392 393 unsigned int 394 dnode_multilist_index_func(multilist_t *ml, void *obj) 395 { 396 dnode_t *dn = obj; 397 return (dnode_hash(dn->dn_objset, dn->dn_object) % 398 multilist_get_num_sublists(ml)); 399 } 400 401 /* 402 * Instantiates the objset_t in-memory structure corresponding to the 403 * objset_phys_t that's pointed to by the specified blkptr_t. 404 */ 405 int 406 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 407 objset_t **osp) 408 { 409 objset_t *os; 410 int i, err; 411 412 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); 413 414 #if 0 415 /* 416 * The $ORIGIN dataset (if it exists) doesn't have an associated 417 * objset, so there's no reason to open it. The $ORIGIN dataset 418 * will not exist on pools older than SPA_VERSION_ORIGIN. 419 */ 420 if (ds != NULL && spa_get_dsl(spa) != NULL && 421 spa_get_dsl(spa)->dp_origin_snap != NULL) { 422 ASSERT3P(ds->ds_dir, !=, 423 spa_get_dsl(spa)->dp_origin_snap->ds_dir); 424 } 425 #endif 426 427 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); 428 os->os_dsl_dataset = ds; 429 os->os_spa = spa; 430 os->os_rootbp = bp; 431 if (!BP_IS_HOLE(os->os_rootbp)) { 432 arc_flags_t aflags = ARC_FLAG_WAIT; 433 zbookmark_phys_t zb; 434 int size; 435 enum zio_flag zio_flags = ZIO_FLAG_CANFAIL; 436 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, 437 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 438 439 if (DMU_OS_IS_L2CACHEABLE(os)) 440 aflags |= ARC_FLAG_L2CACHE; 441 442 if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) { 443 ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); 444 ASSERT(BP_IS_AUTHENTICATED(bp)); 445 zio_flags |= ZIO_FLAG_RAW; 446 } 447 448 dprintf_bp(os->os_rootbp, "reading %s", ""); 449 err = arc_read(NULL, spa, os->os_rootbp, 450 arc_getbuf_func, &os->os_phys_buf, 451 ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb); 452 if (err != 0) { 453 kmem_free(os, sizeof (objset_t)); 454 /* convert checksum errors into IO errors */ 455 if (err == ECKSUM) 456 err = SET_ERROR(EIO); 457 return (err); 458 } 459 460 if (spa_version(spa) < SPA_VERSION_USERSPACE) 461 size = OBJSET_PHYS_SIZE_V1; 462 else if (!spa_feature_is_enabled(spa, 463 SPA_FEATURE_PROJECT_QUOTA)) 464 size = OBJSET_PHYS_SIZE_V2; 465 else 466 size = sizeof (objset_phys_t); 467 468 /* Increase the blocksize if we are permitted. */ 469 if (arc_buf_size(os->os_phys_buf) < size) { 470 arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, 471 ARC_BUFC_METADATA, size); 472 bzero(buf->b_data, size); 473 bcopy(os->os_phys_buf->b_data, buf->b_data, 474 arc_buf_size(os->os_phys_buf)); 475 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 476 os->os_phys_buf = buf; 477 } 478 479 os->os_phys = os->os_phys_buf->b_data; 480 os->os_flags = os->os_phys->os_flags; 481 } else { 482 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? 483 sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1; 484 os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, 485 ARC_BUFC_METADATA, size); 486 os->os_phys = os->os_phys_buf->b_data; 487 bzero(os->os_phys, size); 488 } 489 490 /* 491 * Note: the changed_cb will be called once before the register 492 * func returns, thus changing the checksum/compression from the 493 * default (fletcher2/off). Snapshots don't need to know about 494 * checksum/compression/copies. 495 */ 496 if (ds != NULL) { 497 boolean_t needlock = B_FALSE; 498 499 os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0); 500 501 /* 502 * Note: it's valid to open the objset if the dataset is 503 * long-held, in which case the pool_config lock will not 504 * be held. 505 */ 506 if (!dsl_pool_config_held(dmu_objset_pool(os))) { 507 needlock = B_TRUE; 508 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 509 } 510 511 err = dsl_prop_register(ds, 512 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), 513 primary_cache_changed_cb, os); 514 if (err == 0) { 515 err = dsl_prop_register(ds, 516 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), 517 secondary_cache_changed_cb, os); 518 } 519 if (!ds->ds_is_snapshot) { 520 if (err == 0) { 521 err = dsl_prop_register(ds, 522 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 523 checksum_changed_cb, os); 524 } 525 if (err == 0) { 526 err = dsl_prop_register(ds, 527 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 528 compression_changed_cb, os); 529 } 530 if (err == 0) { 531 err = dsl_prop_register(ds, 532 zfs_prop_to_name(ZFS_PROP_COPIES), 533 copies_changed_cb, os); 534 } 535 if (err == 0) { 536 err = dsl_prop_register(ds, 537 zfs_prop_to_name(ZFS_PROP_DEDUP), 538 dedup_changed_cb, os); 539 } 540 if (err == 0) { 541 err = dsl_prop_register(ds, 542 zfs_prop_to_name(ZFS_PROP_LOGBIAS), 543 logbias_changed_cb, os); 544 } 545 if (err == 0) { 546 err = dsl_prop_register(ds, 547 zfs_prop_to_name(ZFS_PROP_SYNC), 548 sync_changed_cb, os); 549 } 550 if (err == 0) { 551 err = dsl_prop_register(ds, 552 zfs_prop_to_name( 553 ZFS_PROP_REDUNDANT_METADATA), 554 redundant_metadata_changed_cb, os); 555 } 556 if (err == 0) { 557 err = dsl_prop_register(ds, 558 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 559 recordsize_changed_cb, os); 560 } 561 if (err == 0) { 562 err = dsl_prop_register(ds, 563 zfs_prop_to_name(ZFS_PROP_DNODESIZE), 564 dnodesize_changed_cb, os); 565 } 566 if (err == 0) { 567 err = dsl_prop_register(ds, 568 zfs_prop_to_name( 569 ZFS_PROP_SPECIAL_SMALL_BLOCKS), 570 smallblk_changed_cb, os); 571 } 572 } 573 if (needlock) 574 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 575 if (err != 0) { 576 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 577 kmem_free(os, sizeof (objset_t)); 578 return (err); 579 } 580 } else { 581 /* It's the meta-objset. */ 582 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; 583 os->os_compress = ZIO_COMPRESS_ON; 584 os->os_encrypted = B_FALSE; 585 os->os_copies = spa_max_replication(spa); 586 os->os_dedup_checksum = ZIO_CHECKSUM_OFF; 587 os->os_dedup_verify = B_FALSE; 588 os->os_logbias = ZFS_LOGBIAS_LATENCY; 589 os->os_sync = ZFS_SYNC_STANDARD; 590 os->os_primary_cache = ZFS_CACHE_ALL; 591 os->os_secondary_cache = ZFS_CACHE_ALL; 592 os->os_dnodesize = DNODE_MIN_SIZE; 593 } 594 /* 595 * These properties will be filled in by the logic in zfs_get_zplprop() 596 * when they are queried for the first time. 597 */ 598 os->os_version = OBJSET_PROP_UNINITIALIZED; 599 os->os_normalization = OBJSET_PROP_UNINITIALIZED; 600 os->os_utf8only = OBJSET_PROP_UNINITIALIZED; 601 os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED; 602 603 if (ds == NULL || !ds->ds_is_snapshot) 604 os->os_zil_header = os->os_phys->os_zil_header; 605 os->os_zil = zil_alloc(os, &os->os_zil_header); 606 607 for (i = 0; i < TXG_SIZE; i++) { 608 os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t), 609 offsetof(dnode_t, dn_dirty_link[i]), 610 dnode_multilist_index_func); 611 } 612 list_create(&os->os_dnodes, sizeof (dnode_t), 613 offsetof(dnode_t, dn_link)); 614 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), 615 offsetof(dmu_buf_impl_t, db_link)); 616 617 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); 618 mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); 619 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); 620 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); 621 os->os_obj_next_percpu_len = boot_ncpus; 622 os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * 623 sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); 624 625 dnode_special_open(os, &os->os_phys->os_meta_dnode, 626 DMU_META_DNODE_OBJECT, &os->os_meta_dnode); 627 if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) { 628 dnode_special_open(os, &os->os_phys->os_userused_dnode, 629 DMU_USERUSED_OBJECT, &os->os_userused_dnode); 630 dnode_special_open(os, &os->os_phys->os_groupused_dnode, 631 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); 632 if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf)) 633 dnode_special_open(os, 634 &os->os_phys->os_projectused_dnode, 635 DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode); 636 } 637 638 mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL); 639 640 *osp = os; 641 return (0); 642 } 643 644 int 645 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) 646 { 647 int err = 0; 648 649 /* 650 * We shouldn't be doing anything with dsl_dataset_t's unless the 651 * pool_config lock is held, or the dataset is long-held. 652 */ 653 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) || 654 dsl_dataset_long_held(ds)); 655 656 mutex_enter(&ds->ds_opening_lock); 657 if (ds->ds_objset == NULL) { 658 objset_t *os; 659 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 660 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), 661 ds, dsl_dataset_get_blkptr(ds), &os); 662 rrw_exit(&ds->ds_bp_rwlock, FTAG); 663 664 if (err == 0) { 665 mutex_enter(&ds->ds_lock); 666 ASSERT(ds->ds_objset == NULL); 667 ds->ds_objset = os; 668 mutex_exit(&ds->ds_lock); 669 } 670 } 671 *osp = ds->ds_objset; 672 mutex_exit(&ds->ds_opening_lock); 673 return (err); 674 } 675 676 /* 677 * Holds the pool while the objset is held. Therefore only one objset 678 * can be held at a time. 679 */ 680 int 681 dmu_objset_hold_flags(const char *name, boolean_t decrypt, void *tag, 682 objset_t **osp) 683 { 684 dsl_pool_t *dp; 685 dsl_dataset_t *ds; 686 int err; 687 ds_hold_flags_t flags; 688 689 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 690 err = dsl_pool_hold(name, tag, &dp); 691 if (err != 0) 692 return (err); 693 err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds); 694 if (err != 0) { 695 dsl_pool_rele(dp, tag); 696 return (err); 697 } 698 699 err = dmu_objset_from_ds(ds, osp); 700 if (err != 0) { 701 dsl_dataset_rele(ds, tag); 702 dsl_pool_rele(dp, tag); 703 } 704 705 return (err); 706 } 707 708 int 709 dmu_objset_hold(const char *name, void *tag, objset_t **osp) 710 { 711 return (dmu_objset_hold_flags(name, B_FALSE, tag, osp)); 712 } 713 714 /* ARGSUSED */ 715 static int 716 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, 717 boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) 718 { 719 int err; 720 721 err = dmu_objset_from_ds(ds, osp); 722 if (err != 0) { 723 return (err); 724 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { 725 return (SET_ERROR(EINVAL)); 726 } else if (!readonly && dsl_dataset_is_snapshot(ds)) { 727 return (SET_ERROR(EROFS)); 728 } else if (!readonly && decrypt && 729 dsl_dir_incompatible_encryption_version(ds->ds_dir)) { 730 return (SET_ERROR(EROFS)); 731 } 732 733 /* if we are decrypting, we can now check MACs in os->os_phys_buf */ 734 if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) { 735 zbookmark_phys_t zb; 736 737 SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT, 738 ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 739 err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa, 740 &zb, B_FALSE); 741 if (err != 0) 742 return (err); 743 744 ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf)); 745 } 746 747 return (0); 748 } 749 750 /* 751 * dsl_pool must not be held when this is called. 