dmu_objset.c revision 288542
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, 2014 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 2015 Nexenta Systems, Inc. All rights reserved. 27 */ 28 29/* Portions Copyright 2010 Robert Milkowski */ 30 31#include <sys/cred.h> 32#include <sys/zfs_context.h> 33#include <sys/dmu_objset.h> 34#include <sys/dsl_dir.h> 35#include <sys/dsl_dataset.h> 36#include <sys/dsl_prop.h> 37#include <sys/dsl_pool.h> 38#include <sys/dsl_synctask.h> 39#include <sys/dsl_deleg.h> 40#include <sys/dnode.h> 41#include <sys/dbuf.h> 42#include <sys/zvol.h> 43#include <sys/dmu_tx.h> 44#include <sys/zap.h> 45#include <sys/zil.h> 46#include <sys/dmu_impl.h> 47#include <sys/zfs_ioctl.h> 48#include <sys/sa.h> 49#include <sys/zfs_onexit.h> 50#include <sys/dsl_destroy.h> 51 52/* 53 * Needed to close a window in dnode_move() that allows the objset to be freed 54 * before it can be safely accessed. 55 */ 56krwlock_t os_lock; 57 58void 59dmu_objset_init(void) 60{ 61 rw_init(&os_lock, NULL, RW_DEFAULT, NULL); 62} 63 64void 65dmu_objset_fini(void) 66{ 67 rw_destroy(&os_lock); 68} 69 70spa_t * 71dmu_objset_spa(objset_t *os) 72{ 73 return (os->os_spa); 74} 75 76zilog_t * 77dmu_objset_zil(objset_t *os) 78{ 79 return (os->os_zil); 80} 81 82dsl_pool_t * 83dmu_objset_pool(objset_t *os) 84{ 85 dsl_dataset_t *ds; 86 87 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir) 88 return (ds->ds_dir->dd_pool); 89 else 90 return (spa_get_dsl(os->os_spa)); 91} 92 93dsl_dataset_t * 94dmu_objset_ds(objset_t *os) 95{ 96 return (os->os_dsl_dataset); 97} 98 99dmu_objset_type_t 100dmu_objset_type(objset_t *os) 101{ 102 return (os->os_phys->os_type); 103} 104 105void 106dmu_objset_name(objset_t *os, char *buf) 107{ 108 dsl_dataset_name(os->os_dsl_dataset, buf); 109} 110 111uint64_t 112dmu_objset_id(objset_t *os) 113{ 114 dsl_dataset_t *ds = os->os_dsl_dataset; 115 116 return (ds ? ds->ds_object : 0); 117} 118 119zfs_sync_type_t 120dmu_objset_syncprop(objset_t *os) 121{ 122 return (os->os_sync); 123} 124 125zfs_logbias_op_t 126dmu_objset_logbias(objset_t *os) 127{ 128 return (os->os_logbias); 129} 130 131static void 132checksum_changed_cb(void *arg, uint64_t newval) 133{ 134 objset_t *os = arg; 135 136 /* 137 * Inheritance should have been done by now. 138 */ 139 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 140 141 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE); 142} 143 144static void 145compression_changed_cb(void *arg, uint64_t newval) 146{ 147 objset_t *os = arg; 148 149 /* 150 * Inheritance and range checking should have been done by now. 151 */ 152 ASSERT(newval != ZIO_COMPRESS_INHERIT); 153 154 os->os_compress = zio_compress_select(os->os_spa, newval, 155 ZIO_COMPRESS_ON); 156} 157 158static void 159copies_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 > 0); 167 ASSERT(newval <= spa_max_replication(os->os_spa)); 168 169 os->os_copies = newval; 170} 171 172static void 173dedup_changed_cb(void *arg, uint64_t newval) 174{ 175 objset_t *os = arg; 176 spa_t *spa = os->os_spa; 177 enum zio_checksum checksum; 178 179 /* 180 * Inheritance should have been done by now. 181 */ 182 ASSERT(newval != ZIO_CHECKSUM_INHERIT); 183 184 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF); 185 186 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK; 187 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY); 188} 189 190static void 191primary_cache_changed_cb(void *arg, uint64_t newval) 192{ 193 objset_t *os = arg; 194 195 /* 196 * Inheritance and range checking should have been done by now. 197 */ 198 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE || 199 newval == ZFS_CACHE_METADATA); 200 201 os->os_primary_cache = newval; 202} 203 204static void 205secondary_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_secondary_cache = newval; 216} 217 218static void 219sync_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_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS || 227 newval == ZFS_SYNC_DISABLED); 228 229 os->os_sync = newval; 230 if (os->os_zil) 231 zil_set_sync(os->os_zil, newval); 232} 233 234static void 235redundant_metadata_changed_cb(void *arg, uint64_t newval) 236{ 237 objset_t *os = arg; 238 239 /* 240 * Inheritance and range checking should have been done by now. 241 */ 242 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL || 243 newval == ZFS_REDUNDANT_METADATA_MOST); 244 245 os->os_redundant_metadata = newval; 246} 247 248static void 249logbias_changed_cb(void *arg, uint64_t newval) 250{ 251 objset_t *os = arg; 252 253 ASSERT(newval == ZFS_LOGBIAS_LATENCY || 254 newval == ZFS_LOGBIAS_THROUGHPUT); 255 os->os_logbias = newval; 256 if (os->os_zil) 257 zil_set_logbias(os->os_zil, newval); 258} 259 260static void 261recordsize_changed_cb(void *arg, uint64_t newval) 262{ 263 objset_t *os = arg; 264 265 os->os_recordsize = newval; 266} 267 268void 269dmu_objset_byteswap(void *buf, size_t size) 270{ 271 objset_phys_t *osp = buf; 272 273 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t)); 274 dnode_byteswap(&osp->os_meta_dnode); 275 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t)); 276 osp->os_type = BSWAP_64(osp->os_type); 277 osp->os_flags = BSWAP_64(osp->os_flags); 278 if (size == sizeof (objset_phys_t)) { 279 dnode_byteswap(&osp->os_userused_dnode); 280 dnode_byteswap(&osp->os_groupused_dnode); 281 } 282} 283 284int 285dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 286 objset_t **osp) 287{ 288 objset_t *os; 289 int i, err; 290 291 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock)); 292 293 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP); 294 os->os_dsl_dataset = ds; 295 os->os_spa = spa; 296 os->os_rootbp = bp; 297 if (!BP_IS_HOLE(os->os_rootbp)) { 298 arc_flags_t aflags = ARC_FLAG_WAIT; 299 zbookmark_phys_t zb; 300 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, 301 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 302 303 if (DMU_OS_IS_L2CACHEABLE(os)) 304 aflags |= ARC_FLAG_L2CACHE; 305 if (DMU_OS_IS_L2COMPRESSIBLE(os)) 306 aflags |= ARC_FLAG_L2COMPRESS; 307 308 dprintf_bp(os->os_rootbp, "reading %s", ""); 309 err = arc_read(NULL, spa, os->os_rootbp, 310 arc_getbuf_func, &os->os_phys_buf, 311 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb); 312 if (err != 0) { 313 kmem_free(os, sizeof (objset_t)); 314 /* convert checksum errors into IO errors */ 315 if (err == ECKSUM) 316 err = SET_ERROR(EIO); 317 return (err); 318 } 319 320 /* Increase the blocksize if we are permitted. */ 321 if (spa_version(spa) >= SPA_VERSION_USERSPACE && 322 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) { 323 arc_buf_t *buf = arc_buf_alloc(spa, 324 sizeof (objset_phys_t), &os->os_phys_buf, 325 ARC_BUFC_METADATA); 326 bzero(buf->b_data, sizeof (objset_phys_t)); 327 bcopy(os->os_phys_buf->b_data, buf->b_data, 328 arc_buf_size(os->os_phys_buf)); 329 (void) arc_buf_remove_ref(os->os_phys_buf, 330 &os->os_phys_buf); 331 os->os_phys_buf = buf; 332 } 333 334 os->os_phys = os->os_phys_buf->b_data; 335 os->os_flags = os->os_phys->os_flags; 336 } else { 337 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ? 338 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE; 339 os->os_phys_buf = arc_buf_alloc(spa, size, 340 &os->os_phys_buf, ARC_BUFC_METADATA); 341 os->os_phys = os->os_phys_buf->b_data; 342 bzero(os->os_phys, size); 343 } 344 345 /* 346 * Note: the changed_cb will be called once before the register 347 * func returns, thus changing the checksum/compression from the 348 * default (fletcher2/off). Snapshots don't need to know about 349 * checksum/compression/copies. 350 */ 351 if (ds != NULL) { 352 err = dsl_prop_register(ds, 353 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), 354 primary_cache_changed_cb, os); 355 if (err == 0) { 356 err = dsl_prop_register(ds, 357 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), 358 secondary_cache_changed_cb, os); 359 } 360 if (!dsl_dataset_is_snapshot(ds)) { 361 if (err == 0) { 362 err = dsl_prop_register(ds, 363 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 364 checksum_changed_cb, os); 365 } 366 if (err == 0) { 367 err = dsl_prop_register(ds, 368 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 369 compression_changed_cb, os); 370 } 371 if (err == 0) { 372 err = dsl_prop_register(ds, 373 zfs_prop_to_name(ZFS_PROP_COPIES), 374 copies_changed_cb, os); 375 } 376 if (err == 0) { 377 err = dsl_prop_register(ds, 378 zfs_prop_to_name(ZFS_PROP_DEDUP), 379 dedup_changed_cb, os); 380 } 381 if (err == 0) { 382 err = dsl_prop_register(ds, 383 zfs_prop_to_name(ZFS_PROP_LOGBIAS), 384 logbias_changed_cb, os); 385 } 386 if (err == 0) { 387 err = dsl_prop_register(ds, 388 zfs_prop_to_name(ZFS_PROP_SYNC), 389 sync_changed_cb, os); 390 } 391 if (err == 0) { 392 err = dsl_prop_register(ds, 393 zfs_prop_to_name( 394 ZFS_PROP_REDUNDANT_METADATA), 395 redundant_metadata_changed_cb, os); 396 } 397 if (err == 0) { 398 err = dsl_prop_register(ds, 399 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 400 recordsize_changed_cb, os); 401 } 402 } 403 if (err != 0) { 404 VERIFY(arc_buf_remove_ref(os->os_phys_buf, 405 &os->os_phys_buf)); 406 kmem_free(os, sizeof (objset_t)); 407 return (err); 408 } 409 } else { 410 /* It's the meta-objset. */ 411 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4; 412 os->os_compress = ZIO_COMPRESS_ON; 413 os->os_copies = spa_max_replication(spa); 414 os->os_dedup_checksum = ZIO_CHECKSUM_OFF; 415 os->os_dedup_verify = B_FALSE; 416 os->os_logbias = ZFS_LOGBIAS_LATENCY; 417 os->os_sync = ZFS_SYNC_STANDARD; 418 os->os_primary_cache = ZFS_CACHE_ALL; 419 os->os_secondary_cache = ZFS_CACHE_ALL; 420 } 421 422 if (ds == NULL || !dsl_dataset_is_snapshot(ds)) 423 os->os_zil_header = os->os_phys->os_zil_header; 424 os->os_zil = zil_alloc(os, &os->os_zil_header); 425 426 for (i = 0; i < TXG_SIZE; i++) { 427 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t), 428 offsetof(dnode_t, dn_dirty_link[i])); 429 list_create(&os->os_free_dnodes[i], sizeof (dnode_t), 430 offsetof(dnode_t, dn_dirty_link[i])); 431 } 432 list_create(&os->os_dnodes, sizeof (dnode_t), 433 offsetof(dnode_t, dn_link)); 434 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t), 435 offsetof(dmu_buf_impl_t, db_link)); 436 437 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL); 438 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL); 439 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL); 440 441 DMU_META_DNODE(os) = dnode_special_open(os, 442 &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT, 443 &os->os_meta_dnode); 444 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) { 445 DMU_USERUSED_DNODE(os) = dnode_special_open(os, 446 &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT, 447 &os->os_userused_dnode); 448 DMU_GROUPUSED_DNODE(os) = dnode_special_open(os, 449 &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT, 450 &os->os_groupused_dnode); 451 } 452 453 *osp = os; 454 return (0); 455} 456 457int 458dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp) 459{ 460 int err = 0; 461 462 mutex_enter(&ds->ds_opening_lock); 463 if (ds->ds_objset == NULL) { 464 objset_t *os; 465 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds), 466 ds, dsl_dataset_get_blkptr(ds), &os); 467 468 if (err == 0) { 469 mutex_enter(&ds->ds_lock); 470 ASSERT(ds->ds_objset == NULL); 471 ds->ds_objset = os; 472 mutex_exit(&ds->ds_lock); 473 } 474 } 475 *osp = ds->ds_objset; 476 mutex_exit(&ds->ds_opening_lock); 477 return (err); 478} 479 480/* 481 * Holds the pool while the objset is held. Therefore only one objset 482 * can be held at a time. 483 */ 484int 485dmu_objset_hold(const char *name, void *tag, objset_t **osp) 486{ 487 dsl_pool_t *dp; 488 dsl_dataset_t *ds; 489 int err; 490 491 err = dsl_pool_hold(name, tag, &dp); 492 if (err != 0) 493 return (err); 494 err = dsl_dataset_hold(dp, name, tag, &ds); 495 if (err != 0) { 496 dsl_pool_rele(dp, tag); 497 return (err); 498 } 499 500 err = dmu_objset_from_ds(ds, osp); 501 if (err != 0) { 502 dsl_dataset_rele(ds, tag); 503 dsl_pool_rele(dp, tag); 504 } 505 506 return (err); 507} 508 509/* 510 * dsl_pool must not be held when this is called. 