dmu_objset.c revision 307126
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 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2016 by Delphix. All rights reserved. 24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. 25 * Copyright (c) 2013, Joyent, Inc. All rights reserved. 26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 27 * Copyright 2015 Nexenta Systems, Inc. All rights reserved. 28 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved. 29 * Copyright (c) 2014 Integros [integros.com] 30 */ 31 32/* Portions Copyright 2010 Robert Milkowski */ 33 34#include <sys/cred.h> 35#include <sys/zfs_context.h> 36#include <sys/dmu_objset.h> 37#include <sys/dsl_dir.h> 38#include <sys/dsl_dataset.h> 39#include <sys/dsl_prop.h> 40#include <sys/dsl_pool.h> 41#include <sys/dsl_synctask.h> 42#include <sys/dsl_deleg.h> 43#include <sys/dnode.h> 44#include <sys/dbuf.h> 45#include <sys/zvol.h> 46#include <sys/dmu_tx.h> 47#include <sys/zap.h> 48#include <sys/zil.h> 49#include <sys/dmu_impl.h> 50#include <sys/zfs_ioctl.h> 51#include <sys/sa.h> 52#include <sys/zfs_onexit.h> 53#include <sys/dsl_destroy.h> 54#include <sys/vdev.h> 55 56/* 57 * Needed to close a window in dnode_move() that allows the objset to be freed 58 * before it can be safely accessed. 59 */ 60krwlock_t os_lock; 61 62/* 63 * Tunable to overwrite the maximum number of threads for the parallization 64 * of dmu_objset_find_dp, needed to speed up the import of pools with many 65 * datasets. 66 * Default is 4 times the number of leaf vdevs. 67 */ 68int dmu_find_threads = 0; 69 70static void dmu_objset_find_dp_cb(void *arg); 71 72void 73dmu_objset_init(void) 74{ 75 rw_init(&os_lock, NULL, RW_DEFAULT, NULL); 76} 77 78void 79dmu_objset_fini(void) 80{ 81 rw_destroy(&os_lock); 82} 83 84spa_t * 85dmu_objset_spa(objset_t *os) 86{ 87 return (os->os_spa); 88} 89 90zilog_t * 91dmu_objset_zil(objset_t *os) 92{ 93 return (os->os_zil); 94} 95 96dsl_pool_t * 97dmu_objset_pool(objset_t *os) 98{ 99 dsl_dataset_t *ds; 100 101 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) 102 return (ds->ds_dir->dd_pool); 103 else 104 return (spa_get_dsl(os->os_spa)); 105} 106 107dsl_dataset_t * 108dmu_objset_ds(objset_t *os) 109{ 110 return (os->os_dsl_dataset); 111} 112 113dmu_objset_type_t 114dmu_objset_type(objset_t *os) 115{ 116 return (os->os_phys->os_type); 117} 118 119void 120dmu_objset_name(objset_t *os, char *buf) 121{ 122 dsl_dataset_name(os->os_dsl_dataset, buf); 123} 124 125uint64_t 126dmu_objset_id(objset_t *os) 127{ 128 dsl_dataset_t *ds = os->os_dsl_dataset; 129 130 return (ds ? ds->ds_object : 0); 131} 132 133zfs_sync_type_t 134dmu_objset_syncprop(objset_t *os) 135{ 136 return (os->os_sync); 137} 138 139zfs_logbias_op_t 140dmu_objset_logbias(objset_t *os) 141{ 142 return (os->os_logbias); 143} 144 145static void 146checksum_changed_cb(void *arg, uint64_t newval) 147{ 148 objset_t *os = arg; 149 150 /* 151 * Inheritance should have been done by now. 152 */ 153 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 154 155 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); 156} 157 158static void 159compression_changed_cb(void *arg, uint64_t newval) 160{ 161 objset_t *os = arg; 162 163 /* 164 * Inheritance and range checking should have been done by now. 165 */ 166 ASSERT(newval != ZIO_COMPRESS_INHERIT); 167 168 os->os_compress = zio_compress_select(os->os_spa, newval, 169 ZIO_COMPRESS_ON); 170} 171 172static void 173copies_changed_cb(void *arg, uint64_t newval) 174{ 175 objset_t *os = arg; 176 177 /* 178 * Inheritance and range checking should have been done by now. 179 */ 180 ASSERT(newval > 0); 181 ASSERT(newval <= spa_max_replication(os->os_spa)); 182 183 os->os_copies = newval; 184} 185 186static void 187dedup_changed_cb(void *arg, uint64_t newval) 188{ 189 objset_t *os = arg; 190 spa_t *spa = os->os_spa; 191 enum zio_checksum checksum; 192 193 /* 194 * Inheritance should have been done by now. 195 */ 196 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 197 198 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); 199 200 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; 201 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); 202} 203 204static void 205primary_cache_changed_cb(void *arg, uint64_t newval) 206{ 207 objset_t *os = arg; 208 209 /* 210 * Inheritance and range checking should have been done by now. 211 */ 212 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 213 newval == ZFS_CACHE_METADATA); 214 215 os->os_primary_cache = newval; 216} 217 218static void 219secondary_cache_changed_cb(void *arg, uint64_t newval) 220{ 221 objset_t *os = arg; 222 223 /* 224 * Inheritance and range checking should have been done by now. 225 */ 226 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 227 newval == ZFS_CACHE_METADATA); 228 229 os->os_secondary_cache = newval; 230} 231 232static void 233sync_changed_cb(void *arg, uint64_t newval) 234{ 235 objset_t *os = arg; 236 237 /* 238 * Inheritance and range checking should have been done by now. 239 */ 240 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || 241 newval == ZFS_SYNC_DISABLED); 242 243 os->os_sync = newval; 244 if (os->os_zil) 245 zil_set_sync(os->os_zil, newval); 246} 247 248static void 249redundant_metadata_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_REDUNDANT_METADATA_ALL || 257 newval == ZFS_REDUNDANT_METADATA_MOST); 258 259 os->os_redundant_metadata = newval; 260} 261 262static void 263logbias_changed_cb(void *arg, uint64_t newval) 264{ 265 objset_t *os = arg; 266 267 ASSERT(newval == ZFS_LOGBIAS_LATENCY || 268 newval == ZFS_LOGBIAS_THROUGHPUT); 269 os->os_logbias = newval; 270 if (os->os_zil) 271 zil_set_logbias(os->os_zil, newval); 272} 273 274static void 275recordsize_changed_cb(void *arg, uint64_t newval) 276{ 277 objset_t *os = arg; 278 279 os->os_recordsize = newval; 280} 281 282void 283dmu_objset_byteswap(void *buf, size_t size) 284{ 285 objset_phys_t *osp = buf; 286 287 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t)); 288 dnode_byteswap(&osp->os_meta_dnode); 289 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); 290 osp->os_type = BSWAP_64(osp->os_type); 291 osp->os_flags = BSWAP_64(osp->os_flags); 292 if (size == sizeof (objset_phys_t)) { 293 dnode_byteswap(&osp->os_userused_dnode); 294 dnode_byteswap(&osp->os_groupused_dnode); 295 } 296} 297 298int 299dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 300 objset_t **osp) 301{ 302 objset_t *os; 303 int i, err; 304 305 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); 306 307 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); 308 os->os_dsl_dataset = ds; 309 os->os_spa = spa; 310 os->os_rootbp = bp; 311 if (!BP_IS_HOLE(os->os_rootbp)) { 312 arc_flags_t aflags = ARC_FLAG_WAIT; 313 zbookmark_phys_t zb; 314 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, 315 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 316 317 if (DMU_OS_IS_L2CACHEABLE(os)) 318 aflags |= ARC_FLAG_L2CACHE; 319 if (DMU_OS_IS_L2COMPRESSIBLE(os)) 320 aflags |= ARC_FLAG_L2COMPRESS; 321 322 dprintf_bp(os->os_rootbp, "reading %s", ""); 323 err = arc_read(NULL, spa, os->os_rootbp, 324 arc_getbuf_func, &os->os_phys_buf, 325 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb); 326 if (err != 0) { 327 kmem_free(os, sizeof (objset_t)); 328 /* convert checksum errors into IO errors */ 329 if (err == ECKSUM) 330 err = SET_ERROR(EIO); 331 return (err); 332 } 333 334 /* Increase the blocksize if we are permitted. */ 335 if (spa_version(spa) >= SPA_VERSION_USERSPACE && 336 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) { 337 arc_buf_t *buf = arc_buf_alloc(spa, 338 sizeof (objset_phys_t), &os->os_phys_buf, 339 ARC_BUFC_METADATA); 340 bzero(buf->b_data, sizeof (objset_phys_t)); 341 bcopy(os->os_phys_buf->b_data, buf->b_data, 342 arc_buf_size(os->os_phys_buf)); 343 (void) arc_buf_remove_ref(os->os_phys_buf, 344 &os->os_phys_buf); 345 os->os_phys_buf = buf; 346 } 347 348 os->os_phys = os->os_phys_buf->b_data; 349 os->os_flags = os->os_phys->os_flags; 350 } else { 351 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? 352 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE; 353 os->os_phys_buf = arc_buf_alloc(spa, size, 354 &os->os_phys_buf, ARC_BUFC_METADATA); 355 os->os_phys = os->os_phys_buf->b_data; 356 bzero(os->os_phys, size); 357 } 358 359 /* 360 * Note: the changed_cb will be called once before the register 361 * func returns, thus changing the checksum/compression from the 362 * default (fletcher2/off). Snapshots don't need to know about 363 * checksum/compression/copies. 364 */ 365 if (ds != NULL) { 366 boolean_t needlock = B_FALSE; 367 368 /* 369 * Note: it's valid to open the objset if the dataset is 370 * long-held, in which case the pool_config lock will not 371 * be held. 372 */ 373 if (!dsl_pool_config_held(dmu_objset_pool(os))) { 374 needlock = B_TRUE; 375 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 376 } 377 err = dsl_prop_register(ds, 378 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), 379 primary_cache_changed_cb, os); 380 if (err == 0) { 381 err = dsl_prop_register(ds, 382 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), 383 secondary_cache_changed_cb, os); 384 } 385 if (!ds->ds_is_snapshot) { 386 if (err == 0) { 387 err = dsl_prop_register(ds, 388 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 389 checksum_changed_cb, os); 390 } 391 if (err == 0) { 392 err = dsl_prop_register(ds, 393 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 394 compression_changed_cb, os); 395 } 396 if (err == 0) { 397 err = dsl_prop_register(ds, 398 zfs_prop_to_name(ZFS_PROP_COPIES), 399 copies_changed_cb, os); 400 } 401 if (err == 0) { 402 err = dsl_prop_register(ds, 403 zfs_prop_to_name(ZFS_PROP_DEDUP), 404 dedup_changed_cb, os); 405 } 406 if (err == 0) { 407 err = dsl_prop_register(ds, 408 zfs_prop_to_name(ZFS_PROP_LOGBIAS), 409 logbias_changed_cb, os); 410 } 411 if (err == 0) { 412 err = dsl_prop_register(ds, 413 zfs_prop_to_name(ZFS_PROP_SYNC), 414 sync_changed_cb, os); 415 } 416 if (err == 0) { 417 err = dsl_prop_register(ds, 418 zfs_prop_to_name( 419 ZFS_PROP_REDUNDANT_METADATA), 420 redundant_metadata_changed_cb, os); 421 } 422 if (err == 0) { 423 err = dsl_prop_register(ds, 424 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 425 recordsize_changed_cb, os); 426 } 427 } 428 if (needlock) 429 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 430 if (err != 0) { 431 VERIFY(arc_buf_remove_ref(os->os_phys_buf, 432 &os->os_phys_buf)); 433 kmem_free(os, sizeof (objset_t)); 434 return (err); 435 } 436 } else { 437 /* It's the meta-objset. */ 438 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; 439 os->os_compress = ZIO_COMPRESS_ON; 440 os->os_copies = spa_max_replication(spa); 441 os->os_dedup_checksum = ZIO_CHECKSUM_OFF; 442 os->os_dedup_verify = B_FALSE; 443 os->os_logbias = ZFS_LOGBIAS_LATENCY; 444 os->os_sync = ZFS_SYNC_STANDARD; 445 os->os_primary_cache = ZFS_CACHE_ALL; 446 os->os_secondary_cache = ZFS_CACHE_ALL; 447 } 448 449 if (ds == NULL || !ds->ds_is_snapshot) 450 os->os_zil_header = os->os_phys->os_zil_header; 451 os->os_zil = zil_alloc(os, &os->os_zil_header); 452 453 for (i = 0; i < TXG_SIZE; i++) { 454 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t), 455 offsetof(dnode_t, dn_dirty_link[i])); 456 list_create(&os->os_free_dnodes[i], sizeof (dnode_t), 457 offsetof(dnode_t, dn_dirty_link[i])); 458 } 459 list_create(&os->os_dnodes, sizeof (dnode_t), 460 offsetof(dnode_t, dn_link)); 461 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), 462 offsetof(dmu_buf_impl_t, db_link)); 463 464 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); 465 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); 466 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); 467 468 dnode_special_open(os, &os->os_phys->os_meta_dnode, 469 DMU_META_DNODE_OBJECT, &os->os_meta_dnode); 470 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) { 471 dnode_special_open(os, &os->os_phys->os_userused_dnode, 472 DMU_USERUSED_OBJECT, &os->os_userused_dnode); 473 dnode_special_open(os, &os->os_phys->os_groupused_dnode, 474 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode); 475 } 476 477 *osp = os; 478 return (0); 479} 480 481int 482dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) 483{ 484 int err = 0; 485 486 /* 487 * We shouldn't be doing anything with dsl_dataset_t's unless the 488 * pool_config lock is held, or the dataset is long-held. 489 */ 490 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) || 491 dsl_dataset_long_held(ds)); 492 493 mutex_enter(&ds->ds_opening_lock); 494 if (ds->ds_objset == NULL) { 495 objset_t *os; 496 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), 497 ds, dsl_dataset_get_blkptr(ds), &os); 498 499 if (err == 0) { 500 mutex_enter(&ds->ds_lock); 501 ASSERT(ds->ds_objset == NULL); 502 ds->ds_objset = os; 503 mutex_exit(&ds->ds_lock); 504 } 505 } 506 *osp = ds->ds_objset; 507 mutex_exit(&ds->ds_opening_lock); 508 return (err); 509} 510 511/* 512 * Holds the pool while the objset is held. Therefore only one objset 513 * can be held at a time. 514 */ 515int 516dmu_objset_hold(const char *name, void *tag, objset_t **osp) 517{ 518 dsl_pool_t *dp; 519 dsl_dataset_t *ds; 520 int err; 521 522 err = dsl_pool_hold(name, tag, &dp); 523 if (err != 0) 524 return (err); 525 err = dsl_dataset_hold(dp, name, tag, &ds); 526 if (err != 0) { 527 dsl_pool_rele(dp, tag); 528 return (err); 529 } 530 531 err = dmu_objset_from_ds(ds, osp); 532 if (err != 0) { 533 dsl_dataset_rele(ds, tag); 534 dsl_pool_rele(dp, tag); 535 } 536 537 return (err); 538} 539 540static int 541dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type, 542 boolean_t readonly, void *tag, objset_t **osp) 543{ 544 int err; 545 546 err = dmu_objset_from_ds(ds, osp); 547 if (err != 0) { 548 dsl_dataset_disown(ds, tag); 549 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { 550 dsl_dataset_disown(ds, tag); 551 return (SET_ERROR(EINVAL)); 552 } else if (!readonly && dsl_dataset_is_snapshot(ds)) { 553 dsl_dataset_disown(ds, tag); 554 return (SET_ERROR(EROFS)); 555 } 556 return (err); 557} 558 559/* 560 * dsl_pool must not be held when this is called. 