752 * Upon successful return, there will be a longhold on the dataset, 753 * and the dsl_pool will not be held. 754 */ 755 int 756 dmu_objset_own(const char *name, dmu_objset_type_t type, 757 boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) 758 { 759 dsl_pool_t *dp; 760 dsl_dataset_t *ds; 761 int err; 762 ds_hold_flags_t flags; 763 764 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 765 err = dsl_pool_hold(name, FTAG, &dp); 766 if (err != 0) 767 return (err); 768 err = dsl_dataset_own(dp, name, flags, tag, &ds); 769 if (err != 0) { 770 dsl_pool_rele(dp, FTAG); 771 return (err); 772 } 773 err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); 774 if (err != 0) { 775 dsl_dataset_disown(ds, flags, tag); 776 dsl_pool_rele(dp, FTAG); 777 return (err); 778 } 779 780 /* 781 * User accounting requires the dataset to be decrypted and rw. 782 * We also don't begin user accounting during claiming to help 783 * speed up pool import times and to keep this txg reserved 784 * completely for recovery work. 785 */ 786 if ((dmu_objset_userobjspace_upgradable(*osp) || 787 dmu_objset_projectquota_upgradable(*osp)) && 788 !readonly && !dp->dp_spa->spa_claiming && 789 (ds->ds_dir->dd_crypto_obj == 0 || decrypt)) 790 dmu_objset_id_quota_upgrade(*osp); 791 792 dsl_pool_rele(dp, FTAG); 793 return (0); 794 } 795 796 int 797 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, 798 boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp) 799 { 800 dsl_dataset_t *ds; 801 int err; 802 ds_hold_flags_t flags; 803 804 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 805 err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds); 806 if (err != 0) 807 return (err); 808 809 err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp); 810 if (err != 0) { 811 dsl_dataset_disown(ds, flags, tag); 812 return (err); 813 } 814 815 return (0); 816 } 817 818 void 819 dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, void *tag) 820 { 821 ds_hold_flags_t flags; 822 dsl_pool_t *dp = dmu_objset_pool(os); 823 824 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 825 dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag); 826 dsl_pool_rele(dp, tag); 827 } 828 829 void 830 dmu_objset_rele(objset_t *os, void *tag) 831 { 832 dmu_objset_rele_flags(os, B_FALSE, tag); 833 } 834 835 /* 836 * When we are called, os MUST refer to an objset associated with a dataset 837 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner 838 * == tag. We will then release and reacquire ownership of the dataset while 839 * holding the pool config_rwlock to avoid intervening namespace or ownership 840 * changes may occur. 841 * 842 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to 843 * release the hold on its dataset and acquire a new one on the dataset of the 844 * same name so that it can be partially torn down and reconstructed. 845 */ 846 void 847 dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds, 848 boolean_t decrypt, void *tag) 849 { 850 dsl_pool_t *dp; 851 char name[ZFS_MAX_DATASET_NAME_LEN]; 852 ds_hold_flags_t flags; 853 854 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 855 VERIFY3P(ds, !=, NULL); 856 VERIFY3P(ds->ds_owner, ==, tag); 857 VERIFY(dsl_dataset_long_held(ds)); 858 859 dsl_dataset_name(ds, name); 860 dp = ds->ds_dir->dd_pool; 861 dsl_pool_config_enter(dp, FTAG); 862 863 dsl_dataset_disown(ds, flags, tag); 864 VERIFY0(dsl_dataset_own(dp, name, flags, tag, newds)); 865 dsl_pool_config_exit(dp, FTAG); 866 } 867 868 void 869 dmu_objset_disown(objset_t *os, boolean_t decrypt, void *tag) 870 { 871 ds_hold_flags_t flags; 872 873 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE; 874 /* 875 * Stop upgrading thread 876 */ 877 dmu_objset_upgrade_stop(os); 878 dsl_dataset_disown(os->os_dsl_dataset, flags, tag); 879 } 880 881 void 882 dmu_objset_evict_dbufs(objset_t *os) 883 { 884 dnode_t dn_marker; 885 dnode_t *dn; 886 887 mutex_enter(&os->os_lock); 888 dn = list_head(&os->os_dnodes); 889 while (dn != NULL) { 890 /* 891 * Skip dnodes without holds. We have to do this dance 892 * because dnode_add_ref() only works if there is already a 893 * hold. If the dnode has no holds, then it has no dbufs. 894 */ 895 if (dnode_add_ref(dn, FTAG)) { 896 list_insert_after(&os->os_dnodes, dn, &dn_marker); 897 mutex_exit(&os->os_lock); 898 899 dnode_evict_dbufs(dn); 900 dnode_rele(dn, FTAG); 901 902 mutex_enter(&os->os_lock); 903 dn = list_next(&os->os_dnodes, &dn_marker); 904 list_remove(&os->os_dnodes, &dn_marker); 905 } else { 906 dn = list_next(&os->os_dnodes, dn); 907 } 908 } 909 mutex_exit(&os->os_lock); 910 911 if (DMU_USERUSED_DNODE(os) != NULL) { 912 if (DMU_PROJECTUSED_DNODE(os) != NULL) 913 dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os)); 914 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); 915 dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); 916 } 917 dnode_evict_dbufs(DMU_META_DNODE(os)); 918 } 919 920 /* 921 * Objset eviction processing is split into into two pieces. 922 * The first marks the objset as evicting, evicts any dbufs that 923 * have a refcount of zero, and then queues up the objset for the 924 * second phase of eviction. Once os->os_dnodes has been cleared by 925 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. 926 * The second phase closes the special dnodes, dequeues the objset from 927 * the list of those undergoing eviction, and finally frees the objset. 928 * 929 * NOTE: Due to asynchronous eviction processing (invocation of 930 * dnode_buf_pageout()), it is possible for the meta dnode for the 931 * objset to have no holds even though os->os_dnodes is not empty. 932 */ 933 void 934 dmu_objset_evict(objset_t *os) 935 { 936 dsl_dataset_t *ds = os->os_dsl_dataset; 937 938 for (int t = 0; t < TXG_SIZE; t++) 939 ASSERT(!dmu_objset_is_dirty(os, t)); 940 941 if (ds) 942 dsl_prop_unregister_all(ds, os); 943 944 if (os->os_sa) 945 sa_tear_down(os); 946 947 dmu_objset_evict_dbufs(os); 948 949 mutex_enter(&os->os_lock); 950 spa_evicting_os_register(os->os_spa, os); 951 if (list_is_empty(&os->os_dnodes)) { 952 mutex_exit(&os->os_lock); 953 dmu_objset_evict_done(os); 954 } else { 955 mutex_exit(&os->os_lock); 956 } 957 958 959 } 960 961 void 962 dmu_objset_evict_done(objset_t *os) 963 { 964 ASSERT3P(list_head(&os->os_dnodes), ==, NULL); 965 966 dnode_special_close(&os->os_meta_dnode); 967 if (DMU_USERUSED_DNODE(os)) { 968 if (DMU_PROJECTUSED_DNODE(os)) 969 dnode_special_close(&os->os_projectused_dnode); 970 dnode_special_close(&os->os_userused_dnode); 971 dnode_special_close(&os->os_groupused_dnode); 972 } 973 zil_free(os->os_zil); 974 975 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 976 977 /* 978 * This is a barrier to prevent the objset from going away in 979 * dnode_move() until we can safely ensure that the objset is still in 980 * use. We consider the objset valid before the barrier and invalid 981 * after the barrier. 982 */ 983 rw_enter(&os_lock, RW_READER); 984 rw_exit(&os_lock); 985 986 kmem_free(os->os_obj_next_percpu, 987 os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); 988 989 mutex_destroy(&os->os_lock); 990 mutex_destroy(&os->os_userused_lock); 991 mutex_destroy(&os->os_obj_lock); 992 mutex_destroy(&os->os_user_ptr_lock); 993 mutex_destroy(&os->os_upgrade_lock); 994 for (int i = 0; i < TXG_SIZE; i++) { 995 multilist_destroy(os->os_dirty_dnodes[i]); 996 } 997 spa_evicting_os_deregister(os->os_spa, os); 998 kmem_free(os, sizeof (objset_t)); 999 } 1000 1001 timestruc_t 1002 dmu_objset_snap_cmtime(objset_t *os) 1003 { 1004 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); 1005 } 1006 1007 /* ARGSUSED */ 1008 objset_t * 1009 dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 1010 dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx) 1011 { 1012 objset_t *os; 1013 dnode_t *mdn; 1014 1015 ASSERT(dmu_tx_is_syncing(tx)); 1016 1017 if (blksz == 0) 1018 blksz = 1 << DNODE_BLOCK_SHIFT; 1019 if (ibs == 0) 1020 ibs = DN_MAX_INDBLKSHIFT; 1021 1022 if (ds != NULL) 1023 VERIFY0(dmu_objset_from_ds(ds, &os)); 1024 else 1025 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); 1026 1027 mdn = DMU_META_DNODE(os); 1028 1029 dnode_allocate(mdn, DMU_OT_DNODE, DNODE_BLOCK_SIZE, DN_MAX_INDBLKSHIFT, 1030 DMU_OT_NONE, 0, DNODE_MIN_SLOTS, tx); 1031 1032 /* 1033 * We don't want to have to increase the meta-dnode's nlevels 1034 * later, because then we could do it in quescing context while 1035 * we are also accessing it in open context. 