dsl_pool.c revision 308083
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, 2015 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 * Copyright (c) 2014 Integros [integros.com] 27 */ 28 29#include <sys/dsl_pool.h> 30#include <sys/dsl_dataset.h> 31#include <sys/dsl_prop.h> 32#include <sys/dsl_dir.h> 33#include <sys/dsl_synctask.h> 34#include <sys/dsl_scan.h> 35#include <sys/dnode.h> 36#include <sys/dmu_tx.h> 37#include <sys/dmu_objset.h> 38#include <sys/arc.h> 39#include <sys/zap.h> 40#include <sys/zio.h> 41#include <sys/zfs_context.h> 42#include <sys/fs/zfs.h> 43#include <sys/zfs_znode.h> 44#include <sys/spa_impl.h> 45#include <sys/dsl_deadlist.h> 46#include <sys/bptree.h> 47#include <sys/zfeature.h> 48#include <sys/zil_impl.h> 49#include <sys/dsl_userhold.h> 50 51#ifdef __FreeBSD__ 52#include <sys/types.h> 53#include <sys/sysctl.h> 54#endif 55 56/* 57 * ZFS Write Throttle 58 * ------------------ 59 * 60 * ZFS must limit the rate of incoming writes to the rate at which it is able 61 * to sync data modifications to the backend storage. Throttling by too much 62 * creates an artificial limit; throttling by too little can only be sustained 63 * for short periods and would lead to highly lumpy performance. On a per-pool 64 * basis, ZFS tracks the amount of modified (dirty) data. As operations change 65 * data, the amount of dirty data increases; as ZFS syncs out data, the amount 66 * of dirty data decreases. When the amount of dirty data exceeds a 67 * predetermined threshold further modifications are blocked until the amount 68 * of dirty data decreases (as data is synced out). 69 * 70 * The limit on dirty data is tunable, and should be adjusted according to 71 * both the IO capacity and available memory of the system. The larger the 72 * window, the more ZFS is able to aggregate and amortize metadata (and data) 73 * changes. However, memory is a limited resource, and allowing for more dirty 74 * data comes at the cost of keeping other useful data in memory (for example 75 * ZFS data cached by the ARC). 76 * 77 * Implementation 78 * 79 * As buffers are modified dsl_pool_willuse_space() increments both the per- 80 * txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of 81 * dirty space used; dsl_pool_dirty_space() decrements those values as data 82 * is synced out from dsl_pool_sync(). While only the poolwide value is 83 * relevant, the per-txg value is useful for debugging. The tunable 84 * zfs_dirty_data_max determines the dirty space limit. Once that value is 85 * exceeded, new writes are halted until space frees up. 86 * 87 * The zfs_dirty_data_sync tunable dictates the threshold at which we 88 * ensure that there is a txg syncing (see the comment in txg.c for a full 89 * description of transaction group stages). 90 * 91 * The IO scheduler uses both the dirty space limit and current amount of 92 * dirty data as inputs. Those values affect the number of concurrent IOs ZFS 93 * issues. See the comment in vdev_queue.c for details of the IO scheduler. 94 * 95 * The delay is also calculated based on the amount of dirty data. See the 96 * comment above dmu_tx_delay() for details. 97 */ 98 99/* 100 * zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory, 101 * capped at zfs_dirty_data_max_max. It can also be overridden in /etc/system. 102 */ 103uint64_t zfs_dirty_data_max; 104uint64_t zfs_dirty_data_max_max = 4ULL * 1024 * 1024 * 1024; 105int zfs_dirty_data_max_percent = 10; 106 107/* 108 * If there is at least this much dirty data, push out a txg. 109 */ 110uint64_t zfs_dirty_data_sync = 64 * 1024 * 1024; 111 112/* 113 * Once there is this amount of dirty data, the dmu_tx_delay() will kick in 114 * and delay each transaction. 115 * This value should be >= zfs_vdev_async_write_active_max_dirty_percent. 116 */ 117int zfs_delay_min_dirty_percent = 60; 118 119/* 120 * This controls how quickly the delay approaches infinity. 121 * Larger values cause it to delay more for a given amount of dirty data. 122 * Therefore larger values will cause there to be less dirty data for a 123 * given throughput. 124 * 125 * For the smoothest delay, this value should be about 1 billion divided 126 * by the maximum number of operations per second. This will smoothly 127 * handle between 10x and 1/10th this number. 128 * 129 * Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the 130 * multiply in dmu_tx_delay(). 