561 * Upon successful return, there will be a longhold on the dataset, 562 * and the dsl_pool will not be held. 563 */ 564int 565dmu_objset_own(const char *name, dmu_objset_type_t type, 566 boolean_t readonly, void *tag, objset_t **osp) 567{ 568 dsl_pool_t *dp; 569 dsl_dataset_t *ds; 570 int err; 571 572 err = dsl_pool_hold(name, FTAG, &dp); 573 if (err != 0) 574 return (err); 575 err = dsl_dataset_own(dp, name, tag, &ds); 576 if (err != 0) { 577 dsl_pool_rele(dp, FTAG); 578 return (err); 579 } 580 err = dmu_objset_own_impl(ds, type, readonly, tag, osp); 581 dsl_pool_rele(dp, FTAG); 582 583 return (err); 584} 585 586int 587dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type, 588 boolean_t readonly, void *tag, objset_t **osp) 589{ 590 dsl_dataset_t *ds; 591 int err; 592 593 err = dsl_dataset_own_obj(dp, obj, tag, &ds); 594 if (err != 0) 595 return (err); 596 597 return (dmu_objset_own_impl(ds, type, readonly, tag, osp)); 598} 599 600void 601dmu_objset_rele(objset_t *os, void *tag) 602{ 603 dsl_pool_t *dp = dmu_objset_pool(os); 604 dsl_dataset_rele(os->os_dsl_dataset, tag); 605 dsl_pool_rele(dp, tag); 606} 607 608/* 609 * When we are called, os MUST refer to an objset associated with a dataset 610 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner 611 * == tag. We will then release and reacquire ownership of the dataset while 612 * holding the pool config_rwlock to avoid intervening namespace or ownership 613 * changes may occur. 614 * 615 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to 616 * release the hold on its dataset and acquire a new one on the dataset of the 617 * same name so that it can be partially torn down and reconstructed. 618 */ 619void 620dmu_objset_refresh_ownership(objset_t *os, void *tag) 621{ 622 dsl_pool_t *dp; 623 dsl_dataset_t *ds, *newds; 624 char name[ZFS_MAX_DATASET_NAME_LEN]; 625 626 ds = os->os_dsl_dataset; 627 VERIFY3P(ds, !=, NULL); 628 VERIFY3P(ds->ds_owner, ==, tag); 629 VERIFY(dsl_dataset_long_held(ds)); 630 631 dsl_dataset_name(ds, name); 632 dp = dmu_objset_pool(os); 633 dsl_pool_config_enter(dp, FTAG); 634 dmu_objset_disown(os, tag); 635 VERIFY0(dsl_dataset_own(dp, name, tag, &newds)); 636 VERIFY3P(newds, ==, os->os_dsl_dataset); 637 dsl_pool_config_exit(dp, FTAG); 638} 639 640void 641dmu_objset_disown(objset_t *os, void *tag) 642{ 643 dsl_dataset_disown(os->os_dsl_dataset, tag); 644} 645 646void 647dmu_objset_evict_dbufs(objset_t *os) 648{ 649 dnode_t dn_marker; 650 dnode_t *dn; 651 652 mutex_enter(&os->os_lock); 653 dn = list_head(&os->os_dnodes); 654 while (dn != NULL) { 655 /* 656 * Skip dnodes without holds. We have to do this dance 657 * because dnode_add_ref() only works if there is already a 658 * hold. If the dnode has no holds, then it has no dbufs. 659 */ 660 if (dnode_add_ref(dn, FTAG)) { 661 list_insert_after(&os->os_dnodes, dn, &dn_marker); 662 mutex_exit(&os->os_lock); 663 664 dnode_evict_dbufs(dn); 665 dnode_rele(dn, FTAG); 666 667 mutex_enter(&os->os_lock); 668 dn = list_next(&os->os_dnodes, &dn_marker); 669 list_remove(&os->os_dnodes, &dn_marker); 670 } else { 671 dn = list_next(&os->os_dnodes, dn); 672 } 673 } 674 mutex_exit(&os->os_lock); 675 676 if (DMU_USERUSED_DNODE(os) != NULL) { 677 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os)); 678 dnode_evict_dbufs(DMU_USERUSED_DNODE(os)); 679 } 680 dnode_evict_dbufs(DMU_META_DNODE(os)); 681} 682 683/* 684 * Objset eviction processing is split into into two pieces. 685 * The first marks the objset as evicting, evicts any dbufs that 686 * have a refcount of zero, and then queues up the objset for the 687 * second phase of eviction. Once os->os_dnodes has been cleared by 688 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed. 689 * The second phase closes the special dnodes, dequeues the objset from 690 * the list of those undergoing eviction, and finally frees the objset. 691 * 692 * NOTE: Due to asynchronous eviction processing (invocation of 693 * dnode_buf_pageout()), it is possible for the meta dnode for the 694 * objset to have no holds even though os->os_dnodes is not empty. 695 */ 696void 697dmu_objset_evict(objset_t *os) 698{ 699 dsl_dataset_t *ds = os->os_dsl_dataset; 700 701 for (int t = 0; t < TXG_SIZE; t++) 702 ASSERT(!dmu_objset_is_dirty(os, t)); 703 704 if (ds) 705 dsl_prop_unregister_all(ds, os); 706 707 if (os->os_sa) 708 sa_tear_down(os); 709 710 dmu_objset_evict_dbufs(os); 711 712 mutex_enter(&os->os_lock); 713 spa_evicting_os_register(os->os_spa, os); 714 if (list_is_empty(&os->os_dnodes)) { 715 mutex_exit(&os->os_lock); 716 dmu_objset_evict_done(os); 717 } else { 718 mutex_exit(&os->os_lock); 719 } 720} 721 722void 723dmu_objset_evict_done(objset_t *os) 724{ 725 ASSERT3P(list_head(&os->os_dnodes), ==, NULL); 726 727 dnode_special_close(&os->os_meta_dnode); 728 if (DMU_USERUSED_DNODE(os)) { 729 dnode_special_close(&os->os_userused_dnode); 730 dnode_special_close(&os->os_groupused_dnode); 731 } 732 zil_free(os->os_zil); 733 734 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf)); 735 736 /* 737 * This is a barrier to prevent the objset from going away in 738 * dnode_move() until we can safely ensure that the objset is still in 739 * use. We consider the objset valid before the barrier and invalid 740 * after the barrier. 741 */ 742 rw_enter(&os_lock, RW_READER); 743 rw_exit(&os_lock); 744 745 mutex_destroy(&os->os_lock); 746 mutex_destroy(&os->os_obj_lock); 747 mutex_destroy(&os->os_user_ptr_lock); 748 spa_evicting_os_deregister(os->os_spa, os); 749 kmem_free(os, sizeof (objset_t)); 750} 751 752timestruc_t 753dmu_objset_snap_cmtime(objset_t *os) 754{ 755 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); 756} 757 758/* called from dsl for meta-objset */ 759objset_t * 760dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 761 dmu_objset_type_t type, dmu_tx_t *tx) 762{ 763 objset_t *os; 764 dnode_t *mdn; 765 766 ASSERT(dmu_tx_is_syncing(tx)); 767 768 if (ds != NULL) 769 VERIFY0(dmu_objset_from_ds(ds, &os)); 770 else 771 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); 772 773 mdn = DMU_META_DNODE(os); 774 775 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT, 776 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx); 777 778 /* 779 * We don't want to have to increase the meta-dnode's nlevels 780 * later, because then we could do it in quescing context while 781 * we are also accessing it in open context. 782 * 783 * This precaution is not necessary for the MOS (ds == NULL), 784 * because the MOS is only updated in syncing context. 