1036 * 1037 * This precaution is not necessary for the MOS (ds == NULL), 1038 * because the MOS is only updated in syncing context. 1039 * This is most fortunate: the MOS is the only objset that 1040 * needs to be synced multiple times as spa_sync() iterates 1041 * to convergence, so minimizing its dn_nlevels matters. 1042 */ 1043 if (ds != NULL) { 1044 if (levels == 0) { 1045 levels = 1; 1046 1047 /* 1048 * Determine the number of levels necessary for the 1049 * meta-dnode to contain DN_MAX_OBJECT dnodes. Note 1050 * that in order to ensure that we do not overflow 1051 * 64 bits, there has to be a nlevels that gives us a 1052 * number of blocks > DN_MAX_OBJECT but < 2^64. 1053 * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) 1054 * (10) must be less than (64 - log2(DN_MAX_OBJECT)) 1055 * (16). 1056 */ 1057 while ((uint64_t)mdn->dn_nblkptr << 1058 (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) * 1059 (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < 1060 DN_MAX_OBJECT) 1061 levels++; 1062 } 1063 1064 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = 1065 mdn->dn_nlevels = levels; 1066 } 1067 1068 ASSERT(type != DMU_OST_NONE); 1069 ASSERT(type != DMU_OST_ANY); 1070 ASSERT(type < DMU_OST_NUMTYPES); 1071 os->os_phys->os_type = type; 1072 1073 /* 1074 * Enable user accounting if it is enabled and this is not an 1075 * encrypted receive. 1076 */ 1077 if (dmu_objset_userused_enabled(os) && 1078 (!os->os_encrypted || !dmu_objset_is_receiving(os))) { 1079 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 1080 if (dmu_objset_userobjused_enabled(os)) { 1081 ds->ds_feature_activation[ 1082 SPA_FEATURE_USEROBJ_ACCOUNTING] = 1083 (void *)(uintptr_t)B_TRUE; 1084 os->os_phys->os_flags |= 1085 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; 1086 } 1087 if (dmu_objset_projectquota_enabled(os)) { 1088 ds->ds_feature_activation[ 1089 SPA_FEATURE_PROJECT_QUOTA] = 1090 (void *)(uintptr_t)B_TRUE; 1091 os->os_phys->os_flags |= 1092 OBJSET_FLAG_PROJECTQUOTA_COMPLETE; 1093 } 1094 os->os_flags = os->os_phys->os_flags; 1095 } 1096 1097 dsl_dataset_dirty(ds, tx); 1098 1099 return (os); 1100 } 1101 1102 /* called from dsl for meta-objset */ 1103 objset_t * 1104 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 1105 dmu_objset_type_t type, dmu_tx_t *tx) 1106 { 1107 return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx)); 1108 } 1109 1110 typedef struct dmu_objset_create_arg { 1111 const char *doca_name; 1112 cred_t *doca_cred; 1113 void (*doca_userfunc)(objset_t *os, void *arg, 1114 cred_t *cr, dmu_tx_t *tx); 1115 void *doca_userarg; 1116 dmu_objset_type_t doca_type; 1117 uint64_t doca_flags; 1118 dsl_crypto_params_t *doca_dcp; 1119 } dmu_objset_create_arg_t; 1120 1121 /*ARGSUSED*/ 1122 static int 1123 dmu_objset_create_check(void *arg, dmu_tx_t *tx) 1124 { 1125 dmu_objset_create_arg_t *doca = arg; 1126 dsl_pool_t *dp = dmu_tx_pool(tx); 1127 dsl_dir_t *pdd; 1128 const char *tail; 1129 int error; 1130 1131 if (strchr(doca->doca_name, '@') != NULL) 1132 return (SET_ERROR(EINVAL)); 1133 1134 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) 1135 return (SET_ERROR(ENAMETOOLONG)); 1136 1137 if (dataset_nestcheck(doca->doca_name) != 0) 1138 return (SET_ERROR(ENAMETOOLONG)); 1139 1140 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); 1141 if (error != 0) 1142 return (error); 1143 if (tail == NULL) { 1144 dsl_dir_rele(pdd, FTAG); 1145 return (SET_ERROR(EEXIST)); 1146 } 1147 1148 error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp, NULL); 1149 if (error != 0) { 1150 dsl_dir_rele(pdd, FTAG); 1151 return (error); 1152 } 1153 1154 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 1155 doca->doca_cred); 1156 1157 dsl_dir_rele(pdd, FTAG); 1158 1159 return (error); 1160 } 1161 1162 static void 1163 dmu_objset_create_sync(void *arg, dmu_tx_t *tx) 1164 { 1165 dmu_objset_create_arg_t *doca = arg; 1166 dsl_pool_t *dp = dmu_tx_pool(tx); 1167 spa_t *spa = dp->dp_spa; 1168 dsl_dir_t *pdd; 1169 const char *tail; 1170 dsl_dataset_t *ds; 1171 uint64_t obj; 1172 blkptr_t *bp; 1173 objset_t *os; 1174 zio_t *rzio; 1175 1176 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); 1177 1178 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, 1179 doca->doca_cred, doca->doca_dcp, tx); 1180 1181 VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj, 1182 DS_HOLD_FLAG_DECRYPT, FTAG, &ds)); 1183 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 1184 bp = dsl_dataset_get_blkptr(ds); 1185 os = dmu_objset_create_impl(spa, ds, bp, doca->doca_type, tx); 1186 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1187 1188 if (doca->doca_userfunc != NULL) { 1189 doca->doca_userfunc(os, doca->doca_userarg, 1190 doca->doca_cred, tx); 1191 } 1192 1193 /* 1194 * The doca_userfunc() may write out some data that needs to be 1195 * encrypted if the dataset is encrypted (specifically the root 1196 * directory). This data must be written out before the encryption 1197 * key mapping is removed by dsl_dataset_rele_flags(). Force the 1198 * I/O to occur immediately by invoking the relevant sections of 1199 * dsl_pool_sync(). 1200 */ 1201 if (os->os_encrypted) { 1202 dsl_dataset_t *tmpds = NULL; 1203 boolean_t need_sync_done = B_FALSE; 1204 1205 mutex_enter(&ds->ds_lock); 1206 ds->ds_owner = FTAG; 1207 mutex_exit(&ds->ds_lock); 1208 1209 rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1210 tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, 1211 tx->tx_txg); 1212 if (tmpds != NULL) { 1213 dsl_dataset_sync(ds, rzio, tx); 1214 need_sync_done = B_TRUE; 1215 } 1216 VERIFY0(zio_wait(rzio)); 1217 dmu_objset_do_userquota_updates(os, tx); 1218 taskq_wait(dp->dp_sync_taskq); 1219 if (txg_list_member(&dp->dp_dirty_datasets, ds, tx->tx_txg)) { 1220 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1221 key_mapping_rele(spa, ds->ds_key_mapping, ds); 1222 } 1223 1224 rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 1225 tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds, 1226 tx->tx_txg); 1227 if (tmpds != NULL) { 1228 dmu_buf_rele(ds->ds_dbuf, ds); 1229 dsl_dataset_sync(ds, rzio, tx); 1230 } 1231 VERIFY0(zio_wait(rzio)); 1232 1233 if (need_sync_done) { 1234 ASSERT3P(ds->ds_key_mapping, !=, NULL); 1235 key_mapping_rele(spa, ds->ds_key_mapping, ds); 1236 dsl_dataset_sync_done(ds, tx); 1237 } 1238 1239 mutex_enter(&ds->ds_lock); 1240 ds->ds_owner = NULL; 1241 mutex_exit(&ds->ds_lock); 1242 } 1243 1244 spa_history_log_internal_ds(ds, "create", tx, ""); 1245 dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG); 1246 dsl_dir_rele(pdd, FTAG); 1247 } 1248 1249 int 1250 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 1251 dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg) 1252 { 1253 dmu_objset_create_arg_t doca; 1254 dsl_crypto_params_t tmp_dcp = { 0 }; 1255 1256 doca.doca_name = name; 1257 doca.doca_cred = CRED(); 1258 doca.doca_flags = flags; 1259 doca.doca_userfunc = func; 1260 doca.doca_userarg = arg; 1261 doca.doca_type = type; 1262 1263 /* 1264 * Some callers (mostly for testing) do not provide a dcp on their 1265 * own but various code inside the sync task will require it to be 1266 * allocated. Rather than adding NULL checks throughout this code 1267 * or adding dummy dcp's to all of the callers we simply create a 1268 * dummy one here and use that. This zero dcp will have the same 1269 * effect as asking for inheritence of all encryption params. 1270 */ 1271 doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp; 1272 1273 return (dsl_sync_task(name, 1274 dmu_objset_create_check, dmu_objset_create_sync, &doca, 1275 6, ZFS_SPACE_CHECK_NORMAL)); 1276 } 1277 1278 typedef struct dmu_objset_clone_arg { 1279 const char *doca_clone; 1280 const char *doca_origin; 1281 cred_t *doca_cred; 1282 } dmu_objset_clone_arg_t; 1283 1284 /*ARGSUSED*/ 1285 static int 1286 dmu_objset_clone_check(void *arg, dmu_tx_t *tx) 1287 { 1288 dmu_objset_clone_arg_t *doca = arg; 1289 dsl_dir_t *pdd; 1290 const char *tail; 1291 int error; 1292 dsl_dataset_t *origin; 1293 dsl_pool_t *dp = dmu_tx_pool(tx); 1294 1295 if (strchr(doca->doca_clone, '@') != NULL) 1296 return (SET_ERROR(EINVAL)); 1297 1298 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) 1299 return (SET_ERROR(ENAMETOOLONG)); 1300 1301 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); 1302 if (error != 0) 1303 return (error); 1304 if (tail == NULL) { 1305 dsl_dir_rele(pdd, FTAG); 1306 return (SET_ERROR(EEXIST)); 1307 } 1308 1309 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 1310 doca->doca_cred); 1311 if (error != 0) { 1312 dsl_dir_rele(pdd, FTAG); 1313 return (SET_ERROR(EDQUOT)); 1314 } 1315 1316 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); 1317 if (error != 0) { 1318 dsl_dir_rele(pdd, FTAG); 1319 return (error); 1320 } 1321 1322 /* You can only clone snapshots, not the head datasets. */ 1323 if (!origin->ds_is_snapshot) { 1324 dsl_dataset_rele(origin, FTAG); 1325 dsl_dir_rele(pdd, FTAG); 1326 return (SET_ERROR(EINVAL)); 1327 } 1328 1329 dsl_dataset_rele(origin, FTAG); 1330 dsl_dir_rele(pdd, FTAG); 1331 1332 return (0); 1333 } 1334 1335 static void 1336 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) 1337 { 1338 dmu_objset_clone_arg_t *doca = arg; 1339 dsl_pool_t *dp = dmu_tx_pool(tx); 1340 dsl_dir_t *pdd; 1341 const char *tail; 1342 dsl_dataset_t *origin, *ds; 1343 uint64_t obj; 1344 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 1345 1346 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); 1347 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); 1348 1349 obj = dsl_dataset_create_sync(pdd, tail, origin, 0, 1350 doca->doca_cred, NULL, tx); 1351 1352 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 1353 dsl_dataset_name(origin, namebuf); 1354 spa_history_log_internal_ds(ds, "clone", tx, 1355 "origin=%s (%llu)", namebuf, origin->ds_object); 1356 dsl_dataset_rele(ds, FTAG); 1357 dsl_dataset_rele(origin, FTAG); 1358 dsl_dir_rele(pdd, FTAG); 1359 } 1360 1361 int 1362 dmu_objset_clone(const char *clone, const char *origin) 1363 { 1364 dmu_objset_clone_arg_t doca; 1365 1366 doca.doca_clone = clone; 1367 doca.doca_origin = origin; 1368 doca.doca_cred = CRED(); 1369 1370 return (dsl_sync_task(clone, 1371 dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 1372 6, ZFS_SPACE_CHECK_NORMAL)); 1373 } 1374 1375 static int 1376 dmu_objset_remap_indirects_impl(objset_t *os, uint64_t last_removed_txg) 1377 { 1378 int error = 0; 1379 uint64_t object = 0; 1380 while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) { 1381 error = dmu_object_remap_indirects(os, object, 1382 last_removed_txg); 1383 /* 1384 * If the ZPL removed the object before we managed to dnode_hold 1385 * it, we would get an ENOENT. If the ZPL declares its intent 1386 * to remove the object (dnode_free) before we manage to 1387 * dnode_hold it, we would get an EEXIST. In either case, we 1388 * want to continue remapping the other objects in the objset; 1389 * in all other cases, we want to break early. 