dsl_pool.c revision 321567
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, 2017 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 * Copyright 2016 Nexenta Systems, Inc. All rights reserved. 28 */ 29 30#include <sys/dsl_pool.h> 31#include <sys/dsl_dataset.h> 32#include <sys/dsl_prop.h> 33#include <sys/dsl_dir.h> 34#include <sys/dsl_synctask.h> 35#include <sys/dsl_scan.h> 36#include <sys/dnode.h> 37#include <sys/dmu_tx.h> 38#include <sys/dmu_objset.h> 39#include <sys/arc.h> 40#include <sys/zap.h> 41#include <sys/zio.h> 42#include <sys/zfs_context.h> 43#include <sys/fs/zfs.h> 44#include <sys/zfs_znode.h> 45#include <sys/spa_impl.h> 46#include <sys/dsl_deadlist.h> 47#include <sys/bptree.h> 48#include <sys/zfeature.h> 49#include <sys/zil_impl.h> 50#include <sys/dsl_userhold.h> 51 52#if defined(__FreeBSD__) && defined(_KERNEL) 53#include <sys/types.h> 54#include <sys/sysctl.h> 55#endif 56 57/* 58 * ZFS Write Throttle 59 * ------------------ 60 * 61 * ZFS must limit the rate of incoming writes to the rate at which it is able 62 * to sync data modifications to the backend storage. Throttling by too much 63 * creates an artificial limit; throttling by too little can only be sustained 64 * for short periods and would lead to highly lumpy performance. On a per-pool 65 * basis, ZFS tracks the amount of modified (dirty) data. As operations change 66 * data, the amount of dirty data increases; as ZFS syncs out data, the amount 67 * of dirty data decreases. When the amount of dirty data exceeds a 68 * predetermined threshold further modifications are blocked until the amount 69 * of dirty data decreases (as data is synced out). 70 * 71 * The limit on dirty data is tunable, and should be adjusted according to 72 * both the IO capacity and available memory of the system. The larger the 73 * window, the more ZFS is able to aggregate and amortize metadata (and data) 74 * changes. However, memory is a limited resource, and allowing for more dirty 75 * data comes at the cost of keeping other useful data in memory (for example 76 * ZFS data cached by the ARC). 77 * 78 * Implementation 79 * 80 * As buffers are modified dsl_pool_willuse_space() increments both the per- 81 * txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of 82 * dirty space used; dsl_pool_dirty_space() decrements those values as data 83 * is synced out from dsl_pool_sync(). While only the poolwide value is 84 * relevant, the per-txg value is useful for debugging. The tunable 85 * zfs_dirty_data_max determines the dirty space limit. Once that value is 86 * exceeded, new writes are halted until space frees up. 87 * 88 * The zfs_dirty_data_sync tunable dictates the threshold at which we 89 * ensure that there is a txg syncing (see the comment in txg.c for a full 90 * description of transaction group stages). 91 * 92 * The IO scheduler uses both the dirty space limit and current amount of 93 * dirty data as inputs. Those values affect the number of concurrent IOs ZFS 94 * issues. See the comment in vdev_queue.c for details of the IO scheduler. 95 * 96 * The delay is also calculated based on the amount of dirty data. See the 97 * comment above dmu_tx_delay() for details. 98 */ 99 100/* 101 * zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory, 102 * capped at zfs_dirty_data_max_max. It can also be overridden in /etc/system. 103 */ 104uint64_t zfs_dirty_data_max; 105uint64_t zfs_dirty_data_max_max = 4ULL * 1024 * 1024 * 1024; 106int zfs_dirty_data_max_percent = 10; 107 108/* 109 * If there is at least this much dirty data, push out a txg. 110 */ 111uint64_t zfs_dirty_data_sync = 64 * 1024 * 1024; 112 113/* 114 * Once there is this amount of dirty data, the dmu_tx_delay() will kick in 115 * and delay each transaction. 116 * This value should be >= zfs_vdev_async_write_active_max_dirty_percent. 117 */ 118int zfs_delay_min_dirty_percent = 60; 119 120/* 121 * This controls how quickly the delay approaches infinity. 122 * Larger values cause it to delay more for a given amount of dirty data. 123 * Therefore larger values will cause there to be less dirty data for a 124 * given throughput. 125 * 126 * For the smoothest delay, this value should be about 1 billion divided 127 * by the maximum number of operations per second. This will smoothly 128 * handle between 10x and 1/10th this number. 129 * 130 * Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the 131 * multiply in dmu_tx_delay(). 132 */ 133uint64_t zfs_delay_scale = 1000 * 1000 * 1000 / 2000; 134 135/* 136 * This determines the number of threads used by the dp_sync_taskq. 