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, 2017 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) 2014 Integros [integros.com] 30 * Copyright 2017 Nexenta Systems, Inc. 31 */ 32 33 /* Portions Copyright 2010 Robert Milkowski */ 34 35 #include <sys/cred.h> 36 #include <sys/zfs_context.h> 37 #include <sys/dmu_objset.h> 38 #include <sys/dsl_dir.h> 39 #include <sys/dsl_dataset.h> 40 #include <sys/dsl_prop.h> 41 #include <sys/dsl_pool.h> 42 #include <sys/dsl_synctask.h> 43 #include <sys/dsl_deleg.h> 44 #include <sys/dnode.h> 45 #include <sys/dbuf.h> 46 #include <sys/zvol.h> 47 #include <sys/dmu_tx.h> 48 #include <sys/zap.h> 49 #include <sys/zil.h> 50 #include <sys/dmu_impl.h> 51 #include <sys/zfs_ioctl.h> 52 #include <sys/sa.h> 53 #include <sys/zfs_onexit.h> 54 #include <sys/dsl_destroy.h> 55 #include <sys/vdev.h> 56 #include <sys/zfeature.h> 57 #include "zfs_namecheck.h" 58 59 /* 60 * Needed to close a window in dnode_move() that allows the objset to be freed 61 * before it can be safely accessed. 62 */ 63 krwlock_t os_lock; 64 65 /* 66 * Tunable to overwrite the maximum number of threads for the parallization 67 * of dmu_objset_find_dp, needed to speed up the import of pools with many 68 * datasets. 69 * Default is 4 times the number of leaf vdevs. 70 */ 71 int dmu_find_threads = 0; 72 73 /* 74 * Backfill lower metadnode objects after this many have been freed. 75 * Backfilling negatively impacts object creation rates, so only do it 76 * if there are enough holes to fill. 77 */ 78 int dmu_rescan_dnode_threshold = 131072; 79 80 static void dmu_objset_find_dp_cb(void *arg); 81 82 void 83 dmu_objset_init(void) 84 { 85 rw_init(&os_lock, NULL, RW_DEFAULT, NULL); 86 } 87 88 void 89 dmu_objset_fini(void) 90 { 91 rw_destroy(&os_lock); 92 } 93 94 spa_t * 95 dmu_objset_spa(objset_t *os) 96 { 97 return (os->os_spa); 98 } 99 100 zilog_t * 101 dmu_objset_zil(objset_t *os) 102 { 103 return (os->os_zil); 104 } 105 106 dsl_pool_t * 107 dmu_objset_pool(objset_t *os) 108 { 109 dsl_dataset_t *ds; 110 111 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) 112 return (ds->ds_dir->dd_pool); 113 else 114 return (spa_get_dsl(os->os_spa)); 115 } 116 117 dsl_dataset_t * 118 dmu_objset_ds(objset_t *os) 119 { 120 return (os->os_dsl_dataset); 121 } 122 123 dmu_objset_type_t 124 dmu_objset_type(objset_t *os) 125 { 126 return (os->os_phys->os_type); 127 } 128 129 void 130 dmu_objset_name(objset_t *os, char *buf) 131 { 132 dsl_dataset_name(os->os_dsl_dataset, buf); 133 } 134 135 uint64_t 136 dmu_objset_id(objset_t *os) 137 { 138 dsl_dataset_t *ds = os->os_dsl_dataset; 139 140 return (ds ? ds->ds_object : 0); 141 } 142 143 uint64_t 144 dmu_objset_dnodesize(objset_t *os) 145 { 146 return (os->os_dnodesize); 147 } 148 149 zfs_sync_type_t 150 dmu_objset_syncprop(objset_t *os) 151 { 152 return (os->os_sync); 153 } 154 155 zfs_logbias_op_t 156 dmu_objset_logbias(objset_t *os) 157 { 158 return (os->os_logbias); 159 } 160 161 static void 162 checksum_changed_cb(void *arg, uint64_t newval) 163 { 164 objset_t *os = arg; 165 166 /* 167 * Inheritance should have been done by now. 168 */ 169 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 170 171 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); 172 } 173 174 static void 175 compression_changed_cb(void *arg, uint64_t newval) 176 { 177 objset_t *os = arg; 178 179 /* 180 * Inheritance and range checking should have been done by now. 181 */ 182 ASSERT(newval != ZIO_COMPRESS_INHERIT); 183 184 os->os_compress = zio_compress_select(os->os_spa, newval, 185 ZIO_COMPRESS_ON); 186 } 187 188 static void 189 copies_changed_cb(void *arg, uint64_t newval) 190 { 191 objset_t *os = arg; 192 193 /* 194 * Inheritance and range checking should have been done by now. 195 */ 196 ASSERT(newval > 0); 197 ASSERT(newval <= spa_max_replication(os->os_spa)); 198 199 os->os_copies = newval; 200 } 201 202 static void 203 dedup_changed_cb(void *arg, uint64_t newval) 204 { 205 objset_t *os = arg; 206 spa_t *spa = os->os_spa; 207 enum zio_checksum checksum; 208 209 /* 210 * Inheritance should have been done by now. 211 */ 212 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 213 214 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); 215 216 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; 217 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); 218 } 219 220 static void 221 primary_cache_changed_cb(void *arg, uint64_t newval) 222 { 223 objset_t *os = arg; 224 225 /* 226 * Inheritance and range checking should have been done by now. 227 */ 228 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 229 newval == ZFS_CACHE_METADATA); 230 231 os->os_primary_cache = newval; 232 } 233 234 static void 235 secondary_cache_changed_cb(void *arg, uint64_t newval) 236 { 237 objset_t *os = arg; 238 239 /* 240 * Inheritance and range checking should have been done by now. 241 */ 242 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 243 newval == ZFS_CACHE_METADATA); 244 245 os->os_secondary_cache = newval; 246 } 247 248 static void 249 sync_changed_cb(void *arg, uint64_t newval) 250 { 251 objset_t *os = arg; 252 253 /* 254 * Inheritance and range checking should have been done by now. 255 */ 256 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || 257 newval == ZFS_SYNC_DISABLED); 258 259 os->os_sync = newval; 260 if (os->os_zil) 261 zil_set_sync(os->os_zil, newval); 262 } 263 264 static void 265 redundant_metadata_changed_cb(void *arg, uint64_t newval) 266 { 267 objset_t *os = arg; 268 269 /* 270 * Inheritance and range checking should have been done by now. 271 */ 272 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || 273 newval == ZFS_REDUNDANT_METADATA_MOST); 274 275 os->os_redundant_metadata = newval; 276 } 277 278 static void 279 dnodesize_changed_cb(void *arg, uint64_t newval) 280 { 281 objset_t *os = arg; 282 283 switch (newval) { 284 case ZFS_DNSIZE_LEGACY: 285 os->os_dnodesize = DNODE_MIN_SIZE; 286 break; 287 case ZFS_DNSIZE_AUTO: 288 /* 289 * Choose a dnode size that will work well for most 290 * workloads if the user specified "auto". Future code 291 * improvements could dynamically select a dnode size 292 * based on observed workload patterns. 293 */ 294 os->os_dnodesize = DNODE_MIN_SIZE * 2; 295 break; 296 case ZFS_DNSIZE_1K: 297 case ZFS_DNSIZE_2K: 298 case ZFS_DNSIZE_4K: 299 case ZFS_DNSIZE_8K: 300 case ZFS_DNSIZE_16K: 301 os->os_dnodesize = newval; 302 break; 303 } 304 } 305 306 static void 307 logbias_changed_cb(void *arg, uint64_t newval) 308 { 309 objset_t *os = arg; 310 311 ASSERT(newval == ZFS_LOGBIAS_LATENCY || 312 newval == ZFS_LOGBIAS_THROUGHPUT); 313 os->os_logbias = newval; 314 if (os->os_zil) 315 zil_set_logbias(os->os_zil, newval); 316 } 317 318 static void 319 recordsize_changed_cb(void *arg, uint64_t newval) 320 { 321 objset_t *os = arg; 322 323 os->os_recordsize = newval; 324 } 325 326 void 327 dmu_objset_byteswap(void *buf, size_t size) 328 { 329 objset_phys_t *osp = buf; 330 331 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t)); 332 dnode_byteswap(&osp->os_meta_dnode); 333 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); 334 osp->os_type = BSWAP_64(osp->os_type); 335 osp->os_flags = BSWAP_64(osp->os_flags); 336 if (size == sizeof (objset_phys_t)) { 337 dnode_byteswap(&osp->os_userused_dnode); 338 dnode_byteswap(&osp->os_groupused_dnode); 339 } 340 } 341 342 /* 343 * The hash is a CRC-based hash of the objset_t pointer and the object number. 344 */ 345 static uint64_t 346 dnode_hash(const objset_t *os, uint64_t obj) 347 { 348 uintptr_t osv = (uintptr_t)os; 349 uint64_t crc = -1ULL; 350 351 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY); 352 /* 353 * The low 6 bits of the pointer don't have much entropy, because 354 * the objset_t is larger than 2^6 bytes long. 355 */ 356 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF]; 357 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF]; 358 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF]; 359 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF]; 360 361 crc ^= (osv>>14) ^ (obj>>24); 362 363 return (crc); 364 } 365 366 unsigned int 367 dnode_multilist_index_func(multilist_t *ml, void *obj) 368 { 369 dnode_t *dn = obj; 370 return (dnode_hash(dn->dn_objset, dn->dn_object) % 371 multilist_get_num_sublists(ml)); 372 } 373 374 /* 375 * Instantiates the objset_t in-memory structure corresponding to the 376 * objset_phys_t that's pointed to by the specified blkptr_t. 377 */ 378 int 379 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 380 objset_t **osp) 381 { 382 objset_t *os; 383 int i, err; 384 385 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); 386 387 /* 388 * The $ORIGIN dataset (if it exists) doesn't have an associated 389 * objset, so there's no reason to open it. The $ORIGIN dataset 390 * will not exist on pools older than SPA_VERSION_ORIGIN. 