511 * Upon successful return, there will be a longhold on the dataset, 512 * and the dsl_pool will not be held. 513 */ 514int 515dmu_objset_own(const char *name, dmu_objset_type_t type, 516 boolean_t readonly, 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, FTAG, &dp); 523 if (err != 0) 524 return (err); 525 err = dsl_dataset_own(dp, name, tag, &ds); 526 if (err != 0) { 527 dsl_pool_rele(dp, FTAG); 528 return (err); 529 } 530 531 err = dmu_objset_from_ds(ds, osp); 532 dsl_pool_rele(dp, FTAG); 533 if (err != 0) { 534 dsl_dataset_disown(ds, tag); 535 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) { 536 dsl_dataset_disown(ds, tag); 537 return (SET_ERROR(EINVAL)); 538 } else if (!readonly && dsl_dataset_is_snapshot(ds)) { 539 dsl_dataset_disown(ds, tag); 540 return (SET_ERROR(EROFS)); 541 } 542 return (err); 543} 544 545void 546dmu_objset_rele(objset_t *os, void *tag) 547{ 548 dsl_pool_t *dp = dmu_objset_pool(os); 549 dsl_dataset_rele(os->os_dsl_dataset, tag); 550 dsl_pool_rele(dp, tag); 551} 552 553/* 554 * When we are called, os MUST refer to an objset associated with a dataset 555 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner 556 * == tag. We will then release and reacquire ownership of the dataset while 557 * holding the pool config_rwlock to avoid intervening namespace or ownership 558 * changes may occur. 559 * 560 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to 561 * release the hold on its dataset and acquire a new one on the dataset of the 562 * same name so that it can be partially torn down and reconstructed. 563 */ 564void 565dmu_objset_refresh_ownership(objset_t *os, void *tag) 566{ 567 dsl_pool_t *dp; 568 dsl_dataset_t *ds, *newds; 569 char name[MAXNAMELEN]; 570 571 ds = os->os_dsl_dataset; 572 VERIFY3P(ds, !=, NULL); 573 VERIFY3P(ds->ds_owner, ==, tag); 574 VERIFY(dsl_dataset_long_held(ds)); 575 576 dsl_dataset_name(ds, name); 577 dp = dmu_objset_pool(os); 578 dsl_pool_config_enter(dp, FTAG); 579 dmu_objset_disown(os, tag); 580 VERIFY0(dsl_dataset_own(dp, name, tag, &newds)); 581 VERIFY3P(newds, ==, os->os_dsl_dataset); 582 dsl_pool_config_exit(dp, FTAG); 583} 584 585void 586dmu_objset_disown(objset_t *os, void *tag) 587{ 588 dsl_dataset_disown(os->os_dsl_dataset, tag); 589} 590 591void 592dmu_objset_evict_dbufs(objset_t *os) 593{ 594 dnode_t *dn; 595 596 mutex_enter(&os->os_lock); 597 598 /* process the mdn last, since the other dnodes have holds on it */ 599 list_remove(&os->os_dnodes, DMU_META_DNODE(os)); 600 list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os)); 601 602 /* 603 * Find the first dnode with holds. We have to do this dance 604 * because dnode_add_ref() only works if you already have a 605 * hold. If there are no holds then it has no dbufs so OK to 606 * skip. 607 */ 608 for (dn = list_head(&os->os_dnodes); 609 dn && !dnode_add_ref(dn, FTAG); 610 dn = list_next(&os->os_dnodes, dn)) 611 continue; 612 613 while (dn) { 614 dnode_t *next_dn = dn; 615 616 do { 617 next_dn = list_next(&os->os_dnodes, next_dn); 618 } while (next_dn && !dnode_add_ref(next_dn, FTAG)); 619 620 mutex_exit(&os->os_lock); 621 dnode_evict_dbufs(dn); 622 dnode_rele(dn, FTAG); 623 mutex_enter(&os->os_lock); 624 dn = next_dn; 625 } 626 mutex_exit(&os->os_lock); 627} 628 629void 630dmu_objset_evict(objset_t *os) 631{ 632 dsl_dataset_t *ds = os->os_dsl_dataset; 633 634 for (int t = 0; t < TXG_SIZE; t++) 635 ASSERT(!dmu_objset_is_dirty(os, t)); 636 637 if (ds) { 638 if (!dsl_dataset_is_snapshot(ds)) { 639 VERIFY0(dsl_prop_unregister(ds, 640 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 641 checksum_changed_cb, os)); 642 VERIFY0(dsl_prop_unregister(ds, 643 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 644 compression_changed_cb, os)); 645 VERIFY0(dsl_prop_unregister(ds, 646 zfs_prop_to_name(ZFS_PROP_COPIES), 647 copies_changed_cb, os)); 648 VERIFY0(dsl_prop_unregister(ds, 649 zfs_prop_to_name(ZFS_PROP_DEDUP), 650 dedup_changed_cb, os)); 651 VERIFY0(dsl_prop_unregister(ds, 652 zfs_prop_to_name(ZFS_PROP_LOGBIAS), 653 logbias_changed_cb, os)); 654 VERIFY0(dsl_prop_unregister(ds, 655 zfs_prop_to_name(ZFS_PROP_SYNC), 656 sync_changed_cb, os)); 657 VERIFY0(dsl_prop_unregister(ds, 658 zfs_prop_to_name(ZFS_PROP_REDUNDANT_METADATA), 659 redundant_metadata_changed_cb, os)); 660 VERIFY0(dsl_prop_unregister(ds, 661 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 662 recordsize_changed_cb, os)); 663 } 664 VERIFY0(dsl_prop_unregister(ds, 665 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE), 666 primary_cache_changed_cb, os)); 667 VERIFY0(dsl_prop_unregister(ds, 668 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE), 669 secondary_cache_changed_cb, os)); 670 } 671 672 if (os->os_sa) 673 sa_tear_down(os); 674 675 dmu_objset_evict_dbufs(os); 676 677 dnode_special_close(&os->os_meta_dnode); 678 if (DMU_USERUSED_DNODE(os)) { 679 dnode_special_close(&os->os_userused_dnode); 680 dnode_special_close(&os->os_groupused_dnode); 681 } 682 zil_free(os->os_zil); 683 684 ASSERT3P(list_head(&os->os_dnodes), ==, NULL); 685 686 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf)); 687 688 /* 689 * This is a barrier to prevent the objset from going away in 690 * dnode_move() until we can safely ensure that the objset is still in 691 * use. We consider the objset valid before the barrier and invalid 692 * after the barrier. 693 */ 694 rw_enter(&os_lock, RW_READER); 695 rw_exit(&os_lock); 696 697 mutex_destroy(&os->os_lock); 698 mutex_destroy(&os->os_obj_lock); 699 mutex_destroy(&os->os_user_ptr_lock); 700 kmem_free(os, sizeof (objset_t)); 701} 702 703timestruc_t 704dmu_objset_snap_cmtime(objset_t *os) 705{ 706 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir)); 707} 708 709/* called from dsl for meta-objset */ 710objset_t * 711dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp, 712 dmu_objset_type_t type, dmu_tx_t *tx) 713{ 714 objset_t *os; 715 dnode_t *mdn; 716 717 ASSERT(dmu_tx_is_syncing(tx)); 718 719 if (ds != NULL) 720 VERIFY0(dmu_objset_from_ds(ds, &os)); 721 else 722 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os)); 723 724 mdn = DMU_META_DNODE(os); 725 726 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT, 727 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx); 728 729 /* 730 * We don't want to have to increase the meta-dnode's nlevels 731 * later, because then we could do it in quescing context while 732 * we are also accessing it in open context. 