785 * This is most fortunate: the MOS is the only objset that 786 * needs to be synced multiple times as spa_sync() iterates 787 * to convergence, so minimizing its dn_nlevels matters. 788 */ 789 if (ds != NULL) { 790 int levels = 1; 791 792 /* 793 * Determine the number of levels necessary for the meta-dnode 794 * to contain DN_MAX_OBJECT dnodes. Note that in order to 795 * ensure that we do not overflow 64 bits, there has to be 796 * a nlevels that gives us a number of blocks > DN_MAX_OBJECT 797 * but < 2^64. Therefore, 798 * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be 799 * less than (64 - log2(DN_MAX_OBJECT)) (16). 800 */ 801 while ((uint64_t)mdn->dn_nblkptr << 802 (mdn->dn_datablkshift - DNODE_SHIFT + 803 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < 804 DN_MAX_OBJECT) 805 levels++; 806 807 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = 808 mdn->dn_nlevels = levels; 809 } 810 811 ASSERT(type != DMU_OST_NONE); 812 ASSERT(type != DMU_OST_ANY); 813 ASSERT(type < DMU_OST_NUMTYPES); 814 os->os_phys->os_type = type; 815 if (dmu_objset_userused_enabled(os)) { 816 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 817 os->os_flags = os->os_phys->os_flags; 818 } 819 820 dsl_dataset_dirty(ds, tx); 821 822 return (os); 823} 824 825typedef struct dmu_objset_create_arg { 826 const char *doca_name; 827 cred_t *doca_cred; 828 void (*doca_userfunc)(objset_t *os, void *arg, 829 cred_t *cr, dmu_tx_t *tx); 830 void *doca_userarg; 831 dmu_objset_type_t doca_type; 832 uint64_t doca_flags; 833} dmu_objset_create_arg_t; 834 835/*ARGSUSED*/ 836static int 837dmu_objset_create_check(void *arg, dmu_tx_t *tx) 838{ 839 dmu_objset_create_arg_t *doca = arg; 840 dsl_pool_t *dp = dmu_tx_pool(tx); 841 dsl_dir_t *pdd; 842 const char *tail; 843 int error; 844 845 if (strchr(doca->doca_name, '@') != NULL) 846 return (SET_ERROR(EINVAL)); 847 848 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN) 849 return (SET_ERROR(ENAMETOOLONG)); 850 851 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); 852 if (error != 0) 853 return (error); 854 if (tail == NULL) { 855 dsl_dir_rele(pdd, FTAG); 856 return (SET_ERROR(EEXIST)); 857 } 858 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 859 doca->doca_cred); 860 dsl_dir_rele(pdd, FTAG); 861 862 return (error); 863} 864 865static void 866dmu_objset_create_sync(void *arg, dmu_tx_t *tx) 867{ 868 dmu_objset_create_arg_t *doca = arg; 869 dsl_pool_t *dp = dmu_tx_pool(tx); 870 dsl_dir_t *pdd; 871 const char *tail; 872 dsl_dataset_t *ds; 873 uint64_t obj; 874 blkptr_t *bp; 875 objset_t *os; 876 877 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); 878 879 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, 880 doca->doca_cred, tx); 881 882 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 883 bp = dsl_dataset_get_blkptr(ds); 884 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa, 885 ds, bp, doca->doca_type, tx); 886 887 if (doca->doca_userfunc != NULL) { 888 doca->doca_userfunc(os, doca->doca_userarg, 889 doca->doca_cred, tx); 890 } 891 892 spa_history_log_internal_ds(ds, "create", tx, ""); 893 dsl_dataset_rele(ds, FTAG); 894 dsl_dir_rele(pdd, FTAG); 895} 896 897int 898dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 899 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg) 900{ 901 dmu_objset_create_arg_t doca; 902 903 doca.doca_name = name; 904 doca.doca_cred = CRED(); 905 doca.doca_flags = flags; 906 doca.doca_userfunc = func; 907 doca.doca_userarg = arg; 908 doca.doca_type = type; 909 910 return (dsl_sync_task(name, 911 dmu_objset_create_check, dmu_objset_create_sync, &doca, 912 5, ZFS_SPACE_CHECK_NORMAL)); 913} 914 915typedef struct dmu_objset_clone_arg { 916 const char *doca_clone; 917 const char *doca_origin; 918 cred_t *doca_cred; 919} dmu_objset_clone_arg_t; 920 921/*ARGSUSED*/ 922static int 923dmu_objset_clone_check(void *arg, dmu_tx_t *tx) 924{ 925 dmu_objset_clone_arg_t *doca = arg; 926 dsl_dir_t *pdd; 927 const char *tail; 928 int error; 929 dsl_dataset_t *origin; 930 dsl_pool_t *dp = dmu_tx_pool(tx); 931 932 if (strchr(doca->doca_clone, '@') != NULL) 933 return (SET_ERROR(EINVAL)); 934 935 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN) 936 return (SET_ERROR(ENAMETOOLONG)); 937 938 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); 939 if (error != 0) 940 return (error); 941 if (tail == NULL) { 942 dsl_dir_rele(pdd, FTAG); 943 return (SET_ERROR(EEXIST)); 944 } 945 946 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 947 doca->doca_cred); 948 if (error != 0) { 949 dsl_dir_rele(pdd, FTAG); 950 return (SET_ERROR(EDQUOT)); 951 } 952 dsl_dir_rele(pdd, FTAG); 953 954 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); 955 if (error != 0) 956 return (error); 957 958 /* You can only clone snapshots, not the head datasets. */ 959 if (!origin->ds_is_snapshot) { 960 dsl_dataset_rele(origin, FTAG); 961 return (SET_ERROR(EINVAL)); 962 } 963 dsl_dataset_rele(origin, FTAG); 964 965 return (0); 966} 967 968static void 969dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) 970{ 971 dmu_objset_clone_arg_t *doca = arg; 972 dsl_pool_t *dp = dmu_tx_pool(tx); 973 dsl_dir_t *pdd; 974 const char *tail; 975 dsl_dataset_t *origin, *ds; 976 uint64_t obj; 977 char namebuf[ZFS_MAX_DATASET_NAME_LEN]; 978 979 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); 980 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); 981 982 obj = dsl_dataset_create_sync(pdd, tail, origin, 0, 983 doca->doca_cred, tx); 984 985 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 986 dsl_dataset_name(origin, namebuf); 987 spa_history_log_internal_ds(ds, "clone", tx, 988 "origin=%s (%llu)", namebuf, origin->ds_object); 989 dsl_dataset_rele(ds, FTAG); 990 dsl_dataset_rele(origin, FTAG); 991 dsl_dir_rele(pdd, FTAG); 992} 993 994int 995dmu_objset_clone(const char *clone, const char *origin) 996{ 997 dmu_objset_clone_arg_t doca; 998 999 doca.doca_clone = clone; 1000 doca.doca_origin = origin; 1001 doca.doca_cred = CRED(); 1002 1003 return (dsl_sync_task(clone, 1004 dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 1005 5, ZFS_SPACE_CHECK_NORMAL)); 1006} 1007 1008int 1009dmu_objset_snapshot_one(const char *fsname, const char *snapname) 1010{ 1011 int err; 1012 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); 1013 nvlist_t *snaps = fnvlist_alloc(); 1014 1015 fnvlist_add_boolean(snaps, longsnap); 1016 strfree(longsnap); 1017 err = dsl_dataset_snapshot(snaps, NULL, NULL); 1018 fnvlist_free(snaps); 1019 return (err); 1020} 1021 1022static void 1023dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx) 1024{ 1025 dnode_t *dn; 1026 1027 while (dn = list_head(list)) { 1028 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 1029 ASSERT(dn->dn_dbuf->db_data_pending); 1030 /* 1031 * Initialize dn_zio outside dnode_sync() because the 1032 * meta-dnode needs to set it ouside dnode_sync(). 