391 */ 392 if (ds != NULL && spa_get_dsl(spa) != NULL && 393 spa_get_dsl(spa)->dp_origin_snap != NULL) { 394 ASSERT3P(ds->ds_dir, !=, 395 spa_get_dsl(spa)->dp_origin_snap->ds_dir); 396 } 397 398 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); 399 os->os_dsl_dataset = ds; 400 os->os_spa = spa; 401 os->os_rootbp = bp; 402 if (!BP_IS_HOLE(os->os_rootbp)) { 403 arc_flags_t aflags = ARC_FLAG_WAIT; 404 zbookmark_phys_t zb; 405 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, 406 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 407 408 if (DMU_OS_IS_L2CACHEABLE(os)) 409 aflags |= ARC_FLAG_L2CACHE; 410 411 dprintf_bp(os->os_rootbp, "reading %s", ""); 412 err = arc_read(NULL, spa, os->os_rootbp, 413 arc_getbuf_func, &os->os_phys_buf, 414 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb); 415 if (err != 0) { 416 kmem_free(os, sizeof (objset_t)); 417 /* convert checksum errors into IO errors */ 418 if (err == ECKSUM) 419 err = SET_ERROR(EIO); 420 return (err); 421 } 422 423 /* Increase the blocksize if we are permitted. */ 424 if (spa_version(spa) >= SPA_VERSION_USERSPACE && 425 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) { 426 arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf, 427 ARC_BUFC_METADATA, sizeof (objset_phys_t)); 428 bzero(buf->b_data, sizeof (objset_phys_t)); 429 bcopy(os->os_phys_buf->b_data, buf->b_data, 430 arc_buf_size(os->os_phys_buf)); 431 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 432 os->os_phys_buf = buf; 433 } 434 435 os->os_phys = os->os_phys_buf->b_data; 436 os->os_flags = os->os_phys->os_flags; 437 } else { 438 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? 439 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE; 440 os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf, 441 ARC_BUFC_METADATA, size); 442 os->os_phys = os->os_phys_buf->b_data; 443 bzero(os->os_phys, size); 444 } 445 446 /* 447 * Note: the changed_cb will be called once before the register 448 * func returns, thus changing the checksum/compression from the 449 * default (fletcher2/off). Snapshots don't need to know about 450 * checksum/compression/copies. 451 */ 452 if (ds != NULL) { 453 boolean_t needlock = B_FALSE; 454 455 /* 456 * Note: it's valid to open the objset if the dataset is 457 * long-held, in which case the pool_config lock will not 458 * be held. 459 */ 460 if (!dsl_pool_config_held(dmu_objset_pool(os))) { 461 needlock = B_TRUE; 462 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 463 } 464 err = dsl_prop_register(ds, 465 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), 466 primary_cache_changed_cb, os); 467 if (err == 0) { 468 err = dsl_prop_register(ds, 469 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), 470 secondary_cache_changed_cb, os); 471 } 472 if (!ds->ds_is_snapshot) { 473 if (err == 0) { 474 err = dsl_prop_register(ds, 475 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 476 checksum_changed_cb, os); 477 } 478 if (err == 0) { 479 err = dsl_prop_register(ds, 480 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 481 compression_changed_cb, os); 482 } 483 if (err == 0) { 484 err = dsl_prop_register(ds, 485 zfs_prop_to_name(ZFS_PROP_COPIES), 486 copies_changed_cb, os); 487 } 488 if (err == 0) { 489 err = dsl_prop_register(ds, 490 zfs_prop_to_name(ZFS_PROP_DEDUP), 491 dedup_changed_cb, os); 492 } 493 if (err == 0) { 494 err = dsl_prop_register(ds, 495 zfs_prop_to_name(ZFS_PROP_LOGBIAS), 496 logbias_changed_cb, os); 497 } 498 if (err == 0) { 499 err = dsl_prop_register(ds, 500 zfs_prop_to_name(ZFS_PROP_SYNC), 501 sync_changed_cb, os); 502 } 503 if (err == 0) { 504 err = dsl_prop_register(ds, 505 zfs_prop_to_name( 506 ZFS_PROP_REDUNDANT_METADATA), 507 redundant_metadata_changed_cb, os); 508 } 509 if (err == 0) { 510 err = dsl_prop_register(ds, 511 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 512 recordsize_changed_cb, os); 513 } 514 if (err == 0) { 515 err = dsl_prop_register(ds, 516 zfs_prop_to_name(ZFS_PROP_DNODESIZE), 517 dnodesize_changed_cb, os); 518 } 519 } 520 if (needlock) 521 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 522 if (err != 0) { 523 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 524 kmem_free(os, sizeof (objset_t)); 525 return (err); 526 } 527 } else { 528 /* It's the meta-objset. */ 529 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; 530 os->os_compress = ZIO_COMPRESS_ON; 531 os->os_copies = spa_max_replication(spa); 532 os->os_dedup_checksum = ZIO_CHECKSUM_OFF; 533 os->os_dedup_verify = B_FALSE; 534 os->os_logbias = ZFS_LOGBIAS_LATENCY; 535 os->os_sync = ZFS_SYNC_STANDARD; 536 os->os_primary_cache = ZFS_CACHE_ALL; 537 os->os_secondary_cache = ZFS_CACHE_ALL; 538 os->os_dnodesize = DNODE_MIN_SIZE; 539 } 540 /* 541 * These properties will be filled in by the logic in zfs_get_zplprop() 542 * when they are queried for the first time. 543 */ 544 os->os_version = OBJSET_PROP_UNINITIALIZED; 545 os->os_normalization = OBJSET_PROP_UNINITIALIZED; 546 os->os_utf8only = OBJSET_PROP_UNINITIALIZED; 547 os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED; 548 549 if (ds == NULL || !ds->ds_is_snapshot) 550 os->os_zil_header = os->os_phys->os_zil_header; 551 os->os_zil = zil_alloc(os, &os->os_zil_header); 552 553 for (i = 0; i < TXG_SIZE; i++) { 554 os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t), 555 offsetof(dnode_t, dn_dirty_link[i]), 556 dnode_multilist_index_func); 557 } 558 list_create(&os->os_dnodes, sizeof (dnode_t), 559 offsetof(dnode_t, dn_link)); 560 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), 561 offsetof(dmu_buf_impl_t, db_link)); 562 563 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); 564 mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL); 565 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); 566 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); 567 os->os_obj_next_percpu_len = boot_ncpus; 568 os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len * 569 sizeof (os->os_obj_next_percpu[0]), KM_SLEEP); 570 571 dnode_special_open(os, &os->os_phys->os_meta_dnode, 572 DMU_META_DNODE_OBJECT, &os->os_meta_dnode); 573 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) { 574 dnode_special_open(os, &os->os_phys->os_userused_dnode, 575 DMU_USERUSED_OBJECT, &os->os_userused_dnode); 576 dnode_special_open(os, &os->os_phys->os_groupused_dnode, 577 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); 578 } 579 580 *osp = os; 581 return (0); 582 } 583 584 int 585 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) 586 { 587 int err = 0; 588 589 /* 590 * We shouldn't be doing anything with dsl_dataset_t's unless the 591 * pool_config lock is held, or the dataset is long-held. 592 */ 593 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) || 594 dsl_dataset_long_held(ds)); 595 596 mutex_enter(&ds->ds_opening_lock); 597 if (ds->ds_objset == NULL) { 598 objset_t *os; 599 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 600 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), 601 ds, dsl_dataset_get_blkptr(ds), &os); 602 rrw_exit(&ds->ds_bp_rwlock, FTAG); 603 604 if (err == 0) { 605 mutex_enter(&ds->ds_lock); 606 ASSERT(ds->ds_objset == NULL); 607 ds->ds_objset = os; 608 mutex_exit(&ds->ds_lock); 609 } 610 } 611 *osp = ds->ds_objset; 612 mutex_exit(&ds->ds_opening_lock); 613 return (err); 614 } 615 616 /* 617 * Holds the pool while the objset is held. Therefore only one objset 618 * can be held at a time. 619 */ 620 int 621 dmu_objset_hold(const char *name, void *tag, objset_t **osp) 622 { 623 dsl_pool_t *dp; 624 dsl_dataset_t *ds; 625 int err; 626 627 err = dsl_pool_hold(name, tag, &dp); 628 if (err != 0) 629 return (err); 630 err = dsl_dataset_hold(dp, name, tag, &ds); 631 if (err != 0) { 632 dsl_pool_rele(dp, tag); 633 return (err); 634 } 635 636 err = dmu_objset_from_ds(ds, osp); 637 if (err != 0) { 638 dsl_dataset_rele(ds, tag); 639 dsl_pool_rele(dp, tag); 640 } 641 642 return (err); 643 } 644 645 static int 646 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, 647 boolean_t readonly, void *tag, objset_t **osp) 648 { 649 int err; 650 651 err = dmu_objset_from_ds(ds, osp); 652 if (err != 0) { 653 dsl_dataset_disown(ds, tag); 654 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { 655 dsl_dataset_disown(ds, tag); 656 return (SET_ERROR(EINVAL)); 657 } else if (!readonly && dsl_dataset_is_snapshot(ds)) { 658 dsl_dataset_disown(ds, tag); 659 return (SET_ERROR(EROFS)); 660 } 661 return (err); 662 } 663 664 /* 665 * dsl_pool must not be held when this is called. 