131 */ 132uint64_t zfs_delay_scale = 1000 * 1000 * 1000 / 2000; 133 134 135#ifdef __FreeBSD__ 136 137extern int zfs_vdev_async_write_active_max_dirty_percent; 138 139SYSCTL_DECL(_vfs_zfs); 140 141TUNABLE_QUAD("vfs.zfs.dirty_data_max", &zfs_dirty_data_max); 142SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, dirty_data_max, CTLFLAG_RWTUN, 143 &zfs_dirty_data_max, 0, 144 "The maximum amount of dirty data in bytes after which new writes are " 145 "halted until space becomes available"); 146 147TUNABLE_QUAD("vfs.zfs.dirty_data_max_max", &zfs_dirty_data_max_max); 148SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, dirty_data_max_max, CTLFLAG_RDTUN, 149 &zfs_dirty_data_max_max, 0, 150 "The absolute cap on dirty_data_max when auto calculating"); 151 152TUNABLE_INT("vfs.zfs.dirty_data_max_percent", &zfs_dirty_data_max_percent); 153static int sysctl_zfs_dirty_data_max_percent(SYSCTL_HANDLER_ARGS); 154SYSCTL_PROC(_vfs_zfs, OID_AUTO, dirty_data_max_percent, 155 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RWTUN, 0, sizeof(int), 156 sysctl_zfs_dirty_data_max_percent, "I", 157 "The percent of physical memory used to auto calculate dirty_data_max"); 158 159TUNABLE_QUAD("vfs.zfs.dirty_data_sync", &zfs_dirty_data_sync); 160SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, dirty_data_sync, CTLFLAG_RWTUN, 161 &zfs_dirty_data_sync, 0, 162 "Force a txg if the number of dirty buffer bytes exceed this value"); 163 164static int sysctl_zfs_delay_min_dirty_percent(SYSCTL_HANDLER_ARGS); 165/* No zfs_delay_min_dirty_percent tunable due to limit requirements */ 166SYSCTL_PROC(_vfs_zfs, OID_AUTO, delay_min_dirty_percent, 167 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(int), 168 sysctl_zfs_delay_min_dirty_percent, "I", 169 "The limit of outstanding dirty data before transations are delayed"); 170 171static int sysctl_zfs_delay_scale(SYSCTL_HANDLER_ARGS); 172/* No zfs_delay_scale tunable due to limit requirements */ 173SYSCTL_PROC(_vfs_zfs, OID_AUTO, delay_scale, 174 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t), 175 sysctl_zfs_delay_scale, "QU", 176 "Controls how quickly the delay approaches infinity"); 177 178static int 179sysctl_zfs_dirty_data_max_percent(SYSCTL_HANDLER_ARGS) 180{ 181 int val, err; 182 183 val = zfs_dirty_data_max_percent; 184 err = sysctl_handle_int(oidp, &val, 0, req); 185 if (err != 0 || req->newptr == NULL) 186 return (err); 187 188 if (val < 0 || val > 100) 189 return (EINVAL); 190 191 zfs_dirty_data_max_percent = val; 192 193 return (0); 194} 195 196static int 197sysctl_zfs_delay_min_dirty_percent(SYSCTL_HANDLER_ARGS) 198{ 199 int val, err; 200 201 val = zfs_delay_min_dirty_percent; 202 err = sysctl_handle_int(oidp, &val, 0, req); 203 if (err != 0 || req->newptr == NULL) 204 return (err); 205 206 if (val < zfs_vdev_async_write_active_max_dirty_percent) 207 return (EINVAL); 208 209 zfs_delay_min_dirty_percent = val; 210 211 return (0); 212} 213 214static int 215sysctl_zfs_delay_scale(SYSCTL_HANDLER_ARGS) 216{ 217 uint64_t val; 218 int err; 219 220 val = zfs_delay_scale; 221 err = sysctl_handle_64(oidp, &val, 0, req); 222 if (err != 0 || req->newptr == NULL) 223 return (err); 224 225 if (val > UINT64_MAX / zfs_dirty_data_max) 226 return (EINVAL); 227 228 zfs_delay_scale = val; 229 230 return (0); 231} 232#endif 233 234hrtime_t zfs_throttle_delay = MSEC2NSEC(10); 235hrtime_t zfs_throttle_resolution = MSEC2NSEC(10); 236 237int 238dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) 239{ 240 uint64_t obj; 241 int err; 242 243 err = zap_lookup(dp->dp_meta_objset, 244 dsl_dir_phys(dp->dp_root_dir)->dd_child_dir_zapobj, 245 name, sizeof (obj), 1, &obj); 246 if (err) 247 return (err); 248 249 return (dsl_dir_hold_obj(dp, obj, name, dp, ddp)); 250} 251 252static dsl_pool_t * 253dsl_pool_open_impl(spa_t *spa, uint64_t txg) 254{ 255 dsl_pool_t *dp; 256 blkptr_t *bp = spa_get_rootblkptr(spa); 257 258 dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); 259 dp->dp_spa = spa; 260 dp->dp_meta_rootbp = *bp; 261 rrw_init(&dp->dp_config_rwlock, B_TRUE); 262 txg_init(dp, txg); 263 264 txg_list_create(&dp->dp_dirty_datasets, 265 offsetof(dsl_dataset_t, ds_dirty_link)); 266 txg_list_create(&dp->dp_dirty_zilogs, 267 offsetof(zilog_t, zl_dirty_link)); 268 txg_list_create(&dp->dp_dirty_dirs, 269 offsetof(dsl_dir_t, dd_dirty_link)); 270 txg_list_create(&dp->dp_sync_tasks, 271 offsetof(dsl_sync_task_t, dst_node)); 272 273 mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); 274 cv_init(&dp->dp_spaceavail_cv, NULL, CV_DEFAULT, NULL); 275 276 dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri, 277 1, 4, 0); 278 279 return (dp); 280} 281 282int 283dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) 284{ 285 int err; 286 