dsl_pool.c revision 310516
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, 2016 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 void 532dsl_pool_sync_mos(dsl_pool_t *dp, dmu_tx_t *tx) 533{ 534 zio_t *zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 535 dmu_objset_sync(dp->dp_meta_objset, zio, tx); 536 VERIFY0(zio_wait(zio)); 537 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); 538 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); 539} 540 541static void 542dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta) 543{ 544 ASSERT(MUTEX_HELD(&dp->dp_lock)); 545 546 if (delta < 0) 547 ASSERT3U(-delta, <=, dp->dp_dirty_total); 548 549 dp->dp_dirty_total += delta; 550 551 /* 552 * Note: we signal even when increasing dp_dirty_total. 553 * This ensures forward progress -- each thread wakes the next waiter. 554 */ 555 if (dp->dp_dirty_total <= zfs_dirty_data_max) 556 cv_signal(&dp->dp_spaceavail_cv); 557} 558 559void 560dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) 561{ 562 zio_t *zio; 563 dmu_tx_t *tx; 564 dsl_dir_t *dd; 565 dsl_dataset_t *ds; 566 objset_t *mos = dp->dp_meta_objset; 567 list_t synced_datasets; 568 569 list_create(&synced_datasets, sizeof (dsl_dataset_t), 570 offsetof(dsl_dataset_t, ds_synced_link)); 571 572 tx = dmu_tx_create_assigned(dp, txg); 573 574 /* 575 * Write out all dirty blocks of dirty datasets. 576 */ 577 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 578 while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 579 /* 580 * We must not sync any non-MOS datasets twice, because 581 * we may have taken a snapshot of them. However, we 582 * may sync newly-created datasets on pass 2. 583 */ 584 ASSERT(!list_link_active(&ds->ds_synced_link)); 585 list_insert_tail(&synced_datasets, ds); 586 dsl_dataset_sync(ds, zio, tx); 587 } 588 VERIFY0(zio_wait(zio)); 589 590 /* 591 * We have written all of the accounted dirty data, so our 592 * dp_space_towrite should now be zero. However, some seldom-used 593 * code paths do not adhere to this (e.g. dbuf_undirty(), also 594 * rounding error in dbuf_write_physdone). 595 * Shore up the accounting of any dirtied space now. 596 */ 597 dsl_pool_undirty_space(dp, dp->dp_dirty_pertxg[txg & TXG_MASK], txg); 598 599 /* 600 * After the data blocks have been written (ensured by the zio_wait() 601 * above), update the user/group space accounting. 602 */ 603 for (ds = list_head(&synced_datasets); ds != NULL; 604 ds = list_next(&synced_datasets, ds)) { 605 dmu_objset_do_userquota_updates(ds->ds_objset, tx); 606 } 607 608 /* 609 * Sync the datasets again to push out the changes due to 610 * userspace updates. This must be done before we process the 611 * sync tasks, so that any snapshots will have the correct 612 * user accounting information (and we won't get confused 613 * about which blocks are part of the snapshot). 614 */ 615 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 616 while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 617 ASSERT(list_link_active(&ds->ds_synced_link)); 618 dmu_buf_rele(ds->ds_dbuf, ds); 619 dsl_dataset_sync(ds, zio, tx); 620 } 621 VERIFY0(zio_wait(zio)); 622 623 /* 624 * Now that the datasets have been completely synced, we can 625 * clean up our in-memory structures accumulated while syncing: 626 * 627 * - move dead blocks from the pending deadlist to the on-disk deadlist 628 * - release hold from dsl_dataset_dirty() 629 */ 630 while ((ds = list_remove_head(&synced_datasets)) != NULL) { 631 dsl_dataset_sync_done(ds, tx); 632 } 633 while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) != NULL) { 634 dsl_dir_sync(dd, tx); 635 } 636 637 /* 638 * The MOS's space is accounted for in the pool/$MOS 639 * (dp_mos_dir). We can't modify the mos while we're syncing 640 * it, so we remember the deltas and apply them here. 641 */ 642 if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || 643 dp->dp_mos_uncompressed_delta != 0) { 644 dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, 645 dp->dp_mos_used_delta, 646 dp->dp_mos_compressed_delta, 647 dp->dp_mos_uncompressed_delta, tx); 648 dp->dp_mos_used_delta = 0; 649 dp->dp_mos_compressed_delta = 0; 650 dp->dp_mos_uncompressed_delta = 0; 651 } 652 653 if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || 654 list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { 655 dsl_pool_sync_mos(dp, tx); 656 } 657 658 /* 659 * If we modify a dataset in the same txg that we want to destroy it, 660 * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. 