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