1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5ea8dc4b6Seschrock * Common Development and Distribution License (the "License"). 6ea8dc4b6Seschrock * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 21fa9e4066Sahrens /* 223f9d6ad7SLin Ling * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 232a104a52SAlex Reece * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 24a7a845e4SSteven Hartland * Copyright (c) 2013 Steven Hartland. All rights reserved. 25*bc9014e6SJustin Gibbs * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 26fa9e4066Sahrens */ 27fa9e4066Sahrens 28fa9e4066Sahrens #include <sys/dsl_pool.h> 29fa9e4066Sahrens #include <sys/dsl_dataset.h> 303f9d6ad7SLin Ling #include <sys/dsl_prop.h> 31fa9e4066Sahrens #include <sys/dsl_dir.h> 321d452cf5Sahrens #include <sys/dsl_synctask.h> 333f9d6ad7SLin Ling #include <sys/dsl_scan.h> 343f9d6ad7SLin Ling #include <sys/dnode.h> 35fa9e4066Sahrens #include <sys/dmu_tx.h> 36fa9e4066Sahrens #include <sys/dmu_objset.h> 37fa9e4066Sahrens #include <sys/arc.h> 38fa9e4066Sahrens #include <sys/zap.h> 39c717a561Smaybee #include <sys/zio.h> 40fa9e4066Sahrens #include <sys/zfs_context.h> 41fa9e4066Sahrens #include <sys/fs/zfs.h> 42088f3894Sahrens #include <sys/zfs_znode.h> 43088f3894Sahrens #include <sys/spa_impl.h> 44cde58dbcSMatthew Ahrens #include <sys/dsl_deadlist.h> 45ad135b5dSChristopher Siden #include <sys/bptree.h> 46ad135b5dSChristopher Siden #include <sys/zfeature.h> 47ce636f8bSMatthew Ahrens #include <sys/zil_impl.h> 483b2aab18SMatthew Ahrens #include <sys/dsl_userhold.h> 49fa9e4066Sahrens 5069962b56SMatthew Ahrens /* 5169962b56SMatthew Ahrens * ZFS Write Throttle 5269962b56SMatthew Ahrens * ------------------ 5369962b56SMatthew Ahrens * 5469962b56SMatthew Ahrens * ZFS must limit the rate of incoming writes to the rate at which it is able 5569962b56SMatthew Ahrens * to sync data modifications to the backend storage. Throttling by too much 5669962b56SMatthew Ahrens * creates an artificial limit; throttling by too little can only be sustained 5769962b56SMatthew Ahrens * for short periods and would lead to highly lumpy performance. On a per-pool 5869962b56SMatthew Ahrens * basis, ZFS tracks the amount of modified (dirty) data. As operations change 5969962b56SMatthew Ahrens * data, the amount of dirty data increases; as ZFS syncs out data, the amount 6069962b56SMatthew Ahrens * of dirty data decreases. When the amount of dirty data exceeds a 6169962b56SMatthew Ahrens * predetermined threshold further modifications are blocked until the amount 6269962b56SMatthew Ahrens * of dirty data decreases (as data is synced out). 6369962b56SMatthew Ahrens * 6469962b56SMatthew Ahrens * The limit on dirty data is tunable, and should be adjusted according to 6569962b56SMatthew Ahrens * both the IO capacity and available memory of the system. The larger the 6669962b56SMatthew Ahrens * window, the more ZFS is able to aggregate and amortize metadata (and data) 6769962b56SMatthew Ahrens * changes. However, memory is a limited resource, and allowing for more dirty 6869962b56SMatthew Ahrens * data comes at the cost of keeping other useful data in memory (for example 6969962b56SMatthew Ahrens * ZFS data cached by the ARC). 