666 * Upon successful return, there will be a longhold on the dataset, 667 * and the dsl_pool will not be held. 668 */ 669 int 670 dmu_objset_own(const char *name, dmu_objset_type_t type, 671 boolean_t readonly, void *tag, objset_t **osp) 672 { 673 dsl_pool_t *dp; 674 dsl_dataset_t *ds; 675 int err; 676 677 err = dsl_pool_hold(name, FTAG, &dp); 678 if (err != 0) 679 return (err); 680 err = dsl_dataset_own(dp, name, tag, &ds); 681 if (err != 0) { 682 dsl_pool_rele(dp, FTAG); 683 return (err); 684 } 685 err = dmu_objset_own_impl(ds, type, readonly, tag, osp); 686 dsl_pool_rele(dp, FTAG); 687 688 return (err); 689 } 690 691 int 692 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, 693 boolean_t readonly, void *tag, objset_t **osp) 694 { 695 dsl_dataset_t *ds; 696 int err; 697 698 err = dsl_dataset_own_obj(dp, obj, tag, &ds); 699 if (err != 0) 700 return (err); 701 702 return (dmu_objset_own_impl(ds, type, readonly, tag, osp)); 703 } 704 705 void 706 dmu_objset_rele(objset_t *os, void *tag) 707 { 708 dsl_pool_t *dp = dmu_objset_pool(os); 709 dsl_dataset_rele(os->os_dsl_dataset, tag); 710 dsl_pool_rele(dp, tag); 711 } 712 713 /* 714 * When we are called, os MUST refer to an objset associated with a dataset 715 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner 716 * == tag. We will then release and reacquire ownership of the dataset while 717 * holding the pool config_rwlock to avoid intervening namespace or ownership 718 * changes may occur. 719 * 720 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to 721 * release the hold on its dataset and acquire a new one on the dataset of the 722 * same name so that it can be partially torn down and reconstructed. 723 */ 724 void 725 dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds, 726 void *tag) 727 { 728 dsl_pool_t *dp; 729 char name[ZFS_MAX_DATASET_NAME_LEN]; 730 731 VERIFY3P(ds, !=, NULL); 732 VERIFY3P(ds->ds_owner, ==, tag); 733 VERIFY(dsl_dataset_long_held(ds)); 734 735 dsl_dataset_name(ds, name); 736 dp = ds->ds_dir->dd_pool; 737 dsl_pool_config_enter(dp, FTAG); 738 dsl_dataset_disown(ds, tag); 739 VERIFY0(dsl_dataset_own(dp, name, tag, newds)); 740 dsl_pool_config_exit(dp, FTAG); 741 } 742 743 void 744 dmu_objset_disown(objset_t *os, void *tag) 745 { 746 dsl_dataset_disown(os->os_dsl_dataset, tag); 747 } 748 749 void 750 dmu_objset_evict_dbufs(objset_t *os) 751 { 752 dnode_t dn_marker; 753 dnode_t *dn; 754 755 mutex_enter(&os->os_lock); 756 dn = list_head(&os->os_dnodes); 757 while (dn != NULL) { 758 /* 759 * Skip dnodes without holds. We have to do this dance 760 * because dnode_add_ref() only works if there is already a 761 * hold. If the dnode has no holds, then it has no dbufs. 762 */ 763 if (dnode_add_ref(dn, FTAG)) { 764 list_insert_after(&os->os_dnodes, dn, &dn_marker); 765 mutex_exit(&os->os_lock); 766 767 dnode_evict_dbufs(dn); 768 dnode_rele(dn, FTAG); 769 770 mutex_enter(&os->os_lock); 771 dn = list_next(&os->os_dnodes, &dn_marker); 772 list_remove(&os->os_dnodes, &dn_marker); 773 } else { 774 dn = list_next(&os->os_dnodes, dn); 775 } 776 } 777 mutex_exit(&os->os_lock); 778 779 if (DMU_USERUSED_DNODE(os) != NULL) { 780 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); 781 dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); 782 } 783 dnode_evict_dbufs(DMU_META_DNODE(os)); 784 } 785 786 /* 787 * Objset eviction processing is split into into two pieces. 788 * The first marks the objset as evicting, evicts any dbufs that 789 * have a refcount of zero, and then queues up the objset for the 790 * second phase of eviction. Once os->os_dnodes has been cleared by 791 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. 792 * The second phase closes the special dnodes, dequeues the objset from 793 * the list of those undergoing eviction, and finally frees the objset. 794 * 795 * NOTE: Due to asynchronous eviction processing (invocation of 796 * dnode_buf_pageout()), it is possible for the meta dnode for the 797 * objset to have no holds even though os->os_dnodes is not empty. 798 */ 799 void 800 dmu_objset_evict(objset_t *os) 801 { 802 dsl_dataset_t *ds = os->os_dsl_dataset; 803 804 for (int t = 0; t < TXG_SIZE; t++) 805 ASSERT(!dmu_objset_is_dirty(os, t)); 806 807 if (ds) 808 dsl_prop_unregister_all(ds, os); 809 810 if (os->os_sa) 811 sa_tear_down(os); 812 813 dmu_objset_evict_dbufs(os); 814 815 mutex_enter(&os->os_lock); 816 spa_evicting_os_register(os->os_spa, os); 817 if (list_is_empty(&os->os_dnodes)) { 818 mutex_exit(&os->os_lock); 819 dmu_objset_evict_done(os); 820 } else { 821 mutex_exit(&os->os_lock); 822 } 823 } 824 825 void 826 dmu_objset_evict_done(objset_t *os) 827 { 828 ASSERT3P(list_head(&os->os_dnodes), ==, NULL); 829 830 dnode_special_close(&os->os_meta_dnode); 831 if (DMU_USERUSED_DNODE(os)) { 832 dnode_special_close(&os->os_userused_dnode); 833 dnode_special_close(&os->os_groupused_dnode); 834 } 835 zil_free(os->os_zil); 836 837 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf); 838 839 /* 840 * This is a barrier to prevent the objset from going away in 841 * dnode_move() until we can safely ensure that the objset is still in 842 * use. We consider the objset valid before the barrier and invalid 843 * after the barrier. 844 */ 845 rw_enter(&os_lock, RW_READER); 846 rw_exit(&os_lock); 847 848 kmem_free(os->os_obj_next_percpu, 849 os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0])); 850 851 mutex_destroy(&os->os_lock); 852 mutex_destroy(&os->os_userused_lock); 853 mutex_destroy(&os->os_obj_lock); 854 mutex_destroy(&os->os_user_ptr_lock); 855 for (int i = 0; i < TXG_SIZE; i++) { 856 multilist_destroy(os->os_dirty_dnodes[i]); 857 } 858 spa_evicting_os_deregister(os->os_spa, os); 859 kmem_free(os, sizeof (objset_t)); 860 } 861 862 timestruc_t 863 dmu_objset_snap_cmtime(objset_t *os) 864 { 865 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); 866 } 867 868 /* called from dsl for meta-objset */ 869 objset_t * 870 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 871 dmu_objset_type_t type, dmu_tx_t *tx) 872 { 873 objset_t *os; 874 dnode_t *mdn; 875 876 ASSERT(dmu_tx_is_syncing(tx)); 877 878 if (ds != NULL) 879 VERIFY0(dmu_objset_from_ds(ds, &os)); 880 else 881 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); 882 883 mdn = DMU_META_DNODE(os); 884 885 dnode_allocate(mdn, DMU_OT_DNODE, DNODE_BLOCK_SIZE, DN_MAX_INDBLKSHIFT, 886 DMU_OT_NONE, 0, DNODE_MIN_SLOTS, tx); 887 888 /* 889 * We don't want to have to increase the meta-dnode's nlevels 890 * later, because then we could do it in quescing context while 891 * we are also accessing it in open context. 892 * 893 * This precaution is not necessary for the MOS (ds == NULL), 894 * because the MOS is only updated in syncing context. 895 * This is most fortunate: the MOS is the only objset that 896 * needs to be synced multiple times as spa_sync() iterates 897 * to convergence, so minimizing its dn_nlevels matters. 898 */ 899 if (ds != NULL) { 900 int levels = 1; 901 902 /* 903 * Determine the number of levels necessary for the meta-dnode 904 * to contain DN_MAX_OBJECT dnodes. Note that in order to 905 * ensure that we do not overflow 64 bits, there has to be 906 * a nlevels that gives us a number of blocks > DN_MAX_OBJECT 907 * but < 2^64. Therefore, 908 * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be 909 * less than (64 - log2(DN_MAX_OBJECT)) (16). 