1390 */ 1391 if (error != 0 && error != ENOENT && error != EEXIST) { 1392 break; 1393 } 1394 } 1395 if (error == ESRCH) { 1396 error = 0; 1397 } 1398 return (error); 1399 } 1400 1401 int 1402 dmu_objset_remap_indirects(const char *fsname) 1403 { 1404 int error = 0; 1405 objset_t *os = NULL; 1406 uint64_t last_removed_txg; 1407 uint64_t remap_start_txg; 1408 dsl_dir_t *dd; 1409 1410 error = dmu_objset_hold(fsname, FTAG, &os); 1411 if (error != 0) { 1412 return (error); 1413 } 1414 dd = dmu_objset_ds(os)->ds_dir; 1415 1416 if (!spa_feature_is_enabled(dmu_objset_spa(os), 1417 SPA_FEATURE_OBSOLETE_COUNTS)) { 1418 dmu_objset_rele(os, FTAG); 1419 return (SET_ERROR(ENOTSUP)); 1420 } 1421 1422 if (dsl_dataset_is_snapshot(dmu_objset_ds(os))) { 1423 dmu_objset_rele(os, FTAG); 1424 return (SET_ERROR(EINVAL)); 1425 } 1426 1427 /* 1428 * If there has not been a removal, we're done. 1429 */ 1430 last_removed_txg = spa_get_last_removal_txg(dmu_objset_spa(os)); 1431 if (last_removed_txg == -1ULL) { 1432 dmu_objset_rele(os, FTAG); 1433 return (0); 1434 } 1435 1436 /* 1437 * If we have remapped since the last removal, we're done. 1438 */ 1439 if (dsl_dir_is_zapified(dd)) { 1440 uint64_t last_remap_txg; 1441 if (zap_lookup(spa_meta_objset(dmu_objset_spa(os)), 1442 dd->dd_object, DD_FIELD_LAST_REMAP_TXG, 1443 sizeof (last_remap_txg), 1, &last_remap_txg) == 0 && 1444 last_remap_txg > last_removed_txg) { 1445 dmu_objset_rele(os, FTAG); 1446 return (0); 1447 } 1448 } 1449 1450 dsl_dataset_long_hold(dmu_objset_ds(os), FTAG); 1451 dsl_pool_rele(dmu_objset_pool(os), FTAG); 1452 1453 remap_start_txg = spa_last_synced_txg(dmu_objset_spa(os)); 1454 error = dmu_objset_remap_indirects_impl(os, last_removed_txg); 1455 if (error == 0) { 1456 /* 1457 * We update the last_remap_txg to be the start txg so that 1458 * we can guarantee that every block older than last_remap_txg 1459 * that can be remapped has been remapped. 1460 */ 1461 error = dsl_dir_update_last_remap_txg(dd, remap_start_txg); 1462 } 1463 1464 dsl_dataset_long_rele(dmu_objset_ds(os), FTAG); 1465 dsl_dataset_rele(dmu_objset_ds(os), FTAG); 1466 1467 return (error); 1468 } 1469 1470 int 1471 dmu_objset_snapshot_one(const char *fsname, const char *snapname) 1472 { 1473 int err; 1474 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); 1475 nvlist_t *snaps = fnvlist_alloc(); 1476 1477 fnvlist_add_boolean(snaps, longsnap); 1478 strfree(longsnap); 1479 err = dsl_dataset_snapshot(snaps, NULL, NULL); 1480 fnvlist_free(snaps); 1481 return (err); 1482 } 1483 1484 static void 1485 dmu_objset_upgrade_task_cb(void *data) 1486 { 1487 objset_t *os = data; 1488 1489 mutex_enter(&os->os_upgrade_lock); 1490 os->os_upgrade_status = EINTR; 1491 if (!os->os_upgrade_exit) { 1492 int status; 1493 1494 mutex_exit(&os->os_upgrade_lock); 1495 1496 status = os->os_upgrade_cb(os); 1497 1498 mutex_enter(&os->os_upgrade_lock); 1499 1500 os->os_upgrade_status = status; 1501 } 1502 os->os_upgrade_exit = B_TRUE; 1503 os->os_upgrade_id = 0; 1504 mutex_exit(&os->os_upgrade_lock); 1505 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1506 } 1507 1508 static void 1509 dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb) 1510 { 1511 if (os->os_upgrade_id != 0) 1512 return; 1513 1514 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 1515 dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag); 1516 1517 mutex_enter(&os->os_upgrade_lock); 1518 if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) { 1519 os->os_upgrade_exit = B_FALSE; 1520 os->os_upgrade_cb = cb; 1521 os->os_upgrade_id = taskq_dispatch( 1522 os->os_spa->spa_upgrade_taskq, 1523 dmu_objset_upgrade_task_cb, os, TQ_SLEEP); 1524 if (os->os_upgrade_id == TASKQID_INVALID) { 1525 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1526 os->os_upgrade_status = ENOMEM; 1527 } 1528 } else { 1529 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag); 1530 } 1531 mutex_exit(&os->os_upgrade_lock); 1532 } 1533 1534 static void 1535 dmu_objset_upgrade_stop(objset_t *os) 1536 { 1537 mutex_enter(&os->os_upgrade_lock); 1538 os->os_upgrade_exit = B_TRUE; 1539 if (os->os_upgrade_id != 0) { 1540 taskqid_t tid = os->os_upgrade_id; 1541 1542 mutex_exit(&os->os_upgrade_lock); 1543 1544 taskq_wait_id(os->os_spa->spa_upgrade_taskq, tid); 1545 txg_wait_synced(os->os_spa->spa_dsl_pool, 0); 1546 } else { 1547 mutex_exit(&os->os_upgrade_lock); 1548 } 1549 } 1550 1551 static void 1552 dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) 1553 { 1554 dnode_t *dn; 1555 1556 while ((dn = multilist_sublist_head(list)) != NULL) { 1557 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 1558 ASSERT(dn->dn_dbuf->db_data_pending); 1559 /* 1560 * Initialize dn_zio outside dnode_sync() because the 1561 * meta-dnode needs to set it outside dnode_sync(). 1562 */ 1563 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; 1564 ASSERT(dn->dn_zio); 1565 1566 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); 1567 multilist_sublist_remove(list, dn); 1568 1569 /* 1570 * If we are not doing useraccounting (os_synced_dnodes == NULL) 1571 * we are done with this dnode for this txg. Unset dn_dirty_txg 1572 * if later txgs aren't dirtying it so that future holders do 1573 * not get a stale value. Otherwise, we will do this in 1574 * userquota_updates_task() when processing has completely 1575 * finished for this txg. 1576 */ 1577 multilist_t *newlist = dn->dn_objset->os_synced_dnodes; 1578 if (newlist != NULL) { 1579 (void) dnode_add_ref(dn, newlist); 1580 multilist_insert(newlist, dn); 1581 } else { 1582 mutex_enter(&dn->dn_mtx); 1583 if (dn->dn_dirty_txg == tx->tx_txg) 1584 dn->dn_dirty_txg = 0; 1585 mutex_exit(&dn->dn_mtx); 1586 } 1587 1588 dnode_sync(dn, tx); 1589 } 1590 } 1591 1592 /* ARGSUSED */ 1593 static void 1594 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) 1595 { 1596 blkptr_t *bp = zio->io_bp; 1597 objset_t *os = arg; 1598 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; 1599 uint64_t fill = 0; 1600 1601 ASSERT(!BP_IS_EMBEDDED(bp)); 1602 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); 1603 1604 /* 1605 * Update rootbp fill count: it should be the number of objects 1606 * allocated in the object set (not counting the "special" 1607 * objects that are stored in the objset_phys_t -- the meta 1608 * dnode and user/group/project accounting objects). 1609 */ 1610 for (int i = 0; i < dnp->dn_nblkptr; i++) 1611 fill += BP_GET_FILL(&dnp->dn_blkptr[i]); 1612 1613 BP_SET_FILL(bp, fill); 1614 1615 if (os->os_dsl_dataset != NULL) 1616 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); 1617 *os->os_rootbp = *bp; 1618 if (os->os_dsl_dataset != NULL) 1619 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); 1620 } 1621 1622 /* ARGSUSED */ 1623 static void 1624 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) 1625 { 1626 blkptr_t *bp = zio->io_bp; 1627 blkptr_t *bp_orig = &zio->io_bp_orig; 1628 objset_t *os = arg; 1629 1630 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { 1631 ASSERT(BP_EQUAL(bp, bp_orig)); 1632 } else { 1633 dsl_dataset_t *ds = os->os_dsl_dataset; 1634 dmu_tx_t *tx = os->os_synctx; 1635 1636 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); 1637 dsl_dataset_block_born(ds, bp, tx); 1638 } 1639 kmem_free(bp, sizeof (*bp)); 1640 } 1641 1642 typedef struct sync_dnodes_arg { 1643 multilist_t *sda_list; 1644 int sda_sublist_idx; 1645 multilist_t *sda_newlist; 1646 dmu_tx_t *sda_tx; 1647 } sync_dnodes_arg_t; 1648 1649 static void 1650 sync_dnodes_task(void *arg) 1651 { 1652 sync_dnodes_arg_t *sda = arg; 1653 1654 multilist_sublist_t *ms = 1655 multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx); 1656 1657 dmu_objset_sync_dnodes(ms, sda->sda_tx); 1658 1659 multilist_sublist_unlock(ms); 1660 1661 kmem_free(sda, sizeof (*sda)); 1662 } 1663 1664 1665 /* called from dsl */ 1666 void 1667 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) 1668 { 1669 int txgoff; 1670 zbookmark_phys_t zb; 1671 zio_prop_t zp; 1672 zio_t *zio; 1673 list_t *list; 1674 dbuf_dirty_record_t *dr; 1675 blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); 1676 *blkptr_copy = *os->os_rootbp; 1677 1678 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); 1679 1680 ASSERT(dmu_tx_is_syncing(tx)); 1681 /* XXX the write_done callback should really give us the tx... */ 1682 os->os_synctx = tx; 1683 1684 if (os->os_dsl_dataset == NULL) { 1685 /* 1686 * This is the MOS. If we have upgraded, 1687 * spa_max_replication() could change, so reset 1688 * os_copies here. 1689 */ 1690 os->os_copies = spa_max_replication(os->os_spa); 1691 } 1692 1693 /* 1694 * Create the root block IO 1695 */ 1696 SET_BOOKMARK(&zb, os->os_dsl_dataset ? 