dsl_pool.c revision 288581
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) 2011, 2014 by Delphix. All rights reserved. 24 * Copyright (c) 2013 Steven Hartland. All rights reserved. 25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 26 */ 27 28#include <sys/dsl_pool.h> 29#include <sys/dsl_dataset.h> 30#include <sys/dsl_prop.h> 31#include <sys/dsl_dir.h> 32#include <sys/dsl_synctask.h> 33#include <sys/dsl_scan.h> 34#include <sys/dnode.h> 35#include <sys/dmu_tx.h> 36#include <sys/dmu_objset.h> 37#include <sys/arc.h> 38#include <sys/zap.h> 39#include <sys/zio.h> 40#include <sys/zfs_context.h> 41#include <sys/fs/zfs.h> 42#include <sys/zfs_znode.h> 43#include <sys/spa_impl.h> 44#include <sys/dsl_deadlist.h> 45#include <sys/bptree.h> 46#include <sys/zfeature.h> 47#include <sys/zil_impl.h> 48#include <sys/dsl_userhold.h> 49 50#ifdef __FreeBSD__ 51#include <sys/sysctl.h> 52#include <sys/types.h> 53#endif 54 55/* 56 * ZFS Write Throttle 57 * ------------------ 58 * 59 * ZFS must limit the rate of incoming writes to the rate at which it is able 60 * to sync data modifications to the backend storage. Throttling by too much 61 * creates an artificial limit; throttling by too little can only be sustained 62 * for short periods and would lead to highly lumpy performance. On a per-pool 63 * basis, ZFS tracks the amount of modified (dirty) data. As operations change 64 * data, the amount of dirty data increases; as ZFS syncs out data, the amount 65 * of dirty data decreases. When the amount of dirty data exceeds a 66 * predetermined threshold further modifications are blocked until the amount 67 * of dirty data decreases (as data is synced out). 68 * 69 * The limit on dirty data is tunable, and should be adjusted according to 70 * both the IO capacity and available memory of the system. The larger the 71 * window, the more ZFS is able to aggregate and amortize metadata (and data) 72 * changes. However, memory is a limited resource, and allowing for more dirty 73 * data comes at the cost of keeping other useful data in memory (for example 74 * ZFS data cached by the ARC). 75 * 76 * Implementation 77 * 78 * As buffers are modified dsl_pool_willuse_space() increments both the per- 79 * txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of 80 * dirty space used; dsl_pool_dirty_space() decrements those values as data 81 * is synced out from dsl_pool_sync(). While only the poolwide value is 82 * relevant, the per-txg value is useful for debugging. The tunable 83 * zfs_dirty_data_max determines the dirty space limit. Once that value is 84 * exceeded, new writes are halted until space frees up. 85 * 86 * The zfs_dirty_data_sync tunable dictates the threshold at which we 87 * ensure that there is a txg syncing (see the comment in txg.c for a full 88 * description of transaction group stages). 89 * 90 * The IO scheduler uses both the dirty space limit and current amount of 91 * dirty data as inputs. Those values affect the number of concurrent IOs ZFS 92 * issues. See the comment in vdev_queue.c for details of the IO scheduler. 93 * 94 * The delay is also calculated based on the amount of dirty data. See the 95 * comment above dmu_tx_delay() for details. 96 */ 97 98/* 99 * zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory, 100 * capped at zfs_dirty_data_max_max. It can also be overridden in /etc/system. 101 */ 102uint64_t zfs_dirty_data_max; 103uint64_t zfs_dirty_data_max_max = 4ULL * 1024 * 1024 * 1024; 104int zfs_dirty_data_max_percent = 10; 105 106/* 107 * If there is at least this much dirty data, push out a txg. 108 */ 109uint64_t zfs_dirty_data_sync = 64 * 1024 * 1024; 110 111/* 112 * Once there is this amount of dirty data, the dmu_tx_delay() will kick in 113 * and delay each transaction. 114 * This value should be >= zfs_vdev_async_write_active_max_dirty_percent. 115 */ 116int zfs_delay_min_dirty_percent = 60; 117 118/* 119 * This controls how quickly the delay approaches infinity. 120 * Larger values cause it to delay more for a given amount of dirty data. 121 * Therefore larger values will cause there to be less dirty data for a 122 * given throughput. 123 * 124 * For the smoothest delay, this value should be about 1 billion divided 125 * by the maximum number of operations per second. This will smoothly 126 * handle between 10x and 1/10th this number. 127 * 128 * Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the 129 * multiply in dmu_tx_delay(). 