137 */ 138int zfs_sync_taskq_batch_pct = 75; 139 140#if defined(__FreeBSD__) && defined(_KERNEL) 141 142extern int zfs_vdev_async_write_active_max_dirty_percent; 143 144SYSCTL_DECL(_vfs_zfs); 145 146SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, dirty_data_max, CTLFLAG_RWTUN, 147 &zfs_dirty_data_max, 0, 148 "The maximum amount of dirty data in bytes after which new writes are " 149 "halted until space becomes available"); 150 151SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, dirty_data_max_max, CTLFLAG_RDTUN, 152 &zfs_dirty_data_max_max, 0, 153 "The absolute cap on dirty_data_max when auto calculating"); 154 155static int sysctl_zfs_dirty_data_max_percent(SYSCTL_HANDLER_ARGS); 156SYSCTL_PROC(_vfs_zfs, OID_AUTO, dirty_data_max_percent, 157 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RWTUN, 0, sizeof(int), 158 sysctl_zfs_dirty_data_max_percent, "I", 159 "The percent of physical memory used to auto calculate dirty_data_max"); 160 161SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, dirty_data_sync, CTLFLAG_RWTUN, 162 &zfs_dirty_data_sync, 0, 163 "Force a txg if the number of dirty buffer bytes exceed this value"); 164 165static int sysctl_zfs_delay_min_dirty_percent(SYSCTL_HANDLER_ARGS); 166/* No zfs_delay_min_dirty_percent tunable due to limit requirements */ 167SYSCTL_PROC(_vfs_zfs, OID_AUTO, delay_min_dirty_percent, 168 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(int), 169 sysctl_zfs_delay_min_dirty_percent, "I", 170 "The limit of outstanding dirty data before transations are delayed"); 171 172static int sysctl_zfs_delay_scale(SYSCTL_HANDLER_ARGS); 173/* No zfs_delay_scale tunable due to limit requirements */ 174SYSCTL_PROC(_vfs_zfs, OID_AUTO, delay_scale, 175 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t), 176 sysctl_zfs_delay_scale, "QU", 177 "Controls how quickly the delay approaches infinity"); 178 179static int 180sysctl_zfs_dirty_data_max_percent(SYSCTL_HANDLER_ARGS) 181{ 182 int val, err; 183 184 val = zfs_dirty_data_max_percent; 185 err = sysctl_handle_int(oidp, &val, 0, req); 186 if (err != 0 || req->newptr == NULL) 187 return (err); 188 189 if (val < 0 || val > 100) 190 return (EINVAL); 191 192 zfs_dirty_data_max_percent = val; 193 194 return (0); 195} 196 197static int 198sysctl_zfs_delay_min_dirty_percent(SYSCTL_HANDLER_ARGS) 199{ 200 int val, err; 201 202 val = zfs_delay_min_dirty_percent; 203 err = sysctl_handle_int(oidp, &val, 0, req); 204 if (err != 0 || req->newptr == NULL) 205 return (err); 206 207 if (val < zfs_vdev_async_write_active_max_dirty_percent) 208 return (EINVAL); 209 210 zfs_delay_min_dirty_percent = val; 211 212 return (0); 213} 214 215static int 216sysctl_zfs_delay_scale(SYSCTL_HANDLER_ARGS) 217{ 218 uint64_t val; 219 int err; 220 221 val = zfs_delay_scale; 222 err = sysctl_handle_64(oidp, &val, 0, req); 223 if (err != 0 || req->newptr == NULL) 224 return (err); 225 226 if (val > UINT64_MAX / zfs_dirty_data_max) 227 return (EINVAL); 228 229 zfs_delay_scale = val; 230 231 return (0); 232} 233#endif 234 235int 236dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) 237{ 238 uint64_t obj; 239 int err; 240 241 err = zap_lookup(dp->dp_meta_objset, 242 dsl_dir_phys(dp->dp_root_dir)->dd_child_dir_zapobj, 243 name, sizeof (obj), 1, &obj); 244 if (err) 245 return (err); 246 247 return (dsl_dir_hold_obj(dp, obj, name, dp, ddp)); 248} 249 250static dsl_pool_t * 251dsl_pool_open_impl(spa_t *spa, uint64_t txg) 252{ 253 dsl_pool_t *dp; 254 blkptr_t *bp = spa_get_rootblkptr(spa); 255 256 dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); 257 dp->dp_spa = spa; 258 dp->dp_meta_rootbp = *bp; 259 rrw_init(&dp->dp_config_rwlock, B_TRUE); 260 txg_init(dp, txg); 261 262 txg_list_create(&dp->dp_dirty_datasets, spa, 263 offsetof(dsl_dataset_t, ds_dirty_link)); 264 txg_list_create(&dp->dp_dirty_zilogs, spa, 265 offsetof(zilog_t, zl_dirty_link)); 266 txg_list_create(&dp->dp_dirty_dirs, spa, 267 offsetof(dsl_dir_t, dd_dirty_link)); 268 txg_list_create(&dp->dp_sync_tasks, spa, 269 offsetof(dsl_sync_task_t, dst_node)); 270 271 dp->dp_sync_taskq = taskq_create("dp_sync_taskq", 272 zfs_sync_taskq_batch_pct, minclsyspri, 1, INT_MAX, 273 TASKQ_THREADS_CPU_PCT); 274 275 mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); 276 cv_init(&dp->dp_spaceavail_cv, NULL, CV_DEFAULT, NULL); 277 278 dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri, 279 1, 4, 0); 280 281 return (dp); 282} 283 284int 285dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) 286{ 287 int err; 288 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 289 290 err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, 291 &dp->dp_meta_objset); 292 if (err != 0) 293 dsl_pool_close(dp); 294 else 295 *dpp = dp; 296 297 return (err); 298} 299 300int 301dsl_pool_open(dsl_pool_t *dp) 302{ 303 int err; 304 dsl_dir_t *dd; 305 dsl_dataset_t *ds; 306 uint64_t obj; 307 308 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 309 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 310 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, 311 &dp->dp_root_dir_obj); 312 if (err) 313 goto out; 314 315 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, 316 NULL, dp, &dp->dp_root_dir); 317 if (err) 318 goto out; 319 320 err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); 321 if (err) 322 goto out; 323 324 if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { 325 err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); 326 if (err) 327 goto out; 328 err = dsl_dataset_hold_obj(dp, 329 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds); 330 if (err == 0) { 331 err = dsl_dataset_hold_obj(dp, 332 dsl_dataset_phys(ds)->ds_prev_snap_obj, dp, 333 &dp->dp_origin_snap); 334 dsl_dataset_rele(ds, FTAG); 335 } 336 dsl_dir_rele(dd, dp); 337 if (err) 338 goto out; 339 } 340 341 if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { 342 err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, 343 &dp->dp_free_dir); 344 if (err) 345 goto out; 346 347 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 348 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); 349 if (err) 350 goto out; 351 VERIFY0(bpobj_open(&dp->dp_free_bpobj, 352 dp->dp_meta_objset, obj)); 353 } 354 355 /* 356 * Note: errors ignored, because the leak dir will not exist if we 357 * have not encountered a leak yet. 358 */ 359 (void) dsl_pool_open_special_dir(dp, LEAK_DIR_NAME, 360 &dp->dp_leak_dir); 361 362 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) { 363 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 364 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, 365 &dp->dp_bptree_obj); 366 if (err != 0) 367 goto out; 368 } 369 370 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMPTY_BPOBJ)) { 371 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 372 DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, 373 &dp->dp_empty_bpobj); 374 if (err != 0) 375 goto out; 376 } 377 378 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 379 DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, 380 &dp->dp_tmp_userrefs_obj); 381 if (err == ENOENT) 382 err = 0; 383 if (err) 384 goto out; 385 386 err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); 387 388out: 389 rrw_exit(&dp->dp_config_rwlock, FTAG); 390 return (err); 391} 392 393void 394dsl_pool_close(dsl_pool_t *dp) 395{ 396 /* 397 * Drop our references from dsl_pool_open(). 398 * 399 * Since we held the origin_snap from "syncing" context (which 400 * includes pool-opening context), it actually only got a "ref" 401 * and not a hold, so just drop that here. 402 */ 403 if (dp->dp_origin_snap) 404 dsl_dataset_rele(dp->dp_origin_snap, dp); 405 if (dp->dp_mos_dir) 406 dsl_dir_rele(dp->dp_mos_dir, dp); 407 if (dp->dp_free_dir) 408 dsl_dir_rele(dp->dp_free_dir, dp); 409 if (dp->dp_leak_dir) 410 dsl_dir_rele(dp->dp_leak_dir, dp); 411 if (dp->dp_root_dir) 412 dsl_dir_rele(dp->dp_root_dir, dp); 413 414 bpobj_close(&dp->dp_free_bpobj); 415 416 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ 417 if (dp->dp_meta_objset) 418 dmu_objset_evict(dp->dp_meta_objset); 419 420 txg_list_destroy(&dp->dp_dirty_datasets); 421 txg_list_destroy(&dp->dp_dirty_zilogs); 422 txg_list_destroy(&dp->dp_sync_tasks); 423 txg_list_destroy(&dp->dp_dirty_dirs); 424 425 taskq_destroy(dp->dp_sync_taskq); 426 427 /* 428 * We can't set retry to TRUE since we're explicitly specifying 429 * a spa to flush. This is good enough; any missed buffers for 430 * this spa won't cause trouble, and they'll eventually fall 431 * out of the ARC just like any other unused buffer. 432 */ 433 arc_flush(dp->dp_spa, FALSE); 434 435 txg_fini(dp); 436 dsl_scan_fini(dp); 437 dmu_buf_user_evict_wait(); 438 439 rrw_destroy(&dp->dp_config_rwlock); 440 mutex_destroy(&dp->dp_lock); 441 taskq_destroy(dp->dp_vnrele_taskq); 442 if (dp->dp_blkstats) 443 kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); 444 kmem_free(dp, sizeof (dsl_pool_t)); 445} 446 447dsl_pool_t * 448dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) 449{ 450 int err; 451 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 452 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); 453 objset_t *os; 454 dsl_dataset_t *ds; 455 uint64_t obj; 456 457 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 458 459 /* create and open the