7069962b56SMatthew Ahrens * 7169962b56SMatthew Ahrens * Implementation 7269962b56SMatthew Ahrens * 7369962b56SMatthew Ahrens * As buffers are modified dsl_pool_willuse_space() increments both the per- 7469962b56SMatthew Ahrens * txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of 7569962b56SMatthew Ahrens * dirty space used; dsl_pool_dirty_space() decrements those values as data 7669962b56SMatthew Ahrens * is synced out from dsl_pool_sync(). While only the poolwide value is 7769962b56SMatthew Ahrens * relevant, the per-txg value is useful for debugging. The tunable 7869962b56SMatthew Ahrens * zfs_dirty_data_max determines the dirty space limit. Once that value is 7969962b56SMatthew Ahrens * exceeded, new writes are halted until space frees up. 8069962b56SMatthew Ahrens * 8169962b56SMatthew Ahrens * The zfs_dirty_data_sync tunable dictates the threshold at which we 8269962b56SMatthew Ahrens * ensure that there is a txg syncing (see the comment in txg.c for a full 8369962b56SMatthew Ahrens * description of transaction group stages). 8469962b56SMatthew Ahrens * 8569962b56SMatthew Ahrens * The IO scheduler uses both the dirty space limit and current amount of 8669962b56SMatthew Ahrens * dirty data as inputs. Those values affect the number of concurrent IOs ZFS 8769962b56SMatthew Ahrens * issues. See the comment in vdev_queue.c for details of the IO scheduler. 8869962b56SMatthew Ahrens * 8969962b56SMatthew Ahrens * The delay is also calculated based on the amount of dirty data. See the 9069962b56SMatthew Ahrens * comment above dmu_tx_delay() for details. 9169962b56SMatthew Ahrens */ 9269962b56SMatthew Ahrens 9369962b56SMatthew Ahrens /* 9469962b56SMatthew Ahrens * zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory, 9569962b56SMatthew Ahrens * capped at zfs_dirty_data_max_max. It can also be overridden in /etc/system. 9669962b56SMatthew Ahrens */ 9769962b56SMatthew Ahrens uint64_t zfs_dirty_data_max; 9869962b56SMatthew Ahrens uint64_t zfs_dirty_data_max_max = 4ULL * 1024 * 1024 * 1024; 9969962b56SMatthew Ahrens int zfs_dirty_data_max_percent = 10; 10069962b56SMatthew Ahrens 10169962b56SMatthew Ahrens /* 10269962b56SMatthew Ahrens * If there is at least this much dirty data, push out a txg. 10369962b56SMatthew Ahrens */ 10469962b56SMatthew Ahrens uint64_t zfs_dirty_data_sync = 64 * 1024 * 1024; 10569962b56SMatthew Ahrens 10669962b56SMatthew Ahrens /* 10769962b56SMatthew Ahrens * Once there is this amount of dirty data, the dmu_tx_delay() will kick in 10869962b56SMatthew Ahrens * and delay each transaction. 10969962b56SMatthew Ahrens * This value should be >= zfs_vdev_async_write_active_max_dirty_percent. 11069962b56SMatthew Ahrens */ 11169962b56SMatthew Ahrens int zfs_delay_min_dirty_percent = 60; 11269962b56SMatthew Ahrens 11369962b56SMatthew Ahrens /* 11469962b56SMatthew Ahrens * This controls how quickly the delay approaches infinity. 115d85a1e96SMatthew Ahrens * Larger values cause it to delay more for a given amount of dirty data. 116d85a1e96SMatthew Ahrens * Therefore larger values will cause there to be less dirty data for a 11769962b56SMatthew Ahrens * given throughput. 11869962b56SMatthew Ahrens * 11969962b56SMatthew Ahrens * For the smoothest delay, this value should be about 1 billion divided 12069962b56SMatthew Ahrens * by the maximum number of operations per second. This will smoothly 12169962b56SMatthew Ahrens * handle between 10x and 1/10th this number. 12269962b56SMatthew Ahrens * 12369962b56SMatthew Ahrens * Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the 12469962b56SMatthew Ahrens * multiply in dmu_tx_delay(). 