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 287 288 err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, 289 &dp->dp_meta_objset); 290 if (err != 0) 291 dsl_pool_close(dp); 292 else 293 *dpp = dp; 294 295 return (err); 296} 297 298int 299dsl_pool_open(dsl_pool_t *dp) 300{ 301 int err; 302 dsl_dir_t *dd; 303 dsl_dataset_t *ds; 304 uint64_t obj; 305 306 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 307 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 308 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, 309 &dp->dp_root_dir_obj); 310 if (err) 311 goto out; 312 313 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, 314 NULL, dp, &dp->dp_root_dir); 315 if (err) 316 goto out; 317 318 err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); 319 if (err) 320 goto out; 321 322 if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { 323 err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); 324 if (err) 325 goto out; 326 err = dsl_dataset_hold_obj(dp, 327 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds); 328 if (err == 0) { 329 err = dsl_dataset_hold_obj(dp, 330 dsl_dataset_phys(ds)->ds_prev_snap_obj, dp, 331 &dp->dp_origin_snap); 332 dsl_dataset_rele(ds, FTAG); 333 } 334 dsl_dir_rele(dd, dp); 335 if (err) 336 goto out; 337 } 338 339 if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { 340 err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, 341 &dp->dp_free_dir); 342 if (err) 343 goto out; 344 345 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 346 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); 347 if (err) 348 goto out; 349 VERIFY0(bpobj_open(&dp->dp_free_bpobj, 350 dp->dp_meta_objset, obj)); 351 } 352 353 /* 354 * Note: errors ignored, because the leak dir will not exist if we 355 * have not encountered a leak yet. 356 */ 357 (void) dsl_pool_open_special_dir(dp, LEAK_DIR_NAME, 358 &dp->dp_leak_dir); 359 360 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) { 361 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 362 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, 363 &dp->dp_bptree_obj); 364 if (err != 0) 365 goto out; 366 } 367 368 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMPTY_BPOBJ)) { 369 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 370 DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, 371 &dp->dp_empty_bpobj); 372 if (err != 0) 373 goto out; 374 } 375 376 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 377 DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, 378 &dp->dp_tmp_userrefs_obj); 379 if (err == ENOENT) 380 err = 0; 381 if (err) 382 goto out; 383 384 err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); 385 386out: 387 rrw_exit(&dp->dp_config_rwlock, FTAG); 388 return (err); 389} 390 391void 392dsl_pool_close(dsl_pool_t *dp) 393{ 394 /* 395 * Drop our references from dsl_pool_open(). 396 * 397 * Since we held the origin_snap from "syncing" context (which 398 * includes pool-opening context), it actually only got a "ref" 399 * and not a hold, so just drop that here. 400 */ 401 if (dp->dp_origin_snap) 402 dsl_dataset_rele(dp->dp_origin_snap, dp); 403 if (dp->dp_mos_dir) 404 dsl_dir_rele(dp->dp_mos_dir, dp); 405 if (dp->dp_free_dir) 406 dsl_dir_rele(dp->dp_free_dir, dp); 407 if (dp->dp_leak_dir) 408 dsl_dir_rele(dp->dp_leak_dir, dp); 409 if (dp->dp_root_dir) 410 dsl_dir_rele(dp->dp_root_dir, dp); 411 412 bpobj_close(&dp->dp_free_bpobj); 413 414 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ 415 if (dp->dp_meta_objset) 416 dmu_objset_evict(dp->dp_meta_objset); 417 418 txg_list_destroy(&dp->dp_dirty_datasets); 419 txg_list_destroy(&dp->dp_dirty_zilogs); 420 txg_list_destroy(&dp->dp_sync_tasks); 421 txg_list_destroy(&dp->dp_dirty_dirs); 422 423 /* 424 * We can't set retry to TRUE since we're explicitly specifying 425 * a spa to flush. This is good enough; any missed buffers for 426 * this spa won't cause trouble, and they'll eventually fall 427 * out of the ARC just like any other unused buffer. 