733 * 734 * This precaution is not necessary for the MOS (ds == NULL), 735 * because the MOS is only updated in syncing context. 736 * This is most fortunate: the MOS is the only objset that 737 * needs to be synced multiple times as spa_sync() iterates 738 * to convergence, so minimizing its dn_nlevels matters. 739 */ 740 if (ds != NULL) { 741 int levels = 1; 742 743 /* 744 * Determine the number of levels necessary for the meta-dnode 745 * to contain DN_MAX_OBJECT dnodes. 746 */ 747 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift + 748 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) < 749 DN_MAX_OBJECT * sizeof (dnode_phys_t)) 750 levels++; 751 752 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] = 753 mdn->dn_nlevels = levels; 754 } 755 756 ASSERT(type != DMU_OST_NONE); 757 ASSERT(type != DMU_OST_ANY); 758 ASSERT(type < DMU_OST_NUMTYPES); 759 os->os_phys->os_type = type; 760 if (dmu_objset_userused_enabled(os)) { 761 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 762 os->os_flags = os->os_phys->os_flags; 763 } 764 765 dsl_dataset_dirty(ds, tx); 766 767 return (os); 768} 769 770typedef struct dmu_objset_create_arg { 771 const char *doca_name; 772 cred_t *doca_cred; 773 void (*doca_userfunc)(objset_t *os, void *arg, 774 cred_t *cr, dmu_tx_t *tx); 775 void *doca_userarg; 776 dmu_objset_type_t doca_type; 777 uint64_t doca_flags; 778} dmu_objset_create_arg_t; 779 780/*ARGSUSED*/ 781static int 782dmu_objset_create_check(void *arg, dmu_tx_t *tx) 783{ 784 dmu_objset_create_arg_t *doca = arg; 785 dsl_pool_t *dp = dmu_tx_pool(tx); 786 dsl_dir_t *pdd; 787 const char *tail; 788 int error; 789 790 if (strchr(doca->doca_name, '@') != NULL) 791 return (SET_ERROR(EINVAL)); 792 793 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail); 794 if (error != 0) 795 return (error); 796 if (tail == NULL) { 797 dsl_dir_rele(pdd, FTAG); 798 return (SET_ERROR(EEXIST)); 799 } 800 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 801 doca->doca_cred); 802 dsl_dir_rele(pdd, FTAG); 803 804 return (error); 805} 806 807static void 808dmu_objset_create_sync(void *arg, dmu_tx_t *tx) 809{ 810 dmu_objset_create_arg_t *doca = arg; 811 dsl_pool_t *dp = dmu_tx_pool(tx); 812 dsl_dir_t *pdd; 813 const char *tail; 814 dsl_dataset_t *ds; 815 uint64_t obj; 816 blkptr_t *bp; 817 objset_t *os; 818 819 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail)); 820 821 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags, 822 doca->doca_cred, tx); 823 824 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 825 bp = dsl_dataset_get_blkptr(ds); 826 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa, 827 ds, bp, doca->doca_type, tx); 828 829 if (doca->doca_userfunc != NULL) { 830 doca->doca_userfunc(os, doca->doca_userarg, 831 doca->doca_cred, tx); 832 } 833 834 spa_history_log_internal_ds(ds, "create", tx, ""); 835 dsl_dataset_rele(ds, FTAG); 836 dsl_dir_rele(pdd, FTAG); 837} 838 839int 840dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags, 841 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg) 842{ 843 dmu_objset_create_arg_t doca; 844 845 doca.doca_name = name; 846 doca.doca_cred = CRED(); 847 doca.doca_flags = flags; 848 doca.doca_userfunc = func; 849 doca.doca_userarg = arg; 850 doca.doca_type = type; 851 852 return (dsl_sync_task(name, 853 dmu_objset_create_check, dmu_objset_create_sync, &doca, 854 5, ZFS_SPACE_CHECK_NORMAL)); 855} 856 857typedef struct dmu_objset_clone_arg { 858 const char *doca_clone; 859 const char *doca_origin; 860 cred_t *doca_cred; 861} dmu_objset_clone_arg_t; 862 863/*ARGSUSED*/ 864static int 865dmu_objset_clone_check(void *arg, dmu_tx_t *tx) 866{ 867 dmu_objset_clone_arg_t *doca = arg; 868 dsl_dir_t *pdd; 869 const char *tail; 870 int error; 871 dsl_dataset_t *origin; 872 dsl_pool_t *dp = dmu_tx_pool(tx); 873 874 if (strchr(doca->doca_clone, '@') != NULL) 875 return (SET_ERROR(EINVAL)); 876 877 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail); 878 if (error != 0) 879 return (error); 880 if (tail == NULL) { 881 dsl_dir_rele(pdd, FTAG); 882 return (SET_ERROR(EEXIST)); 883 } 884 885 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL, 886 doca->doca_cred); 887 if (error != 0) { 888 dsl_dir_rele(pdd, FTAG); 889 return (SET_ERROR(EDQUOT)); 890 } 891 dsl_dir_rele(pdd, FTAG); 892 893 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin); 894 if (error != 0) 895 return (error); 896 897 /* You can only clone snapshots, not the head datasets. */ 898 if (!dsl_dataset_is_snapshot(origin)) { 899 dsl_dataset_rele(origin, FTAG); 900 return (SET_ERROR(EINVAL)); 901 } 902 dsl_dataset_rele(origin, FTAG); 903 904 return (0); 905} 906 907static void 908dmu_objset_clone_sync(void *arg, dmu_tx_t *tx) 909{ 910 dmu_objset_clone_arg_t *doca = arg; 911 dsl_pool_t *dp = dmu_tx_pool(tx); 912 dsl_dir_t *pdd; 913 const char *tail; 914 dsl_dataset_t *origin, *ds; 915 uint64_t obj; 916 char namebuf[MAXNAMELEN]; 917 918 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail)); 919 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin)); 920 921 obj = dsl_dataset_create_sync(pdd, tail, origin, 0, 922 doca->doca_cred, tx); 923 924 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds)); 925 dsl_dataset_name(origin, namebuf); 926 spa_history_log_internal_ds(ds, "clone", tx, 927 "origin=%s (%llu)", namebuf, origin->ds_object); 928 dsl_dataset_rele(ds, FTAG); 929 dsl_dataset_rele(origin, FTAG); 930 dsl_dir_rele(pdd, FTAG); 931} 932 933int 934dmu_objset_clone(const char *clone, const char *origin) 935{ 936 dmu_objset_clone_arg_t doca; 937 938 doca.doca_clone = clone; 939 doca.doca_origin = origin; 940 doca.doca_cred = CRED(); 941 942 return (dsl_sync_task(clone, 943 dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 944 5, ZFS_SPACE_CHECK_NORMAL)); 945} 946 947int 948dmu_objset_snapshot_one(const char *fsname, const char *snapname) 949{ 950 int err; 951 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname); 952 nvlist_t *snaps = fnvlist_alloc(); 953 954 fnvlist_add_boolean(snaps, longsnap); 955 strfree(longsnap); 956 err = dsl_dataset_snapshot(snaps, NULL, NULL); 957 fnvlist_free(snaps); 958 return (err); 959} 960 961static void 962dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx) 963{ 964 dnode_t *dn; 965 966 while (dn = list_head(list)) { 967 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT); 968 ASSERT(dn->dn_dbuf->db_data_pending); 969 /* 970 * Initialize dn_zio outside dnode_sync() because the 971 * meta-dnode needs to set it ouside dnode_sync(). 