130 */ 131uint64_t zfs_delay_scale = 1000 * 1000 * 1000 / 2000; 132 133 134#ifdef __FreeBSD__ 135 136extern int zfs_vdev_async_write_active_max_dirty_percent; 137 138SYSCTL_DECL(_vfs_zfs); 139 140TUNABLE_QUAD("vfs.zfs.dirty_data_max", &zfs_dirty_data_max); 141SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, dirty_data_max, CTLFLAG_RWTUN, 142 &zfs_dirty_data_max, 0, 143 "The maximum amount of dirty data in bytes after which new writes are " 144 "halted until space becomes available"); 145 146TUNABLE_QUAD("vfs.zfs.dirty_data_max_max", &zfs_dirty_data_max_max); 147SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, dirty_data_max_max, CTLFLAG_RDTUN, 148 &zfs_dirty_data_max_max, 0, 149 "The absolute cap on dirty_data_max when auto calculating"); 150 151TUNABLE_INT("vfs.zfs.dirty_data_max_percent", &zfs_dirty_data_max_percent); 152static int sysctl_zfs_dirty_data_max_percent(SYSCTL_HANDLER_ARGS); 153SYSCTL_PROC(_vfs_zfs, OID_AUTO, dirty_data_max_percent, 154 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RWTUN, 0, sizeof(int), 155 sysctl_zfs_dirty_data_max_percent, "I", 156 "The percent of physical memory used to auto calculate dirty_data_max"); 157 158TUNABLE_QUAD("vfs.zfs.dirty_data_sync", &zfs_dirty_data_sync); 159SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, dirty_data_sync, CTLFLAG_RWTUN, 160 &zfs_dirty_data_sync, 0, 161 "Force a txg if the number of dirty buffer bytes exceed this value"); 162 163static int sysctl_zfs_delay_min_dirty_percent(SYSCTL_HANDLER_ARGS); 164/* No zfs_delay_min_dirty_percent tunable due to limit requirements */ 165SYSCTL_PROC(_vfs_zfs, OID_AUTO, delay_min_dirty_percent, 166 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(int), 167 sysctl_zfs_delay_min_dirty_percent, "I", 168 "The limit of outstanding dirty data before transations are delayed"); 169 170static int sysctl_zfs_delay_scale(SYSCTL_HANDLER_ARGS); 171/* No zfs_delay_scale tunable due to limit requirements */ 172SYSCTL_PROC(_vfs_zfs, OID_AUTO, delay_scale, 173 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t), 174 sysctl_zfs_delay_scale, "QU", 175 "Controls how quickly the delay approaches infinity"); 176 177static int 178sysctl_zfs_dirty_data_max_percent(SYSCTL_HANDLER_ARGS) 179{ 180 int val, err; 181 182 val = zfs_dirty_data_max_percent; 183 err = sysctl_handle_int(oidp, &val, 0, req); 184 if (err != 0 || req->newptr == NULL) 185 return (err); 186 187 if (val < 0 || val > 100) 188 return (EINVAL); 189 190 zfs_dirty_data_max_percent = val; 191 192 return (0); 193} 194 195static int 196sysctl_zfs_delay_min_dirty_percent(SYSCTL_HANDLER_ARGS) 197{ 198 int val, err; 199 200 val = zfs_delay_min_dirty_percent; 201 err = sysctl_handle_int(oidp, &val, 0, req); 202 if (err != 0 || req->newptr == NULL) 203 return (err); 204 205 if (val < zfs_vdev_async_write_active_max_dirty_percent) 206 return (EINVAL); 207 208 zfs_delay_min_dirty_percent = val; 209 210 return (0); 211} 212 213static int 214sysctl_zfs_delay_scale(SYSCTL_HANDLER_ARGS) 215{ 216 uint64_t val; 217 int err; 218 219 val = zfs_delay_scale; 220 err = sysctl_handle_64(oidp, &val, 0, req); 221 if (err != 0 || req->newptr == NULL) 222 return (err); 223 224 if (val > UINT64_MAX / zfs_dirty_data_max) 225 return (EINVAL); 226 227 zfs_delay_scale = val; 228 229 return (0); 230} 231#endif 232 233hrtime_t zfs_throttle_delay = MSEC2NSEC(10); 234hrtime_t zfs_throttle_resolution = MSEC2NSEC(10); 235 236int 237dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) 238{ 239 uint64_t obj; 240 int err; 241 242 err = zap_lookup(dp->dp_meta_objset, 243 dsl_dir_phys(dp->dp_root_dir)->dd_child_dir_zapobj, 244 name, sizeof (obj), 1, &obj); 245 if (err) 246 return (err); 247 248 return (dsl_dir_hold_obj(dp, obj, name, dp, ddp)); 249} 250 251static dsl_pool_t * 252dsl_pool_open_impl(spa_t *spa, uint64_t txg) 253{ 254 dsl_pool_t *dp; 255 blkptr_t *bp = spa_get_rootblkptr(spa); 256 257 dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); 258 dp->dp_spa = spa; 259 dp->dp_meta_rootbp = *bp; 260 rrw_init(&dp->dp_config_rwlock, B_TRUE); 261 txg_init(dp, txg); 262 263 txg_list_create(&dp->dp_dirty_datasets, 264 offsetof(dsl_dataset_t, ds_dirty_link)); 265 txg_list_create(&dp->dp_dirty_zilogs, 266 offsetof(zilog_t, zl_dirty_link)); 267 txg_list_create(&dp->dp_dirty_dirs, 268 offsetof(dsl_dir_t, dd_dirty_link)); 269 txg_list_create(&dp->dp_sync_tasks, 270 offsetof(dsl_sync_task_t, dst_node)); 271 272 mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); 273 cv_init(&dp->dp_spaceavail_cv, NULL, CV_DEFAULT, NULL); 274 275 dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri, 276 1, 4, 0); 277 278 return (dp); 279} 280 281int 282dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) 283{ 284 int err; 285 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 286 287 err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, 288 &dp->dp_meta_objset); 289 if (err != 0) 290 dsl_pool_close(dp); 291 else 292 *dpp = dp; 293 294 return (err); 295} 296 297int 298dsl_pool_open(dsl_pool_t *dp) 299{ 300 int err; 301 dsl_dir_t *dd; 302 dsl_dataset_t *ds; 303 uint64_t obj; 304 305 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 306 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 307 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, 308 &dp->dp_root_dir_obj); 309 if (err) 310 goto out; 311 312 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, 313 NULL, dp, &dp->dp_root_dir); 314 if (err) 315 goto out; 316 317 err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); 318 if (err) 319 goto out; 320 321 if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { 322 err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); 323 if (err) 324 goto out; 325 err = dsl_dataset_hold_obj(dp, 326 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds); 327 if (err == 0) { 328 err = dsl_dataset_hold_obj(dp, 329 dsl_dataset_phys(ds)->ds_prev_snap_obj, dp, 330 &dp->dp_origin_snap); 331 dsl_dataset_rele(ds, FTAG); 332 } 333 dsl_dir_rele(dd, dp); 334 if (err) 335 goto out; 336 } 337 338 if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { 339 err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, 340 &dp->dp_free_dir); 341 if (err) 342 goto out; 343 344 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 345 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); 346 if (err) 347 goto out; 348 VERIFY0(bpobj_open(&dp->dp_free_bpobj, 349 dp->dp_meta_objset, obj)); 350 } 351 352 /* 353 * Note: errors ignored, because the leak dir will not exist if we 354 * have not encountered a leak yet. 355 */ 356 (void) dsl_pool_open_special_dir(dp, LEAK_DIR_NAME, 357 &dp->dp_leak_dir); 358 359 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) { 360 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 361 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, 362 &dp->dp_bptree_obj); 363 if (err != 0) 364 goto out; 365 } 366 367 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMPTY_BPOBJ)) { 368 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 369 DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, 370 &dp->dp_empty_bpobj); 371 if (err != 0) 372 goto out; 373 } 374 375 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 376 DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, 377 &dp->dp_tmp_userrefs_obj); 378 if (err == ENOENT) 379 err = 0; 380 if (err) 381 goto out; 382 383 err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); 384 385out: 386 rrw_exit(&dp->dp_config_rwlock, FTAG); 387 return (err); 388} 389 390void 391dsl_pool_close(dsl_pool_t *dp) 392{ 393 /* 394 * Drop our references from dsl_pool_open(). 395 * 396 * Since we held the origin_snap from "syncing" context (which 397 * includes pool-opening context), it actually only got a "ref" 398 * and not a hold, so just drop that here. 399 */ 400 if (dp->dp_origin_snap) 401 dsl_dataset_rele(dp->dp_origin_snap, dp); 402 if (dp->dp_mos_dir) 403 dsl_dir_rele(dp->dp_mos_dir, dp); 404 if (dp->dp_free_dir) 405 dsl_dir_rele(dp->dp_free_dir, dp); 406 if (dp->dp_leak_dir) 407 dsl_dir_rele(dp->dp_leak_dir, dp); 408 if (dp->dp_root_dir) 409 dsl_dir_rele(dp->dp_root_dir, dp); 410 411 bpobj_close(&dp->dp_free_bpobj); 412 413 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ 414 if (dp->dp_meta_objset) 415 dmu_objset_evict(dp->dp_meta_objset); 416 417 txg_list_destroy(&dp->dp_dirty_datasets); 418 txg_list_destroy(&dp->dp_dirty_zilogs); 419 txg_list_destroy(&dp->dp_sync_tasks); 420 txg_list_destroy(&dp->dp_dirty_dirs); 421 422 /* 423 * We can't set retry to TRUE since we're explicitly specifying 424 * a spa to flush. This is good enough; any missed buffers for 425 * this spa won't cause trouble, and they'll eventually fall 426 * out of the ARC just like any other unused buffer. 