1033 */ 1034 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; 1035 ASSERT(dn->dn_zio); 1036 1037 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); 1038 list_remove(list, dn); 1039 1040 if (newlist) { 1041 (void) dnode_add_ref(dn, newlist); 1042 list_insert_tail(newlist, dn); 1043 } 1044 1045 dnode_sync(dn, tx); 1046 } 1047} 1048 1049/* ARGSUSED */ 1050static void 1051dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) 1052{ 1053 blkptr_t *bp = zio->io_bp; 1054 objset_t *os = arg; 1055 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; 1056 1057 ASSERT(!BP_IS_EMBEDDED(bp)); 1058 ASSERT3P(bp, ==, os->os_rootbp); 1059 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); 1060 ASSERT0(BP_GET_LEVEL(bp)); 1061 1062 /* 1063 * Update rootbp fill count: it should be the number of objects 1064 * allocated in the object set (not counting the "special" 1065 * objects that are stored in the objset_phys_t -- the meta 1066 * dnode and user/group accounting objects). 1067 */ 1068 bp->blk_fill = 0; 1069 for (int i = 0; i < dnp->dn_nblkptr; i++) 1070 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]); 1071} 1072 1073/* ARGSUSED */ 1074static void 1075dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) 1076{ 1077 blkptr_t *bp = zio->io_bp; 1078 blkptr_t *bp_orig = &zio->io_bp_orig; 1079 objset_t *os = arg; 1080 1081 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { 1082 ASSERT(BP_EQUAL(bp, bp_orig)); 1083 } else { 1084 dsl_dataset_t *ds = os->os_dsl_dataset; 1085 dmu_tx_t *tx = os->os_synctx; 1086 1087 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); 1088 dsl_dataset_block_born(ds, bp, tx); 1089 } 1090} 1091 1092/* called from dsl */ 1093void 1094dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) 1095{ 1096 int txgoff; 1097 zbookmark_phys_t zb; 1098 zio_prop_t zp; 1099 zio_t *zio; 1100 list_t *list; 1101 list_t *newlist = NULL; 1102 dbuf_dirty_record_t *dr; 1103 1104 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); 1105 1106 ASSERT(dmu_tx_is_syncing(tx)); 1107 /* XXX the write_done callback should really give us the tx... */ 1108 os->os_synctx = tx; 1109 1110 if (os->os_dsl_dataset == NULL) { 1111 /* 1112 * This is the MOS. If we have upgraded, 1113 * spa_max_replication() could change, so reset 1114 * os_copies here. 1115 */ 1116 os->os_copies = spa_max_replication(os->os_spa); 1117 } 1118 1119 /* 1120 * Create the root block IO 1121 */ 1122 SET_BOOKMARK(&zb, os->os_dsl_dataset ? 1123 os->os_dsl_dataset->ds_object : DMU_META_OBJSET, 1124 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 1125 arc_release(os->os_phys_buf, &os->os_phys_buf); 1126 1127 dmu_write_policy(os, NULL, 0, 0, &zp); 1128 1129 zio = arc_write(pio, os->os_spa, tx->tx_txg, 1130 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), 1131 DMU_OS_IS_L2COMPRESSIBLE(os), 1132 &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done, 1133 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb); 1134 1135 /* 1136 * Sync special dnodes - the parent IO for the sync is the root block 1137 */ 1138 DMU_META_DNODE(os)->dn_zio = zio; 1139 dnode_sync(DMU_META_DNODE(os), tx); 1140 1141 os->os_phys->os_flags = os->os_flags; 1142 1143 if (DMU_USERUSED_DNODE(os) && 1144 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1145 DMU_USERUSED_DNODE(os)->dn_zio = zio; 1146 dnode_sync(DMU_USERUSED_DNODE(os), tx); 1147 DMU_GROUPUSED_DNODE(os)->dn_zio = zio; 1148 dnode_sync(DMU_GROUPUSED_DNODE(os), tx); 1149 } 1150 1151 txgoff = tx->tx_txg & TXG_MASK; 1152 1153 if (dmu_objset_userused_enabled(os)) { 1154 newlist = &os->os_synced_dnodes; 1155 /* 1156 * We must create the list here because it uses the 1157 * dn_dirty_link[] of this txg. 1158 */ 1159 list_create(newlist, sizeof (dnode_t), 1160 offsetof(dnode_t, dn_dirty_link[txgoff])); 1161 } 1162 1163 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx); 1164 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx); 1165 1166 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; 1167 while (dr = list_head(list)) { 1168 ASSERT0(dr->dr_dbuf->db_level); 1169 list_remove(list, dr); 1170 if (dr->dr_zio) 1171 zio_nowait(dr->dr_zio); 1172 } 1173 /* 1174 * Free intent log blocks up to this tx. 1175 */ 1176 zil_sync(os->os_zil, tx); 1177 os->os_phys->os_zil_header = os->os_zil_header; 1178 zio_nowait(zio); 1179} 1180 1181boolean_t 1182dmu_objset_is_dirty(objset_t *os, uint64_t txg) 1183{ 1184 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) || 1185 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK])); 1186} 1187 1188static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES]; 1189 1190void 1191dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb) 1192{ 1193 used_cbs[ost] = cb; 1194} 1195 1196boolean_t 1197dmu_objset_userused_enabled(objset_t *os) 1198{ 1199 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && 1200 used_cbs[os->os_phys->os_type] != NULL && 1201 DMU_USERUSED_DNODE(os) != NULL); 1202} 1203 1204static void 1205do_userquota_update(objset_t *os, uint64_t used, uint64_t flags, 1206 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx) 1207{ 1208 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) { 1209 int64_t delta = DNODE_SIZE + used; 1210 if (subtract) 1211 delta = -delta; 1212 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT, 1213 user, delta, tx)); 1214 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT, 1215 group, delta, tx)); 1216 } 1217} 1218 1219void 1220dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) 1221{ 1222 dnode_t *dn; 1223 list_t *list = &os->os_synced_dnodes; 1224 1225 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os)); 1226 1227 while (dn = list_head(list)) { 1228 int flags; 1229 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); 1230 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || 1231 dn->dn_phys->dn_flags & 1232 DNODE_FLAG_USERUSED_ACCOUNTED); 1233 1234 /* Allocate the user/groupused objects if necessary. */ 1235 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 1236 VERIFY(0 == zap_create_claim(os, 1237 DMU_USERUSED_OBJECT, 1238 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 1239 VERIFY(0 == zap_create_claim(os, 1240 DMU_GROUPUSED_OBJECT, 1241 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 1242 } 1243 1244 /* 1245 * We intentionally modify the zap object even if the 1246 * net delta is zero. Otherwise 1247 * the block of the zap obj could be shared between 1248 * datasets but need to be different between them after 1249 * a bprewrite. 