1697 os->os_dsl_dataset->ds_object : DMU_META_OBJSET, 1698 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 1699 arc_release(os->os_phys_buf, &os->os_phys_buf); 1700 1701 dmu_write_policy(os, NULL, 0, 0, &zp); 1702 1703 /* 1704 * If we are either claiming the ZIL or doing a raw receive, write 1705 * out the os_phys_buf raw. Neither of these actions will effect the 1706 * MAC at this point. 1707 */ 1708 if (os->os_raw_receive || 1709 os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { 1710 ASSERT(os->os_encrypted); 1711 arc_convert_to_raw(os->os_phys_buf, 1712 os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER, 1713 DMU_OT_OBJSET, NULL, NULL, NULL); 1714 } 1715 1716 zio = arc_write(pio, os->os_spa, tx->tx_txg, 1717 blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), 1718 &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, 1719 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); 1720 1721 /* 1722 * Sync special dnodes - the parent IO for the sync is the root block 1723 */ 1724 DMU_META_DNODE(os)->dn_zio = zio; 1725 dnode_sync(DMU_META_DNODE(os), tx); 1726 1727 os->os_phys->os_flags = os->os_flags; 1728 1729 if (DMU_USERUSED_DNODE(os) && 1730 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1731 DMU_USERUSED_DNODE(os)->dn_zio = zio; 1732 dnode_sync(DMU_USERUSED_DNODE(os), tx); 1733 DMU_GROUPUSED_DNODE(os)->dn_zio = zio; 1734 dnode_sync(DMU_GROUPUSED_DNODE(os), tx); 1735 } 1736 1737 if (DMU_PROJECTUSED_DNODE(os) && 1738 DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1739 DMU_PROJECTUSED_DNODE(os)->dn_zio = zio; 1740 dnode_sync(DMU_PROJECTUSED_DNODE(os), tx); 1741 } 1742 1743 txgoff = tx->tx_txg & TXG_MASK; 1744 1745 if (dmu_objset_userused_enabled(os) && 1746 (!os->os_encrypted || !dmu_objset_is_receiving(os))) { 1747 /* 1748 * We must create the list here because it uses the 1749 * dn_dirty_link[] of this txg. But it may already 1750 * exist because we call dsl_dataset_sync() twice per txg. 1751 */ 1752 if (os->os_synced_dnodes == NULL) { 1753 os->os_synced_dnodes = 1754 multilist_create(sizeof (dnode_t), 1755 offsetof(dnode_t, dn_dirty_link[txgoff]), 1756 dnode_multilist_index_func); 1757 } else { 1758 ASSERT3U(os->os_synced_dnodes->ml_offset, ==, 1759 offsetof(dnode_t, dn_dirty_link[txgoff])); 1760 } 1761 } 1762 1763 for (int i = 0; 1764 i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) { 1765 sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP); 1766 sda->sda_list = os->os_dirty_dnodes[txgoff]; 1767 sda->sda_sublist_idx = i; 1768 sda->sda_tx = tx; 1769 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, 1770 sync_dnodes_task, sda, 0); 1771 /* callback frees sda */ 1772 } 1773 taskq_wait(dmu_objset_pool(os)->dp_sync_taskq); 1774 1775 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; 1776 while ((dr = list_head(list)) != NULL) { 1777 ASSERT0(dr->dr_dbuf->db_level); 1778 list_remove(list, dr); 1779 if (dr->dr_zio) 1780 zio_nowait(dr->dr_zio); 1781 } 1782 1783 /* Enable dnode backfill if enough objects have been freed. */ 1784 if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { 1785 os->os_rescan_dnodes = B_TRUE; 1786 os->os_freed_dnodes = 0; 1787 } 1788 1789 /* 1790 * Free intent log blocks up to this tx. 1791 */ 1792 zil_sync(os->os_zil, tx); 1793 os->os_phys->os_zil_header = os->os_zil_header; 1794 zio_nowait(zio); 1795 } 1796 1797 boolean_t 1798 dmu_objset_is_dirty(objset_t *os, uint64_t txg) 1799 { 1800 return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK])); 1801 } 1802 1803 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES]; 1804 1805 void 1806 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb) 1807 { 1808 used_cbs[ost] = cb; 1809 } 1810 1811 boolean_t 1812 dmu_objset_userused_enabled(objset_t *os) 1813 { 1814 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && 1815 used_cbs[os->os_phys->os_type] != NULL && 1816 DMU_USERUSED_DNODE(os) != NULL); 1817 } 1818 1819 boolean_t 1820 dmu_objset_userobjused_enabled(objset_t *os) 1821 { 1822 return (dmu_objset_userused_enabled(os) && 1823 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING)); 1824 } 1825 1826 boolean_t 1827 dmu_objset_projectquota_enabled(objset_t *os) 1828 { 1829 return (used_cbs[os->os_phys->os_type] != NULL && 1830 DMU_PROJECTUSED_DNODE(os) != NULL && 1831 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA)); 1832 } 1833 1834 typedef struct userquota_node { 1835 /* must be in the first field, see userquota_update_cache() */ 1836 char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN]; 1837 int64_t uqn_delta; 1838 avl_node_t uqn_node; 1839 } userquota_node_t; 1840 1841 typedef struct userquota_cache { 1842 avl_tree_t uqc_user_deltas; 1843 avl_tree_t uqc_group_deltas; 1844 avl_tree_t uqc_project_deltas; 1845 } userquota_cache_t; 1846 1847 static int 1848 userquota_compare(const void *l, const void *r) 1849 { 1850 const userquota_node_t *luqn = l; 1851 const userquota_node_t *ruqn = r; 1852 int rv; 1853 1854 /* 1855 * NB: can only access uqn_id because userquota_update_cache() doesn't 1856 * pass in an entire userquota_node_t. 1857 */ 1858 rv = strcmp(luqn->uqn_id, ruqn->uqn_id); 1859 1860 return (TREE_ISIGN(rv)); 1861 } 1862 1863 static void 1864 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) 1865 { 1866 void *cookie; 1867 userquota_node_t *uqn; 1868 1869 ASSERT(dmu_tx_is_syncing(tx)); 1870 1871 cookie = NULL; 1872 while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, 1873 &cookie)) != NULL) { 1874 /* 1875 * os_userused_lock protects against concurrent calls to 1876 * zap_increment_int(). It's needed because zap_increment_int() 1877 * is not thread-safe (i.e. not atomic). 1878 */ 1879 mutex_enter(&os->os_userused_lock); 1880 VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT, 1881 uqn->uqn_id, uqn->uqn_delta, tx)); 1882 mutex_exit(&os->os_userused_lock); 1883 kmem_free(uqn, sizeof (*uqn)); 1884 } 1885 avl_destroy(&cache->uqc_user_deltas); 1886 1887 cookie = NULL; 1888 while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, 1889 &cookie)) != NULL) { 1890 mutex_enter(&os->os_userused_lock); 1891 VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT, 1892 uqn->uqn_id, uqn->uqn_delta, tx)); 1893 mutex_exit(&os->os_userused_lock); 1894 kmem_free(uqn, sizeof (*uqn)); 1895 } 1896 avl_destroy(&cache->uqc_group_deltas); 1897 1898 if (dmu_objset_projectquota_enabled(os)) { 1899 cookie = NULL; 1900 while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas, 1901 &cookie)) != NULL) { 1902 mutex_enter(&os->os_userused_lock); 1903 VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT, 1904 uqn->uqn_id, uqn->uqn_delta, tx)); 1905 mutex_exit(&os->os_userused_lock); 1906 kmem_free(uqn, sizeof (*uqn)); 1907 } 1908 avl_destroy(&cache->uqc_project_deltas); 1909 } 1910 } 1911 1912 static void 1913 userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta) 1914 { 1915 userquota_node_t *uqn; 1916 avl_index_t idx; 1917 1918 ASSERT(strlen(id) < sizeof (uqn->uqn_id)); 1919 /* 1920 * Use id directly for searching because uqn_id is the first field of 1921 * userquota_node_t and fields after uqn_id won't be accessed in 1922 * avl_find(). 1923 */ 1924 uqn = avl_find(avl, (const void *)id, &idx); 1925 if (uqn == NULL) { 1926 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); 1927 (void) strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id)); 1928 avl_insert(avl, uqn, idx); 1929 } 1930 uqn->uqn_delta += delta; 1931 } 1932 1933 static void 1934 do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used, 1935 uint64_t flags, uint64_t user, uint64_t group, uint64_t project, 1936 boolean_t subtract) 1937 { 1938 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) { 1939 int64_t delta = DNODE_MIN_SIZE + used; 1940 char name[20]; 1941 1942 if (subtract) 1943 delta = -delta; 1944 1945 (void) sprintf(name, "%llx", (longlong_t)user); 1946 userquota_update_cache(&cache->uqc_user_deltas, name, delta); 1947 1948 (void) sprintf(name, "%llx", (longlong_t)group); 1949 userquota_update_cache(&cache->uqc_group_deltas, name, delta); 1950 1951 if (dmu_objset_projectquota_enabled(os)) { 1952 (void) sprintf(name, "%llx", (longlong_t)project); 1953 userquota_update_cache(&cache->uqc_project_deltas, 1954 name, delta); 1955 } 1956 } 1957 } 1958 1959 static void 1960 do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags, 1961 uint64_t user, uint64_t group, uint64_t project, boolean_t subtract) 1962 { 1963 if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) { 1964 char name[20 + DMU_OBJACCT_PREFIX_LEN]; 1965 int delta = subtract ? -1 : 1; 1966 1967 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", 1968 (longlong_t)user); 1969 userquota_update_cache(&cache->uqc_user_deltas, name, delta); 1970 1971 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx", 1972 (longlong_t)group); 1973 userquota_update_cache(&cache->uqc_group_deltas, name, delta); 1974 1975 if (dmu_objset_projectquota_enabled(os)) { 1976 (void) snprintf(name, sizeof (name), 1977 DMU_OBJACCT_PREFIX "%llx", (longlong_t)project); 1978 userquota_update_cache(&cache->uqc_project_deltas, 1979 name, delta); 1980 } 1981 } 1982 } 1983 1984 typedef struct userquota_updates_arg { 1985 objset_t *uua_os; 1986 int uua_sublist_idx; 1987 dmu_tx_t *uua_tx; 1988 } userquota_updates_arg_t; 1989 1990 static void 1991 userquota_updates_task(void *arg) 1992 { 1993 userquota_updates_arg_t *uua = arg; 1994 objset_t *os = uua->uua_os; 1995 dmu_tx_t *tx = uua->uua_tx; 1996 dnode_t *dn; 1997 userquota_cache_t cache = { 0 }; 1998 1999 multilist_sublist_t *list = 2000 multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx); 2001 2002 ASSERT(multilist_sublist_head(list) == NULL || 2003 dmu_objset_userused_enabled(os)); 2004 avl_create(&cache.uqc_user_deltas, userquota_compare, 2005 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 2006 avl_create(&cache.uqc_group_deltas, userquota_compare, 2007 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 2008 if (dmu_objset_projectquota_enabled(os)) 2009 avl_create(&cache.uqc_project_deltas, userquota_compare, 2010 sizeof (userquota_node_t), offsetof(userquota_node_t, 2011 uqn_node)); 2012 2013 while ((dn = multilist_sublist_head(list)) != NULL) { 2014 int flags; 2015 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); 2016 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || 2017 dn->dn_phys->dn_flags & 2018 DNODE_FLAG_USERUSED_ACCOUNTED); 2019 2020 flags = dn->dn_id_flags; 2021 ASSERT(flags); 2022 if (flags & DN_ID_OLD_EXIST) { 2023 do_userquota_update(os, &cache, dn->dn_oldused, 2024 dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid, 2025 dn->dn_oldprojid, B_TRUE); 2026 do_userobjquota_update(os, &cache, dn->dn_oldflags, 2027 dn->dn_olduid, dn->dn_oldgid, 2028 dn->dn_oldprojid, B_TRUE); 2029 } 2030 if (flags & DN_ID_NEW_EXIST) { 2031 do_userquota_update(os, &cache, 2032 DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags, 2033 dn->dn_newuid, dn->dn_newgid, 2034 dn->dn_newprojid, B_FALSE); 2035 do_userobjquota_update(os, &cache, 2036 dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid, 2037 dn->dn_newprojid, B_FALSE); 2038 } 2039 2040 mutex_enter(&dn->dn_mtx); 2041 dn->dn_oldused = 0; 2042 dn->dn_oldflags = 0; 2043 if (dn->dn_id_flags & DN_ID_NEW_EXIST) { 2044 dn->dn_olduid = dn->dn_newuid; 2045 dn->dn_oldgid = dn->dn_newgid; 2046 dn->dn_oldprojid = dn->dn_newprojid; 2047 dn->dn_id_flags |= DN_ID_OLD_EXIST; 2048 if (dn->dn_bonuslen == 0) 2049 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 2050 else 2051 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2052 } 2053 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); 2054 if (dn->dn_dirty_txg == spa_syncing_txg(os->os_spa)) 2055 dn->dn_dirty_txg = 0; 2056 mutex_exit(&dn->dn_mtx); 2057 2058 multilist_sublist_remove(list, dn); 2059 dnode_rele(dn, os->os_synced_dnodes); 2060 } 2061 do_userquota_cacheflush(os, &cache, tx); 2062 multilist_sublist_unlock(list); 2063 kmem_free(uua, sizeof (*uua)); 2064 } 2065 2066 void 2067 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) 2068 { 2069 if (!dmu_objset_userused_enabled(os)) 2070 return; 2071 2072 /* 2073 * If this is a raw receive just return and handle accounting 2074 * later when we have the keys loaded. We also don't do user 2075 * accounting during claiming since the datasets are not owned 2076 * for the duration of claiming and this txg should only be 2077 * used for recovery. 2078 */ 2079 if (os->os_encrypted && dmu_objset_is_receiving(os)) 2080 return; 2081 2082 if (tx->tx_txg <= os->os_spa->spa_claim_max_txg) 2083 return; 2084 2085 /* Allocate the user/group/project used objects if necessary. */ 2086 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 2087 VERIFY0(zap_create_claim(os, 2088 DMU_USERUSED_OBJECT, 2089 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2090 VERIFY0(zap_create_claim(os, 2091 DMU_GROUPUSED_OBJECT, 2092 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2093 } 2094 2095 if (dmu_objset_projectquota_enabled(os) && 2096 DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 2097 VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT, 2098 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 2099 } 2100 2101 for (int i = 0; 2102 i < multilist_get_num_sublists(os->os_synced_dnodes); i++) { 2103 userquota_updates_arg_t *uua = 2104 kmem_alloc(sizeof (*uua), KM_SLEEP); 2105 uua->uua_os = os; 2106 uua->uua_sublist_idx = i; 2107 uua->uua_tx = tx; 2108 /* note: caller does taskq_wait() */ 2109 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, 2110 userquota_updates_task, uua, 0); 2111 /* callback frees uua */ 2112 } 2113 } 2114 2115 /* 2116 * Returns a pointer to data to find uid/gid from 2117 * 2118 * If a dirty record for transaction group that is syncing can't 2119 * be found then NULL is returned. In the NULL case it is assumed 2120 * the uid/gid aren't changing. 2121 */ 2122 static void * 2123 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) 2124 { 2125 dbuf_dirty_record_t *dr, **drp; 2126 void *data; 2127 2128 if (db->db_dirtycnt == 0) 2129 return (db->db.db_data); /* Nothing is changing */ 2130 2131 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) 2132 if (dr->dr_txg == tx->tx_txg) 2133 break; 2134 2135 if (dr == NULL) { 2136 data = NULL; 2137 } else { 2138 dnode_t *dn; 2139 2140 DB_DNODE_ENTER(dr->dr_dbuf); 2141 dn = DB_DNODE(dr->dr_dbuf); 2142 2143 if (dn->dn_bonuslen == 0 && 2144 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) 2145 data = dr->dt.dl.dr_data->b_data; 2146 else 2147 data = dr->dt.dl.dr_data; 2148 2149 DB_DNODE_EXIT(dr->dr_dbuf); 2150 } 2151 2152 return (data); 2153 } 2154 2155 void 2156 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) 2157 { 2158 objset_t *os = dn->dn_objset; 2159 void *data = NULL; 2160 dmu_buf_impl_t *db = NULL; 2161 uint64_t *user = NULL; 2162 uint64_t *group = NULL; 2163 uint64_t *project = NULL; 2164 int flags = dn->dn_id_flags; 2165 int error; 2166 boolean_t have_spill = B_FALSE; 2167 2168 if (!dmu_objset_userused_enabled(dn->dn_objset)) 2169 return; 2170 2171 /* 2172 * Raw receives introduce a problem with user accounting. Raw 2173 * receives cannot update the user accounting info because the 2174 * user ids and the sizes are encrypted. To guarantee that we 2175 * never end up with bad user accounting, we simply disable it 2176 * during raw receives. We also disable this for normal receives 2177 * so that an incremental raw receive may be done on top of an 2178 * existing non-raw receive. 2179 */ 2180 if (os->os_encrypted && dmu_objset_is_receiving(os)) 2181 return; 2182 2183 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| 2184 DN_ID_CHKED_SPILL))) 2185 return; 2186 2187 if (before && dn->dn_bonuslen != 0) 2188 data = DN_BONUS(dn->dn_phys); 2189 else if (!before && dn->dn_bonuslen != 0) { 2190 if (dn->dn_bonus) { 2191 db = dn->dn_bonus; 2192 mutex_enter(&db->db_mtx); 2193 data = dmu_objset_userquota_find_data(db, tx); 2194 } else { 2195 data = DN_BONUS(dn->dn_phys); 2196 } 2197 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { 2198 int rf = 0; 2199 2200 if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) 2201 rf |= DB_RF_HAVESTRUCT; 2202 error = dmu_spill_hold_by_dnode(dn, 2203 rf | DB_RF_MUST_SUCCEED, 2204 FTAG, (dmu_buf_t **)&db); 2205 ASSERT(error == 0); 2206 mutex_enter(&db->db_mtx); 2207 data = (before) ? db->db.db_data : 2208 dmu_objset_userquota_find_data(db, tx); 2209 have_spill = B_TRUE; 2210 } else { 2211 mutex_enter(&dn->dn_mtx); 2212 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2213 mutex_exit(&dn->dn_mtx); 2214 return; 2215 } 2216 2217 if (before) { 2218 ASSERT(data); 2219 user = &dn->dn_olduid; 2220 group = &dn->dn_oldgid; 2221 project = &dn->dn_oldprojid; 2222 } else if (data) { 2223 user = &dn->dn_newuid; 2224 group = &dn->dn_newgid; 2225 project = &dn->dn_newprojid; 2226 } 2227 2228 /* 2229 * Must always call the callback in case the object 2230 * type has changed and that type isn't an object type to track 2231 */ 2232 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data, 2233 user, group, project); 2234 2235 /* 2236 * Preserve existing uid/gid when the callback can't determine 2237 * what the new uid/gid are and the callback returned EEXIST. 2238 * The EEXIST error tells us to just use the existing uid/gid. 2239 * If we don't know what the old values are then just assign 2240 * them to 0, since that is a new file being created. 2241 */ 2242 if (!before && data == NULL && error == EEXIST) { 2243 if (flags & DN_ID_OLD_EXIST) { 2244 dn->dn_newuid = dn->dn_olduid; 2245 dn->dn_newgid = dn->dn_oldgid; 2246 dn->dn_newprojid = dn->dn_oldprojid; 2247 } else { 2248 dn->dn_newuid = 0; 2249 dn->dn_newgid = 0; 2250 dn->dn_newprojid = ZFS_DEFAULT_PROJID; 2251 } 2252 error = 0; 2253 } 2254 2255 if (db) 2256 mutex_exit(&db->db_mtx); 2257 2258 mutex_enter(&dn->dn_mtx); 2259 if (error == 0 && before) 2260 dn->dn_id_flags |= DN_ID_OLD_EXIST; 2261 if (error == 0 && !before) 2262 dn->dn_id_flags |= DN_ID_NEW_EXIST; 2263 2264 if (have_spill) { 2265 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 2266 } else { 2267 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 2268 } 2269 mutex_exit(&dn->dn_mtx); 2270 if (have_spill) 2271 dmu_buf_rele((dmu_buf_t *)db, FTAG); 2272 } 2273 2274 boolean_t 2275 dmu_objset_userspace_present(objset_t *os) 2276 { 2277 return (os->os_phys->os_flags & 2278 OBJSET_FLAG_USERACCOUNTING_COMPLETE); 2279 } 2280 2281 boolean_t 2282 dmu_objset_userobjspace_present(objset_t *os) 2283 { 2284 return (os->os_phys->os_flags & 2285 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE); 2286 } 2287 2288 boolean_t 2289 dmu_objset_projectquota_present(objset_t *os) 2290 { 2291 return (os->os_phys->os_flags & 2292 OBJSET_FLAG_PROJECTQUOTA_COMPLETE); 2293 } 2294 2295 static int 2296 dmu_objset_space_upgrade(objset_t *os) 2297 { 2298 uint64_t obj; 2299 int err = 0; 2300 2301 /* 2302 * We simply need to mark every object dirty, so that it will be 2303 * synced out and now accounted. If this is called 2304 * concurrently, or if we already did some work before crashing, 2305 * that's fine, since we track each object's accounted state 2306 * independently. 