972 */ 973 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio; 974 ASSERT(dn->dn_zio); 975 976 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS); 977 list_remove(list, dn); 978 979 if (newlist) { 980 (void) dnode_add_ref(dn, newlist); 981 list_insert_tail(newlist, dn); 982 } 983 984 dnode_sync(dn, tx); 985 } 986} 987 988/* ARGSUSED */ 989static void 990dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg) 991{ 992 blkptr_t *bp = zio->io_bp; 993 objset_t *os = arg; 994 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode; 995 996 ASSERT(!BP_IS_EMBEDDED(bp)); 997 ASSERT3P(bp, ==, os->os_rootbp); 998 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET); 999 ASSERT0(BP_GET_LEVEL(bp)); 1000 1001 /* 1002 * Update rootbp fill count: it should be the number of objects 1003 * allocated in the object set (not counting the "special" 1004 * objects that are stored in the objset_phys_t -- the meta 1005 * dnode and user/group accounting objects). 1006 */ 1007 bp->blk_fill = 0; 1008 for (int i = 0; i < dnp->dn_nblkptr; i++) 1009 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]); 1010} 1011 1012/* ARGSUSED */ 1013static void 1014dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg) 1015{ 1016 blkptr_t *bp = zio->io_bp; 1017 blkptr_t *bp_orig = &zio->io_bp_orig; 1018 objset_t *os = arg; 1019 1020 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) { 1021 ASSERT(BP_EQUAL(bp, bp_orig)); 1022 } else { 1023 dsl_dataset_t *ds = os->os_dsl_dataset; 1024 dmu_tx_t *tx = os->os_synctx; 1025 1026 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE); 1027 dsl_dataset_block_born(ds, bp, tx); 1028 } 1029} 1030 1031/* called from dsl */ 1032void 1033dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx) 1034{ 1035 int txgoff; 1036 zbookmark_phys_t zb; 1037 zio_prop_t zp; 1038 zio_t *zio; 1039 list_t *list; 1040 list_t *newlist = NULL; 1041 dbuf_dirty_record_t *dr; 1042 1043 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg); 1044 1045 ASSERT(dmu_tx_is_syncing(tx)); 1046 /* XXX the write_done callback should really give us the tx... */ 1047 os->os_synctx = tx; 1048 1049 if (os->os_dsl_dataset == NULL) { 1050 /* 1051 * This is the MOS. If we have upgraded, 1052 * spa_max_replication() could change, so reset 1053 * os_copies here. 1054 */ 1055 os->os_copies = spa_max_replication(os->os_spa); 1056 } 1057 1058 /* 1059 * Create the root block IO 1060 */ 1061 SET_BOOKMARK(&zb, os->os_dsl_dataset ? 1062 os->os_dsl_dataset->ds_object : DMU_META_OBJSET, 1063 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); 1064 arc_release(os->os_phys_buf, &os->os_phys_buf); 1065 1066 dmu_write_policy(os, NULL, 0, 0, &zp); 1067 1068 zio = arc_write(pio, os->os_spa, tx->tx_txg, 1069 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), 1070 DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready, 1071 NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE, 1072 ZIO_FLAG_MUSTSUCCEED, &zb); 1073 1074 /* 1075 * Sync special dnodes - the parent IO for the sync is the root block 1076 */ 1077 DMU_META_DNODE(os)->dn_zio = zio; 1078 dnode_sync(DMU_META_DNODE(os), tx); 1079 1080 os->os_phys->os_flags = os->os_flags; 1081 1082 if (DMU_USERUSED_DNODE(os) && 1083 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) { 1084 DMU_USERUSED_DNODE(os)->dn_zio = zio; 1085 dnode_sync(DMU_USERUSED_DNODE(os), tx); 1086 DMU_GROUPUSED_DNODE(os)->dn_zio = zio; 1087 dnode_sync(DMU_GROUPUSED_DNODE(os), tx); 1088 } 1089 1090 txgoff = tx->tx_txg & TXG_MASK; 1091 1092 if (dmu_objset_userused_enabled(os)) { 1093 newlist = &os->os_synced_dnodes; 1094 /* 1095 * We must create the list here because it uses the 1096 * dn_dirty_link[] of this txg. 1097 */ 1098 list_create(newlist, sizeof (dnode_t), 1099 offsetof(dnode_t, dn_dirty_link[txgoff])); 1100 } 1101 1102 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx); 1103 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx); 1104 1105 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff]; 1106 while (dr = list_head(list)) { 1107 ASSERT0(dr->dr_dbuf->db_level); 1108 list_remove(list, dr); 1109 if (dr->dr_zio) 1110 zio_nowait(dr->dr_zio); 1111 } 1112 /* 1113 * Free intent log blocks up to this tx. 1114 */ 1115 zil_sync(os->os_zil, tx); 1116 os->os_phys->os_zil_header = os->os_zil_header; 1117 zio_nowait(zio); 1118} 1119 1120boolean_t 1121dmu_objset_is_dirty(objset_t *os, uint64_t txg) 1122{ 1123 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) || 1124 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK])); 1125} 1126 1127static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES]; 1128 1129void 1130dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb) 1131{ 1132 used_cbs[ost] = cb; 1133} 1134 1135boolean_t 1136dmu_objset_userused_enabled(objset_t *os) 1137{ 1138 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE && 1139 used_cbs[os->os_phys->os_type] != NULL && 1140 DMU_USERUSED_DNODE(os) != NULL); 1141} 1142 1143static void 1144do_userquota_update(objset_t *os, uint64_t used, uint64_t flags, 1145 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx) 1146{ 1147 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) { 1148 int64_t delta = DNODE_SIZE + used; 1149 if (subtract) 1150 delta = -delta; 1151 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT, 1152 user, delta, tx)); 1153 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT, 