427 */ 428 arc_flush(dp->dp_spa, FALSE); 429 430 txg_fini(dp); 431 dsl_scan_fini(dp); 432 dmu_buf_user_evict_wait(); 433 434 rrw_destroy(&dp->dp_config_rwlock); 435 mutex_destroy(&dp->dp_lock); 436 taskq_destroy(dp->dp_vnrele_taskq); 437 if (dp->dp_blkstats) 438 kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); 439 kmem_free(dp, sizeof (dsl_pool_t)); 440} 441 442dsl_pool_t * 443dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) 444{ 445 int err; 446 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 447 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); 448 objset_t *os; 449 dsl_dataset_t *ds; 450 uint64_t obj; 451 452 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 453 454 /* create and open the MOS (meta-objset) */ 455 dp->dp_meta_objset = dmu_objset_create_impl(spa, 456 NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); 457 458 /* create the pool directory */ 459 err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 460 DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); 461 ASSERT0(err); 462 463 /* Initialize scan structures */ 464 VERIFY0(dsl_scan_init(dp, txg)); 465 466 /* create and open the root dir */ 467 dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); 468 VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, 469 NULL, dp, &dp->dp_root_dir)); 470 471 /* create and open the meta-objset dir */ 472 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); 473 VERIFY0(dsl_pool_open_special_dir(dp, 474 MOS_DIR_NAME, &dp->dp_mos_dir)); 475 476 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { 477 /* create and open the free dir */ 478 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, 479 FREE_DIR_NAME, tx); 480 VERIFY0(dsl_pool_open_special_dir(dp, 481 FREE_DIR_NAME, &dp->dp_free_dir)); 482 483 /* create and open the free_bplist */ 484 obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx); 485 VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 486 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); 487 VERIFY0(bpobj_open(&dp->dp_free_bpobj, 488 dp->dp_meta_objset, obj)); 489 } 490 491 if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) 492 dsl_pool_create_origin(dp, tx); 493 494 /* create the root dataset */ 495 obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); 496 497 /* create the root objset */ 498 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); 499 os = dmu_objset_create_impl(dp->dp_spa, ds, 500 dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx); 501#ifdef _KERNEL 502 zfs_create_fs(os, kcred, zplprops, tx); 503#endif 504 dsl_dataset_rele(ds, FTAG); 505 506 dmu_tx_commit(tx); 507 508 rrw_exit(&dp->dp_config_rwlock, FTAG); 509 510 return (dp); 511} 512 513/* 514 * Account for the meta-objset space in its placeholder dsl_dir. 515 */ 516void 517dsl_pool_mos_diduse_space(dsl_pool_t *dp, 518 int64_t used, int64_t comp, int64_t uncomp) 519{ 520 ASSERT3U(comp, ==, uncomp); /* it's all metadata */ 521 mutex_enter(&dp->dp_lock); 522 dp->dp_mos_used_delta += used; 523 dp->dp_mos_compressed_delta += comp; 524 dp->dp_mos_uncompressed_delta += uncomp; 525 mutex_exit(&dp->dp_lock); 526} 527 528static int 529deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 530{ 531 dsl_deadlist_t *dl = arg; 532 dsl_deadlist_insert(dl, bp, tx); 533 return (0); 534} 535 536static void 537dsl_pool_sync_mos(dsl_pool_t *dp, dmu_tx_t *tx) 538{ 539 zio_t *zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 540 dmu_objset_sync(dp->dp_meta_objset, zio, tx); 541 VERIFY0(zio_wait(zio)); 542 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); 543 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); 544} 545 546static void 547dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta) 548{ 549 ASSERT(MUTEX_HELD(&dp->dp_lock)); 550 551 if (delta < 0) 552 ASSERT3U(-delta, <=, dp->dp_dirty_total); 553 554 dp->dp_dirty_total += delta; 555 556 /* 557 * Note: we signal even when increasing dp_dirty_total. 558 * This ensures forward progress -- each thread wakes the next waiter. 559 */ 560 if (dp->dp_dirty_total <= zfs_dirty_data_max) 561 cv_signal(&dp->dp_spaceavail_cv); 562} 563 564void 565dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) 566{ 567 zio_t *zio; 568 dmu_tx_t *tx; 569 dsl_dir_t *dd; 570 dsl_dataset_t *ds; 571 objset_t *mos = dp->dp_meta_objset; 572 list_t synced_datasets; 573 574 list_create(&synced_datasets, sizeof (dsl_dataset_t), 575 offsetof(dsl_dataset_t, ds_synced_link)); 576 577 tx = dmu_tx_create_assigned(dp, txg); 578 579 /* 580 * Write out all dirty blocks of dirty datasets. 581 */ 582 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 583 while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 584 /* 585 * We must not sync any non-MOS datasets twice, because 586 * we may have taken a snapshot of them. However, we 587 * may sync newly-created datasets on pass 2. 