1250 */ 1251 1252 flags = dn->dn_id_flags; 1253 ASSERT(flags); 1254 if (flags & DN_ID_OLD_EXIST) { 1255 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags, 1256 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx); 1257 } 1258 if (flags & DN_ID_NEW_EXIST) { 1259 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys), 1260 dn->dn_phys->dn_flags, dn->dn_newuid, 1261 dn->dn_newgid, B_FALSE, tx); 1262 } 1263 1264 mutex_enter(&dn->dn_mtx); 1265 dn->dn_oldused = 0; 1266 dn->dn_oldflags = 0; 1267 if (dn->dn_id_flags & DN_ID_NEW_EXIST) { 1268 dn->dn_olduid = dn->dn_newuid; 1269 dn->dn_oldgid = dn->dn_newgid; 1270 dn->dn_id_flags |= DN_ID_OLD_EXIST; 1271 if (dn->dn_bonuslen == 0) 1272 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 1273 else 1274 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 1275 } 1276 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); 1277 mutex_exit(&dn->dn_mtx); 1278 1279 list_remove(list, dn); 1280 dnode_rele(dn, list); 1281 } 1282} 1283 1284/* 1285 * Returns a pointer to data to find uid/gid from 1286 * 1287 * If a dirty record for transaction group that is syncing can't 1288 * be found then NULL is returned. In the NULL case it is assumed 1289 * the uid/gid aren't changing. 1290 */ 1291static void * 1292dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) 1293{ 1294 dbuf_dirty_record_t *dr, **drp; 1295 void *data; 1296 1297 if (db->db_dirtycnt == 0) 1298 return (db->db.db_data); /* Nothing is changing */ 1299 1300 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) 1301 if (dr->dr_txg == tx->tx_txg) 1302 break; 1303 1304 if (dr == NULL) { 1305 data = NULL; 1306 } else { 1307 dnode_t *dn; 1308 1309 DB_DNODE_ENTER(dr->dr_dbuf); 1310 dn = DB_DNODE(dr->dr_dbuf); 1311 1312 if (dn->dn_bonuslen == 0 && 1313 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) 1314 data = dr->dt.dl.dr_data->b_data; 1315 else 1316 data = dr->dt.dl.dr_data; 1317 1318 DB_DNODE_EXIT(dr->dr_dbuf); 1319 } 1320 1321 return (data); 1322} 1323 1324void 1325dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) 1326{ 1327 objset_t *os = dn->dn_objset; 1328 void *data = NULL; 1329 dmu_buf_impl_t *db = NULL; 1330 uint64_t *user = NULL; 1331 uint64_t *group = NULL; 1332 int flags = dn->dn_id_flags; 1333 int error; 1334 boolean_t have_spill = B_FALSE; 1335 1336 if (!dmu_objset_userused_enabled(dn->dn_objset)) 1337 return; 1338 1339 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| 1340 DN_ID_CHKED_SPILL))) 1341 return; 1342 1343 if (before && dn->dn_bonuslen != 0) 1344 data = DN_BONUS(dn->dn_phys); 1345 else if (!before && dn->dn_bonuslen != 0) { 1346 if (dn->dn_bonus) { 1347 db = dn->dn_bonus; 1348 mutex_enter(&db->db_mtx); 1349 data = dmu_objset_userquota_find_data(db, tx); 1350 } else { 1351 data = DN_BONUS(dn->dn_phys); 1352 } 1353 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { 1354 int rf = 0; 1355 1356 if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) 1357 rf |= DB_RF_HAVESTRUCT; 1358 error = dmu_spill_hold_by_dnode(dn, 1359 rf | DB_RF_MUST_SUCCEED, 1360 FTAG, (dmu_buf_t **)&db); 1361 ASSERT(error == 0); 1362 mutex_enter(&db->db_mtx); 1363 data = (before) ? db->db.db_data : 1364 dmu_objset_userquota_find_data(db, tx); 1365 have_spill = B_TRUE; 1366 } else { 1367 mutex_enter(&dn->dn_mtx); 1368 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 1369 mutex_exit(&dn->dn_mtx); 1370 return; 1371 } 1372 1373 if (before) { 1374 ASSERT(data); 1375 user = &dn->dn_olduid; 1376 group = &dn->dn_oldgid; 1377 } else if (data) { 1378 user = &dn->dn_newuid; 1379 group = &dn->dn_newgid; 1380 } 1381 1382 /* 1383 * Must always call the callback in case the object 1384 * type has changed and that type isn't an object type to track 1385 */ 1386 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data, 1387 user, group); 1388 1389 /* 1390 * Preserve existing uid/gid when the callback can't determine 1391 * what the new uid/gid are and the callback returned EEXIST. 1392 * The EEXIST error tells us to just use the existing uid/gid. 1393 * If we don't know what the old values are then just assign 1394 * them to 0, since that is a new file being created. 1395 */ 1396 if (!before && data == NULL && error == EEXIST) { 1397 if (flags & DN_ID_OLD_EXIST) { 1398 dn->dn_newuid = dn->dn_olduid; 1399 dn->dn_newgid = dn->dn_oldgid; 1400 } else { 1401 dn->dn_newuid = 0; 1402 dn->dn_newgid = 0; 1403 } 1404 error = 0; 1405 } 1406 1407 if (db) 1408 mutex_exit(&db->db_mtx); 1409 1410 mutex_enter(&dn->dn_mtx); 1411 if (error == 0 && before) 1412 dn->dn_id_flags |= DN_ID_OLD_EXIST; 1413 if (error == 0 && !before) 1414 dn->dn_id_flags |= DN_ID_NEW_EXIST; 1415 1416 if (have_spill) { 1417 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 1418 } else { 1419 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 1420 } 1421 mutex_exit(&dn->dn_mtx); 1422 if (have_spill) 1423 dmu_buf_rele((dmu_buf_t *)db, FTAG); 1424} 1425 1426boolean_t 1427dmu_objset_userspace_present(objset_t *os) 1428{ 1429 return (os->os_phys->os_flags & 1430 OBJSET_FLAG_USERACCOUNTING_COMPLETE); 1431} 1432 1433int 1434dmu_objset_userspace_upgrade(objset_t *os) 1435{ 1436 uint64_t obj; 1437 int err = 0; 1438 1439 if (dmu_objset_userspace_present(os)) 1440 return (0); 1441 if (!dmu_objset_userused_enabled(os)) 1442 return (SET_ERROR(ENOTSUP)); 1443 if (dmu_objset_is_snapshot(os)) 1444 return (SET_ERROR(EINVAL)); 1445 1446 /* 1447 * We simply need to mark every object dirty, so that it will be 1448 * synced out and now accounted. If this is called 1449 * concurrently, or if we already did some work before crashing, 1450 * that's fine, since we track each object's accounted state 1451 * independently. 1452 */ 1453 1454 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { 1455 dmu_tx_t *tx; 1456 dmu_buf_t *db; 1457 int objerr; 1458 1459 if (issig(JUSTLOOKING) && issig(FORREAL)) 1460 return (SET_ERROR(EINTR)); 1461 1462 objerr = dmu_bonus_hold(os, obj, FTAG, &db); 1463 if (objerr != 0) 1464 continue; 1465 tx = dmu_tx_create(os); 1466 dmu_tx_hold_bonus(tx, obj); 1467 objerr = dmu_tx_assign(tx, TXG_WAIT); 1468 if (objerr != 0) { 1469 dmu_tx_abort(tx); 1470 continue; 1471 } 1472 dmu_buf_will_dirty(db, tx); 1473 dmu_buf_rele(db, FTAG); 1474 dmu_tx_commit(tx); 1475 } 1476 1477 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 1478 txg_wait_synced(dmu_objset_pool(os), 0); 1479 return (0); 1480} 1481 1482void 1483dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 1484 uint64_t *usedobjsp, uint64_t *availobjsp) 1485{ 1486 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, 1487 usedobjsp, availobjsp); 1488} 1489 1490uint64_t 1491dmu_objset_fsid_guid(objset_t *os) 1492{ 1493 return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); 1494} 1495 1496void 1497dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) 1498{ 1499 stat->dds_type = os->os_phys->os_type; 1500 if (os->os_dsl_dataset) 1501 dsl_dataset_fast_stat(os->os_dsl_dataset, stat); 1502} 1503 1504void 1505dmu_objset_stats(objset_t *os, nvlist_t *nv) 1506{ 1507 ASSERT(os->os_dsl_dataset || 1508 os->os_phys->os_type == DMU_OST_META); 1509 1510 if (os->os_dsl_dataset != NULL) 1511 dsl_dataset_stats(os->os_dsl_dataset, nv); 1512 1513 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, 1514 os->os_phys->os_type); 1515 