910 */ 911 while ((uint64_t)mdn->dn_nblkptr << 912 (mdn->dn_datablkshift - DNODE_SHIFT + 913 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < 914 DN_MAX_OBJECT) 915 levels++; 916 917 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = 918 mdn->dn_nlevels = levels; 919 } 920 921 ASSERT(type != DMU_OST_NONE); 922 ASSERT(type != DMU_OST_ANY); 923 ASSERT(type < DMU_OST_NUMTYPES); 924 os->os_phys->os_type = type; 925 if (dmu_objset_userused_enabled(os)) { 926 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 927 os->os_flags = os->os_phys->os_flags; 928 } 929 930 dsl_dataset_dirty(ds, tx); 931 932 return (os); 933 } 934 935 typedef struct dmu_objset_create_arg { 936 const char *doca_name; 937 cred_t *doca_cred; 938 void (*doca_userfunc)(objset_t *os, void *arg, 939 cred_t *cr, dmu_tx_t *tx); 940 void *doca_userarg; 941 dmu_objset_type_t doca_type; 942 uint64_t doca_flags; 943 } dmu_objset_create_arg_t; 944 945 /*ARGSUSED*/ 946 static int 947 dmu_objset_create_check(void *arg, dmu_tx_t *tx) 948 { 949 dmu_objset_create_arg_t *doca = arg; 950 dsl_pool_t *dp = dmu_tx_pool(tx); 951 dsl_dir_t *pdd; 952 const char *tail; 953 int error; 954 955 if (strchr(doca->doca_name, '@') != NULL) 956 return (SET_ERROR(EINVAL)); 957 958 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) 959 return (SET_ERROR(ENAMETOOLONG)); 960 961 if (dataset_nestcheck(doca->doca_name) != 0) 962 return (SET_ERROR(ENAMETOOLONG)); 963 964 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); 965 if (error != 0) 966 return (error); 967 if (tail == NULL) { 968 dsl_dir_rele(pdd, FTAG); 969 return (SET_ERROR(EEXIST)); 970 } 971 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 972 doca->doca_cred); 973 dsl_dir_rele(pdd, FTAG); 974 975 return (error); 976 } 977 978 static void 979 dmu_objset_create_sync(void *arg, dmu_tx_t *tx) 980 { 981 dmu_objset_create_arg_t *doca = arg; 982 dsl_pool_t *dp = dmu_tx_pool(tx); 983 dsl_dir_t *pdd; 984 const char *tail; 985 dsl_dataset_t *ds; 986 uint64_t obj; 987 blkptr_t *bp; 988 objset_t *os; 989 990 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); 991 992 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, 993 doca->doca_cred, tx); 994 995 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 996 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 997 bp = dsl_dataset_get_blkptr(ds); 998 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa, 999 ds, bp, doca->doca_type, tx); 1000 rrw_exit(&ds->ds_bp_rwlock, FTAG); 1001 1002 if (doca->doca_userfunc != NULL) { 1003 doca->doca_userfunc(os, doca->doca_userarg, 1004 doca->doca_cred, tx); 1005 } 1006 1007 spa_history_log_internal_ds(ds, "create", tx, ""); 1008 dsl_dataset_rele(ds, FTAG); 1009 dsl_dir_rele(pdd, FTAG); 1010 } 1011 1012 int 1013 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 1014 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg) 1015 { 1016 dmu_objset_create_arg_t doca; 1017 1018 doca.doca_name = name; 1019 doca.doca_cred = CRED(); 1020 doca.doca_flags = flags; 1021 doca.doca_userfunc = func; 1022 doca.doca_userarg = arg; 1023 doca.doca_type = type; 1024 1025 return (dsl_sync_task(name, 1026 dmu_objset_create_check, dmu_objset_create_sync, &doca, 1027 5, ZFS_SPACE_CHECK_NORMAL)); 1028 } 1029 1030 typedef struct dmu_objset_clone_arg { 1031 const char *doca_clone; 1032 const char *doca_origin; 1033 cred_t *doca_cred; 1034 } dmu_objset_clone_arg_t; 1035 1036 /*ARGSUSED*/ 1037 static int 1038 dmu_objset_clone_check(void *arg, dmu_tx_t *tx) 1039 { 1040 dmu_objset_clone_arg_t *doca = arg; 1041 dsl_dir_t *pdd; 1042 const char *tail; 1043 int error; 1044 dsl_dataset_t *origin; 1045 dsl_pool_t *dp = dmu_tx_pool(tx); 1046 1047 if (strchr(doca->doca_clone, '@') != NULL) 1048 return (SET_ERROR(EINVAL)); 1049 1050 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) 1051 return (SET_ERROR(ENAMETOOLONG)); 1052 1053 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); 1054 if (error != 0) 1055 return (error); 1056 if (tail == NULL) { 1057 dsl_dir_rele(pdd, FTAG); 1058 return (SET_ERROR(EEXIST)); 1059 } 1060 1061 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 1062 doca->doca_cred); 1063 if (error != 0) { 1064 dsl_dir_rele(pdd, FTAG); 1065 return (SET_ERROR(EDQUOT)); 1066 } 1067 dsl_dir_rele(pdd, FTAG); 1068 1069 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); 1070 if (error != 0) 1071 return (error); 1072 1073 /* You can only clone snapshots, not the head datasets. */ 1074 if (!origin->ds_is_snapshot) { 1075 dsl_dataset_rele(origin, FTAG); 1076 return (SET_ERROR(EINVAL)); 1077 } 1078 dsl_dataset_rele(origin, FTAG); 1079 1080 return (0); 1081 } 1082 1083 static void 1084 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) 1085 { 1086 dmu_objset_clone_arg_t *doca = arg; 1087 dsl_pool_t *dp = dmu_tx_pool(tx); 1088 dsl_dir_t *pdd; 1089 const char *tail; 1090 dsl_dataset_t *origin, *ds; 1091 uint64_t obj; 1092 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 1093 1094 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); 1095 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); 1096 1097 obj = dsl_dataset_create_sync(pdd, tail, origin, 0, 1098 doca->doca_cred, tx); 1099 1100 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 1101 dsl_dataset_name(origin, namebuf); 1102 spa_history_log_internal_ds(ds, "clone", tx, 1103 "origin=%s (%llu)", namebuf, origin->ds_object); 1104 dsl_dataset_rele(ds, FTAG); 1105 dsl_dataset_rele(origin, FTAG); 1106 dsl_dir_rele(pdd, FTAG); 1107 } 1108 1109 int 1110 dmu_objset_clone(const char *clone, const char *origin) 1111 { 1112 dmu_objset_clone_arg_t doca; 1113 1114 doca.doca_clone = clone; 1115 doca.doca_origin = origin; 1116 doca.doca_cred = CRED(); 1117 1118 return (dsl_sync_task(clone, 1119 dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 1120 5, ZFS_SPACE_CHECK_NORMAL)); 1121 } 1122 1123 static int 1124 dmu_objset_remap_indirects_impl(objset_t *os, uint64_t last_removed_txg) 1125 { 1126 int error = 0; 1127 uint64_t object = 0; 1128 while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) { 1129 error = dmu_object_remap_indirects(os, object, 1130 last_removed_txg); 1131 /* 1132 * If the ZPL removed the object before we managed to dnode_hold 1133 * it, we would get an ENOENT. If the ZPL declares its intent 1134 * to remove the object (dnode_free) before we manage to 1135 * dnode_hold it, we would get an EEXIST. In either case, we 1136 * want to continue remapping the other objects in the objset; 1137 * in all other cases, we want to break early. 1138 */ 1139 if (error != 0 && error != ENOENT && error != EEXIST) { 1140 break; 1141 } 1142 } 1143 if (error == ESRCH) { 1144 error = 0; 1145 } 1146 return (error); 1147 } 1148 1149 int 1150 dmu_objset_remap_indirects(const char *fsname) 1151 { 1152 int error = 0; 1153 objset_t *os = NULL; 1154 uint64_t last_removed_txg; 1155 uint64_t remap_start_txg; 1156 dsl_dir_t *dd; 1157 1158 error = dmu_objset_hold(fsname, FTAG, &os); 1159 if (error != 0) { 1160 return (error); 1161 } 1162 dd = dmu_objset_ds(os)->ds_dir; 1163 1164 if (!spa_feature_is_enabled(dmu_objset_spa(os), 1165 SPA_FEATURE_OBSOLETE_COUNTS)) { 1166 dmu_objset_rele(os, FTAG); 1167 return (SET_ERROR(ENOTSUP)); 1168 } 1169 1170 if (dsl_dataset_is_snapshot(dmu_objset_ds(os))) { 1171 dmu_objset_rele(os, FTAG); 1172 return (SET_ERROR(EINVAL)); 1173 } 1174 1175 /* 1176 * If there has not been a removal, we're done. 1177 */ 1178 last_removed_txg = spa_get_last_removal_txg(dmu_objset_spa(os)); 1179 if (last_removed_txg == -1ULL) { 1180 dmu_objset_rele(os, FTAG); 1181 return (0); 1182 } 1183 1184 /* 1185 * If we have remapped since the last removal, we're done. 1186 */ 1187 if (dsl_dir_is_zapified(dd)) { 1188 uint64_t last_remap_txg; 1189 if (zap_lookup(spa_meta_objset(dmu_objset_spa(os)), 1190 dd->dd_object, DD_FIELD_LAST_REMAP_TXG, 1191 sizeof (last_remap_txg), 1, &last_remap_txg) == 0 && 1192 last_remap_txg > last_removed_txg) { 1193 dmu_objset_rele(os, FTAG); 1194 return (0); 1195 } 1196 } 1197 1198 dsl_dataset_long_hold(dmu_objset_ds(os), FTAG); 1199 dsl_pool_rele(dmu_objset_pool(os), FTAG); 1200 1201 remap_start_txg = spa_last_synced_txg(dmu_objset_spa(os)); 1202 error = dmu_objset_remap_indirects_impl(os, last_removed_txg); 1203 if (error == 0) { 1204 /* 1205 * We update the last_remap_txg to be the start txg so that 1206 * we can guarantee that every block older than last_remap_txg 1207 * that can be remapped has been remapped. 1208 */ 1209 error = dsl_dir_update_last_remap_txg(dd, remap_start_txg); 1210 } 1211 1212 dsl_dataset_long_rele(dmu_objset_ds(os), FTAG); 1213 dsl_dataset_rele(dmu_objset_ds(os), FTAG); 1214 1215 return (error); 1216 } 1217 1218 int 1219 dmu_objset_snapshot_one(const char *fsname, const char *snapname) 1220 { 1221 int err; 1222 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); 1223 nvlist_t *snaps = fnvlist_alloc(); 1224 1225 fnvlist_add_boolean(snaps, longsnap); 1226 strfree(longsnap); 1227 err = dsl_dataset_snapshot(snaps, NULL, NULL); 1228 fnvlist_free(snaps); 1229 return (err); 1230 } 1231 1232 static void 1233 dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx) 1234 { 1235 dnode_t *dn; 1236 1237 while ((dn = multilist_sublist_head(list)) != NULL) { 1238 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 1239 ASSERT(dn->dn_dbuf->db_data_pending); 1240 /* 1241 * Initialize dn_zio outside dnode_sync() because the 1242 * meta-dnode needs to set it ouside dnode_sync(). 