2307 */ 2308 2309 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { 2310 dmu_tx_t *tx; 2311 dmu_buf_t *db; 2312 int objerr; 2313 2314 mutex_enter(&os->os_upgrade_lock); 2315 if (os->os_upgrade_exit) 2316 err = SET_ERROR(EINTR); 2317 mutex_exit(&os->os_upgrade_lock); 2318 if (err != 0) 2319 return (err); 2320 2321 /* 2322 * The following is only valid on Linux since we cannot send 2323 * a signal to a kernel thread on illumos (because we have no 2324 * lwp and never return to user-land). 2325 * 2326 * if (issig(JUSTLOOKING) && issig(FORREAL)) 2327 * return (SET_ERROR(EINTR)); 2328 */ 2329 2330 objerr = dmu_bonus_hold(os, obj, FTAG, &db); 2331 if (objerr != 0) 2332 continue; 2333 tx = dmu_tx_create(os); 2334 dmu_tx_hold_bonus(tx, obj); 2335 objerr = dmu_tx_assign(tx, TXG_WAIT); 2336 if (objerr != 0) { 2337 dmu_buf_rele(db, FTAG); 2338 dmu_tx_abort(tx); 2339 continue; 2340 } 2341 dmu_buf_will_dirty(db, tx); 2342 dmu_buf_rele(db, FTAG); 2343 dmu_tx_commit(tx); 2344 } 2345 return (0); 2346 } 2347 2348 int 2349 dmu_objset_userspace_upgrade(objset_t *os) 2350 { 2351 int err = 0; 2352 2353 if (dmu_objset_userspace_present(os)) 2354 return (0); 2355 if (dmu_objset_is_snapshot(os)) 2356 return (SET_ERROR(EINVAL)); 2357 if (!dmu_objset_userused_enabled(os)) 2358 return (SET_ERROR(ENOTSUP)); 2359 2360 err = dmu_objset_space_upgrade(os); 2361 if (err) 2362 return (err); 2363 2364 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 2365 txg_wait_synced(dmu_objset_pool(os), 0); 2366 return (0); 2367 } 2368 2369 static int 2370 dmu_objset_id_quota_upgrade_cb(objset_t *os) 2371 { 2372 int err = 0; 2373 2374 if (dmu_objset_userobjspace_present(os) && 2375 dmu_objset_projectquota_present(os)) 2376 return (0); 2377 if (dmu_objset_is_snapshot(os)) 2378 return (SET_ERROR(EINVAL)); 2379 if (!dmu_objset_userobjused_enabled(os)) 2380 return (SET_ERROR(ENOTSUP)); 2381 if (!dmu_objset_projectquota_enabled(os) && 2382 dmu_objset_userobjspace_present(os)) 2383 return (SET_ERROR(ENOTSUP)); 2384 2385 dmu_objset_ds(os)->ds_feature_activation[ 2386 SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)(uintptr_t)B_TRUE; 2387 if (dmu_objset_projectquota_enabled(os)) 2388 dmu_objset_ds(os)->ds_feature_activation[ 2389 SPA_FEATURE_PROJECT_QUOTA] = (void *)(uintptr_t)B_TRUE; 2390 2391 err = dmu_objset_space_upgrade(os); 2392 if (err) 2393 return (err); 2394 2395 os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE; 2396 if (dmu_objset_projectquota_enabled(os)) 2397 os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE; 2398 2399 txg_wait_synced(dmu_objset_pool(os), 0); 2400 return (0); 2401 } 2402 2403 void 2404 dmu_objset_id_quota_upgrade(objset_t *os) 2405 { 2406 dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb); 2407 } 2408 2409 boolean_t 2410 dmu_objset_userobjspace_upgradable(objset_t *os) 2411 { 2412 return (dmu_objset_type(os) == DMU_OST_ZFS && 2413 !dmu_objset_is_snapshot(os) && 2414 dmu_objset_userobjused_enabled(os) && 2415 !dmu_objset_userobjspace_present(os) && 2416 spa_writeable(dmu_objset_spa(os))); 2417 } 2418 2419 boolean_t 2420 dmu_objset_projectquota_upgradable(objset_t *os) 2421 { 2422 return (dmu_objset_type(os) == DMU_OST_ZFS && 2423 !dmu_objset_is_snapshot(os) && 2424 dmu_objset_projectquota_enabled(os) && 2425 !dmu_objset_projectquota_present(os) && 2426 spa_writeable(dmu_objset_spa(os))); 2427 } 2428 2429 void 2430 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 2431 uint64_t *usedobjsp, uint64_t *availobjsp) 2432 { 2433 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, 2434 usedobjsp, availobjsp); 2435 } 2436 2437 uint64_t 2438 dmu_objset_fsid_guid(objset_t *os) 2439 { 2440 return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); 2441 } 2442 2443 void 2444 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) 2445 { 2446 stat->dds_type = os->os_phys->os_type; 2447 if (os->os_dsl_dataset) 2448 dsl_dataset_fast_stat(os->os_dsl_dataset, stat); 2449 } 2450 2451 void 2452 dmu_objset_stats(objset_t *os, nvlist_t *nv) 2453 { 2454 ASSERT(os->os_dsl_dataset || 2455 os->os_phys->os_type == DMU_OST_META); 2456 2457 if (os->os_dsl_dataset != NULL) 2458 dsl_dataset_stats(os->os_dsl_dataset, nv); 2459 2460 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, 2461 os->os_phys->os_type); 2462 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, 2463 dmu_objset_userspace_present(os)); 2464 } 2465 2466 int 2467 dmu_objset_is_snapshot(objset_t *os) 2468 { 2469 if (os->os_dsl_dataset != NULL) 2470 return (os->os_dsl_dataset->ds_is_snapshot); 2471 else 2472 return (B_FALSE); 2473 } 2474 2475 int 2476 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen, 2477 boolean_t *conflict) 2478 { 2479 dsl_dataset_t *ds = os->os_dsl_dataset; 2480 uint64_t ignored; 2481 2482 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 2483 return (SET_ERROR(ENOENT)); 2484 2485 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, 2486 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, 2487 MT_NORMALIZE, real, maxlen, conflict)); 2488 } 2489 2490 int 2491 dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 2492 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) 2493 { 2494 dsl_dataset_t *ds = os->os_dsl_dataset; 2495 zap_cursor_t cursor; 2496 zap_attribute_t attr; 2497 2498 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 2499 2500 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 2501 return (SET_ERROR(ENOENT)); 2502 2503 zap_cursor_init_serialized(&cursor, 2504 ds->ds_dir->dd_pool->dp_meta_objset, 2505 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); 2506 2507 if (zap_cursor_retrieve(&cursor, &attr) != 0) { 2508 zap_cursor_fini(&cursor); 2509 return (SET_ERROR(ENOENT)); 2510 } 2511 2512 if (strlen(attr.za_name) + 1 > namelen) { 2513 zap_cursor_fini(&cursor); 2514 return (SET_ERROR(ENAMETOOLONG)); 2515 } 2516 2517 (void) strcpy(name, attr.za_name); 2518 if (idp) 2519 *idp = attr.za_first_integer; 2520 if (case_conflict) 2521 *case_conflict = attr.za_normalization_conflict; 2522 zap_cursor_advance(&cursor); 2523 *offp = zap_cursor_serialize(&cursor); 2524 zap_cursor_fini(&cursor); 2525 2526 return (0); 2527 } 2528 2529 int 2530 dmu_dir_list_next(objset_t *os, int namelen, char *name, 2531 uint64_t *idp, uint64_t *offp) 2532 { 2533 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 2534 zap_cursor_t cursor; 2535 zap_attribute_t attr; 2536 2537 /* there is no next dir on a snapshot! */ 2538 if (os->os_dsl_dataset->ds_object != 2539 dsl_dir_phys(dd)->dd_head_dataset_obj) 2540 return (SET_ERROR(ENOENT)); 2541 2542 zap_cursor_init_serialized(&cursor, 2543 dd->dd_pool->dp_meta_objset, 2544 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); 2545 2546 if (zap_cursor_retrieve(&cursor, &attr) != 0) { 2547 zap_cursor_fini(&cursor); 2548 return (SET_ERROR(ENOENT)); 2549 } 2550 2551 if (strlen(attr.za_name) + 1 > namelen) { 2552 zap_cursor_fini(&cursor); 2553 return (SET_ERROR(ENAMETOOLONG)); 2554 } 2555 2556 (void) strcpy(name, attr.za_name); 2557 if (idp) 2558 *idp = attr.za_first_integer; 2559 zap_cursor_advance(&cursor); 2560 *offp = zap_cursor_serialize(&cursor); 2561 zap_cursor_fini(&cursor); 2562 2563 return (0); 2564 } 2565 2566 typedef struct dmu_objset_find_ctx { 2567 taskq_t *dc_tq; 2568 dsl_pool_t *dc_dp; 2569 uint64_t dc_ddobj; 2570 char *dc_ddname; /* last component of ddobj's name */ 2571 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); 2572 void *dc_arg; 2573 int dc_flags; 2574 kmutex_t *dc_error_lock; 2575 int *dc_error; 2576 } dmu_objset_find_ctx_t; 2577 2578 static void 2579 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) 2580 { 2581 dsl_pool_t *dp = dcp->dc_dp; 2582 dsl_dir_t *dd; 2583 dsl_dataset_t *ds; 2584 zap_cursor_t zc; 2585 zap_attribute_t *attr; 2586 uint64_t thisobj; 2587 int err = 0; 2588 2589 /* don't process if there already was an error */ 2590 if (*dcp->dc_error != 0) 2591 goto out; 2592 2593 /* 2594 * Note: passing the name (dc_ddname) here is optional, but it 2595 * improves performance because we don't need to call 2596 * zap_value_search() to determine the name. 2597 */ 2598 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); 2599 if (err != 0) 2600 goto out; 2601 2602 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2603 if (dd->dd_myname[0] == '$') { 2604 dsl_dir_rele(dd, FTAG); 2605 goto out; 2606 } 2607 2608 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2609 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 2610 2611 /* 2612 * Iterate over all children. 2613 */ 2614 if (dcp->dc_flags & DS_FIND_CHILDREN) { 2615 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2616 dsl_dir_phys(dd)->dd_child_dir_zapobj); 2617 zap_cursor_retrieve(&zc, attr) == 0; 2618 (void) zap_cursor_advance(&zc)) { 2619 ASSERT3U(attr->za_integer_length, ==, 2620 sizeof (uint64_t)); 2621 ASSERT3U(attr->za_num_integers, ==, 1); 2622 2623 dmu_objset_find_ctx_t *child_dcp = 2624 kmem_alloc(sizeof (*child_dcp), KM_SLEEP); 2625 *child_dcp = *dcp; 2626 child_dcp->dc_ddobj = attr->za_first_integer; 2627 child_dcp->dc_ddname = spa_strdup(attr->za_name); 2628 if (dcp->dc_tq != NULL) 2629 (void) taskq_dispatch(dcp->dc_tq, 2630 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); 2631 else 2632 dmu_objset_find_dp_impl(child_dcp); 2633 } 2634 zap_cursor_fini(&zc); 2635 } 2636 2637 /* 2638 * Iterate over all snapshots. 