12569962b56SMatthew Ahrens */ 12669962b56SMatthew Ahrens uint64_t zfs_delay_scale = 1000 * 1000 * 1000 / 2000; 12705715f94SMark Maybee 1281ab7f2deSmaybee 1290689f76cSAdam Leventhal hrtime_t zfs_throttle_delay = MSEC2NSEC(10); 1300689f76cSAdam Leventhal hrtime_t zfs_throttle_resolution = MSEC2NSEC(10); 1310689f76cSAdam Leventhal 1323f9d6ad7SLin Ling int 133088f3894Sahrens dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) 134fa9e4066Sahrens { 135fa9e4066Sahrens uint64_t obj; 136fa9e4066Sahrens int err; 137fa9e4066Sahrens 138fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, 139c1379625SJustin T. Gibbs dsl_dir_phys(dp->dp_root_dir)->dd_child_dir_zapobj, 140088f3894Sahrens name, sizeof (obj), 1, &obj); 141ea8dc4b6Seschrock if (err) 142ea8dc4b6Seschrock return (err); 143fa9e4066Sahrens 1443b2aab18SMatthew Ahrens return (dsl_dir_hold_obj(dp, obj, name, dp, ddp)); 145fa9e4066Sahrens } 146fa9e4066Sahrens 147fa9e4066Sahrens static dsl_pool_t * 148fa9e4066Sahrens dsl_pool_open_impl(spa_t *spa, uint64_t txg) 149fa9e4066Sahrens { 150fa9e4066Sahrens dsl_pool_t *dp; 151fa9e4066Sahrens blkptr_t *bp = spa_get_rootblkptr(spa); 152fa9e4066Sahrens 153fa9e4066Sahrens dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); 154fa9e4066Sahrens dp->dp_spa = spa; 155fa9e4066Sahrens dp->dp_meta_rootbp = *bp; 1563b2aab18SMatthew Ahrens rrw_init(&dp->dp_config_rwlock, B_TRUE); 157fa9e4066Sahrens txg_init(dp, txg); 158fa9e4066Sahrens 159fa9e4066Sahrens txg_list_create(&dp->dp_dirty_datasets, 160fa9e4066Sahrens offsetof(dsl_dataset_t, ds_dirty_link)); 161ce636f8bSMatthew Ahrens txg_list_create(&dp->dp_dirty_zilogs, 162ce636f8bSMatthew Ahrens offsetof(zilog_t, zl_dirty_link)); 163fa9e4066Sahrens txg_list_create(&dp->dp_dirty_dirs, 164fa9e4066Sahrens offsetof(dsl_dir_t, dd_dirty_link)); 1651d452cf5Sahrens txg_list_create(&dp->dp_sync_tasks, 1663b2aab18SMatthew Ahrens offsetof(dsl_sync_task_t, dst_node)); 167fa9e4066Sahrens 1681ab7f2deSmaybee mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); 16969962b56SMatthew Ahrens cv_init(&dp->dp_spaceavail_cv, NULL, CV_DEFAULT, NULL); 1701ab7f2deSmaybee 1719d3574bfSNeil Perrin dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri, 1729d3574bfSNeil Perrin 1, 4, 0); 1739d3574bfSNeil Perrin 174fa9e4066Sahrens return (dp); 175fa9e4066Sahrens } 176fa9e4066Sahrens 177ea8dc4b6Seschrock int 178ad135b5dSChristopher Siden dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) 179fa9e4066Sahrens { 180fa9e4066Sahrens int err; 181fa9e4066Sahrens dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 182ad135b5dSChristopher Siden 183ad135b5dSChristopher Siden err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, 184ad135b5dSChristopher Siden &dp->dp_meta_objset); 185ad135b5dSChristopher Siden if (err != 0) 186ad135b5dSChristopher Siden dsl_pool_close(dp); 187ad135b5dSChristopher Siden else 188ad135b5dSChristopher Siden *dpp = dp; 189ad135b5dSChristopher Siden 190ad135b5dSChristopher Siden return (err); 191ad135b5dSChristopher Siden } 192ad135b5dSChristopher Siden 193ad135b5dSChristopher Siden int 194ad135b5dSChristopher Siden dsl_pool_open(dsl_pool_t *dp) 195ad135b5dSChristopher Siden { 196ad135b5dSChristopher Siden int err; 197088f3894Sahrens dsl_dir_t *dd; 198088f3894Sahrens dsl_dataset_t *ds; 199cde58dbcSMatthew Ahrens uint64_t obj; 200fa9e4066Sahrens 2013b2aab18SMatthew Ahrens rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 202fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 203fa9e4066Sahrens DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, 204fa9e4066Sahrens &dp->dp_root_dir_obj); 205ea8dc4b6Seschrock if (err) 206ea8dc4b6Seschrock goto out; 207ea8dc4b6Seschrock 2083b2aab18SMatthew Ahrens err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, 209ea8dc4b6Seschrock NULL, dp, &dp->dp_root_dir); 210ea8dc4b6Seschrock if (err) 211ea8dc4b6Seschrock goto out; 212fa9e4066Sahrens 213088f3894Sahrens err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); 214ea8dc4b6Seschrock if (err) 215ea8dc4b6Seschrock goto out; 216ea8dc4b6Seschrock 217ad135b5dSChristopher Siden if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { 218088f3894Sahrens