MOS (meta-objset) */ 460 dp->dp_meta_objset = dmu_objset_create_impl(spa, 461 NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); 462 463 /* create the pool directory */ 464 err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 465 DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); 466 ASSERT0(err); 467 468 /* Initialize scan structures */ 469 VERIFY0(dsl_scan_init(dp, txg)); 470 471 /* create and open the root dir */ 472 dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); 473 VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, 474 NULL, dp, &dp->dp_root_dir)); 475 476 /* create and open the meta-objset dir */ 477 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); 478 VERIFY0(dsl_pool_open_special_dir(dp, 479 MOS_DIR_NAME, &dp->dp_mos_dir)); 480 481 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { 482 /* create and open the free dir */ 483 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, 484 FREE_DIR_NAME, tx); 485 VERIFY0(dsl_pool_open_special_dir(dp, 486 FREE_DIR_NAME, &dp->dp_free_dir)); 487 488 /* create and open the free_bplist */ 489 obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx); 490 VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 491 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); 492 VERIFY0(bpobj_open(&dp->dp_free_bpobj, 493 dp->dp_meta_objset, obj)); 494 } 495 496 if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) 497 dsl_pool_create_origin(dp, tx); 498 499 /* create the root dataset */ 500 obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); 501 502 /* create the root objset */ 503 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); 504 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 505 os = dmu_objset_create_impl(dp->dp_spa, ds, 506 dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx); 507 rrw_exit(&ds->ds_bp_rwlock, FTAG); 508#ifdef _KERNEL 509 zfs_create_fs(os, kcred, zplprops, tx); 510#endif 511 dsl_dataset_rele(ds, FTAG); 512 513 dmu_tx_commit(tx); 514 515 rrw_exit(&dp->dp_config_rwlock, FTAG); 516 517 return (dp); 518} 519 520/* 521 * Account for the meta-objset space in its placeholder dsl_dir. 522 */ 523void 524dsl_pool_mos_diduse_space(dsl_pool_t *dp, 525 int64_t used, int64_t comp, int64_t uncomp) 526{ 527 ASSERT3U(comp, ==, uncomp); /* it's all metadata */ 528 mutex_enter(&dp->dp_lock); 529 dp->dp_mos_used_delta += used; 530 dp->dp_mos_compressed_delta += comp; 531 dp->dp_mos_uncompressed_delta += uncomp; 532 mutex_exit(&dp->dp_lock); 533} 534 535static void 536dsl_pool_sync_mos(dsl_pool_t *dp, dmu_tx_t *tx) 537{ 538 zio_t *zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 539 dmu_objset_sync(dp->dp_meta_objset, zio, tx); 540 VERIFY0(zio_wait(zio)); 541 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); 542 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); 543} 544 545static void 546dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta) 547{ 548 ASSERT(MUTEX_HELD(&dp->dp_lock)); 549 550 if (delta < 0) 551 ASSERT3U(-delta, <=, dp->dp_dirty_total); 552 553 dp->dp_dirty_total += delta; 554 555 /* 556 * Note: we signal even when increasing dp_dirty_total. 557 * This ensures forward progress -- each thread wakes the next waiter. 558 */ 559 if (dp->dp_dirty_total < zfs_dirty_data_max) 560 cv_signal(&dp->dp_spaceavail_cv); 561} 562 563void 564dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) 565{ 566 zio_t *zio; 567 dmu_tx_t *tx; 568 dsl_dir_t *dd; 569 dsl_dataset_t *ds; 570 objset_t *mos = dp->dp_meta_objset; 571 list_t synced_datasets; 572 573 list_create(&synced_datasets, sizeof (dsl_dataset_t), 574 offsetof(dsl_dataset_t, ds_synced_link)); 575 576 tx = dmu_tx_create_assigned(dp, txg); 577 578 /* 579 * Write out all dirty blocks of dirty datasets. 580 */ 581 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 582 while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 583 /* 584 * We must not sync any non-MOS datasets twice, because 585 * we may have taken a snapshot of them. However, we 586 * may sync newly-created datasets on pass 2. 587 */ 588 ASSERT(!list_link_active(&ds->ds_synced_link)); 589 list_insert_tail(&synced_datasets, ds); 590 dsl_dataset_sync(ds, zio, tx); 591 } 592 VERIFY0(zio_wait(zio)); 593 594 /* 595 * We have written all of the accounted dirty data, so our 596 * dp_space_towrite should now be zero. However, some seldom-used 597 * code paths do not adhere to this (e.g. dbuf_undirty(), also 598 * rounding error in dbuf_write_physdone). 