1243 */ 1244 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; 1245 ASSERT(dn->dn_zio); 1246 1247 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); 1248 multilist_sublist_remove(list, dn); 1249 1250 /* 1251 * If we are not doing useraccounting (os_synced_dnodes == NULL) 1252 * we are done with this dnode for this txg. Unset dn_dirty_txg 1253 * if later txgs aren't dirtying it so that future holders do 1254 * not get a stale value. Otherwise, we will do this in 1255 * userquota_updates_task() when processing has completely 1256 * finished for this txg. 1257 */ 1258 multilist_t *newlist = dn->dn_objset->os_synced_dnodes; 1259 if (newlist != NULL) { 1260 (void) dnode_add_ref(dn, newlist); 1261 multilist_insert(newlist, dn); 1262 } else { 1263 mutex_enter(&dn->dn_mtx); 1264 if (dn->dn_dirty_txg == tx->tx_txg) 1265 dn->dn_dirty_txg = 0; 1266 mutex_exit(&dn->dn_mtx); 1267 } 1268 1269 dnode_sync(dn, tx); 1270 } 1271 } 1272 1273 /* ARGSUSED */ 1274 static void 1275 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) 1276 { 1277 blkptr_t *bp = zio->io_bp; 1278 objset_t *os = arg; 1279 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; 1280 1281 ASSERT(!BP_IS_EMBEDDED(bp)); 1282 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); 1283 ASSERT0(BP_GET_LEVEL(bp)); 1284 1285 /* 1286 * Update rootbp fill count: it should be the number of objects 1287 * allocated in the object set (not counting the "special" 1288 * objects that are stored in the objset_phys_t -- the meta 1289 * dnode and user/group accounting objects). 1290 */ 1291 bp->blk_fill = 0; 1292 for (int i = 0; i < dnp->dn_nblkptr; i++) 1293 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]); 1294 if (os->os_dsl_dataset != NULL) 1295 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG); 1296 *os->os_rootbp = *bp; 1297 if (os->os_dsl_dataset != NULL) 1298 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG); 1299 } 1300 1301 /* ARGSUSED */ 1302 static void 1303 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) 1304 { 1305 blkptr_t *bp = zio->io_bp; 1306 blkptr_t *bp_orig = &zio->io_bp_orig; 1307 objset_t *os = arg; 1308 1309 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { 1310 ASSERT(BP_EQUAL(bp, bp_orig)); 1311 } else { 1312 dsl_dataset_t *ds = os->os_dsl_dataset; 1313 dmu_tx_t *tx = os->os_synctx; 1314 1315 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); 1316 dsl_dataset_block_born(ds, bp, tx); 1317 } 1318 kmem_free(bp, sizeof (*bp)); 1319 } 1320 1321 typedef struct sync_dnodes_arg { 1322 multilist_t *sda_list; 1323 int sda_sublist_idx; 1324 multilist_t *sda_newlist; 1325 dmu_tx_t *sda_tx; 1326 } sync_dnodes_arg_t; 1327 1328 static void 1329 sync_dnodes_task(void *arg) 1330 { 1331 sync_dnodes_arg_t *sda = arg; 1332 1333 multilist_sublist_t *ms = 1334 multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx); 1335 1336 dmu_objset_sync_dnodes(ms, sda->sda_tx); 1337 1338 multilist_sublist_unlock(ms); 1339 1340 kmem_free(sda, sizeof (*sda)); 1341 } 1342 1343 1344 /* called from dsl */ 1345 void 1346 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) 1347 { 1348 int txgoff; 1349 zbookmark_phys_t zb; 1350 zio_prop_t zp; 1351 zio_t *zio; 1352 list_t *list; 1353 dbuf_dirty_record_t *dr; 1354 blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP); 1355 *blkptr_copy = *os->os_rootbp; 1356 1357 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); 1358 1359 ASSERT(dmu_tx_is_syncing(tx)); 1360 /* XXX the write_done callback should really give us the tx... */ 1361 os->os_synctx = tx; 1362 1363 if (os->os_dsl_dataset == NULL) { 1364 /* 1365 * This is the MOS. If we have upgraded, 1366 * spa_max_replication() could change, so reset 1367 * os_copies here. 1368 */ 1369 os->os_copies = spa_max_replication(os->os_spa); 1370 } 1371 1372 /* 1373 * Create the root block IO 1374 */ 1375 SET_BOOKMARK(&zb, os->os_dsl_dataset ? 1376 os->os_dsl_dataset->ds_object : DMU_META_OBJSET, 1377 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 1378 arc_release(os->os_phys_buf, &os->os_phys_buf); 1379 1380 dmu_write_policy(os, NULL, 0, 0, &zp); 1381 1382 zio = arc_write(pio, os->os_spa, tx->tx_txg, 1383 blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), 1384 &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, 1385 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); 1386 1387 /* 1388 * Sync special dnodes - the parent IO for the sync is the root block 1389 */ 1390 DMU_META_DNODE(os)->dn_zio = zio; 1391 dnode_sync(DMU_META_DNODE(os), tx); 1392 1393 os->os_phys->os_flags = os->os_flags; 1394 1395 if (DMU_USERUSED_DNODE(os) && 1396 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1397 DMU_USERUSED_DNODE(os)->dn_zio = zio; 1398 dnode_sync(DMU_USERUSED_DNODE(os), tx); 1399 DMU_GROUPUSED_DNODE(os)->dn_zio = zio; 1400 dnode_sync(DMU_GROUPUSED_DNODE(os), tx); 1401 } 1402 1403 txgoff = tx->tx_txg & TXG_MASK; 1404 1405 if (dmu_objset_userused_enabled(os)) { 1406 /* 1407 * We must create the list here because it uses the 1408 * dn_dirty_link[] of this txg. But it may already 1409 * exist because we call dsl_dataset_sync() twice per txg. 1410 */ 1411 if (os->os_synced_dnodes == NULL) { 1412 os->os_synced_dnodes = 1413 multilist_create(sizeof (dnode_t), 1414 offsetof(dnode_t, dn_dirty_link[txgoff]), 1415 dnode_multilist_index_func); 1416 } else { 1417 ASSERT3U(os->os_synced_dnodes->ml_offset, ==, 1418 offsetof(dnode_t, dn_dirty_link[txgoff])); 1419 } 1420 } 1421 1422 for (int i = 0; 1423 i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) { 1424 sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP); 1425 sda->sda_list = os->os_dirty_dnodes[txgoff]; 1426 sda->sda_sublist_idx = i; 1427 sda->sda_tx = tx; 1428 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, 1429 sync_dnodes_task, sda, 0); 1430 /* callback frees sda */ 1431 } 1432 taskq_wait(dmu_objset_pool(os)->dp_sync_taskq); 1433 1434 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; 1435 while ((dr = list_head(list)) != NULL) { 1436 ASSERT0(dr->dr_dbuf->db_level); 1437 list_remove(list, dr); 1438 if (dr->dr_zio) 1439 zio_nowait(dr->dr_zio); 1440 } 1441 1442 /* Enable dnode backfill if enough objects have been freed. */ 1443 if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) { 1444 os->os_rescan_dnodes = B_TRUE; 1445 os->os_freed_dnodes = 0; 1446 } 1447 1448 /* 1449 * Free intent log blocks up to this tx. 1450 */ 1451 zil_sync(os->os_zil, tx); 1452 os->os_phys->os_zil_header = os->os_zil_header; 1453 zio_nowait(zio); 1454 } 1455 1456 boolean_t 1457 dmu_objset_is_dirty(objset_t *os, uint64_t txg) 1458 { 1459 return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK])); 1460 } 1461 1462 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES]; 1463 1464 void 1465 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb) 1466 { 1467 used_cbs[ost] = cb; 1468 } 1469 1470 boolean_t 1471 dmu_objset_userused_enabled(objset_t *os) 1472 { 1473 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && 1474 used_cbs[os->os_phys->os_type] != NULL && 1475 DMU_USERUSED_DNODE(os) != NULL); 1476 } 1477 1478 typedef struct userquota_node { 1479 uint64_t uqn_id; 1480 int64_t uqn_delta; 1481 avl_node_t uqn_node; 1482 } userquota_node_t; 1483 1484 typedef struct userquota_cache { 1485 avl_tree_t uqc_user_deltas; 1486 avl_tree_t uqc_group_deltas; 1487 } userquota_cache_t; 1488 1489 static int 1490 userquota_compare(const void *l, const void *r) 1491 { 1492 const userquota_node_t *luqn = l; 1493 const userquota_node_t *ruqn = r; 1494 1495 if (luqn->uqn_id < ruqn->uqn_id) 1496 return (-1); 1497 if (luqn->uqn_id > ruqn->uqn_id) 1498 return (1); 1499 return (0); 1500 } 1501 1502 static void 1503 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx) 1504 { 1505 void *cookie; 1506 userquota_node_t *uqn; 1507 1508 ASSERT(dmu_tx_is_syncing(tx)); 1509 1510 cookie = NULL; 1511 while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas, 1512 &cookie)) != NULL) { 1513 /* 1514 * os_userused_lock protects against concurrent calls to 1515 * zap_increment_int(). It's needed because zap_increment_int() 1516 * is not thread-safe (i.e. not atomic). 