661 * dsl_dir_destroy_check() will fail if there are unexpected holds. 662 * Therefore, we want to sync the MOS (thus syncing the dd_dbuf 663 * and clearing the hold on it) before we process the sync_tasks. 664 * The MOS data dirtied by the sync_tasks will be synced on the next 665 * pass. 666 */ 667 if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { 668 dsl_sync_task_t *dst; 669 /* 670 * No more sync tasks should have been added while we 671 * were syncing. 672 */ 673 ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1); 674 while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)) != NULL) 675 dsl_sync_task_sync(dst, tx); 676 } 677 678 dmu_tx_commit(tx); 679 680 DTRACE_PROBE2(dsl_pool_sync__done, dsl_pool_t *dp, dp, uint64_t, txg); 681} 682 683void 684dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) 685{ 686 zilog_t *zilog; 687 688 while (zilog = txg_list_head(&dp->dp_dirty_zilogs, txg)) { 689 dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); 690 /* 691 * We don't remove the zilog from the dp_dirty_zilogs 692 * list until after we've cleaned it. This ensures that 693 * callers of zilog_is_dirty() receive an accurate 694 * answer when they are racing with the spa sync thread. 695 */ 696 zil_clean(zilog, txg); 697 (void) txg_list_remove_this(&dp->dp_dirty_zilogs, zilog, txg); 698 ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); 699 dmu_buf_rele(ds->ds_dbuf, zilog); 700 } 701 ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); 702} 703 704/* 705 * TRUE if the current thread is the tx_sync_thread or if we 706 * are being called from SPA context during pool initialization. 707 */ 708int 709dsl_pool_sync_context(dsl_pool_t *dp) 710{ 711 return (curthread == dp->dp_tx.tx_sync_thread || 712 spa_is_initializing(dp->dp_spa)); 713} 714 715uint64_t 716dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) 717{ 718 uint64_t space, resv; 719 720 /* 721 * If we're trying to assess whether it's OK to do a free, 722 * cut the reservation in half to allow forward progress 723 * (e.g. make it possible to rm(1) files from a full pool). 724 */ 725 space = spa_get_dspace(dp->dp_spa); 726 resv = spa_get_slop_space(dp->dp_spa); 727 if (netfree) 728 resv >>= 1; 729 730 return (space - resv); 731} 732 733boolean_t 734dsl_pool_need_dirty_delay(dsl_pool_t *dp) 735{ 736 uint64_t delay_min_bytes = 737 zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100; 738 boolean_t rv; 739 740 mutex_enter(&dp->dp_lock); 741 if (dp->dp_dirty_total > zfs_dirty_data_sync) 742 txg_kick(dp); 743 rv = (dp->dp_dirty_total > delay_min_bytes); 744 mutex_exit(&dp->dp_lock); 745 return (rv); 746} 747 748void 749dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 750{ 751 if (space > 0) { 752 mutex_enter(&dp->dp_lock); 753 dp->dp_dirty_pertxg[tx->tx_txg & TXG_MASK] += space; 754 dsl_pool_dirty_delta(dp, space); 755 mutex_exit(&dp->dp_lock); 756 } 757} 758 759void 760dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg) 761{ 762 ASSERT3S(space, >=, 0); 763 if (space == 0) 764 return; 765 mutex_enter(&dp->dp_lock); 766 if (dp->dp_dirty_pertxg[txg & TXG_MASK] < space) { 767 /* XXX writing something we didn't dirty? */ 768 space = dp->dp_dirty_pertxg[txg & TXG_MASK]; 769 } 770 ASSERT3U(dp->dp_dirty_pertxg[txg & TXG_MASK], >=, space); 771 dp->dp_dirty_pertxg[txg & TXG_MASK] -= space; 772 ASSERT3U(dp->dp_dirty_total, >=, space); 773 dsl_pool_dirty_delta(dp, -space); 774 mutex_exit(&dp->dp_lock); 775} 776 777/* ARGSUSED */ 778static int 779upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) 780{ 781 dmu_tx_t *tx = arg; 782 dsl_dataset_t *ds, *prev = NULL; 783 int err; 784 785 err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); 786 if (err) 787 return (err); 788 789 while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 790 err = dsl_dataset_hold_obj(dp, 791 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 792 if (err) { 793 dsl_dataset_rele(ds, FTAG); 794 return (err); 795 } 796 797 if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) 798 break; 799 dsl_dataset_rele(ds, FTAG); 800 ds = prev; 801 prev = NULL; 802 } 803 804 if (prev == NULL) { 805 prev = dp->dp_origin_snap; 806 807 /* 808 * The $ORIGIN can't have any data, or the accounting 809 * will be wrong. 810 */ 811 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 812 ASSERT0(dsl_dataset_phys(prev)->ds_bp.blk_birth); 813 rrw_exit(&ds->ds_bp_rwlock, FTAG); 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