2639 */ 2640 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { 2641 dsl_dataset_t *ds; 2642 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2643 2644 if (err == 0) { 2645 uint64_t snapobj; 2646 2647 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 2648 dsl_dataset_rele(ds, FTAG); 2649 2650 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 2651 zap_cursor_retrieve(&zc, attr) == 0; 2652 (void) zap_cursor_advance(&zc)) { 2653 ASSERT3U(attr->za_integer_length, ==, 2654 sizeof (uint64_t)); 2655 ASSERT3U(attr->za_num_integers, ==, 1); 2656 2657 err = dsl_dataset_hold_obj(dp, 2658 attr->za_first_integer, FTAG, &ds); 2659 if (err != 0) 2660 break; 2661 err = dcp->dc_func(dp, ds, dcp->dc_arg); 2662 dsl_dataset_rele(ds, FTAG); 2663 if (err != 0) 2664 break; 2665 } 2666 zap_cursor_fini(&zc); 2667 } 2668 } 2669 2670 kmem_free(attr, sizeof (zap_attribute_t)); 2671 2672 if (err != 0) { 2673 dsl_dir_rele(dd, FTAG); 2674 goto out; 2675 } 2676 2677 /* 2678 * Apply to self. 2679 */ 2680 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2681 2682 /* 2683 * Note: we hold the dir while calling dsl_dataset_hold_obj() so 2684 * that the dir will remain cached, and we won't have to re-instantiate 2685 * it (which could be expensive due to finding its name via 2686 * zap_value_search()). 2687 */ 2688 dsl_dir_rele(dd, FTAG); 2689 if (err != 0) 2690 goto out; 2691 err = dcp->dc_func(dp, ds, dcp->dc_arg); 2692 dsl_dataset_rele(ds, FTAG); 2693 2694 out: 2695 if (err != 0) { 2696 mutex_enter(dcp->dc_error_lock); 2697 /* only keep first error */ 2698 if (*dcp->dc_error == 0) 2699 *dcp->dc_error = err; 2700 mutex_exit(dcp->dc_error_lock); 2701 } 2702 2703 if (dcp->dc_ddname != NULL) 2704 spa_strfree(dcp->dc_ddname); 2705 kmem_free(dcp, sizeof (*dcp)); 2706 } 2707 2708 static void 2709 dmu_objset_find_dp_cb(void *arg) 2710 { 2711 dmu_objset_find_ctx_t *dcp = arg; 2712 dsl_pool_t *dp = dcp->dc_dp; 2713 2714 /* 2715 * We need to get a pool_config_lock here, as there are several 2716 * asssert(pool_config_held) down the stack. Getting a lock via 2717 * dsl_pool_config_enter is risky, as it might be stalled by a 2718 * pending writer. This would deadlock, as the write lock can 2719 * only be granted when our parent thread gives up the lock. 2720 * The _prio interface gives us priority over a pending writer. 2721 */ 2722 dsl_pool_config_enter_prio(dp, FTAG); 2723 2724 dmu_objset_find_dp_impl(dcp); 2725 2726 dsl_pool_config_exit(dp, FTAG); 2727 } 2728 2729 /* 2730 * Find objsets under and including ddobj, call func(ds) on each. 2731 * The order for the enumeration is completely undefined. 2732 * func is called with dsl_pool_config held. 2733 */ 2734 int 2735 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, 2736 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) 2737 { 2738 int error = 0; 2739 taskq_t *tq = NULL; 2740 int ntasks; 2741 dmu_objset_find_ctx_t *dcp; 2742 kmutex_t err_lock; 2743 2744 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); 2745 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); 2746 dcp->dc_tq = NULL; 2747 dcp->dc_dp = dp; 2748 dcp->dc_ddobj = ddobj; 2749 dcp->dc_ddname = NULL; 2750 dcp->dc_func = func; 2751 dcp->dc_arg = arg; 2752 dcp->dc_flags = flags; 2753 dcp->dc_error_lock = &err_lock; 2754 dcp->dc_error = &error; 2755 2756 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { 2757 /* 2758 * In case a write lock is held we can't make use of 2759 * parallelism, as down the stack of the worker threads 2760 * the lock is asserted via dsl_pool_config_held. 2761 * In case of a read lock this is solved by getting a read 2762 * lock in each worker thread, which isn't possible in case 2763 * of a writer lock. So we fall back to the synchronous path 2764 * here. 2765 * In the future it might be possible to get some magic into 2766 * dsl_pool_config_held in a way that it returns true for 2767 * the worker threads so that a single lock held from this 2768 * thread suffices. For now, stay single threaded. 2769 */ 2770 dmu_objset_find_dp_impl(dcp); 2771 mutex_destroy(&err_lock); 2772 2773 return (error); 2774 } 2775 2776 ntasks = dmu_find_threads; 2777 if (ntasks == 0) 2778 ntasks = vdev_count_leaves(dp->dp_spa) * 4; 2779 tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks, 2780 INT_MAX, 0); 2781 if (tq == NULL) { 2782 kmem_free(dcp, sizeof (*dcp)); 2783 mutex_destroy(&err_lock); 2784 2785 return (SET_ERROR(ENOMEM)); 2786 } 2787 dcp->dc_tq = tq; 2788 2789 /* dcp will be freed by task */ 2790 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); 2791 2792 /* 2793 * PORTING: this code relies on the property of taskq_wait to wait 2794 * until no more tasks are queued and no more tasks are active. As 2795 * we always queue new tasks from within other tasks, task_wait 2796 * reliably waits for the full recursion to finish, even though we 2797 * enqueue new tasks after taskq_wait has been called. 2798 * On platforms other than illumos, taskq_wait may not have this 2799 * property. 2800 */ 2801 taskq_wait(tq); 2802 taskq_destroy(tq); 2803 mutex_destroy(&err_lock); 2804 2805 return (error); 2806 } 2807 2808 /* 2809 * Find all objsets under name, and for each, call 'func(child_name, arg)'. 2810 * The dp_config_rwlock must not be held when this is called, and it 2811 * will not be held when the callback is called. 2812 * Therefore this function should only be used when the pool is not changing 2813 * (e.g. in syncing context), or the callback can deal with the possible races. 2814 */ 2815 static int 2816 dmu_objset_find_impl(spa_t *spa, const char *name, 2817 int func(const char *, void *), void *arg, int flags) 2818 { 2819 dsl_dir_t *dd; 2820 dsl_pool_t *dp = spa_get_dsl(spa); 2821 dsl_dataset_t *ds; 2822 zap_cursor_t zc; 2823 zap_attribute_t *attr; 2824 char *child; 2825 uint64_t thisobj; 2826 int err; 2827 2828 dsl_pool_config_enter(dp, FTAG); 2829 2830 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); 2831 if (err != 0) { 2832 dsl_pool_config_exit(dp, FTAG); 2833 return (err); 2834 } 2835 2836 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2837 if (dd->dd_myname[0] == '$') { 2838 dsl_dir_rele(dd, FTAG); 2839 dsl_pool_config_exit(dp, FTAG); 2840 return (0); 2841 } 2842 2843 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2844 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 2845 2846 /* 2847 * Iterate over all children. 2848 */ 2849 if (flags & DS_FIND_CHILDREN) { 2850 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2851 dsl_dir_phys(dd)->dd_child_dir_zapobj); 2852 zap_cursor_retrieve(&zc, attr) == 0; 2853 (void) zap_cursor_advance(&zc)) { 2854 ASSERT3U(attr->za_integer_length, ==, 2855 sizeof (uint64_t)); 2856 ASSERT3U(attr->za_num_integers, ==, 1); 2857 2858 child = kmem_asprintf("%s/%s", name, attr->za_name); 2859 dsl_pool_config_exit(dp, FTAG); 2860 err = dmu_objset_find_impl(spa, child, 2861 func, arg, flags); 2862 dsl_pool_config_enter(dp, FTAG); 2863 strfree(child); 2864 if (err != 0) 2865 break; 2866 } 2867 zap_cursor_fini(&zc); 2868 2869 if (err != 0) { 2870 dsl_dir_rele(dd, FTAG); 2871 dsl_pool_config_exit(dp, FTAG); 2872 kmem_free(attr, sizeof (zap_attribute_t)); 2873 return (err); 2874 } 2875 } 2876 2877 /* 2878 * Iterate over all snapshots. 2879 */ 2880 if (flags & DS_FIND_SNAPSHOTS) { 2881 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2882 2883 if (err == 0) { 2884 uint64_t snapobj; 2885 2886 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 2887 dsl_dataset_rele(ds, FTAG); 2888 2889 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 2890 zap_cursor_retrieve(&zc, attr) == 0; 2891 (void) zap_cursor_advance(&zc)) { 2892 ASSERT3U(attr->za_integer_length, ==, 2893 sizeof (uint64_t)); 2894 ASSERT3U(attr->za_num_integers, ==, 1); 2895 2896 child = kmem_asprintf("%s@%s", 2897 name, attr->za_name); 2898 dsl_pool_config_exit(dp, FTAG); 2899 err = func(child, arg); 2900 dsl_pool_config_enter(dp, FTAG); 2901 strfree(child); 2902 if (err != 0) 2903 break; 2904 } 2905 zap_cursor_fini(&zc); 2906 } 2907 } 2908 2909 dsl_dir_rele(dd, FTAG); 2910 kmem_free(attr, sizeof (zap_attribute_t)); 2911 dsl_pool_config_exit(dp, FTAG); 2912 2913 if (err != 0) 2914 return (err); 2915 2916 /* Apply to self. */ 2917 return (func(name, arg)); 2918 } 2919 2920 /* 2921 * See comment above dmu_objset_find_impl(). 2922 */ 2923 int 2924 dmu_objset_find(char *name, int func(const char *, void *), void *arg, 2925 int flags) 2926 { 2927 spa_t *spa; 2928 int error; 2929 2930 error = spa_open(name, &spa, FTAG); 2931 if (error != 0) 2932 return (error); 2933 error = dmu_objset_find_impl(spa, name, func, arg, flags); 2934 spa_close(spa, FTAG); 2935 return (error); 2936 } 2937 2938 boolean_t 2939 dmu_objset_incompatible_encryption_version(objset_t *os) 2940 { 2941 return (dsl_dir_incompatible_encryption_version( 2942 os->os_dsl_dataset->ds_dir)); 2943 } 2944 2945 void 2946 dmu_objset_set_user(objset_t *os, void *user_ptr) 2947 { 2948 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 2949 os->os_user_ptr = user_ptr; 2950 } 2951 2952 void * 2953 dmu_objset_get_user(objset_t *os) 2954 { 2955 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 2956 return (os->os_user_ptr); 2957 } 2958 2959 /* 2960 * Determine name of filesystem, given name of snapshot. 2961 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes 2962 */ 2963 int 2964 dmu_fsname(const char *snapname, char *buf) 2965 { 2966 char *atp = strchr(snapname, '@'); 2967 if (atp == NULL) 2968 return (SET_ERROR(EINVAL)); 2969 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) 2970 return (SET_ERROR(ENAMETOOLONG)); 2971 (void) strlcpy(buf, snapname, atp - snapname + 1); 2972 return (0); 2973 } 2974 2975 /* 2976 * Call when we think we're going to write/free space in open context to track 2977 * the amount of dirty data in the open txg, which is also the amount 2978 * of memory that can not be evicted until this txg syncs. 2979 */ 2980 void 2981 dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) 2982 { 2983 dsl_dataset_t *ds = os->os_dsl_dataset; 2984 int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); 2985 2986 if (ds != NULL) { 2987 dsl_dir_willuse_space(ds->ds_dir, aspace, tx); 2988 dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); 2989 } 2990 } 2991