1517 */ 1518 mutex_enter(&os->os_userused_lock); 1519 VERIFY0(zap_increment_int(os, DMU_USERUSED_OBJECT, 1520 uqn->uqn_id, uqn->uqn_delta, tx)); 1521 mutex_exit(&os->os_userused_lock); 1522 kmem_free(uqn, sizeof (*uqn)); 1523 } 1524 avl_destroy(&cache->uqc_user_deltas); 1525 1526 cookie = NULL; 1527 while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas, 1528 &cookie)) != NULL) { 1529 mutex_enter(&os->os_userused_lock); 1530 VERIFY0(zap_increment_int(os, DMU_GROUPUSED_OBJECT, 1531 uqn->uqn_id, uqn->uqn_delta, tx)); 1532 mutex_exit(&os->os_userused_lock); 1533 kmem_free(uqn, sizeof (*uqn)); 1534 } 1535 avl_destroy(&cache->uqc_group_deltas); 1536 } 1537 1538 static void 1539 userquota_update_cache(avl_tree_t *avl, uint64_t id, int64_t delta) 1540 { 1541 userquota_node_t search = { .uqn_id = id }; 1542 avl_index_t idx; 1543 1544 userquota_node_t *uqn = avl_find(avl, &search, &idx); 1545 if (uqn == NULL) { 1546 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP); 1547 uqn->uqn_id = id; 1548 avl_insert(avl, uqn, idx); 1549 } 1550 uqn->uqn_delta += delta; 1551 } 1552 1553 static void 1554 do_userquota_update(userquota_cache_t *cache, uint64_t used, uint64_t flags, 1555 uint64_t user, uint64_t group, boolean_t subtract) 1556 { 1557 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) { 1558 int64_t delta = DNODE_MIN_SIZE + used; 1559 if (subtract) 1560 delta = -delta; 1561 1562 userquota_update_cache(&cache->uqc_user_deltas, user, delta); 1563 userquota_update_cache(&cache->uqc_group_deltas, group, delta); 1564 } 1565 } 1566 1567 typedef struct userquota_updates_arg { 1568 objset_t *uua_os; 1569 int uua_sublist_idx; 1570 dmu_tx_t *uua_tx; 1571 } userquota_updates_arg_t; 1572 1573 static void 1574 userquota_updates_task(void *arg) 1575 { 1576 userquota_updates_arg_t *uua = arg; 1577 objset_t *os = uua->uua_os; 1578 dmu_tx_t *tx = uua->uua_tx; 1579 dnode_t *dn; 1580 userquota_cache_t cache = { 0 }; 1581 1582 multilist_sublist_t *list = 1583 multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx); 1584 1585 ASSERT(multilist_sublist_head(list) == NULL || 1586 dmu_objset_userused_enabled(os)); 1587 avl_create(&cache.uqc_user_deltas, userquota_compare, 1588 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 1589 avl_create(&cache.uqc_group_deltas, userquota_compare, 1590 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node)); 1591 1592 while ((dn = multilist_sublist_head(list)) != NULL) { 1593 int flags; 1594 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); 1595 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || 1596 dn->dn_phys->dn_flags & 1597 DNODE_FLAG_USERUSED_ACCOUNTED); 1598 1599 flags = dn->dn_id_flags; 1600 ASSERT(flags); 1601 if (flags & DN_ID_OLD_EXIST) { 1602 do_userquota_update(&cache, 1603 dn->dn_oldused, dn->dn_oldflags, 1604 dn->dn_olduid, dn->dn_oldgid, B_TRUE); 1605 } 1606 if (flags & DN_ID_NEW_EXIST) { 1607 do_userquota_update(&cache, 1608 DN_USED_BYTES(dn->dn_phys), 1609 dn->dn_phys->dn_flags, dn->dn_newuid, 1610 dn->dn_newgid, B_FALSE); 1611 } 1612 1613 mutex_enter(&dn->dn_mtx); 1614 dn->dn_oldused = 0; 1615 dn->dn_oldflags = 0; 1616 if (dn->dn_id_flags & DN_ID_NEW_EXIST) { 1617 dn->dn_olduid = dn->dn_newuid; 1618 dn->dn_oldgid = dn->dn_newgid; 1619 dn->dn_id_flags |= DN_ID_OLD_EXIST; 1620 if (dn->dn_bonuslen == 0) 1621 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 1622 else 1623 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 1624 } 1625 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); 1626 if (dn->dn_dirty_txg == spa_syncing_txg(os->os_spa)) 1627 dn->dn_dirty_txg = 0; 1628 mutex_exit(&dn->dn_mtx); 1629 1630 multilist_sublist_remove(list, dn); 1631 dnode_rele(dn, os->os_synced_dnodes); 1632 } 1633 do_userquota_cacheflush(os, &cache, tx); 1634 multilist_sublist_unlock(list); 1635 kmem_free(uua, sizeof (*uua)); 1636 } 1637 1638 void 1639 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) 1640 { 1641 if (!dmu_objset_userused_enabled(os)) 1642 return; 1643 1644 /* Allocate the user/groupused objects if necessary. */ 1645 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 1646 VERIFY0(zap_create_claim(os, 1647 DMU_USERUSED_OBJECT, 1648 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 1649 VERIFY0(zap_create_claim(os, 1650 DMU_GROUPUSED_OBJECT, 1651 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 1652 } 1653 1654 for (int i = 0; 1655 i < multilist_get_num_sublists(os->os_synced_dnodes); i++) { 1656 userquota_updates_arg_t *uua = 1657 kmem_alloc(sizeof (*uua), KM_SLEEP); 1658 uua->uua_os = os; 1659 uua->uua_sublist_idx = i; 1660 uua->uua_tx = tx; 1661 /* note: caller does taskq_wait() */ 1662 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq, 1663 userquota_updates_task, uua, 0); 1664 /* callback frees uua */ 1665 } 1666 } 1667 1668 /* 1669 * Returns a pointer to data to find uid/gid from 1670 * 1671 * If a dirty record for transaction group that is syncing can't 1672 * be found then NULL is returned. In the NULL case it is assumed 1673 * the uid/gid aren't changing. 1674 */ 1675 static void * 1676 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) 1677 { 1678 dbuf_dirty_record_t *dr, **drp; 1679 void *data; 1680 1681 if (db->db_dirtycnt == 0) 1682 return (db->db.db_data); /* Nothing is changing */ 1683 1684 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) 1685 if (dr->dr_txg == tx->tx_txg) 1686 break; 1687 1688 if (dr == NULL) { 1689 data = NULL; 1690 } else { 1691 dnode_t *dn; 1692 1693 DB_DNODE_ENTER(dr->dr_dbuf); 1694 dn = DB_DNODE(dr->dr_dbuf); 1695 1696 if (dn->dn_bonuslen == 0 && 1697 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) 1698 data = dr->dt.dl.dr_data->b_data; 1699 else 1700 data = dr->dt.dl.dr_data; 1701 1702 DB_DNODE_EXIT(dr->dr_dbuf); 1703 } 1704 1705 return (data); 1706 } 1707 1708 void 1709 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) 1710 { 1711 objset_t *os = dn->dn_objset; 1712 void *data = NULL; 1713 dmu_buf_impl_t *db = NULL; 1714 uint64_t *user = NULL; 1715 uint64_t *group = NULL; 1716 int flags = dn->dn_id_flags; 1717 int error; 1718 boolean_t have_spill = B_FALSE; 1719 1720 if (!dmu_objset_userused_enabled(dn->dn_objset)) 1721 return; 1722 1723 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| 1724 DN_ID_CHKED_SPILL))) 1725 return; 1726 1727 if (before && dn->dn_bonuslen != 0) 1728 data = DN_BONUS(dn->dn_phys); 1729 else if (!before && dn->dn_bonuslen != 0) { 1730 if (dn->dn_bonus) { 1731 db = dn->dn_bonus; 1732 mutex_enter(&db->db_mtx); 1733 data = dmu_objset_userquota_find_data(db, tx); 1734 } else { 1735 data = DN_BONUS(dn->dn_phys); 1736 } 1737 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { 1738 int rf = 0; 1739 1740 if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) 1741 rf |= DB_RF_HAVESTRUCT; 1742 error = dmu_spill_hold_by_dnode(dn, 1743 rf | DB_RF_MUST_SUCCEED, 1744 FTAG, (dmu_buf_t **)&db); 1745 ASSERT(error == 0); 1746 mutex_enter(&db->db_mtx); 1747 data = (before) ? db->db.db_data : 1748 dmu_objset_userquota_find_data(db, tx); 1749 have_spill = B_TRUE; 1750 } else { 1751 mutex_enter(&dn->dn_mtx); 1752 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 1753 mutex_exit(&dn->dn_mtx); 1754 return; 1755 } 1756 1757 if (before) { 1758 ASSERT(data); 1759 user = &dn->dn_olduid; 1760 group = &dn->dn_oldgid; 1761 } else if (data) { 1762 user = &dn->dn_newuid; 1763 group = &dn->dn_newgid; 1764 } 1765 1766 /* 1767 * Must always call the callback in case the object 1768 * type has changed and that type isn't an object type to track 1769 */ 1770 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data, 1771 user, group); 1772 1773 /* 1774 * Preserve existing uid/gid when the callback can't determine 1775 * what the new uid/gid are and the callback returned EEXIST. 1776 * The EEXIST error tells us to just use the existing uid/gid. 1777 * If we don't know what the old values are then just assign 1778 * them to 0, since that is a new file being created. 1779 */ 1780 if (!before && data == NULL && error == EEXIST) { 1781 if (flags & DN_ID_OLD_EXIST) { 1782 dn->dn_newuid = dn->dn_olduid; 1783 dn->dn_newgid = dn->dn_oldgid; 1784 } else { 1785 dn->dn_newuid = 0; 1786 dn->dn_newgid = 0; 1787 } 1788 error = 0; 1789 } 1790 1791 if (db) 1792 mutex_exit(&db->db_mtx); 1793 1794 mutex_enter(&dn->dn_mtx); 1795 if (error == 0 && before) 1796 dn->dn_id_flags |= DN_ID_OLD_EXIST; 1797 if (error == 0 && !before) 1798 dn->dn_id_flags |= DN_ID_NEW_EXIST; 1799 1800 if (have_spill) { 1801 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 1802 } else { 1803 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 1804 } 1805 mutex_exit(&dn->dn_mtx); 1806 if (have_spill) 1807 dmu_buf_rele((dmu_buf_t *)db, FTAG); 1808 } 1809 1810 boolean_t 1811 dmu_objset_userspace_present(objset_t *os) 1812 { 1813 return (os->os_phys->os_flags & 1814 OBJSET_FLAG_USERACCOUNTING_COMPLETE); 1815 } 1816 1817 int 1818 dmu_objset_userspace_upgrade(objset_t *os) 1819 { 1820 uint64_t obj; 1821 int err = 0; 1822 1823 if (dmu_objset_userspace_present(os)) 1824 return (0); 1825 if (!dmu_objset_userused_enabled(os)) 1826 return (SET_ERROR(ENOTSUP)); 1827 if (dmu_objset_is_snapshot(os)) 1828 return (SET_ERROR(EINVAL)); 1829 1830 /* 1831 * We simply need to mark every object dirty, so that it will be 1832 * synced out and now accounted. If this is called 1833 * concurrently, or if we already did some work before crashing, 1834 * that's fine, since we track each object's accounted state 1835 * independently. 