599 * Shore up the accounting of any dirtied space now. 600 */ 601 dsl_pool_undirty_space(dp, dp->dp_dirty_pertxg[txg & TXG_MASK], txg); 602 603 /* 604 * Update the long range free counter after 605 * we're done syncing user data 606 */ 607 mutex_enter(&dp->dp_lock); 608 ASSERT(spa_sync_pass(dp->dp_spa) == 1 || 609 dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] == 0); 610 dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] = 0; 611 mutex_exit(&dp->dp_lock); 612 613 /* 614 * After the data blocks have been written (ensured by the zio_wait() 615 * above), update the user/group space accounting. This happens 616 * in tasks dispatched to dp_sync_taskq, so wait for them before 617 * continuing. 618 */ 619 for (ds = list_head(&synced_datasets); ds != NULL; 620 ds = list_next(&synced_datasets, ds)) { 621 dmu_objset_do_userquota_updates(ds->ds_objset, tx); 622 } 623 taskq_wait(dp->dp_sync_taskq); 624 625 /* 626 * Sync the datasets again to push out the changes due to 627 * userspace updates. This must be done before we process the 628 * sync tasks, so that any snapshots will have the correct 629 * user accounting information (and we won't get confused 630 * about which blocks are part of the snapshot). 631 */ 632 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 633 while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 634 ASSERT(list_link_active(&ds->ds_synced_link)); 635 dmu_buf_rele(ds->ds_dbuf, ds); 636 dsl_dataset_sync(ds, zio, tx); 637 } 638 VERIFY0(zio_wait(zio)); 639 640 /* 641 * Now that the datasets have been completely synced, we can 642 * clean up our in-memory structures accumulated while syncing: 643 * 644 * - move dead blocks from the pending deadlist to the on-disk deadlist 645 * - release hold from dsl_dataset_dirty() 646 */ 647 while ((ds = list_remove_head(&synced_datasets)) != NULL) { 648 dsl_dataset_sync_done(ds, tx); 649 } 650 while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) != NULL) { 651 dsl_dir_sync(dd, tx); 652 } 653 654 /* 655 * The MOS's space is accounted for in the pool/$MOS 656 * (dp_mos_dir). We can't modify the mos while we're syncing 657 * it, so we remember the deltas and apply them here. 658 */ 659 if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || 660 dp->dp_mos_uncompressed_delta != 0) { 661 dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, 662 dp->dp_mos_used_delta, 663 dp->dp_mos_compressed_delta, 664 dp->dp_mos_uncompressed_delta, tx); 665 dp->dp_mos_used_delta = 0; 666 dp->dp_mos_compressed_delta = 0; 667 dp->dp_mos_uncompressed_delta = 0; 668 } 669 670 if (!multilist_is_empty(mos->os_dirty_dnodes[txg & TXG_MASK])) { 671 dsl_pool_sync_mos(dp, tx); 672 } 673 674 /* 675 * If we modify a dataset in the same txg that we want to destroy it, 676 * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. 677 * dsl_dir_destroy_check() will fail if there are unexpected holds. 678 * Therefore, we want to sync the MOS (thus syncing the dd_dbuf 679 * and clearing the hold on it) before we process the sync_tasks. 680 * The MOS data dirtied by the sync_tasks will be synced on the next 681 * pass. 682 */ 683 if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { 684 dsl_sync_task_t *dst; 685 /* 686 * No more sync tasks should have been added while we 687 * were syncing. 688 */ 689 ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1); 690 while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)) != NULL) 691 dsl_sync_task_sync(dst, tx); 692 } 693 694 dmu_tx_commit(tx); 695 696 DTRACE_PROBE2(dsl_pool_sync__done, dsl_pool_t *dp, dp, uint64_t, txg); 697} 698 699void 700dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) 701{ 702 zilog_t *zilog; 703 704 while (zilog = txg_list_head(&dp->dp_dirty_zilogs, txg)) { 705 dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); 706 /* 707 * We don't remove the zilog from the dp_dirty_zilogs 708 * list until after we've cleaned it. This ensures that 709 * callers of zilog_is_dirty() receive an accurate 710 * answer when they are racing with the spa sync thread. 711 */ 712 zil_clean(zilog, txg); 713 (void) txg_list_remove_this(&dp->dp_dirty_zilogs, zilog, txg); 714 ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); 715 dmu_buf_rele(ds->ds_dbuf, zilog); 716 } 717 ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); 718} 719 720/* 721 * TRUE if the current thread is the tx_sync_thread or if we 722 * are being called from SPA context during pool initialization. 723 */ 724int 725dsl_pool_sync_context(dsl_pool_t *dp) 726{ 727 return (curthread == dp->dp_tx.tx_sync_thread || 728 spa_is_initializing(dp->dp_spa) || 729 taskq_member(dp->dp_sync_taskq, curthread)); 730} 731 732uint64_t 733dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) 734{ 735 uint64_t space, resv; 736 737 /* 738 * If we're trying to assess whether it's OK to do a free, 739 * cut the reservation in half to allow forward progress 740 * (e.g. make it possible to rm(1) files from a full pool). 