588 */ 589 ASSERT(!list_link_active(&ds->ds_synced_link)); 590 list_insert_tail(&synced_datasets, ds); 591 dsl_dataset_sync(ds, zio, tx); 592 } 593 VERIFY0(zio_wait(zio)); 594 595 /* 596 * We have written all of the accounted dirty data, so our 597 * dp_space_towrite should now be zero. However, some seldom-used 598 * code paths do not adhere to this (e.g. dbuf_undirty(), also 599 * rounding error in dbuf_write_physdone). 600 * Shore up the accounting of any dirtied space now. 601 */ 602 dsl_pool_undirty_space(dp, dp->dp_dirty_pertxg[txg & TXG_MASK], txg); 603 604 /* 605 * After the data blocks have been written (ensured by the zio_wait() 606 * above), update the user/group space accounting. 607 */ 608 for (ds = list_head(&synced_datasets); ds != NULL; 609 ds = list_next(&synced_datasets, ds)) { 610 dmu_objset_do_userquota_updates(ds->ds_objset, tx); 611 } 612 613 /* 614 * Sync the datasets again to push out the changes due to 615 * userspace updates. This must be done before we process the 616 * sync tasks, so that any snapshots will have the correct 617 * user accounting information (and we won't get confused 618 * about which blocks are part of the snapshot). 619 */ 620 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 621 while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 622 ASSERT(list_link_active(&ds->ds_synced_link)); 623 dmu_buf_rele(ds->ds_dbuf, ds); 624 dsl_dataset_sync(ds, zio, tx); 625 } 626 VERIFY0(zio_wait(zio)); 627 628 /* 629 * Now that the datasets have been completely synced, we can 630 * clean up our in-memory structures accumulated while syncing: 631 * 632 * - move dead blocks from the pending deadlist to the on-disk deadlist 633 * - release hold from dsl_dataset_dirty() 634 */ 635 while ((ds = list_remove_head(&synced_datasets)) != NULL) { 636 objset_t *os = ds->ds_objset; 637 bplist_iterate(&ds->ds_pending_deadlist, 638 deadlist_enqueue_cb, &ds->ds_deadlist, tx); 639 ASSERT(!dmu_objset_is_dirty(os, txg)); 640 dmu_buf_rele(ds->ds_dbuf, ds); 641 } 642 while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) != NULL) { 643 dsl_dir_sync(dd, tx); 644 } 645 646 /* 647 * The MOS's space is accounted for in the pool/$MOS 648 * (dp_mos_dir). We can't modify the mos while we're syncing 649 * it, so we remember the deltas and apply them here. 650 */ 651 if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || 652 dp->dp_mos_uncompressed_delta != 0) { 653 dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, 654 dp->dp_mos_used_delta, 655 dp->dp_mos_compressed_delta, 656 dp->dp_mos_uncompressed_delta, tx); 657 dp->dp_mos_used_delta = 0; 658 dp->dp_mos_compressed_delta = 0; 659 dp->dp_mos_uncompressed_delta = 0; 660 } 661 662 if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || 663 list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { 664 dsl_pool_sync_mos(dp, tx); 665 } 666 667 /* 668 * If we modify a dataset in the same txg that we want to destroy it, 669 * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. 670 * dsl_dir_destroy_check() will fail if there are unexpected holds. 671 * Therefore, we want to sync the MOS (thus syncing the dd_dbuf 672 * and clearing the hold on it) before we process the sync_tasks. 673 * The MOS data dirtied by the sync_tasks will be synced on the next 674 * pass. 675 */ 676 if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { 677 dsl_sync_task_t *dst; 678 /* 679 * No more sync tasks should have been added while we 680 * were syncing. 681 */ 682 ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1); 683 while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)) != NULL) 684 dsl_sync_task_sync(dst, tx); 685 } 686 687 dmu_tx_commit(tx); 688 689 DTRACE_PROBE2(dsl_pool_sync__done, dsl_pool_t *dp, dp, uint64_t, txg); 690} 691 692void 693dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) 694{ 695 zilog_t *zilog; 696 697 while (zilog = txg_list_remove(&dp->dp_dirty_zilogs, txg)) { 698 dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); 699 zil_clean(zilog, txg); 700 ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); 701 dmu_buf_rele(ds->ds_dbuf, zilog); 702 } 703 ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); 704} 705 706/* 707 * TRUE if the current thread is the tx_sync_thread or if we 708 * are being called from SPA context during pool initialization. 709 */ 710int 711dsl_pool_sync_context(dsl_pool_t *dp) 712{ 713 return (curthread == dp->dp_tx.tx_sync_thread || 714 spa_is_initializing(dp->dp_spa)); 715} 716 717uint64_t 718dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) 719{ 720 uint64_t space, resv; 721 722 /* 723 * If we're trying to assess whether it's OK to do a free, 724 * cut the reservation in half to allow forward progress 725 * (e.g. make it possible to rm(1) files from a full pool). 