1836 */ 1837 1838 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { 1839 dmu_tx_t *tx; 1840 dmu_buf_t *db; 1841 int objerr; 1842 1843 if (issig(JUSTLOOKING) && issig(FORREAL)) 1844 return (SET_ERROR(EINTR)); 1845 1846 objerr = dmu_bonus_hold(os, obj, FTAG, &db); 1847 if (objerr != 0) 1848 continue; 1849 tx = dmu_tx_create(os); 1850 dmu_tx_hold_bonus(tx, obj); 1851 objerr = dmu_tx_assign(tx, TXG_WAIT); 1852 if (objerr != 0) { 1853 dmu_tx_abort(tx); 1854 continue; 1855 } 1856 dmu_buf_will_dirty(db, tx); 1857 dmu_buf_rele(db, FTAG); 1858 dmu_tx_commit(tx); 1859 } 1860 1861 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 1862 txg_wait_synced(dmu_objset_pool(os), 0); 1863 return (0); 1864 } 1865 1866 void 1867 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 1868 uint64_t *usedobjsp, uint64_t *availobjsp) 1869 { 1870 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, 1871 usedobjsp, availobjsp); 1872 } 1873 1874 uint64_t 1875 dmu_objset_fsid_guid(objset_t *os) 1876 { 1877 return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); 1878 } 1879 1880 void 1881 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) 1882 { 1883 stat->dds_type = os->os_phys->os_type; 1884 if (os->os_dsl_dataset) 1885 dsl_dataset_fast_stat(os->os_dsl_dataset, stat); 1886 } 1887 1888 void 1889 dmu_objset_stats(objset_t *os, nvlist_t *nv) 1890 { 1891 ASSERT(os->os_dsl_dataset || 1892 os->os_phys->os_type == DMU_OST_META); 1893 1894 if (os->os_dsl_dataset != NULL) 1895 dsl_dataset_stats(os->os_dsl_dataset, nv); 1896 1897 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, 1898 os->os_phys->os_type); 1899 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, 1900 dmu_objset_userspace_present(os)); 1901 } 1902 1903 int 1904 dmu_objset_is_snapshot(objset_t *os) 1905 { 1906 if (os->os_dsl_dataset != NULL) 1907 return (os->os_dsl_dataset->ds_is_snapshot); 1908 else 1909 return (B_FALSE); 1910 } 1911 1912 int 1913 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen, 1914 boolean_t *conflict) 1915 { 1916 dsl_dataset_t *ds = os->os_dsl_dataset; 1917 uint64_t ignored; 1918 1919 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 1920 return (SET_ERROR(ENOENT)); 1921 1922 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, 1923 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, 1924 MT_NORMALIZE, real, maxlen, conflict)); 1925 } 1926 1927 int 1928 dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 1929 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) 1930 { 1931 dsl_dataset_t *ds = os->os_dsl_dataset; 1932 zap_cursor_t cursor; 1933 zap_attribute_t attr; 1934 1935 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 1936 1937 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 1938 return (SET_ERROR(ENOENT)); 1939 1940 zap_cursor_init_serialized(&cursor, 1941 ds->ds_dir->dd_pool->dp_meta_objset, 1942 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); 1943 1944 if (zap_cursor_retrieve(&cursor, &attr) != 0) { 1945 zap_cursor_fini(&cursor); 1946 return (SET_ERROR(ENOENT)); 1947 } 1948 1949 if (strlen(attr.za_name) + 1 > namelen) { 1950 zap_cursor_fini(&cursor); 1951 return (SET_ERROR(ENAMETOOLONG)); 1952 } 1953 1954 (void) strcpy(name, attr.za_name); 1955 if (idp) 1956 *idp = attr.za_first_integer; 1957 if (case_conflict) 1958 *case_conflict = attr.za_normalization_conflict; 1959 zap_cursor_advance(&cursor); 1960 *offp = zap_cursor_serialize(&cursor); 1961 zap_cursor_fini(&cursor); 1962 1963 return (0); 1964 } 1965 1966 int 1967 dmu_dir_list_next(objset_t *os, int namelen, char *name, 1968 uint64_t *idp, uint64_t *offp) 1969 { 1970 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 1971 zap_cursor_t cursor; 1972 zap_attribute_t attr; 1973 1974 /* there is no next dir on a snapshot! */ 1975 if (os->os_dsl_dataset->ds_object != 1976 dsl_dir_phys(dd)->dd_head_dataset_obj) 1977 return (SET_ERROR(ENOENT)); 1978 1979 zap_cursor_init_serialized(&cursor, 1980 dd->dd_pool->dp_meta_objset, 1981 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); 1982 1983 if (zap_cursor_retrieve(&cursor, &attr) != 0) { 1984 zap_cursor_fini(&cursor); 1985 return (SET_ERROR(ENOENT)); 1986 } 1987 1988 if (strlen(attr.za_name) + 1 > namelen) { 1989 zap_cursor_fini(&cursor); 1990 return (SET_ERROR(ENAMETOOLONG)); 1991 } 1992 1993 (void) strcpy(name, attr.za_name); 1994 if (idp) 1995 *idp = attr.za_first_integer; 1996 zap_cursor_advance(&cursor); 1997 *offp = zap_cursor_serialize(&cursor); 1998 zap_cursor_fini(&cursor); 1999 2000 return (0); 2001 } 2002 2003 typedef struct dmu_objset_find_ctx { 2004 taskq_t *dc_tq; 2005 dsl_pool_t *dc_dp; 2006 uint64_t dc_ddobj; 2007 char *dc_ddname; /* last component of ddobj's name */ 2008 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); 2009 void *dc_arg; 2010 int dc_flags; 2011 kmutex_t *dc_error_lock; 2012 int *dc_error; 2013 } dmu_objset_find_ctx_t; 2014 2015 static void 2016 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) 2017 { 2018 dsl_pool_t *dp = dcp->dc_dp; 2019 dsl_dir_t *dd; 2020 dsl_dataset_t *ds; 2021 zap_cursor_t zc; 2022 zap_attribute_t *attr; 2023 uint64_t thisobj; 2024 int err = 0; 2025 2026 /* don't process if there already was an error */ 2027 if (*dcp->dc_error != 0) 2028 goto out; 2029 2030 /* 2031 * Note: passing the name (dc_ddname) here is optional, but it 2032 * improves performance because we don't need to call 2033 * zap_value_search() to determine the name. 2034 */ 2035 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd); 2036 if (err != 0) 2037 goto out; 2038 2039 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2040 if (dd->dd_myname[0] == '$') { 2041 dsl_dir_rele(dd, FTAG); 2042 goto out; 2043 } 2044 2045 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2046 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 2047 2048 /* 2049 * Iterate over all children. 2050 */ 2051 if (dcp->dc_flags & DS_FIND_CHILDREN) { 2052 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2053 dsl_dir_phys(dd)->dd_child_dir_zapobj); 2054 zap_cursor_retrieve(&zc, attr) == 0; 2055 (void) zap_cursor_advance(&zc)) { 2056 ASSERT3U(attr->za_integer_length, ==, 2057 sizeof (uint64_t)); 2058 ASSERT3U(attr->za_num_integers, ==, 1); 2059 2060 dmu_objset_find_ctx_t *child_dcp = 2061 kmem_alloc(sizeof (*child_dcp), KM_SLEEP); 2062 *child_dcp = *dcp; 2063 child_dcp->dc_ddobj = attr->za_first_integer; 2064 child_dcp->dc_ddname = spa_strdup(attr->za_name); 2065 if (dcp->dc_tq != NULL) 2066 (void) taskq_dispatch(dcp->dc_tq, 2067 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); 2068 else 2069 dmu_objset_find_dp_impl(child_dcp); 2070 } 2071 zap_cursor_fini(&zc); 2072 } 2073 2074 /* 2075 * Iterate over all snapshots. 2076 */ 2077 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { 2078 dsl_dataset_t *ds; 2079 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2080 2081 if (err == 0) { 2082 uint64_t snapobj; 2083 2084 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 2085 dsl_dataset_rele(ds, FTAG); 2086 2087 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 2088 zap_cursor_retrieve(&zc, attr) == 0; 2089 (void) zap_cursor_advance(&zc)) { 2090 ASSERT3U(attr->za_integer_length, ==, 2091 sizeof (uint64_t)); 2092 ASSERT3U(attr->za_num_integers, ==, 1); 2093 2094 err = dsl_dataset_hold_obj(dp, 2095 attr->za_first_integer, FTAG, &ds); 2096 if (err != 0) 2097 break; 2098 err = dcp->dc_func(dp, ds, dcp->dc_arg); 2099 dsl_dataset_rele(ds, FTAG); 2100 if (err != 0) 2101 break; 2102 } 2103 zap_cursor_fini(&zc); 2104 } 2105 } 2106 2107 kmem_free(attr, sizeof (zap_attribute_t)); 2108 2109 if (err != 0) { 2110 dsl_dir_rele(dd, FTAG); 2111 goto out; 2112 } 2113 2114 /* 2115 * Apply to self. 2116 */ 2117 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2118 2119 /* 2120 * Note: we hold the dir while calling dsl_dataset_hold_obj() so 2121 * that the dir will remain cached, and we won't have to re-instantiate 2122 * it (which could be expensive due to finding its name via 2123 * zap_value_search()). 