741 */ 742 space = spa_get_dspace(dp->dp_spa); 743 resv = spa_get_slop_space(dp->dp_spa); 744 if (netfree) 745 resv >>= 1; 746 747 return (space - resv); 748} 749 750boolean_t 751dsl_pool_need_dirty_delay(dsl_pool_t *dp) 752{ 753 uint64_t delay_min_bytes = 754 zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100; 755 boolean_t rv; 756 757 mutex_enter(&dp->dp_lock); 758 if (dp->dp_dirty_total > zfs_dirty_data_sync) 759 txg_kick(dp); 760 rv = (dp->dp_dirty_total > delay_min_bytes); 761 mutex_exit(&dp->dp_lock); 762 return (rv); 763} 764 765void 766dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 767{ 768 if (space > 0) { 769 mutex_enter(&dp->dp_lock); 770 dp->dp_dirty_pertxg[tx->tx_txg & TXG_MASK] += space; 771 dsl_pool_dirty_delta(dp, space); 772 mutex_exit(&dp->dp_lock); 773 } 774} 775 776void 777dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg) 778{ 779 ASSERT3S(space, >=, 0); 780 if (space == 0) 781 return; 782 mutex_enter(&dp->dp_lock); 783 if (dp->dp_dirty_pertxg[txg & TXG_MASK] < space) { 784 /* XXX writing something we didn't dirty? */ 785 space = dp->dp_dirty_pertxg[txg & TXG_MASK]; 786 } 787 ASSERT3U(dp->dp_dirty_pertxg[txg & TXG_MASK], >=, space); 788 dp->dp_dirty_pertxg[txg & TXG_MASK] -= space; 789 ASSERT3U(dp->dp_dirty_total, >=, space); 790 dsl_pool_dirty_delta(dp, -space); 791 mutex_exit(&dp->dp_lock); 792} 793 794/* ARGSUSED */ 795static int 796upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) 797{ 798 dmu_tx_t *tx = arg; 799 dsl_dataset_t *ds, *prev = NULL; 800 int err; 801 802 err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); 803 if (err) 804 return (err); 805 806 while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 807 err = dsl_dataset_hold_obj(dp, 808 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 809 if (err) { 810 dsl_dataset_rele(ds, FTAG); 811 return (err); 812 } 813 814 if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) 815 break; 816 dsl_dataset_rele(ds, FTAG); 817 ds = prev; 818 prev = NULL; 819 } 820 821 if (prev == NULL) { 822 prev = dp->dp_origin_snap; 823 824 /* 825 * The $ORIGIN can't have any data, or the accounting 826 * will be wrong. 827 */ 828 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 829 ASSERT0(dsl_dataset_phys(prev)->ds_bp.blk_birth); 830 rrw_exit(&ds->ds_bp_rwlock, FTAG); 831 832 /* The origin doesn't get attached to itself */ 833 if (ds->ds_object == prev->ds_object) { 834 dsl_dataset_rele(ds, FTAG); 835 return (0); 836 } 837 838 dmu_buf_will_dirty(ds->ds_dbuf, tx); 839 dsl_dataset_phys(ds)->ds_prev_snap_obj = prev->ds_object; 840 dsl_dataset_phys(ds)->ds_prev_snap_txg = 841 dsl_dataset_phys(prev)->ds_creation_txg; 842 843 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 844 dsl_dir_phys(ds->ds_dir)->dd_origin_obj = prev->ds_object; 845 846 dmu_buf_will_dirty(prev->ds_dbuf, tx); 847 dsl_dataset_phys(prev)->ds_num_children++; 848 849 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) { 850 ASSERT(ds->ds_prev == NULL); 851 VERIFY0(dsl_dataset_hold_obj(dp, 852 dsl_dataset_phys(ds)->ds_prev_snap_obj, 853 ds, &ds->ds_prev)); 854 } 855 } 856 857 ASSERT3U(dsl_dir_phys(ds->ds_dir)->dd_origin_obj, ==, prev->ds_object); 858 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_obj, ==, prev->ds_object); 859 860 if (dsl_dataset_phys(prev)->ds_next_clones_obj == 0) { 861 dmu_buf_will_dirty(prev->ds_dbuf, tx); 862 dsl_dataset_phys(prev)->ds_next_clones_obj = 863 zap_create(dp->dp_meta_objset, 864 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 865 } 866 VERIFY0(zap_add_int(dp->dp_meta_objset, 867 dsl_dataset_phys(prev)->ds_next_clones_obj, ds->ds_object, tx)); 868 869 dsl_dataset_rele(ds, FTAG); 870 if (prev != dp->dp_origin_snap) 871 dsl_dataset_rele(prev, FTAG); 872 return (0); 873} 874 875void 876dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) 877{ 878 ASSERT(dmu_tx_is_syncing(tx)); 879 ASSERT(dp->dp_origin_snap != NULL); 880 881 VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb, 882 tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); 883} 884 885/* ARGSUSED */ 886static int 887upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 888{ 889 dmu_tx_t *tx = arg; 890 objset_t *mos = dp->dp_meta_objset; 891 892 if (dsl_dir_phys(ds->ds_dir)->dd_origin_obj != 0) { 893 dsl_dataset_t *origin; 894 895 VERIFY0(dsl_dataset_hold_obj(dp, 896 dsl_dir_phys(ds->ds_dir)->dd_origin_obj, FTAG, &origin)); 897 898 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 