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, 1516 dmu_objset_userspace_present(os)); 1517} 1518 1519int 1520dmu_objset_is_snapshot(objset_t *os) 1521{ 1522 if (os->os_dsl_dataset != NULL) 1523 return (os->os_dsl_dataset->ds_is_snapshot); 1524 else 1525 return (B_FALSE); 1526} 1527 1528int 1529dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen, 1530 boolean_t *conflict) 1531{ 1532 dsl_dataset_t *ds = os->os_dsl_dataset; 1533 uint64_t ignored; 1534 1535 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 1536 return (SET_ERROR(ENOENT)); 1537 1538 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, 1539 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, 1540 MT_FIRST, real, maxlen, conflict)); 1541} 1542 1543int 1544dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 1545 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) 1546{ 1547 dsl_dataset_t *ds = os->os_dsl_dataset; 1548 zap_cursor_t cursor; 1549 zap_attribute_t attr; 1550 1551 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 1552 1553 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 1554 return (SET_ERROR(ENOENT)); 1555 1556 zap_cursor_init_serialized(&cursor, 1557 ds->ds_dir->dd_pool->dp_meta_objset, 1558 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); 1559 1560 if (zap_cursor_retrieve(&cursor, &attr) != 0) { 1561 zap_cursor_fini(&cursor); 1562 return (SET_ERROR(ENOENT)); 1563 } 1564 1565 if (strlen(attr.za_name) + 1 > namelen) { 1566 zap_cursor_fini(&cursor); 1567 return (SET_ERROR(ENAMETOOLONG)); 1568 } 1569 1570 (void) strcpy(name, attr.za_name); 1571 if (idp) 1572 *idp = attr.za_first_integer; 1573 if (case_conflict) 1574 *case_conflict = attr.za_normalization_conflict; 1575 zap_cursor_advance(&cursor); 1576 *offp = zap_cursor_serialize(&cursor); 1577 zap_cursor_fini(&cursor); 1578 1579 return (0); 1580} 1581 1582int 1583dmu_dir_list_next(objset_t *os, int namelen, char *name, 1584 uint64_t *idp, uint64_t *offp) 1585{ 1586 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 1587 zap_cursor_t cursor; 1588 zap_attribute_t attr; 1589 1590 /* there is no next dir on a snapshot! */ 1591 if (os->os_dsl_dataset->ds_object != 1592 dsl_dir_phys(dd)->dd_head_dataset_obj) 1593 return (SET_ERROR(ENOENT)); 1594 1595 zap_cursor_init_serialized(&cursor, 1596 dd->dd_pool->dp_meta_objset, 1597 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); 1598 1599 if (zap_cursor_retrieve(&cursor, &attr) != 0) { 1600 zap_cursor_fini(&cursor); 1601 return (SET_ERROR(ENOENT)); 1602 } 1603 1604 if (strlen(attr.za_name) + 1 > namelen) { 1605 zap_cursor_fini(&cursor); 1606 return (SET_ERROR(ENAMETOOLONG)); 1607 } 1608 1609 (void) strcpy(name, attr.za_name); 1610 if (idp) 1611 *idp = attr.za_first_integer; 1612 zap_cursor_advance(&cursor); 1613 *offp = zap_cursor_serialize(&cursor); 1614 zap_cursor_fini(&cursor); 1615 1616 return (0); 1617} 1618 1619typedef struct dmu_objset_find_ctx { 1620 taskq_t *dc_tq; 1621 dsl_pool_t *dc_dp; 1622 uint64_t dc_ddobj; 1623 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *); 1624 void *dc_arg; 1625 int dc_flags; 1626 kmutex_t *dc_error_lock; 1627 int *dc_error; 1628} dmu_objset_find_ctx_t; 1629 1630static void 1631dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp) 1632{ 1633 dsl_pool_t *dp = dcp->dc_dp; 1634 dmu_objset_find_ctx_t *child_dcp; 1635 dsl_dir_t *dd; 1636 dsl_dataset_t *ds; 1637 zap_cursor_t zc; 1638 zap_attribute_t *attr; 1639 uint64_t thisobj; 1640 int err = 0; 1641 1642 /* don't process if there already was an error */ 1643 if (*dcp->dc_error != 0) 1644 goto out; 1645 1646 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, NULL, FTAG, &dd); 1647 if (err != 0) 1648 goto out; 1649 1650 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 1651 if (dd->dd_myname[0] == '$') { 1652 dsl_dir_rele(dd, FTAG); 1653 goto out; 1654 } 1655 1656 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 1657 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 1658 1659 /* 1660 * Iterate over all children. 1661 */ 1662 if (dcp->dc_flags & DS_FIND_CHILDREN) { 1663 for (zap_cursor_init(&zc, dp->dp_meta_objset, 1664 dsl_dir_phys(dd)->dd_child_dir_zapobj); 1665 zap_cursor_retrieve(&zc, attr) == 0; 1666 (void) zap_cursor_advance(&zc)) { 1667 ASSERT3U(attr->za_integer_length, ==, 1668 sizeof (uint64_t)); 1669 ASSERT3U(attr->za_num_integers, ==, 1); 1670 1671 child_dcp = kmem_alloc(sizeof (*child_dcp), KM_SLEEP); 1672 *child_dcp = *dcp; 1673 child_dcp->dc_ddobj = attr->za_first_integer; 1674 if (dcp->dc_tq != NULL) 1675 (void) taskq_dispatch(dcp->dc_tq, 1676 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP); 1677 else 1678 dmu_objset_find_dp_impl(child_dcp); 1679 } 1680 zap_cursor_fini(&zc); 1681 } 1682 1683 /* 1684 * Iterate over all snapshots. 1685 */ 1686 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) { 1687 dsl_dataset_t *ds; 1688 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 1689 1690 if (err == 0) { 1691 uint64_t snapobj; 1692 1693 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 1694 dsl_dataset_rele(ds, FTAG); 1695 1696 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 1697 zap_cursor_retrieve(&zc, attr) == 0; 1698 (void) zap_cursor_advance(&zc)) { 1699 ASSERT3U(attr->za_integer_length, ==, 1700 sizeof (uint64_t)); 1701 ASSERT3U(attr->za_num_integers, ==, 1); 1702 1703 err = dsl_dataset_hold_obj(dp, 1704 attr->za_first_integer, FTAG, &ds); 1705 if (err != 0) 1706 break; 1707 err = dcp->dc_func(dp, ds, dcp->dc_arg); 1708 dsl_dataset_rele(ds, FTAG); 1709 if (err != 0) 1710 break; 1711 } 1712 zap_cursor_fini(&zc); 1713 } 1714 } 1715 1716 dsl_dir_rele(dd, FTAG); 1717 kmem_free(attr, sizeof (zap_attribute_t)); 1718 1719 if (err != 0) 1720 goto out; 1721 1722 /* 1723 * Apply to self. 1724 */ 1725 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 1726 if (err != 0) 1727 goto out; 1728 err = dcp->dc_func(dp, ds, dcp->dc_arg); 1729 dsl_dataset_rele(ds, FTAG); 1730 1731out: 1732 if (err != 0) { 1733 mutex_enter(dcp->dc_error_lock); 1734 /* only keep first error */ 1735 if (*dcp->dc_error == 0) 1736 *dcp->dc_error = err; 1737 mutex_exit(dcp->dc_error_lock); 1738 } 1739 1740 kmem_free(dcp, sizeof (*dcp)); 1741} 1742 1743static void 1744dmu_objset_find_dp_cb(void *arg) 1745{ 1746 dmu_objset_find_ctx_t *dcp = arg; 1747 dsl_pool_t *dp = dcp->dc_dp; 1748 1749 /* 1750 * We need to get a pool_config_lock here, as there are several 1751 * asssert(pool_config_held) down the stack. Getting a lock via 1752 * dsl_pool_config_enter is risky, as it might be stalled by a 1753 * pending writer. This would deadlock, as the write lock can 1754 * only be granted when our parent thread gives up the lock. 1755 * The _prio interface gives us priority over a pending writer. 1756 */ 1757 dsl_pool_config_enter_prio(dp, FTAG); 1758 1759 dmu_objset_find_dp_impl(dcp); 1760 1761 dsl_pool_config_exit(dp, FTAG); 1762} 1763 1764/* 1765 * Find objsets under and including ddobj, call func(ds) on each. 1766 * The order for the enumeration is completely undefined. 