2124 */ 2125 dsl_dir_rele(dd, FTAG); 2126 if (err != 0) 2127 goto out; 2128 err = dcp->dc_func(dp, ds, dcp->dc_arg); 2129 dsl_dataset_rele(ds, FTAG); 2130 2131 out: 2132 if (err != 0) { 2133 mutex_enter(dcp->dc_error_lock); 2134 /* only keep first error */ 2135 if (*dcp->dc_error == 0) 2136 *dcp->dc_error = err; 2137 mutex_exit(dcp->dc_error_lock); 2138 } 2139 2140 if (dcp->dc_ddname != NULL) 2141 spa_strfree(dcp->dc_ddname); 2142 kmem_free(dcp, sizeof (*dcp)); 2143 } 2144 2145 static void 2146 dmu_objset_find_dp_cb(void *arg) 2147 { 2148 dmu_objset_find_ctx_t *dcp = arg; 2149 dsl_pool_t *dp = dcp->dc_dp; 2150 2151 /* 2152 * We need to get a pool_config_lock here, as there are several 2153 * asssert(pool_config_held) down the stack. Getting a lock via 2154 * dsl_pool_config_enter is risky, as it might be stalled by a 2155 * pending writer. This would deadlock, as the write lock can 2156 * only be granted when our parent thread gives up the lock. 2157 * The _prio interface gives us priority over a pending writer. 2158 */ 2159 dsl_pool_config_enter_prio(dp, FTAG); 2160 2161 dmu_objset_find_dp_impl(dcp); 2162 2163 dsl_pool_config_exit(dp, FTAG); 2164 } 2165 2166 /* 2167 * Find objsets under and including ddobj, call func(ds) on each. 2168 * The order for the enumeration is completely undefined. 2169 * func is called with dsl_pool_config held. 2170 */ 2171 int 2172 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, 2173 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) 2174 { 2175 int error = 0; 2176 taskq_t *tq = NULL; 2177 int ntasks; 2178 dmu_objset_find_ctx_t *dcp; 2179 kmutex_t err_lock; 2180 2181 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); 2182 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); 2183 dcp->dc_tq = NULL; 2184 dcp->dc_dp = dp; 2185 dcp->dc_ddobj = ddobj; 2186 dcp->dc_ddname = NULL; 2187 dcp->dc_func = func; 2188 dcp->dc_arg = arg; 2189 dcp->dc_flags = flags; 2190 dcp->dc_error_lock = &err_lock; 2191 dcp->dc_error = &error; 2192 2193 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { 2194 /* 2195 * In case a write lock is held we can't make use of 2196 * parallelism, as down the stack of the worker threads 2197 * the lock is asserted via dsl_pool_config_held. 2198 * In case of a read lock this is solved by getting a read 2199 * lock in each worker thread, which isn't possible in case 2200 * of a writer lock. So we fall back to the synchronous path 2201 * here. 2202 * In the future it might be possible to get some magic into 2203 * dsl_pool_config_held in a way that it returns true for 2204 * the worker threads so that a single lock held from this 2205 * thread suffices. For now, stay single threaded. 2206 */ 2207 dmu_objset_find_dp_impl(dcp); 2208 mutex_destroy(&err_lock); 2209 2210 return (error); 2211 } 2212 2213 ntasks = dmu_find_threads; 2214 if (ntasks == 0) 2215 ntasks = vdev_count_leaves(dp->dp_spa) * 4; 2216 tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks, 2217 INT_MAX, 0); 2218 if (tq == NULL) { 2219 kmem_free(dcp, sizeof (*dcp)); 2220 mutex_destroy(&err_lock); 2221 2222 return (SET_ERROR(ENOMEM)); 2223 } 2224 dcp->dc_tq = tq; 2225 2226 /* dcp will be freed by task */ 2227 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); 2228 2229 /* 2230 * PORTING: this code relies on the property of taskq_wait to wait 2231 * until no more tasks are queued and no more tasks are active. As 2232 * we always queue new tasks from within other tasks, task_wait 2233 * reliably waits for the full recursion to finish, even though we 2234 * enqueue new tasks after taskq_wait has been called. 2235 * On platforms other than illumos, taskq_wait may not have this 2236 * property. 2237 */ 2238 taskq_wait(tq); 2239 taskq_destroy(tq); 2240 mutex_destroy(&err_lock); 2241 2242 return (error); 2243 } 2244 2245 /* 2246 * Find all objsets under name, and for each, call 'func(child_name, arg)'. 2247 * The dp_config_rwlock must not be held when this is called, and it 2248 * will not be held when the callback is called. 2249 * Therefore this function should only be used when the pool is not changing 2250 * (e.g. in syncing context), or the callback can deal with the possible races. 2251 */ 2252 static int 2253 dmu_objset_find_impl(spa_t *spa, const char *name, 2254 int func(const char *, void *), void *arg, int flags) 2255 { 2256 dsl_dir_t *dd; 2257 dsl_pool_t *dp = spa_get_dsl(spa); 2258 dsl_dataset_t *ds; 2259 zap_cursor_t zc; 2260 zap_attribute_t *attr; 2261 char *child; 2262 uint64_t thisobj; 2263 int err; 2264 2265 dsl_pool_config_enter(dp, FTAG); 2266 2267 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); 2268 if (err != 0) { 2269 dsl_pool_config_exit(dp, FTAG); 2270 return (err); 2271 } 2272 2273 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 2274 if (dd->dd_myname[0] == '$') { 2275 dsl_dir_rele(dd, FTAG); 2276 dsl_pool_config_exit(dp, FTAG); 2277 return (0); 2278 } 2279 2280 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 2281 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 2282 2283 /* 2284 * Iterate over all children. 2285 */ 2286 if (flags & DS_FIND_CHILDREN) { 2287 for (zap_cursor_init(&zc, dp->dp_meta_objset, 2288 dsl_dir_phys(dd)->dd_child_dir_zapobj); 2289 zap_cursor_retrieve(&zc, attr) == 0; 2290 (void) zap_cursor_advance(&zc)) { 2291 ASSERT3U(attr->za_integer_length, ==, 2292 sizeof (uint64_t)); 2293 ASSERT3U(attr->za_num_integers, ==, 1); 2294 2295 child = kmem_asprintf("%s/%s", name, attr->za_name); 2296 dsl_pool_config_exit(dp, FTAG); 2297 err = dmu_objset_find_impl(spa, child, 2298 func, arg, flags); 2299 dsl_pool_config_enter(dp, FTAG); 2300 strfree(child); 2301 if (err != 0) 2302 break; 2303 } 2304 zap_cursor_fini(&zc); 2305 2306 if (err != 0) { 2307 dsl_dir_rele(dd, FTAG); 2308 dsl_pool_config_exit(dp, FTAG); 2309 kmem_free(attr, sizeof (zap_attribute_t)); 2310 return (err); 2311 } 2312 } 2313 2314 /* 2315 * Iterate over all snapshots. 2316 */ 2317 if (flags & DS_FIND_SNAPSHOTS) { 2318 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 2319 2320 if (err == 0) { 2321 uint64_t snapobj; 2322 2323 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 2324 dsl_dataset_rele(ds, FTAG); 2325 2326 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 2327 zap_cursor_retrieve(&zc, attr) == 0; 2328 (void) zap_cursor_advance(&zc)) { 2329 ASSERT3U(attr->za_integer_length, ==, 2330 sizeof (uint64_t)); 2331 ASSERT3U(attr->za_num_integers, ==, 1); 2332 2333 child = kmem_asprintf("%s@%s", 2334 name, attr->za_name); 2335 dsl_pool_config_exit(dp, FTAG); 2336 err = func(child, arg); 2337 dsl_pool_config_enter(dp, FTAG); 2338 strfree(child); 2339 if (err != 0) 2340 break; 2341 } 2342 zap_cursor_fini(&zc); 2343 } 2344 } 2345 2346 dsl_dir_rele(dd, FTAG); 2347 kmem_free(attr, sizeof (zap_attribute_t)); 2348 dsl_pool_config_exit(dp, FTAG); 2349 2350 if (err != 0) 2351 return (err); 2352 2353 /* Apply to self. */ 2354 return (func(name, arg)); 2355 } 2356 2357 /* 2358 * See comment above dmu_objset_find_impl(). 2359 */ 2360 int 2361 dmu_objset_find(char *name, int func(const char *, void *), void *arg, 2362 int flags) 2363 { 2364 spa_t *spa; 2365 int error; 2366 2367 error = spa_open(name, &spa, FTAG); 2368 if (error != 0) 2369 return (error); 2370 error = dmu_objset_find_impl(spa, name, func, arg, flags); 2371 spa_close(spa, FTAG); 2372 return (error); 2373 } 2374 2375 void 2376 dmu_objset_set_user(objset_t *os, void *user_ptr) 2377 { 2378 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 2379 os->os_user_ptr = user_ptr; 2380 } 2381 2382 void * 2383 dmu_objset_get_user(objset_t *os) 2384 { 2385 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 2386 return (os->os_user_ptr); 2387 } 2388 2389 /* 2390 * Determine name of filesystem, given name of snapshot. 2391 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes 2392 */ 2393 int 2394 dmu_fsname(const char *snapname, char *buf) 2395 { 2396 char *atp = strchr(snapname, '@'); 2397 if (atp == NULL) 2398 return (SET_ERROR(EINVAL)); 2399 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) 2400 return (SET_ERROR(ENAMETOOLONG)); 2401 (void) strlcpy(buf, snapname, atp - snapname + 1); 2402 return (0); 2403 } 2404 2405 /* 2406 * Call when we think we're going to write/free space in open context to track 2407 * the amount of dirty data in the open txg, which is also the amount 2408 * of memory that can not be evicted until this txg syncs. 2409 */ 2410 void 2411 dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx) 2412 { 2413 dsl_dataset_t *ds = os->os_dsl_dataset; 2414 int64_t aspace = spa_get_worst_case_asize(os->os_spa, space); 2415 2416 if (ds != NULL) { 2417 dsl_dir_willuse_space(ds->ds_dir, aspace, tx); 2418 dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx); 2419 } 2420 } 2421