1154 group, delta, tx)); 1155 } 1156} 1157 1158void 1159dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx) 1160{ 1161 dnode_t *dn; 1162 list_t *list = &os->os_synced_dnodes; 1163 1164 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os)); 1165 1166 while (dn = list_head(list)) { 1167 int flags; 1168 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object)); 1169 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE || 1170 dn->dn_phys->dn_flags & 1171 DNODE_FLAG_USERUSED_ACCOUNTED); 1172 1173 /* Allocate the user/groupused objects if necessary. */ 1174 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) { 1175 VERIFY(0 == zap_create_claim(os, 1176 DMU_USERUSED_OBJECT, 1177 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 1178 VERIFY(0 == zap_create_claim(os, 1179 DMU_GROUPUSED_OBJECT, 1180 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx)); 1181 } 1182 1183 /* 1184 * We intentionally modify the zap object even if the 1185 * net delta is zero. Otherwise 1186 * the block of the zap obj could be shared between 1187 * datasets but need to be different between them after 1188 * a bprewrite. 1189 */ 1190 1191 flags = dn->dn_id_flags; 1192 ASSERT(flags); 1193 if (flags & DN_ID_OLD_EXIST) { 1194 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags, 1195 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx); 1196 } 1197 if (flags & DN_ID_NEW_EXIST) { 1198 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys), 1199 dn->dn_phys->dn_flags, dn->dn_newuid, 1200 dn->dn_newgid, B_FALSE, tx); 1201 } 1202 1203 mutex_enter(&dn->dn_mtx); 1204 dn->dn_oldused = 0; 1205 dn->dn_oldflags = 0; 1206 if (dn->dn_id_flags & DN_ID_NEW_EXIST) { 1207 dn->dn_olduid = dn->dn_newuid; 1208 dn->dn_oldgid = dn->dn_newgid; 1209 dn->dn_id_flags |= DN_ID_OLD_EXIST; 1210 if (dn->dn_bonuslen == 0) 1211 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 1212 else 1213 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 1214 } 1215 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST); 1216 mutex_exit(&dn->dn_mtx); 1217 1218 list_remove(list, dn); 1219 dnode_rele(dn, list); 1220 } 1221} 1222 1223/* 1224 * Returns a pointer to data to find uid/gid from 1225 * 1226 * If a dirty record for transaction group that is syncing can't 1227 * be found then NULL is returned. In the NULL case it is assumed 1228 * the uid/gid aren't changing. 1229 */ 1230static void * 1231dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx) 1232{ 1233 dbuf_dirty_record_t *dr, **drp; 1234 void *data; 1235 1236 if (db->db_dirtycnt == 0) 1237 return (db->db.db_data); /* Nothing is changing */ 1238 1239 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next) 1240 if (dr->dr_txg == tx->tx_txg) 1241 break; 1242 1243 if (dr == NULL) { 1244 data = NULL; 1245 } else { 1246 dnode_t *dn; 1247 1248 DB_DNODE_ENTER(dr->dr_dbuf); 1249 dn = DB_DNODE(dr->dr_dbuf); 1250 1251 if (dn->dn_bonuslen == 0 && 1252 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID) 1253 data = dr->dt.dl.dr_data->b_data; 1254 else 1255 data = dr->dt.dl.dr_data; 1256 1257 DB_DNODE_EXIT(dr->dr_dbuf); 1258 } 1259 1260 return (data); 1261} 1262 1263void 1264dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx) 1265{ 1266 objset_t *os = dn->dn_objset; 1267 void *data = NULL; 1268 dmu_buf_impl_t *db = NULL; 1269 uint64_t *user = NULL; 1270 uint64_t *group = NULL; 1271 int flags = dn->dn_id_flags; 1272 int error; 1273 boolean_t have_spill = B_FALSE; 1274 1275 if (!dmu_objset_userused_enabled(dn->dn_objset)) 1276 return; 1277 1278 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST| 1279 DN_ID_CHKED_SPILL))) 1280 return; 1281 1282 if (before && dn->dn_bonuslen != 0) 1283 data = DN_BONUS(dn->dn_phys); 1284 else if (!before && dn->dn_bonuslen != 0) { 1285 if (dn->dn_bonus) { 1286 db = dn->dn_bonus; 1287 mutex_enter(&db->db_mtx); 1288 data = dmu_objset_userquota_find_data(db, tx); 1289 } else { 1290 data = DN_BONUS(dn->dn_phys); 1291 } 1292 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) { 1293 int rf = 0; 1294 1295 if (RW_WRITE_HELD(&dn->dn_struct_rwlock)) 1296 rf |= DB_RF_HAVESTRUCT; 1297 error = dmu_spill_hold_by_dnode(dn, 1298 rf | DB_RF_MUST_SUCCEED, 1299 FTAG, (dmu_buf_t **)&db); 1300 ASSERT(error == 0); 1301 mutex_enter(&db->db_mtx); 1302 data = (before) ? db->db.db_data : 1303 dmu_objset_userquota_find_data(db, tx); 1304 have_spill = B_TRUE; 1305 } else { 1306 mutex_enter(&dn->dn_mtx); 1307 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 1308 mutex_exit(&dn->dn_mtx); 1309 return; 1310 } 1311 1312 if (before) { 1313 ASSERT(data); 1314 user = &dn->dn_olduid; 1315 group = &dn->dn_oldgid; 1316 } else if (data) { 1317 user = &dn->dn_newuid; 1318 group = &dn->dn_newgid; 1319 } 1320 1321 /* 1322 * Must always call the callback in case the object 1323 * type has changed and that type isn't an object type to track 1324 */ 1325 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data, 1326 user, group); 1327 1328 /* 1329 * Preserve existing uid/gid when the callback can't determine 1330 * what the new uid/gid are and the callback returned EEXIST. 1331 * The EEXIST error tells us to just use the existing uid/gid. 1332 * If we don't know what the old values are then just assign 1333 * them to 0, since that is a new file being created. 1334 */ 1335 if (!before && data == NULL && error == EEXIST) { 1336 if (flags & DN_ID_OLD_EXIST) { 1337 dn->dn_newuid = dn->dn_olduid; 1338 dn->dn_newgid = dn->dn_oldgid; 1339 } else { 1340 dn->dn_newuid = 0; 1341 dn->dn_newgid = 0; 1342 } 1343 error = 0; 1344 } 1345 1346 if (db) 1347 mutex_exit(&db->db_mtx); 1348 1349 mutex_enter(&dn->dn_mtx); 1350 if (error == 0 && before) 1351 dn->dn_id_flags |= DN_ID_OLD_EXIST; 1352 if (error == 0 && !before) 1353 dn->dn_id_flags |= DN_ID_NEW_EXIST; 1354 1355 if (have_spill) { 1356 dn->dn_id_flags |= DN_ID_CHKED_SPILL; 1357 } else { 1358 dn->dn_id_flags |= DN_ID_CHKED_BONUS; 1359 } 1360 mutex_exit(&dn->dn_mtx); 1361 if (have_spill) 1362 dmu_buf_rele((dmu_buf_t *)db, FTAG); 1363} 1364 1365boolean_t 1366dmu_objset_userspace_present(objset_t *os) 1367{ 1368 return (os->os_phys->os_flags & 1369 OBJSET_FLAG_USERACCOUNTING_COMPLETE); 1370} 1371 1372int 1373dmu_objset_userspace_upgrade(objset_t *os) 1374{ 1375 uint64_t obj; 1376 int err = 0; 1377 1378 if (dmu_objset_userspace_present(os)) 1379 return (0); 1380 if (!dmu_objset_userused_enabled(os)) 1381 return (SET_ERROR(ENOTSUP)); 1382 if (dmu_objset_is_snapshot(os)) 1383 return (SET_ERROR(EINVAL)); 1384 1385 /* 1386 * We simply need to mark every object dirty, so that it will be 1387 * synced out and now accounted. If this is called 1388 * concurrently, or if we already did some work before crashing, 1389 * that's fine, since we track each object's accounted state 1390 * independently. 