428 */ 429 arc_flush(dp->dp_spa, FALSE); 430 431 txg_fini(dp); 432 dsl_scan_fini(dp); 433 dmu_buf_user_evict_wait(); 434 435 rrw_destroy(&dp->dp_config_rwlock); 436 mutex_destroy(&dp->dp_lock); 437 taskq_destroy(dp->dp_vnrele_taskq); 438 if (dp->dp_blkstats) 439 kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); 440 kmem_free(dp, sizeof (dsl_pool_t)); 441} 442 443dsl_pool_t * 444dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) 445{ 446 int err; 447 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 448 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); 449 objset_t *os; 450 dsl_dataset_t *ds; 451 uint64_t obj; 452 453 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 454 455 /* create and open the MOS (meta-objset) */ 456 dp->dp_meta_objset = dmu_objset_create_impl(spa, 457 NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); 458 459 /* create the pool directory */ 460 err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 461 DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); 462 ASSERT0(err); 463 464 /* Initialize scan structures */ 465 VERIFY0(dsl_scan_init(dp, txg)); 466 467 /* create and open the root dir */ 468 dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); 469 VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, 470 NULL, dp, &dp->dp_root_dir)); 471 472 /* create and open the meta-objset dir */ 473 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); 474 VERIFY0(dsl_pool_open_special_dir(dp, 475 MOS_DIR_NAME, &dp->dp_mos_dir)); 476 477 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { 478 /* create and open the free dir */ 479 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, 480 FREE_DIR_NAME, tx); 481 VERIFY0(dsl_pool_open_special_dir(dp, 482 FREE_DIR_NAME, &dp->dp_free_dir)); 483 484 /* create and open the free_bplist */ 485 obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx); 486 VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 487 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); 488 VERIFY0(bpobj_open(&dp->dp_free_bpobj, 489 dp->dp_meta_objset, obj)); 490 } 491 492 if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) 493 dsl_pool_create_origin(dp, tx); 494 495 /* create the root dataset */ 496 obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); 497 498 /* create the root objset */ 499 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); 500 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 501 os = dmu_objset_create_impl(dp->dp_spa, ds, 502 dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx); 503 rrw_exit(&ds->ds_bp_rwlock, FTAG); 504#ifdef _KERNEL 505 zfs_create_fs(os, kcred, zplprops, tx); 506#endif 507 dsl_dataset_rele(ds, FTAG); 508 509 dmu_tx_commit(tx); 510 511 rrw_exit(&dp->dp_config_rwlock, FTAG); 512 513 return (dp); 514} 515 516/* 517 * Account for the meta-objset space in its placeholder dsl_dir. 518 */ 519void 520dsl_pool_mos_diduse_space(dsl_pool_t *dp, 521 int64_t used, int64_t comp, int64_t uncomp) 522{ 523 ASSERT3U(comp, ==, uncomp); /* it's all metadata */ 524 mutex_enter(&dp->dp_lock); 525 dp->dp_mos_used_delta += used; 526 dp->dp_mos_compressed_delta += comp; 527 dp->dp_mos_uncompressed_delta += uncomp; 528 mutex_exit(&dp->dp_lock); 529} 530 531static int 532deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 533{ 534 dsl_deadlist_t *dl = arg; 535 dsl_deadlist_insert(dl, bp, tx); 536 return (0); 537} 538 539static void 540dsl_pool_sync_mos(dsl_pool_t *dp, dmu_tx_t *tx) 541{ 542 zio_t *zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 543 dmu_objset_sync(dp->dp_meta_objset, zio, tx); 544 VERIFY0(zio_wait(zio)); 545 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); 546 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); 547} 548 549static void 550dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta) 551{ 552 ASSERT(MUTEX_HELD(&dp->dp_lock)); 553 554 if (delta < 0) 555 ASSERT3U(-delta, <=, dp->dp_dirty_total); 556 557 dp->dp_dirty_total += delta; 558 559 /* 560 * Note: we signal even when increasing dp_dirty_total. 561 * This ensures forward progress -- each thread wakes the next waiter. 562 */ 563 if (dp->dp_dirty_total <= zfs_dirty_data_max) 564 cv_signal(&dp->dp_spaceavail_cv); 565} 566 567void 568dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) 569{ 570 zio_t *zio; 571 dmu_tx_t *tx; 572 dsl_dir_t *dd; 573 dsl_dataset_t *ds; 574 objset_t *mos = dp->dp_meta_objset; 575 list_t synced_datasets; 576 577 list_create(&synced_datasets, sizeof (dsl_dataset_t), 578 offsetof(dsl_dataset_t, ds_synced_link)); 579 580 tx = dmu_tx_create_assigned(dp, txg); 581 582 /* 583 * Write out all dirty blocks of dirty datasets. 584 */ 585 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 586 while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 587 /* 588 * We must not sync any non-MOS datasets twice, because 589 * we may have taken a snapshot of them. However, we 590 * may sync newly-created datasets on pass 2. 