726 */ 727 space = spa_get_dspace(dp->dp_spa); 728 resv = spa_get_slop_space(dp->dp_spa); 729 if (netfree) 730 resv >>= 1; 731 732 return (space - resv); 733} 734 735boolean_t 736dsl_pool_need_dirty_delay(dsl_pool_t *dp) 737{ 738 uint64_t delay_min_bytes = 739 zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100; 740 boolean_t rv; 741 742 mutex_enter(&dp->dp_lock); 743 if (dp->dp_dirty_total > zfs_dirty_data_sync) 744 txg_kick(dp); 745 rv = (dp->dp_dirty_total > delay_min_bytes); 746 mutex_exit(&dp->dp_lock); 747 return (rv); 748} 749 750void 751dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 752{ 753 if (space > 0) { 754 mutex_enter(&dp->dp_lock); 755 dp->dp_dirty_pertxg[tx->tx_txg & TXG_MASK] += space; 756 dsl_pool_dirty_delta(dp, space); 757 mutex_exit(&dp->dp_lock); 758 } 759} 760 761void 762dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg) 763{ 764 ASSERT3S(space, >=, 0); 765 if (space == 0) 766 return; 767 mutex_enter(&dp->dp_lock); 768 if (dp->dp_dirty_pertxg[txg & TXG_MASK] < space) { 769 /* XXX writing something we didn't dirty? */ 770 space = dp->dp_dirty_pertxg[txg & TXG_MASK]; 771 } 772 ASSERT3U(dp->dp_dirty_pertxg[txg & TXG_MASK], >=, space); 773 dp->dp_dirty_pertxg[txg & TXG_MASK] -= space; 774 ASSERT3U(dp->dp_dirty_total, >=, space); 775 dsl_pool_dirty_delta(dp, -space); 776 mutex_exit(&dp->dp_lock); 777} 778 779/* ARGSUSED */ 780static int 781upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) 782{ 783 dmu_tx_t *tx = arg; 784 dsl_dataset_t *ds, *prev = NULL; 785 int err; 786 787 err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); 788 if (err) 789 return (err); 790 791 while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 792 err = dsl_dataset_hold_obj(dp, 793 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 794 if (err) { 795 dsl_dataset_rele(ds, FTAG); 796 return (err); 797 } 798 799 if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) 800 break; 801 dsl_dataset_rele(ds, FTAG); 802 ds = prev; 803 prev = NULL; 804 } 805 806 if (prev == NULL) { 807 prev = dp->dp_origin_snap; 808 809 /* 810 * The $ORIGIN can't have any data, or the accounting 811 * will be wrong. 812 */ 813 ASSERT0(dsl_dataset_phys(prev)->ds_bp.blk_birth); 814 815 /* The origin doesn't get attached to itself */ 816 if (ds->ds_object == prev->ds_object) { 817 dsl_dataset_rele(ds, FTAG); 818 return (0); 819 } 820 821 dmu_buf_will_dirty(ds->ds_dbuf, tx); 822 dsl_dataset_phys(ds)->ds_prev_snap_obj = prev->ds_object; 823 dsl_dataset_phys(ds)->ds_prev_snap_txg = 824 dsl_dataset_phys(prev)->ds_creation_txg; 825 826 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 827 dsl_dir_phys(ds->ds_dir)->dd_origin_obj = prev->ds_object; 828 829 dmu_buf_will_dirty(prev->ds_dbuf, tx); 830 dsl_dataset_phys(prev)->ds_num_children++; 831 832 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) { 833 ASSERT(ds->ds_prev == NULL); 834 VERIFY0(dsl_dataset_hold_obj(dp, 835 dsl_dataset_phys(ds)->ds_prev_snap_obj, 836 ds, &ds->ds_prev)); 837 } 838 } 839 840 ASSERT3U(dsl_dir_phys(ds->ds_dir)->dd_origin_obj, ==, prev->ds_object); 841 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_obj, ==, prev->ds_object); 842 843 if (dsl_dataset_phys(prev)->ds_next_clones_obj == 0) { 844 dmu_buf_will_dirty(prev->ds_dbuf, tx); 845 dsl_dataset_phys(prev)->ds_next_clones_obj = 846 zap_create(dp->dp_meta_objset, 847 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 848 } 849 VERIFY0(zap_add_int(dp->dp_meta_objset, 850 dsl_dataset_phys(prev)->ds_next_clones_obj, ds->ds_object, tx)); 851 852 dsl_dataset_rele(ds, FTAG); 853 if (prev != dp->dp_origin_snap) 854 dsl_dataset_rele(prev, FTAG); 855 return (0); 856} 857 858void 859dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) 860{ 861 ASSERT(dmu_tx_is_syncing(tx)); 862 ASSERT(dp->dp_origin_snap != NULL); 863 864 VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb, 865 tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); 866} 867 868/* ARGSUSED */ 869static int 870upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 871{ 872 dmu_tx_t *tx = arg; 873 objset_t *mos = dp->dp_meta_objset; 874 875 if (dsl_dir_phys(ds->ds_dir)->dd_origin_obj != 0) { 876 dsl_dataset_t *origin; 877 878 VERIFY0(dsl_dataset_hold_obj(dp, 879 dsl_dir_phys(ds->ds_dir)->dd_origin_obj, FTAG, &origin)); 880 881 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 882 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 883 dsl_dir_phys(origin->ds_dir)->dd_clones = 884 