err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); 219088f3894Sahrens if (err) 220088f3894Sahrens goto out; 221c1379625SJustin T. Gibbs err = dsl_dataset_hold_obj(dp, 222c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds); 2238f63aa46SLin Ling if (err == 0) { 2248f63aa46SLin Ling err = dsl_dataset_hold_obj(dp, 225c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_obj, dp, 2268f63aa46SLin Ling &dp->dp_origin_snap); 2278f63aa46SLin Ling dsl_dataset_rele(ds, FTAG); 2288f63aa46SLin Ling } 2293b2aab18SMatthew Ahrens dsl_dir_rele(dd, dp); 230088f3894Sahrens if (err) 231088f3894Sahrens goto out; 232088f3894Sahrens } 233088f3894Sahrens 234ad135b5dSChristopher Siden if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { 235cde58dbcSMatthew Ahrens err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, 236cde58dbcSMatthew Ahrens &dp->dp_free_dir); 237cde58dbcSMatthew Ahrens if (err) 238cde58dbcSMatthew Ahrens goto out; 239cde58dbcSMatthew Ahrens 240cde58dbcSMatthew Ahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 241cde58dbcSMatthew Ahrens DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); 242cde58dbcSMatthew Ahrens if (err) 243cde58dbcSMatthew Ahrens goto out; 2443b2aab18SMatthew Ahrens VERIFY0(bpobj_open(&dp->dp_free_bpobj, 245cde58dbcSMatthew Ahrens dp->dp_meta_objset, obj)); 246cde58dbcSMatthew Ahrens } 247cde58dbcSMatthew Ahrens 2487fd05ac4SMatthew Ahrens /* 2497fd05ac4SMatthew Ahrens * Note: errors ignored, because the leak dir will not exist if we 2507fd05ac4SMatthew Ahrens * have not encountered a leak yet. 2517fd05ac4SMatthew Ahrens */ 2527fd05ac4SMatthew Ahrens (void) dsl_pool_open_special_dir(dp, LEAK_DIR_NAME, 2537fd05ac4SMatthew Ahrens &dp->dp_leak_dir); 2547fd05ac4SMatthew Ahrens 2552acef22dSMatthew Ahrens if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) { 256ad135b5dSChristopher Siden err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 257ad135b5dSChristopher Siden DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, 258ad135b5dSChristopher Siden &dp->dp_bptree_obj); 259ad135b5dSChristopher Siden if (err != 0) 260ad135b5dSChristopher Siden goto out; 261ad135b5dSChristopher Siden } 262ad135b5dSChristopher Siden 2632acef22dSMatthew Ahrens if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMPTY_BPOBJ)) { 264f1745736SMatthew Ahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 265f1745736SMatthew Ahrens DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, 266f1745736SMatthew Ahrens &dp->dp_empty_bpobj); 267f1745736SMatthew Ahrens if (err != 0) 268f1745736SMatthew Ahrens goto out; 269f1745736SMatthew Ahrens } 270f1745736SMatthew Ahrens 271ca45db41SChris Kirby err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 272ca45db41SChris Kirby DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, 273ca45db41SChris Kirby &dp->dp_tmp_userrefs_obj); 274ca45db41SChris Kirby if (err == ENOENT) 275ca45db41SChris Kirby err = 0; 276ca45db41SChris Kirby if (err) 277ca45db41SChris Kirby goto out; 278ca45db41SChris Kirby 279ad135b5dSChristopher Siden err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); 280088f3894Sahrens 281ea8dc4b6Seschrock out: 2823b2aab18SMatthew Ahrens rrw_exit(&dp->dp_config_rwlock, FTAG); 283ea8dc4b6Seschrock return (err); 284fa9e4066Sahrens } 285fa9e4066Sahrens 286fa9e4066Sahrens void 287fa9e4066Sahrens dsl_pool_close(dsl_pool_t *dp) 288fa9e4066Sahrens { 289088f3894Sahrens /* 29069962b56SMatthew Ahrens * Drop our references from dsl_pool_open(). 