591 */ 592 ASSERT(!list_link_active(&ds->ds_synced_link)); 593 list_insert_tail(&synced_datasets, ds); 594 dsl_dataset_sync(ds, zio, tx); 595 } 596 VERIFY0(zio_wait(zio)); 597 598 /* 599 * We have written all of the accounted dirty data, so our 600 * dp_space_towrite should now be zero. However, some seldom-used 601 * code paths do not adhere to this (e.g. dbuf_undirty(), also 602 * rounding error in dbuf_write_physdone). 603 * Shore up the accounting of any dirtied space now. 604 */ 605 dsl_pool_undirty_space(dp, dp->dp_dirty_pertxg[txg & TXG_MASK], txg); 606 607 /* 608 * After the data blocks have been written (ensured by the zio_wait() 609 * above), update the user/group space accounting. 610 */ 611 for (ds = list_head(&synced_datasets); ds != NULL; 612 ds = list_next(&synced_datasets, ds)) { 613 dmu_objset_do_userquota_updates(ds->ds_objset, tx); 614 } 615 616 /* 617 * Sync the datasets again to push out the changes due to 618 * userspace updates. This must be done before we process the 619 * sync tasks, so that any snapshots will have the correct 620 * user accounting information (and we won't get confused 621 * about which blocks are part of the snapshot). 622 */ 623 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 624 while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 625 ASSERT(list_link_active(&ds->ds_synced_link)); 626 dmu_buf_rele(ds->ds_dbuf, ds); 627 dsl_dataset_sync(ds, zio, tx); 628 } 629 VERIFY0(zio_wait(zio)); 630 631 /* 632 * Now that the datasets have been completely synced, we can 633 * clean up our in-memory structures accumulated while syncing: 634 * 635 * - move dead blocks from the pending deadlist to the on-disk deadlist 636 * - release hold from dsl_dataset_dirty() 637 */ 638 while ((ds = list_remove_head(&synced_datasets)) != NULL) { 639 objset_t *os = ds->ds_objset; 640 bplist_iterate(&ds->ds_pending_deadlist, 641 deadlist_enqueue_cb, &ds->ds_deadlist, tx); 642 ASSERT(!dmu_objset_is_dirty(os, txg)); 643 dmu_buf_rele(ds->ds_dbuf, ds); 644 } 645 while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) != NULL) { 646 dsl_dir_sync(dd, tx); 647 } 648 649 /* 650 * The MOS's space is accounted for in the pool/$MOS 651 * (dp_mos_dir). We can't modify the mos while we're syncing 652 * it, so we remember the deltas and apply them here. 653 */ 654 if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || 655 dp->dp_mos_uncompressed_delta != 0) { 656 dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, 657 dp->dp_mos_used_delta, 658 dp->dp_mos_compressed_delta, 659 dp->dp_mos_uncompressed_delta, tx); 660 dp->dp_mos_used_delta = 0; 661 dp->dp_mos_compressed_delta = 0; 662 dp->dp_mos_uncompressed_delta = 0; 663 } 664 665 if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || 666 list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { 667 dsl_pool_sync_mos(dp, tx); 668 } 669 670 /* 671 * If we modify a dataset in the same txg that we want to destroy it, 672 * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. 673 * dsl_dir_destroy_check() will fail if there are unexpected holds. 674 * Therefore, we want to sync the MOS (thus syncing the dd_dbuf 675 * and clearing the hold on it) before we process the sync_tasks. 676 * The MOS data dirtied by the sync_tasks will be synced on the next 677 * pass. 678 */ 679 if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { 680 dsl_sync_task_t *dst; 681 /* 682 * No more sync tasks should have been added while we 683 * were syncing. 684 */ 685 ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1); 686 while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)) != NULL) 687 dsl_sync_task_sync(dst, tx); 688 } 689 690 dmu_tx_commit(tx); 691 692 DTRACE_PROBE2(dsl_pool_sync__done, dsl_pool_t *dp, dp, uint64_t, txg); 693} 694 695void 696dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) 697{ 698 zilog_t *zilog; 699 700 while (zilog = txg_list_remove(&dp->dp_dirty_zilogs, txg)) { 701 dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); 702 zil_clean(zilog, txg); 703 ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); 704 dmu_buf_rele(ds->ds_dbuf, zilog); 705 } 706 ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); 707} 708 709/* 710 * TRUE if the current thread is the tx_sync_thread or if we 711 * are being called from SPA context during pool initialization. 712 */ 713int 714dsl_pool_sync_context(dsl_pool_t *dp) 715{ 716 return (curthread == dp->dp_tx.tx_sync_thread || 717 spa_is_initializing(dp->dp_spa)); 718} 719 720uint64_t 721dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) 722{ 723 uint64_t space, resv; 724 725 /* 726 * If we're trying to assess whether it's OK to do a free, 727 * cut the reservation in half to allow forward progress 728 * (e.g. make it possible to rm(1) files from a full pool). 