899 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 900 dsl_dir_phys(origin->ds_dir)->dd_clones = 901 zap_create(mos, DMU_OT_DSL_CLONES, DMU_OT_NONE, 902 0, tx); 903 } 904 905 VERIFY0(zap_add_int(dp->dp_meta_objset, 906 dsl_dir_phys(origin->ds_dir)->dd_clones, 907 ds->ds_object, tx)); 908 909 dsl_dataset_rele(origin, FTAG); 910 } 911 return (0); 912} 913 914void 915dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) 916{ 917 ASSERT(dmu_tx_is_syncing(tx)); 918 uint64_t obj; 919 920 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); 921 VERIFY0(dsl_pool_open_special_dir(dp, 922 FREE_DIR_NAME, &dp->dp_free_dir)); 923 924 /* 925 * We can't use bpobj_alloc(), because spa_version() still 926 * returns the old version, and we need a new-version bpobj with 927 * subobj support. So call dmu_object_alloc() directly. 928 */ 929 obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, 930 SPA_OLD_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); 931 VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 932 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); 933 VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); 934 935 VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 936 upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); 937} 938 939void 940dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) 941{ 942 uint64_t dsobj; 943 dsl_dataset_t *ds; 944 945 ASSERT(dmu_tx_is_syncing(tx)); 946 ASSERT(dp->dp_origin_snap == NULL); 947 ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER)); 948 949 /* create the origin dir, ds, & snap-ds */ 950 dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, 951 NULL, 0, kcred, tx); 952 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 953 dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx); 954 VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, 955 dp, &dp->dp_origin_snap)); 956 dsl_dataset_rele(ds, FTAG); 957} 958 959taskq_t * 960dsl_pool_vnrele_taskq(dsl_pool_t *dp) 961{ 962 return (dp->dp_vnrele_taskq); 963} 964 965/* 966 * Walk through the pool-wide zap object of temporary snapshot user holds 967 * and release them. 968 */ 969void 970dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) 971{ 972 zap_attribute_t za; 973 zap_cursor_t zc; 974 objset_t *mos = dp->dp_meta_objset; 975 uint64_t zapobj = dp->dp_tmp_userrefs_obj; 976 nvlist_t *holds; 977 978 if (zapobj == 0) 979 return; 980 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 981 982 holds = fnvlist_alloc(); 983 984 for (zap_cursor_init(&zc, mos, zapobj); 985 zap_cursor_retrieve(&zc, &za) == 0; 986 zap_cursor_advance(&zc)) { 987 char *htag; 988 nvlist_t *tags; 989 990 htag = strchr(za.za_name, '-'); 991 *htag = '\0'; 992 ++htag; 993 if (nvlist_lookup_nvlist(holds, za.za_name, &tags) != 0) { 994 tags = fnvlist_alloc(); 995 fnvlist_add_boolean(tags, htag); 996 fnvlist_add_nvlist(holds, za.za_name, tags); 997 fnvlist_free(tags); 998 } else { 999 fnvlist_add_boolean(tags, htag); 1000 } 1001 } 1002 dsl_dataset_user_release_tmp(dp, holds); 1003 fnvlist_free(holds); 1004 zap_cursor_fini(&zc); 1005} 1006 1007/* 1008 * Create the pool-wide zap object for storing temporary snapshot holds. 1009 */ 1010void 1011dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) 1012{ 1013 objset_t *mos = dp->dp_meta_objset; 1014 1015 ASSERT(dp->dp_tmp_userrefs_obj == 0); 1016 ASSERT(dmu_tx_is_syncing(tx)); 1017 1018 dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, 1019 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); 1020} 1021 1022static int 1023dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, 1024 const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding) 1025{ 1026 objset_t *mos = dp->dp_meta_objset; 1027 uint64_t zapobj = dp->dp_tmp_userrefs_obj; 1028 char *name; 1029 int error; 1030 1031 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 1032 ASSERT(dmu_tx_is_syncing(tx)); 1033 1034 /* 1035 * If the pool was created prior to SPA_VERSION_USERREFS, the 1036 * zap object for temporary holds might not exist yet. 1037 */ 1038 if (zapobj == 0) { 1039 if (holding) { 1040 dsl_pool_user_hold_create_obj(dp, tx); 1041 zapobj = dp->dp_tmp_userrefs_obj; 1042 } else { 1043 return (SET_ERROR(ENOENT)); 1044 } 1045 } 1046 1047 name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); 1048 if (holding) 1049 error = zap_add(mos, zapobj, name, 8, 1, &now, tx); 1050 else 1051 error = zap_remove(mos, zapobj, name, tx); 1052 strfree(name); 1053 1054 return (error); 1055} 1056 1057/* 1058 * Add a temporary hold for the given dataset object and tag. 1059 */ 1060int 