1391 */ 1392 1393 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) { 1394 dmu_tx_t *tx; 1395 dmu_buf_t *db; 1396 int objerr; 1397 1398 if (issig(JUSTLOOKING) && issig(FORREAL)) 1399 return (SET_ERROR(EINTR)); 1400 1401 objerr = dmu_bonus_hold(os, obj, FTAG, &db); 1402 if (objerr != 0) 1403 continue; 1404 tx = dmu_tx_create(os); 1405 dmu_tx_hold_bonus(tx, obj); 1406 objerr = dmu_tx_assign(tx, TXG_WAIT); 1407 if (objerr != 0) { 1408 dmu_tx_abort(tx); 1409 continue; 1410 } 1411 dmu_buf_will_dirty(db, tx); 1412 dmu_buf_rele(db, FTAG); 1413 dmu_tx_commit(tx); 1414 } 1415 1416 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE; 1417 txg_wait_synced(dmu_objset_pool(os), 0); 1418 return (0); 1419} 1420 1421void 1422dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp, 1423 uint64_t *usedobjsp, uint64_t *availobjsp) 1424{ 1425 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp, 1426 usedobjsp, availobjsp); 1427} 1428 1429uint64_t 1430dmu_objset_fsid_guid(objset_t *os) 1431{ 1432 return (dsl_dataset_fsid_guid(os->os_dsl_dataset)); 1433} 1434 1435void 1436dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat) 1437{ 1438 stat->dds_type = os->os_phys->os_type; 1439 if (os->os_dsl_dataset) 1440 dsl_dataset_fast_stat(os->os_dsl_dataset, stat); 1441} 1442 1443void 1444dmu_objset_stats(objset_t *os, nvlist_t *nv) 1445{ 1446 ASSERT(os->os_dsl_dataset || 1447 os->os_phys->os_type == DMU_OST_META); 1448 1449 if (os->os_dsl_dataset != NULL) 1450 dsl_dataset_stats(os->os_dsl_dataset, nv); 1451 1452 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE, 1453 os->os_phys->os_type); 1454 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING, 1455 dmu_objset_userspace_present(os)); 1456} 1457 1458int 1459dmu_objset_is_snapshot(objset_t *os) 1460{ 1461 if (os->os_dsl_dataset != NULL) 1462 return (dsl_dataset_is_snapshot(os->os_dsl_dataset)); 1463 else 1464 return (B_FALSE); 1465} 1466 1467int 1468dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen, 1469 boolean_t *conflict) 1470{ 1471 dsl_dataset_t *ds = os->os_dsl_dataset; 1472 uint64_t ignored; 1473 1474 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 1475 return (SET_ERROR(ENOENT)); 1476 1477 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset, 1478 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored, 1479 MT_FIRST, real, maxlen, conflict)); 1480} 1481 1482int 1483dmu_snapshot_list_next(objset_t *os, int namelen, char *name, 1484 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict) 1485{ 1486 dsl_dataset_t *ds = os->os_dsl_dataset; 1487 zap_cursor_t cursor; 1488 zap_attribute_t attr; 1489 1490 ASSERT(dsl_pool_config_held(dmu_objset_pool(os))); 1491 1492 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0) 1493 return (SET_ERROR(ENOENT)); 1494 1495 zap_cursor_init_serialized(&cursor, 1496 ds->ds_dir->dd_pool->dp_meta_objset, 1497 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp); 1498 1499 if (zap_cursor_retrieve(&cursor, &attr) != 0) { 1500 zap_cursor_fini(&cursor); 1501 return (SET_ERROR(ENOENT)); 1502 } 1503 1504 if (strlen(attr.za_name) + 1 > namelen) { 1505 zap_cursor_fini(&cursor); 1506 return (SET_ERROR(ENAMETOOLONG)); 1507 } 1508 1509 (void) strcpy(name, attr.za_name); 1510 if (idp) 1511 *idp = attr.za_first_integer; 1512 if (case_conflict) 1513 *case_conflict = attr.za_normalization_conflict; 1514 zap_cursor_advance(&cursor); 1515 *offp = zap_cursor_serialize(&cursor); 1516 zap_cursor_fini(&cursor); 1517 1518 return (0); 1519} 1520 1521int 1522dmu_dir_list_next(objset_t *os, int namelen, char *name, 1523 uint64_t *idp, uint64_t *offp) 1524{ 1525 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 1526 zap_cursor_t cursor; 1527 zap_attribute_t attr; 1528 1529 /* there is no next dir on a snapshot! */ 1530 if (os->os_dsl_dataset->ds_object != 1531 dsl_dir_phys(dd)->dd_head_dataset_obj) 1532 return (SET_ERROR(ENOENT)); 1533 1534 zap_cursor_init_serialized(&cursor, 1535 dd->dd_pool->dp_meta_objset, 1536 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp); 1537 1538 if (zap_cursor_retrieve(&cursor, &attr) != 0) { 1539 zap_cursor_fini(&cursor); 1540 return (SET_ERROR(ENOENT)); 1541 } 1542 1543 if (strlen(attr.za_name) + 1 > namelen) { 1544 zap_cursor_fini(&cursor); 1545 return (SET_ERROR(ENAMETOOLONG)); 1546 } 1547 1548 (void) strcpy(name, attr.za_name); 1549 if (idp) 1550 *idp = attr.za_first_integer; 1551 zap_cursor_advance(&cursor); 1552 *offp = zap_cursor_serialize(&cursor); 1553 zap_cursor_fini(&cursor); 1554 1555 return (0); 1556} 1557 1558/* 1559 * Find objsets under and including ddobj, call func(ds) on each. 1560 */ 1561int 1562dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj, 1563 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags) 1564{ 1565 dsl_dir_t *dd; 1566 dsl_dataset_t *ds; 1567 zap_cursor_t zc; 1568 zap_attribute_t *attr; 1569 uint64_t thisobj; 1570 int err; 1571 1572 ASSERT(dsl_pool_config_held(dp)); 1573 1574 err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd); 1575 if (err != 0) 1576 return (err); 1577 1578 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 1579 if (dd->dd_myname[0] == '$') { 1580 dsl_dir_rele(dd, FTAG); 1581 return (0); 1582 } 1583 1584 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 1585 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 1586 1587 /* 1588 * Iterate over all children. 