zap_create(mos, DMU_OT_DSL_CLONES, DMU_OT_NONE, 885 0, tx); 886 } 887 888 VERIFY0(zap_add_int(dp->dp_meta_objset, 889 dsl_dir_phys(origin->ds_dir)->dd_clones, 890 ds->ds_object, tx)); 891 892 dsl_dataset_rele(origin, FTAG); 893 } 894 return (0); 895} 896 897void 898dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) 899{ 900 ASSERT(dmu_tx_is_syncing(tx)); 901 uint64_t obj; 902 903 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); 904 VERIFY0(dsl_pool_open_special_dir(dp, 905 FREE_DIR_NAME, &dp->dp_free_dir)); 906 907 /* 908 * We can't use bpobj_alloc(), because spa_version() still 909 * returns the old version, and we need a new-version bpobj with 910 * subobj support. So call dmu_object_alloc() directly. 911 */ 912 obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, 913 SPA_OLD_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); 914 VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 915 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); 916 VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); 917 918 VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 919 upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); 920} 921 922void 923dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) 924{ 925 uint64_t dsobj; 926 dsl_dataset_t *ds; 927 928 ASSERT(dmu_tx_is_syncing(tx)); 929 ASSERT(dp->dp_origin_snap == NULL); 930 ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER)); 931 932 /* create the origin dir, ds, & snap-ds */ 933 dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, 934 NULL, 0, kcred, tx); 935 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 936 dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx); 937 VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, 938 dp, &dp->dp_origin_snap)); 939 dsl_dataset_rele(ds, FTAG); 940} 941 942taskq_t * 943dsl_pool_vnrele_taskq(dsl_pool_t *dp) 944{ 945 return (dp->dp_vnrele_taskq); 946} 947 948/* 949 * Walk through the pool-wide zap object of temporary snapshot user holds 950 * and release them. 951 */ 952void 953dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) 954{ 955 zap_attribute_t za; 956 zap_cursor_t zc; 957 objset_t *mos = dp->dp_meta_objset; 958 uint64_t zapobj = dp->dp_tmp_userrefs_obj; 959 nvlist_t *holds; 960 961 if (zapobj == 0) 962 return; 963 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 964 965 holds = fnvlist_alloc(); 966 967 for (zap_cursor_init(&zc, mos, zapobj); 968 zap_cursor_retrieve(&zc, &za) == 0; 969 zap_cursor_advance(&zc)) { 970 char *htag; 971 nvlist_t *tags; 972 973 htag = strchr(za.za_name, '-'); 974 *htag = '\0'; 975 ++htag; 976 if (nvlist_lookup_nvlist(holds, za.za_name, &tags) != 0) { 977 tags = fnvlist_alloc(); 978 fnvlist_add_boolean(tags, htag); 979 fnvlist_add_nvlist(holds, za.za_name, tags); 980 fnvlist_free(tags); 981 } else { 982 fnvlist_add_boolean(tags, htag); 983 } 984 } 985 dsl_dataset_user_release_tmp(dp, holds); 986 fnvlist_free(holds); 987 zap_cursor_fini(&zc); 988} 989 990/* 991 * Create the pool-wide zap object for storing temporary snapshot holds. 992 */ 993void 994dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) 995{ 996 objset_t *mos = dp->dp_meta_objset; 997 998 ASSERT(dp->dp_tmp_userrefs_obj == 0); 999 ASSERT(dmu_tx_is_syncing(tx)); 1000 1001 dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, 1002 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); 1003} 1004 1005static int 1006dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, 1007 const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding) 1008{ 1009 objset_t *mos = dp->dp_meta_objset; 1010 uint64_t zapobj = dp->dp_tmp_userrefs_obj; 1011 char *name; 1012 int error; 1013 1014 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 1015 ASSERT(dmu_tx_is_syncing(tx)); 1016 1017 /* 1018 * If the pool was created prior to SPA_VERSION_USERREFS, the 1019 * zap object for temporary holds might not exist yet. 1020 */ 1021 if (zapobj == 0) { 1022 if (holding) { 1023 dsl_pool_user_hold_create_obj(dp, tx); 1024 zapobj = dp->dp_tmp_userrefs_obj; 1025 } else { 1026 return (SET_ERROR(ENOENT)); 1027 } 1028 } 1029 1030 name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); 1031 if (holding) 1032 error = zap_add(mos, zapobj, name, 8, 1, &now, tx); 1033 else 1034 error = zap_remove(mos, zapobj, name, tx); 1035 strfree(name); 1036 1037 return (error); 1038} 1039 1040/* 1041 * Add a temporary hold for the given dataset object and tag. 1042 */ 1043int 1044dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 1045 uint64_t now, dmu_tx_t *tx) 1046{ 1047 