1767 * func is called with dsl_pool_config held. 1768 */ 1769int 1770dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, 1771 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) 1772{ 1773 int error = 0; 1774 taskq_t *tq = NULL; 1775 int ntasks; 1776 dmu_objset_find_ctx_t *dcp; 1777 kmutex_t err_lock; 1778 1779 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL); 1780 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP); 1781 dcp->dc_tq = NULL; 1782 dcp->dc_dp = dp; 1783 dcp->dc_ddobj = ddobj; 1784 dcp->dc_func = func; 1785 dcp->dc_arg = arg; 1786 dcp->dc_flags = flags; 1787 dcp->dc_error_lock = &err_lock; 1788 dcp->dc_error = &error; 1789 1790 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) { 1791 /* 1792 * In case a write lock is held we can't make use of 1793 * parallelism, as down the stack of the worker threads 1794 * the lock is asserted via dsl_pool_config_held. 1795 * In case of a read lock this is solved by getting a read 1796 * lock in each worker thread, which isn't possible in case 1797 * of a writer lock. So we fall back to the synchronous path 1798 * here. 1799 * In the future it might be possible to get some magic into 1800 * dsl_pool_config_held in a way that it returns true for 1801 * the worker threads so that a single lock held from this 1802 * thread suffices. For now, stay single threaded. 1803 */ 1804 dmu_objset_find_dp_impl(dcp); 1805 mutex_destroy(&err_lock); 1806 1807 return (error); 1808 } 1809 1810 ntasks = dmu_find_threads; 1811 if (ntasks == 0) 1812 ntasks = vdev_count_leaves(dp->dp_spa) * 4; 1813 tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks, 1814 INT_MAX, 0); 1815 if (tq == NULL) { 1816 kmem_free(dcp, sizeof (*dcp)); 1817 mutex_destroy(&err_lock); 1818 1819 return (SET_ERROR(ENOMEM)); 1820 } 1821 dcp->dc_tq = tq; 1822 1823 /* dcp will be freed by task */ 1824 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP); 1825 1826 /* 1827 * PORTING: this code relies on the property of taskq_wait to wait 1828 * until no more tasks are queued and no more tasks are active. As 1829 * we always queue new tasks from within other tasks, task_wait 1830 * reliably waits for the full recursion to finish, even though we 1831 * enqueue new tasks after taskq_wait has been called. 1832 * On platforms other than illumos, taskq_wait may not have this 1833 * property. 1834 */ 1835 taskq_wait(tq); 1836 taskq_destroy(tq); 1837 mutex_destroy(&err_lock); 1838 1839 return (error); 1840} 1841 1842/* 1843 * Find all objsets under name, and for each, call 'func(child_name, arg)'. 1844 * The dp_config_rwlock must not be held when this is called, and it 1845 * will not be held when the callback is called. 1846 * Therefore this function should only be used when the pool is not changing 1847 * (e.g. in syncing context), or the callback can deal with the possible races. 1848 */ 1849static int 1850dmu_objset_find_impl(spa_t *spa, const char *name, 1851 int func(const char *, void *), void *arg, int flags) 1852{ 1853 dsl_dir_t *dd; 1854 dsl_pool_t *dp = spa_get_dsl(spa); 1855 dsl_dataset_t *ds; 1856 zap_cursor_t zc; 1857 zap_attribute_t *attr; 1858 char *child; 1859 uint64_t thisobj; 1860 int err; 1861 1862 dsl_pool_config_enter(dp, FTAG); 1863 1864 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); 1865 if (err != 0) { 1866 dsl_pool_config_exit(dp, FTAG); 1867 return (err); 1868 } 1869 1870 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 1871 if (dd->dd_myname[0] == '$') { 1872 dsl_dir_rele(dd, FTAG); 1873 dsl_pool_config_exit(dp, FTAG); 1874 return (0); 1875 } 1876 1877 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 1878 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 1879 1880 /* 1881 * Iterate over all children. 1882 */ 1883 if (flags & DS_FIND_CHILDREN) { 1884 for (zap_cursor_init(&zc, dp->dp_meta_objset, 1885 dsl_dir_phys(dd)->dd_child_dir_zapobj); 1886 zap_cursor_retrieve(&zc, attr) == 0; 1887 (void) zap_cursor_advance(&zc)) { 1888 ASSERT3U(attr->za_integer_length, ==, 1889 sizeof (uint64_t)); 1890 ASSERT3U(attr->za_num_integers, ==, 1); 1891 1892 child = kmem_asprintf("%s/%s", name, attr->za_name); 1893 dsl_pool_config_exit(dp, FTAG); 1894 err = dmu_objset_find_impl(spa, child, 1895 func, arg, flags); 1896 dsl_pool_config_enter(dp, FTAG); 1897 strfree(child); 1898 if (err != 0) 1899 break; 1900 } 1901 zap_cursor_fini(&zc); 1902 1903 if (err != 0) { 1904 dsl_dir_rele(dd, FTAG); 1905 dsl_pool_config_exit(dp, FTAG); 1906 kmem_free(attr, sizeof (zap_attribute_t)); 1907 return (err); 1908 } 1909 } 1910 1911 /* 1912 * Iterate over all snapshots. 1913 */ 1914 if (flags & DS_FIND_SNAPSHOTS) { 1915 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 1916 1917 if (err == 0) { 1918 uint64_t snapobj; 1919 1920 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 1921 dsl_dataset_rele(ds, FTAG); 1922 1923 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 1924 zap_cursor_retrieve(&zc, attr) == 0; 1925 (void) zap_cursor_advance(&zc)) { 1926 ASSERT3U(attr->za_integer_length, ==, 1927 sizeof (uint64_t)); 1928 ASSERT3U(attr->za_num_integers, ==, 1); 1929 1930 child = kmem_asprintf("%s@%s", 1931 name, attr->za_name); 1932 dsl_pool_config_exit(dp, FTAG); 1933 err = func(child, arg); 1934 dsl_pool_config_enter(dp, FTAG); 1935 strfree(child); 1936 if (err != 0) 1937 break; 1938 } 1939 zap_cursor_fini(&zc); 1940 } 1941 } 1942 1943 dsl_dir_rele(dd, FTAG); 1944 kmem_free(attr, sizeof (zap_attribute_t)); 1945 dsl_pool_config_exit(dp, FTAG); 1946 1947 if (err != 0) 1948 return (err); 1949 1950 /* Apply to self. */ 1951 return (func(name, arg)); 1952} 1953 1954/* 1955 * See comment above dmu_objset_find_impl(). 1956 */ 1957int 1958dmu_objset_find(char *name, int func(const char *, void *), void *arg, 1959 int flags) 1960{ 1961 spa_t *spa; 1962 int error; 1963 1964 error = spa_open(name, &spa, FTAG); 1965 if (error != 0) 1966 return (error); 1967 error = dmu_objset_find_impl(spa, name, func, arg, flags); 1968 spa_close(spa, FTAG); 1969 return (error); 1970} 1971 1972void 1973dmu_objset_set_user(objset_t *os, void *user_ptr) 1974{ 1975 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 1976 os->os_user_ptr = user_ptr; 1977} 1978 1979void * 1980dmu_objset_get_user(objset_t *os) 1981{ 1982 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 1983 return (os->os_user_ptr); 1984} 1985 1986/* 1987 * Determine name of filesystem, given name of snapshot. 1988 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes 1989 */ 1990int 1991dmu_fsname(const char *snapname, char *buf) 1992{ 1993 char *atp = strchr(snapname, '@'); 1994 if (atp == NULL) 1995 return (SET_ERROR(EINVAL)); 1996 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN) 1997 return (SET_ERROR(ENAMETOOLONG)); 1998 (void) strlcpy(buf, snapname, atp - snapname + 1); 1999 return (0); 2000} 2001