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