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); 1048} 1049 1050/* 1051 * Release a temporary hold for the given dataset object and tag. 1052 */ 1053int 1054dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 1055 dmu_tx_t *tx) 1056{ 1057 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, 0, 1058 tx, B_FALSE)); 1059} 1060 1061/* 1062 * DSL Pool Configuration Lock 1063 * 1064 * The dp_config_rwlock protects against changes to DSL state (e.g. dataset 1065 * creation / destruction / rename / property setting). It must be held for 1066 * read to hold a dataset or dsl_dir. I.e. you must call 1067 * dsl_pool_config_enter() or dsl_pool_hold() before calling 1068 * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock 1069 * must be held continuously until all datasets and dsl_dirs are released. 1070 * 1071 * The only exception to this rule is that if a "long hold" is placed on 1072 * a dataset, then the dp_config_rwlock may be dropped while the dataset 1073 * is still held. The long hold will prevent the dataset from being 1074 * destroyed -- the destroy will fail with EBUSY. A long hold can be 1075 * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset 1076 * (by calling dsl_{dataset,objset}_{try}own{_obj}). 1077 * 1078 * Legitimate long-holders (including owners) should be long-running, cancelable 1079 * tasks that should cause "zfs destroy" to fail. This includes DMU 1080 * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open), 1081 * "zfs send", and "zfs diff". There are several other long-holders whose 1082 * uses are suboptimal (e.g. "zfs promote", and zil_suspend()). 1083 * 1084 * The usual formula for long-holding would be: 1085 * dsl_pool_hold() 1086 * dsl_dataset_hold() 1087 * ... perform checks ... 1088 * dsl_dataset_long_hold() 1089 * dsl_pool_rele() 1090 * ... perform long-running task ... 1091 * dsl_dataset_long_rele() 1092 * dsl_dataset_rele() 1093 * 1094 * Note that when the long hold is released, the dataset is still held but 1095 * the pool is not held. The dataset may change arbitrarily during this time 1096 * (e.g. it could be destroyed). Therefore you shouldn't do anything to the 1097 * dataset except release it. 1098 * 1099 * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only 1100 * or modifying operations. 1101 * 1102 * Modifying operations should generally use dsl_sync_task(). The synctask 1103 * infrastructure enforces proper locking strategy with respect to the 1104 * dp_config_rwlock. See the comment above dsl_sync_task() for details. 1105 * 1106 * Read-only operations will manually hold the pool, then the dataset, obtain 1107 * information from the dataset, then release the pool and dataset. 1108 * dmu_objset_{hold,rele}() are convenience routines that also do the pool 1109 * hold/rele. 1110 */ 1111 1112int 1113dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp) 1114{ 1115 spa_t *spa; 1116 int error; 1117 1118 error = spa_open(name, &spa, tag); 1119 if (error == 0) { 1120 *dp = spa_get_dsl(spa); 1121 dsl_pool_config_enter(*dp, tag); 1122 } 1123 return (error); 1124} 1125 1126void 1127dsl_pool_rele(dsl_pool_t *dp, void *tag) 1128{ 1129 dsl_pool_config_exit(dp, tag); 1130 spa_close(dp->dp_spa, tag); 1131} 1132 1133void 1134dsl_pool_config_enter(dsl_pool_t *dp, void *tag) 1135{ 1136 /* 1137 * We use a "reentrant" reader-writer lock, but not reentrantly. 1138 * 1139 * The rrwlock can (with the track_all flag) track all reading threads, 1140 * which is very useful for debugging which code path failed to release 1141 * the lock, and for verifying that the *current* thread does hold 1142 * the lock. 1143 * 1144 * (Unlike a rwlock, which knows that N threads hold it for 1145 * read, but not *which* threads, so rw_held(RW_READER) returns TRUE 1146 * if any thread holds it for read, even if this thread doesn't). 1147 */ 1148 ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); 1149 rrw_enter(&dp->dp_config_rwlock, RW_READER, tag); 1150} 1151 1152void 1153dsl_pool_config_enter_prio(dsl_pool_t *dp, void *tag) 1154{ 1155 ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); 1156 rrw_enter_read_prio(&dp->dp_config_rwlock, tag); 1157} 1158 1159void 1160dsl_pool_config_exit(dsl_pool_t *dp, void *tag) 1161{ 1162 rrw_exit(&dp->dp_config_rwlock, tag); 1163} 1164 1165boolean_t 1166dsl_pool_config_held(dsl_pool_t *dp) 1167{ 1168 return (RRW_LOCK_HELD(&dp->dp_config_rwlock)); 1169} 1170 1171boolean_t 1172dsl_pool_config_held_writer(dsl_pool_t *dp) 1173{ 1174 return (RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1175} 1176