729 */ 730 space = spa_get_dspace(dp->dp_spa); 731 resv = spa_get_slop_space(dp->dp_spa); 732 if (netfree) 733 resv >>= 1; 734 735 return (space - resv); 736} 737 738boolean_t 739dsl_pool_need_dirty_delay(dsl_pool_t *dp) 740{ 741 uint64_t delay_min_bytes = 742 zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100; 743 boolean_t rv; 744 745 mutex_enter(&dp->dp_lock); 746 if (dp->dp_dirty_total > zfs_dirty_data_sync) 747 txg_kick(dp); 748 rv = (dp->dp_dirty_total > delay_min_bytes); 749 mutex_exit(&dp->dp_lock); 750 return (rv); 751} 752 753void 754dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 755{ 756 if (space > 0) { 757 mutex_enter(&dp->dp_lock); 758 dp->dp_dirty_pertxg[tx->tx_txg & TXG_MASK] += space; 759 dsl_pool_dirty_delta(dp, space); 760 mutex_exit(&dp->dp_lock); 761 } 762} 763 764void 765dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg) 766{ 767 ASSERT3S(space, >=, 0); 768 if (space == 0) 769 return; 770 mutex_enter(&dp->dp_lock); 771 if (dp->dp_dirty_pertxg[txg & TXG_MASK] < space) { 772 /* XXX writing something we didn't dirty? */ 773 space = dp->dp_dirty_pertxg[txg & TXG_MASK]; 774 } 775 ASSERT3U(dp->dp_dirty_pertxg[txg & TXG_MASK], >=, space); 776 dp->dp_dirty_pertxg[txg & TXG_MASK] -= space; 777 ASSERT3U(dp->dp_dirty_total, >=, space); 778 dsl_pool_dirty_delta(dp, -space); 779 mutex_exit(&dp->dp_lock); 780} 781 782/* ARGSUSED */ 783static int 784upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) 785{ 786 dmu_tx_t *tx = arg; 787 dsl_dataset_t *ds, *prev = NULL; 788 int err; 789 790 err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); 791 if (err) 792 return (err); 793 794 while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 795 err = dsl_dataset_hold_obj(dp, 796 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 797 if (err) { 798 dsl_dataset_rele(ds, FTAG); 799 return (err); 800 } 801 802 if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) 803 break; 804 dsl_dataset_rele(ds, FTAG); 805 ds = prev; 806 prev = NULL; 807 } 808 809 if (prev == NULL) { 810 prev = dp->dp_origin_snap; 811 812 /* 813 * The $ORIGIN can't have any data, or the accounting 814 * will be wrong. 815 */ 816 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 817 ASSERT0(dsl_dataset_phys(prev)->ds_bp.blk_birth); 818 rrw_exit(&ds->ds_bp_rwlock, FTAG); 819 820 /* The origin doesn't get attached to itself */ 821 if (ds->ds_object == prev->ds_object) { 822 dsl_dataset_rele(ds, FTAG); 823 return (0); 824 } 825 826 dmu_buf_will_dirty(ds->ds_dbuf, tx); 827 dsl_dataset_phys(ds)->ds_prev_snap_obj = prev->ds_object; 828 dsl_dataset_phys(ds)->ds_prev_snap_txg = 829 dsl_dataset_phys(prev)->ds_creation_txg; 830 831 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 832 dsl_dir_phys(ds->ds_dir)->dd_origin_obj = prev->ds_object; 833 834 dmu_buf_will_dirty(prev->ds_dbuf, tx); 835 dsl_dataset_phys(prev)->ds_num_children++; 836 837 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) { 838 ASSERT(ds->ds_prev == NULL); 839 VERIFY0(dsl_dataset_hold_obj(dp, 840 dsl_dataset_phys(ds)->ds_prev_snap_obj, 841 ds, &ds->ds_prev)); 842 } 843 } 844 845 ASSERT3U(dsl_dir_phys(ds->ds_dir)->dd_origin_obj, ==, prev->ds_object); 846 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_obj, ==, prev->ds_object); 847 848 if (dsl_dataset_phys(prev)->ds_next_clones_obj == 0) { 849 dmu_buf_will_dirty(prev->ds_dbuf, tx); 850 dsl_dataset_phys(prev)->ds_next_clones_obj = 851 zap_create(dp->dp_meta_objset, 852 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 853 } 854 VERIFY0(zap_add_int(dp->dp_meta_objset, 855 dsl_dataset_phys(prev)->ds_next_clones_obj, ds->ds_object, tx)); 856 857 dsl_dataset_rele(ds, FTAG); 858 if (prev != dp->dp_origin_snap) 859 dsl_dataset_rele(prev, FTAG); 860 return (0); 861} 862 863void 864dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) 865{ 866 ASSERT(dmu_tx_is_syncing(tx)); 867 ASSERT(dp->dp_origin_snap != NULL); 868 869 VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb, 870 tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); 871} 872 873/* ARGSUSED */ 874static int 875upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 876{ 877 dmu_tx_t *tx = arg; 878 objset_t *mos = dp->dp_meta_objset; 879 880 if (dsl_dir_phys(ds->ds_dir)->dd_origin_obj != 0) { 881 dsl_dataset_t *origin; 882 883 VERIFY0(dsl_dataset_hold_obj(dp, 884 dsl_dir_phys(ds->ds_dir)->dd_origin_obj, FTAG, &origin)); 885 886 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 887 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 888 dsl_dir_phys(origin->ds_dir)->dd_clones = 889 zap_create(mos, DMU_OT_DSL_CLONES, DMU_OT_NONE, 890 0, tx); 891 } 892 893 VERIFY0(zap_add_int(dp->dp_meta_objset, 894 dsl_dir_phys(origin->ds_dir)->dd_clones, 895 ds->ds_object, tx)); 896 897 dsl_dataset_rele(origin, FTAG); 898 } 899 return (0); 900} 901 902void 903dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) 904{ 905 ASSERT(dmu_tx_is_syncing(tx)); 906 uint64_t obj; 907 908 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); 909 VERIFY0(dsl_pool_open_special_dir(dp, 910 FREE_DIR_NAME, &dp->dp_free_dir)); 911 912 /* 913 * We can't use bpobj_alloc(), because spa_version() still 914 * returns the old version, and we need a new-version bpobj with 915 * subobj support. So call dmu_object_alloc() directly. 916 */ 917 obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, 918 SPA_OLD_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); 919 VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 920 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); 921 VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); 922 923 VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 924 upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); 925} 926 927void 928dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) 929{ 930 uint64_t dsobj; 931 dsl_dataset_t *ds; 932 933 ASSERT(dmu_tx_is_syncing(tx)); 934 ASSERT(dp->dp_origin_snap == NULL); 935 ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER)); 936 937 /* create the origin dir, ds, & snap-ds */ 938 dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, 939 NULL, 0, kcred, tx); 940 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 941 dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx); 942 VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, 943 dp, &dp->dp_origin_snap)); 944 dsl_dataset_rele(ds, FTAG); 945} 946 947taskq_t * 948dsl_pool_vnrele_taskq(dsl_pool_t *dp) 949{ 950 return (dp->dp_vnrele_taskq); 951} 952 953/* 954 * Walk through the pool-wide zap object of temporary snapshot user holds 955 * and release them. 956 */ 957void 958dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) 959{ 960 zap_attribute_t za; 961 zap_cursor_t zc; 962 objset_t *mos = dp->dp_meta_objset; 963 uint64_t zapobj = dp->dp_tmp_userrefs_obj; 964 nvlist_t *holds; 965 966 if (zapobj == 0) 967 return; 968 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 969 970 holds = fnvlist_alloc(); 971 972 for (zap_cursor_init(&zc, mos, zapobj); 973 zap_cursor_retrieve(&zc, &za) == 0; 974 zap_cursor_advance(&zc)) { 975 char *htag; 976 nvlist_t *tags; 977 978 htag = strchr(za.za_name, '-'); 979 *htag = '\0'; 980 ++htag; 981 if (nvlist_lookup_nvlist(holds, za.za_name, &tags) != 0) { 982 tags = fnvlist_alloc(); 983 fnvlist_add_boolean(tags, htag); 984 fnvlist_add_nvlist(holds, za.za_name, tags); 985 fnvlist_free(tags); 986 } else { 987 fnvlist_add_boolean(tags, htag); 988 } 989 } 990 dsl_dataset_user_release_tmp(dp, holds); 991 fnvlist_free(holds); 992 zap_cursor_fini(&zc); 993} 994 995/* 996 * Create the pool-wide zap object for storing temporary snapshot holds. 997 */ 998void 999dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) 1000{ 1001 objset_t *mos = dp->dp_meta_objset; 1002 1003 ASSERT(dp->dp_tmp_userrefs_obj == 0); 1004 ASSERT(dmu_tx_is_syncing(tx)); 1005 1006 dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, 1007 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); 1008} 1009 1010static int 1011dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, 1012 const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding) 1013{ 1014 objset_t *mos = dp->dp_meta_objset; 1015 uint64_t zapobj = dp->dp_tmp_userrefs_obj; 1016 char *name; 1017 int error; 1018 1019 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 1020 ASSERT(dmu_tx_is_syncing(tx)); 1021 1022 /* 1023 * If the pool was created prior to SPA_VERSION_USERREFS, the 1024 * zap object for temporary holds might not exist yet. 1025 */ 1026 if (zapobj == 0) { 1027 if (holding) { 1028 dsl_pool_user_hold_create_obj(dp, tx); 1029 zapobj = dp->dp_tmp_userrefs_obj; 1030 } else { 1031 return (SET_ERROR(ENOENT)); 1032 } 1033 } 1034 1035 name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); 1036 if (holding) 1037 error = zap_add(mos, zapobj, name, 8, 1, &now, tx); 1038 else 1039 error = zap_remove(mos, zapobj, name, tx); 1040 strfree(name); 1041 1042 return (error); 1043} 1044 1045/* 1046 * Add a temporary hold for the given dataset object and tag. 1047 */ 1048int 1049dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 1050 uint64_t now, dmu_tx_t *tx) 1051{ 1052 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); 1053} 1054 1055/* 1056 * Release a temporary hold for the given dataset object and tag. 1057 */ 1058int 1059dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 1060 dmu_tx_t *tx) 1061{ 1062 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, 0, 1063 tx, B_FALSE)); 1064} 1065 1066/* 1067 * DSL Pool Configuration Lock 1068 * 1069 * The dp_config_rwlock protects against changes to DSL state (e.g. dataset 1070 * creation / destruction / rename / property setting). It must be held for 1071 * read to hold a dataset or dsl_dir. I.e. you must call 1072 * dsl_pool_config_enter() or dsl_pool_hold() before calling 1073 * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock 1074 * must be held continuously until all datasets and dsl_dirs are released. 1075 * 1076 * The only exception to this rule is that if a "long hold" is placed on 1077 * a dataset, then the dp_config_rwlock may be dropped while the dataset 1078 * is still held. The long hold will prevent the dataset from being 1079 * destroyed -- the destroy will fail with EBUSY. A long hold can be 1080 * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset 1081 * (by calling dsl_{dataset,objset}_{try}own{_obj}). 1082 * 1083 * Legitimate long-holders (including owners) should be long-running, cancelable 1084 * tasks that should cause "zfs destroy" to fail. This includes DMU 1085 * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open), 1086 * "zfs send", and "zfs diff". There are several other long-holders whose 1087 * uses are suboptimal (e.g. "zfs promote", and zil_suspend()). 1088 * 1089 * The usual formula for long-holding would be: 1090 * dsl_pool_hold() 1091 * dsl_dataset_hold() 1092 * ... perform checks ... 1093 * dsl_dataset_long_hold() 1094 * dsl_pool_rele() 1095 * ... perform long-running task ... 1096 * dsl_dataset_long_rele() 1097 * dsl_dataset_rele() 1098 * 1099 * Note that when the long hold is released, the dataset is still held but 1100 * the pool is not held. The dataset may change arbitrarily during this time 1101 * (e.g. it could be destroyed). Therefore you shouldn't do anything to the 1102 * dataset except release it. 1103 * 1104 * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only 1105 * or modifying operations. 1106 * 1107 * Modifying operations should generally use dsl_sync_task(). The synctask 1108 * infrastructure enforces proper locking strategy with respect to the 1109 * dp_config_rwlock. See the comment above dsl_sync_task() for details. 1110 * 1111 * Read-only operations will manually hold the pool, then the dataset, obtain 1112 * information from the dataset, then release the pool and dataset. 1113 * dmu_objset_{hold,rele}() are convenience routines that also do the pool 1114 * hold/rele. 1115 */ 1116 1117int 1118dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp) 1119{ 1120 spa_t *spa; 1121 int error; 1122 1123 error = spa_open(name, &spa, tag); 1124 if (error == 0) { 1125 *dp = spa_get_dsl(spa); 1126 dsl_pool_config_enter(*dp, tag); 1127 } 1128 return (error); 1129} 1130 1131void 1132dsl_pool_rele(dsl_pool_t *dp, void *tag) 1133{ 1134 dsl_pool_config_exit(dp, tag); 1135 spa_close(dp->dp_spa, tag); 1136} 1137 1138void 1139dsl_pool_config_enter(dsl_pool_t *dp, void *tag) 1140{ 1141 /* 1142 * We use a "reentrant" reader-writer lock, but not reentrantly. 1143 * 1144 * The rrwlock can (with the track_all flag) track all reading threads, 1145 * which is very useful for debugging which code path failed to release 1146 * the lock, and for verifying that the *current* thread does hold 1147 * the lock. 1148 * 1149 * (Unlike a rwlock, which knows that N threads hold it for 1150 * read, but not *which* threads, so rw_held(RW_READER) returns TRUE 1151 * if any thread holds it for read, even if this thread doesn't). 1152 */ 1153 ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); 1154 rrw_enter(&dp->dp_config_rwlock, RW_READER, tag); 1155} 1156 1157void 1158dsl_pool_config_enter_prio(dsl_pool_t *dp, void *tag) 1159{ 1160 ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); 1161 rrw_enter_read_prio(&dp->dp_config_rwlock, tag); 1162} 1163 1164void 1165dsl_pool_config_exit(dsl_pool_t *dp, void *tag) 1166{ 1167 rrw_exit(&dp->dp_config_rwlock, tag); 1168} 1169 1170boolean_t 1171dsl_pool_config_held(dsl_pool_t *dp) 1172{ 1173 return (RRW_LOCK_HELD(&dp->dp_config_rwlock)); 1174} 1175 1176boolean_t 1177dsl_pool_config_held_writer(dsl_pool_t *dp) 1178{ 1179 return (RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1180} 1181