1589 */ 1590 if (flags & DS_FIND_CHILDREN) { 1591 for (zap_cursor_init(&zc, dp->dp_meta_objset, 1592 dsl_dir_phys(dd)->dd_child_dir_zapobj); 1593 zap_cursor_retrieve(&zc, attr) == 0; 1594 (void) zap_cursor_advance(&zc)) { 1595 ASSERT3U(attr->za_integer_length, ==, 1596 sizeof (uint64_t)); 1597 ASSERT3U(attr->za_num_integers, ==, 1); 1598 1599 err = dmu_objset_find_dp(dp, attr->za_first_integer, 1600 func, arg, flags); 1601 if (err != 0) 1602 break; 1603 } 1604 zap_cursor_fini(&zc); 1605 1606 if (err != 0) { 1607 dsl_dir_rele(dd, FTAG); 1608 kmem_free(attr, sizeof (zap_attribute_t)); 1609 return (err); 1610 } 1611 } 1612 1613 /* 1614 * Iterate over all snapshots. 1615 */ 1616 if (flags & DS_FIND_SNAPSHOTS) { 1617 dsl_dataset_t *ds; 1618 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 1619 1620 if (err == 0) { 1621 uint64_t snapobj; 1622 1623 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 1624 dsl_dataset_rele(ds, FTAG); 1625 1626 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 1627 zap_cursor_retrieve(&zc, attr) == 0; 1628 (void) zap_cursor_advance(&zc)) { 1629 ASSERT3U(attr->za_integer_length, ==, 1630 sizeof (uint64_t)); 1631 ASSERT3U(attr->za_num_integers, ==, 1); 1632 1633 err = dsl_dataset_hold_obj(dp, 1634 attr->za_first_integer, FTAG, &ds); 1635 if (err != 0) 1636 break; 1637 err = func(dp, ds, arg); 1638 dsl_dataset_rele(ds, FTAG); 1639 if (err != 0) 1640 break; 1641 } 1642 zap_cursor_fini(&zc); 1643 } 1644 } 1645 1646 dsl_dir_rele(dd, FTAG); 1647 kmem_free(attr, sizeof (zap_attribute_t)); 1648 1649 if (err != 0) 1650 return (err); 1651 1652 /* 1653 * Apply to self. 1654 */ 1655 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 1656 if (err != 0) 1657 return (err); 1658 err = func(dp, ds, arg); 1659 dsl_dataset_rele(ds, FTAG); 1660 return (err); 1661} 1662 1663/* 1664 * Find all objsets under name, and for each, call 'func(child_name, arg)'. 1665 * The dp_config_rwlock must not be held when this is called, and it 1666 * will not be held when the callback is called. 1667 * Therefore this function should only be used when the pool is not changing 1668 * (e.g. in syncing context), or the callback can deal with the possible races. 1669 */ 1670static int 1671dmu_objset_find_impl(spa_t *spa, const char *name, 1672 int func(const char *, void *), void *arg, int flags) 1673{ 1674 dsl_dir_t *dd; 1675 dsl_pool_t *dp = spa_get_dsl(spa); 1676 dsl_dataset_t *ds; 1677 zap_cursor_t zc; 1678 zap_attribute_t *attr; 1679 char *child; 1680 uint64_t thisobj; 1681 int err; 1682 1683 dsl_pool_config_enter(dp, FTAG); 1684 1685 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL); 1686 if (err != 0) { 1687 dsl_pool_config_exit(dp, FTAG); 1688 return (err); 1689 } 1690 1691 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */ 1692 if (dd->dd_myname[0] == '$') { 1693 dsl_dir_rele(dd, FTAG); 1694 dsl_pool_config_exit(dp, FTAG); 1695 return (0); 1696 } 1697 1698 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj; 1699 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); 1700 1701 /* 1702 * Iterate over all children. 1703 */ 1704 if (flags & DS_FIND_CHILDREN) { 1705 for (zap_cursor_init(&zc, dp->dp_meta_objset, 1706 dsl_dir_phys(dd)->dd_child_dir_zapobj); 1707 zap_cursor_retrieve(&zc, attr) == 0; 1708 (void) zap_cursor_advance(&zc)) { 1709 ASSERT3U(attr->za_integer_length, ==, 1710 sizeof (uint64_t)); 1711 ASSERT3U(attr->za_num_integers, ==, 1); 1712 1713 child = kmem_asprintf("%s/%s", name, attr->za_name); 1714 dsl_pool_config_exit(dp, FTAG); 1715 err = dmu_objset_find_impl(spa, child, 1716 func, arg, flags); 1717 dsl_pool_config_enter(dp, FTAG); 1718 strfree(child); 1719 if (err != 0) 1720 break; 1721 } 1722 zap_cursor_fini(&zc); 1723 1724 if (err != 0) { 1725 dsl_dir_rele(dd, FTAG); 1726 dsl_pool_config_exit(dp, FTAG); 1727 kmem_free(attr, sizeof (zap_attribute_t)); 1728 return (err); 1729 } 1730 } 1731 1732 /* 1733 * Iterate over all snapshots. 1734 */ 1735 if (flags & DS_FIND_SNAPSHOTS) { 1736 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds); 1737 1738 if (err == 0) { 1739 uint64_t snapobj; 1740 1741 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj; 1742 dsl_dataset_rele(ds, FTAG); 1743 1744 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj); 1745 zap_cursor_retrieve(&zc, attr) == 0; 1746 (void) zap_cursor_advance(&zc)) { 1747 ASSERT3U(attr->za_integer_length, ==, 1748 sizeof (uint64_t)); 1749 ASSERT3U(attr->za_num_integers, ==, 1); 1750 1751 child = kmem_asprintf("%s@%s", 1752 name, attr->za_name); 1753 dsl_pool_config_exit(dp, FTAG); 1754 err = func(child, arg); 1755 dsl_pool_config_enter(dp, FTAG); 1756 strfree(child); 1757 if (err != 0) 1758 break; 1759 } 1760 zap_cursor_fini(&zc); 1761 } 1762 } 1763 1764 dsl_dir_rele(dd, FTAG); 1765 kmem_free(attr, sizeof (zap_attribute_t)); 1766 dsl_pool_config_exit(dp, FTAG); 1767 1768 if (err != 0) 1769 return (err); 1770 1771 /* Apply to self. */ 1772 return (func(name, arg)); 1773} 1774 1775/* 1776 * See comment above dmu_objset_find_impl(). 1777 */ 1778int 1779dmu_objset_find(char *name, int func(const char *, void *), void *arg, 1780 int flags) 1781{ 1782 spa_t *spa; 1783 int error; 1784 1785 error = spa_open(name, &spa, FTAG); 1786 if (error != 0) 1787 return (error); 1788 error = dmu_objset_find_impl(spa, name, func, arg, flags); 1789 spa_close(spa, FTAG); 1790 return (error); 1791} 1792 1793void 1794dmu_objset_set_user(objset_t *os, void *user_ptr) 1795{ 1796 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 1797 os->os_user_ptr = user_ptr; 1798} 1799 1800void * 1801dmu_objset_get_user(objset_t *os) 1802{ 1803 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock)); 1804 return (os->os_user_ptr); 1805} 1806 1807/* 1808 * Determine name of filesystem, given name of snapshot. 1809 * buf must be at least MAXNAMELEN bytes 1810 */ 1811int 1812dmu_fsname(const char *snapname, char *buf) 1813{ 1814 char *atp = strchr(snapname, '@'); 1815 if (atp == NULL) 1816 return (SET_ERROR(EINVAL)); 1817 if (atp - snapname >= MAXNAMELEN) 1818 return (SET_ERROR(ENAMETOOLONG)); 1819 (void) strlcpy(buf, snapname, atp - snapname + 1); 1820 return (0); 1821} 1822