1061dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 1062 uint64_t now, dmu_tx_t *tx) 1063{ 1064 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); 1065} 1066 1067/* 1068 * Release a temporary hold for the given dataset object and tag. 1069 */ 1070int 1071dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 1072 dmu_tx_t *tx) 1073{ 1074 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, 0, 1075 tx, B_FALSE)); 1076} 1077 1078/* 1079 * DSL Pool Configuration Lock 1080 * 1081 * The dp_config_rwlock protects against changes to DSL state (e.g. dataset 1082 * creation / destruction / rename / property setting). It must be held for 1083 * read to hold a dataset or dsl_dir. I.e. you must call 1084 * dsl_pool_config_enter() or dsl_pool_hold() before calling 1085 * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock 1086 * must be held continuously until all datasets and dsl_dirs are released. 1087 * 1088 * The only exception to this rule is that if a "long hold" is placed on 1089 * a dataset, then the dp_config_rwlock may be dropped while the dataset 1090 * is still held. The long hold will prevent the dataset from being 1091 * destroyed -- the destroy will fail with EBUSY. A long hold can be 1092 * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset 1093 * (by calling dsl_{dataset,objset}_{try}own{_obj}). 1094 * 1095 * Legitimate long-holders (including owners) should be long-running, cancelable 1096 * tasks that should cause "zfs destroy" to fail. This includes DMU 1097 * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open), 1098 * "zfs send", and "zfs diff". There are several other long-holders whose 1099 * uses are suboptimal (e.g. "zfs promote", and zil_suspend()). 1100 * 1101 * The usual formula for long-holding would be: 1102 * dsl_pool_hold() 1103 * dsl_dataset_hold() 1104 * ... perform checks ... 1105 * dsl_dataset_long_hold() 1106 * dsl_pool_rele() 1107 * ... perform long-running task ... 1108 * dsl_dataset_long_rele() 1109 * dsl_dataset_rele() 1110 * 1111 * Note that when the long hold is released, the dataset is still held but 1112 * the pool is not held. The dataset may change arbitrarily during this time 1113 * (e.g. it could be destroyed). Therefore you shouldn't do anything to the 1114 * dataset except release it. 1115 * 1116 * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only 1117 * or modifying operations. 1118 * 1119 * Modifying operations should generally use dsl_sync_task(). The synctask 1120 * infrastructure enforces proper locking strategy with respect to the 1121 * dp_config_rwlock. See the comment above dsl_sync_task() for details. 1122 * 1123 * Read-only operations will manually hold the pool, then the dataset, obtain 1124 * information from the dataset, then release the pool and dataset. 1125 * dmu_objset_{hold,rele}() are convenience routines that also do the pool 1126 * hold/rele. 1127 */ 1128 1129int 1130dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp) 1131{ 1132 spa_t *spa; 1133 int error; 1134 1135 error = spa_open(name, &spa, tag); 1136 if (error == 0) { 1137 *dp = spa_get_dsl(spa); 1138 dsl_pool_config_enter(*dp, tag); 1139 } 1140 return (error); 1141} 1142 1143void 1144dsl_pool_rele(dsl_pool_t *dp, void *tag) 1145{ 1146 dsl_pool_config_exit(dp, tag); 1147 spa_close(dp->dp_spa, tag); 1148} 1149 1150void 1151dsl_pool_config_enter(dsl_pool_t *dp, void *tag) 1152{ 1153 /* 1154 * We use a "reentrant" reader-writer lock, but not reentrantly. 1155 * 1156 * The rrwlock can (with the track_all flag) track all reading threads, 1157 * which is very useful for debugging which code path failed to release 1158 * the lock, and for verifying that the *current* thread does hold 1159 * the lock. 1160 * 1161 * (Unlike a rwlock, which knows that N threads hold it for 1162 * read, but not *which* threads, so rw_held(RW_READER) returns TRUE 1163 * if any thread holds it for read, even if this thread doesn't). 1164 */ 1165 ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); 1166 rrw_enter(&dp->dp_config_rwlock, RW_READER, tag); 1167} 1168 1169void 1170dsl_pool_config_enter_prio(dsl_pool_t *dp, void *tag) 1171{ 1172 ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); 1173 rrw_enter_read_prio(&dp->dp_config_rwlock, tag); 1174} 1175 1176void 1177dsl_pool_config_exit(dsl_pool_t *dp, void *tag) 1178{ 1179 rrw_exit(&dp->dp_config_rwlock, tag); 1180} 1181 1182boolean_t 1183dsl_pool_config_held(dsl_pool_t *dp) 1184{ 1185 return (RRW_LOCK_HELD(&dp->dp_config_rwlock)); 1186} 1187 1188boolean_t 1189dsl_pool_config_held_writer(dsl_pool_t *dp) 1190{ 1191 return (RRW_WRITE_HELD(&dp->dp_config_rwlock)); 1192} 1193