29169962b56SMatthew Ahrens * 292088f3894Sahrens * Since we held the origin_snap from "syncing" context (which 293088f3894Sahrens * includes pool-opening context), it actually only got a "ref" 294088f3894Sahrens * and not a hold, so just drop that here. 295088f3894Sahrens */ 296088f3894Sahrens if (dp->dp_origin_snap) 2973b2aab18SMatthew Ahrens dsl_dataset_rele(dp->dp_origin_snap, dp); 298ea8dc4b6Seschrock if (dp->dp_mos_dir) 2993b2aab18SMatthew Ahrens dsl_dir_rele(dp->dp_mos_dir, dp); 300cde58dbcSMatthew Ahrens if (dp->dp_free_dir) 3013b2aab18SMatthew Ahrens dsl_dir_rele(dp->dp_free_dir, dp); 3027fd05ac4SMatthew Ahrens if (dp->dp_leak_dir) 3037fd05ac4SMatthew Ahrens dsl_dir_rele(dp->dp_leak_dir, dp); 304ea8dc4b6Seschrock if (dp->dp_root_dir) 3053b2aab18SMatthew Ahrens dsl_dir_rele(dp->dp_root_dir, dp); 306fa9e4066Sahrens 307cde58dbcSMatthew Ahrens bpobj_close(&dp->dp_free_bpobj); 308cde58dbcSMatthew Ahrens 309fa9e4066Sahrens /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ 310ea8dc4b6Seschrock if (dp->dp_meta_objset) 311503ad85cSMatthew Ahrens dmu_objset_evict(dp->dp_meta_objset); 312fa9e4066Sahrens 313fa9e4066Sahrens txg_list_destroy(&dp->dp_dirty_datasets); 314ce636f8bSMatthew Ahrens txg_list_destroy(&dp->dp_dirty_zilogs); 31554a91118SChris Kirby txg_list_destroy(&dp->dp_sync_tasks); 316fa9e4066Sahrens txg_list_destroy(&dp->dp_dirty_dirs); 317fa9e4066Sahrens 318874395d5Smaybee arc_flush(dp->dp_spa); 319fa9e4066Sahrens txg_fini(dp); 3203f9d6ad7SLin Ling dsl_scan_fini(dp); 321*bc9014e6SJustin Gibbs dmu_buf_user_evict_wait(); 322*bc9014e6SJustin Gibbs 3233b2aab18SMatthew Ahrens rrw_destroy(&dp->dp_config_rwlock); 3241ab7f2deSmaybee mutex_destroy(&dp->dp_lock); 3259d3574bfSNeil Perrin taskq_destroy(dp->dp_vnrele_taskq); 32688b7b0f2SMatthew Ahrens if (dp->dp_blkstats) 32788b7b0f2SMatthew Ahrens kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); 328fa9e4066Sahrens kmem_free(dp, sizeof (dsl_pool_t)); 329fa9e4066Sahrens } 330fa9e4066Sahrens 331fa9e4066Sahrens dsl_pool_t * 3320a48a24eStimh dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) 333fa9e4066Sahrens { 334fa9e4066Sahrens int err; 335fa9e4066Sahrens dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 336fa9e4066Sahrens dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); 337503ad85cSMatthew Ahrens objset_t *os; 338088f3894Sahrens dsl_dataset_t *ds; 339cde58dbcSMatthew Ahrens uint64_t obj; 340088f3894Sahrens 3413b2aab18SMatthew Ahrens rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 3423b2aab18SMatthew Ahrens 343088f3894Sahrens /* create and open the MOS (meta-objset) */ 344503ad85cSMatthew Ahrens dp->dp_meta_objset = dmu_objset_create_impl(spa, 345503ad85cSMatthew Ahrens NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); 346fa9e4066Sahrens 347fa9e4066Sahrens /* create the pool directory */ 348fa9e4066Sahrens err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 349fa9e4066Sahrens DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); 350fb09f5aaSMadhav Suresh ASSERT0(err); 351fa9e4066Sahrens 3523f9d6ad7SLin Ling /* Initialize scan structures */ 3533b2aab18SMatthew Ahrens VERIFY0(dsl_scan_init(dp, txg)); 3543f9d6ad7SLin Ling 355fa9e4066Sahrens /* create and open the root dir */ 356088f3894Sahrens dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); 3573b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, 358ea8dc4b6Seschrock NULL, dp, &dp->dp_root_dir)); 359fa9e4066Sahrens 360fa9e4066Sahrens /* create and open the meta-objset dir */ 361088f3894Sahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); 3623b2aab18SMatthew Ahrens VERIFY0(dsl_pool_open_special_dir(dp, 363088f3894Sahrens MOS_DIR_NAME, &dp->dp_mos_dir)); 364088f3894Sahrens 365cde58dbcSMatthew Ahrens if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { 366cde58dbcSMatthew Ahrens /* create and open the free dir */ 367cde58dbcSMatthew Ahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, 368cde58dbcSMatthew Ahrens FREE_DIR_NAME, tx); 3693b2aab18SMatthew Ahrens VERIFY0(dsl_pool_open_special_dir(dp, 370cde58dbcSMatthew Ahrens FREE_DIR_NAME, &dp->dp_free_dir)); 371cde58dbcSMatthew Ahrens 372cde58dbcSMatthew Ahrens /* create and open the free_bplist */ 373b5152584SMatthew Ahrens obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx); 374cde58dbcSMatthew Ahrens VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 375cde58dbcSMatthew Ahrens DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); 3763b2aab18SMatthew Ahrens VERIFY0(bpobj_open(&dp->dp_free_bpobj, 377cde58dbcSMatthew Ahrens dp->dp_meta_objset, obj)); 378cde58dbcSMatthew Ahrens } 379cde58dbcSMatthew Ahrens 380088f3894Sahrens if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) 381088f3894Sahrens dsl_pool_create_origin(dp, tx); 382088f3894Sahrens 383088f3894Sahrens /* create the root dataset */ 384cde58dbcSMatthew Ahrens obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); 385088f3894Sahrens 386088f3894Sahrens /* create the root objset */ 3873b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); 388503ad85cSMatthew Ahrens os = dmu_objset_create_impl(dp->dp_spa, ds, 389088f3894Sahrens dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx); 390088f3894Sahrens #ifdef _KERNEL 391503ad85cSMatthew Ahrens zfs_create_fs(os, kcred, zplprops, tx); 392088f3894Sahrens #endif 393088f3894Sahrens dsl_dataset_rele(ds, FTAG); 394fa9e4066Sahrens 395fa9e4066Sahrens dmu_tx_commit(tx); 396fa9e4066Sahrens 3973b2aab18SMatthew Ahrens rrw_exit(&dp->dp_config_rwlock, FTAG); 3983b2aab18SMatthew Ahrens 399fa9e4066Sahrens return (dp); 400fa9e4066Sahrens } 401fa9e4066Sahrens 402ce636f8bSMatthew Ahrens /* 403ce636f8bSMatthew Ahrens * Account for the meta-objset space in its placeholder dsl_dir. 404ce636f8bSMatthew Ahrens */ 405ce636f8bSMatthew Ahrens void 406ce636f8bSMatthew Ahrens dsl_pool_mos_diduse_space(dsl_pool_t *dp, 407ce636f8bSMatthew Ahrens int64_t used, int64_t comp, int64_t uncomp) 408ce636f8bSMatthew Ahrens { 409ce636f8bSMatthew Ahrens ASSERT3U(comp, ==, uncomp); /* it's all metadata */ 410ce636f8bSMatthew Ahrens mutex_enter(&dp->dp_lock); 411ce636f8bSMatthew Ahrens dp->dp_mos_used_delta += used; 412ce636f8bSMatthew Ahrens dp->dp_mos_compressed_delta += comp; 413ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta += uncomp; 414ce636f8bSMatthew Ahrens mutex_exit(&dp->dp_lock); 415ce636f8bSMatthew Ahrens } 416ce636f8bSMatthew Ahrens 417cde58dbcSMatthew Ahrens static int 418cde58dbcSMatthew Ahrens deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 419cde58dbcSMatthew Ahrens { 420cde58dbcSMatthew Ahrens dsl_deadlist_t *dl = arg; 421cde58dbcSMatthew Ahrens dsl_deadlist_insert(dl, bp, tx); 422cde58dbcSMatthew Ahrens return (0); 423cde58dbcSMatthew Ahrens } 424cde58dbcSMatthew Ahrens 42569962b56SMatthew Ahrens static void 42669962b56SMatthew Ahrens dsl_pool_sync_mos(dsl_pool_t *dp, dmu_tx_t *tx) 42769962b56SMatthew Ahrens { 42869962b56SMatthew Ahrens zio_t *zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 42969962b56SMatthew Ahrens dmu_objset_sync(dp->dp_meta_objset, zio, tx); 43069962b56SMatthew Ahrens VERIFY0(zio_wait(zio)); 43169962b56SMatthew Ahrens dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); 43269962b56SMatthew Ahrens spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); 43369962b56SMatthew Ahrens } 43469962b56SMatthew Ahrens 43569962b56SMatthew Ahrens static void 43669962b56SMatthew Ahrens dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta) 43769962b56SMatthew Ahrens { 43869962b56SMatthew Ahrens ASSERT(MUTEX_HELD(&dp->dp_lock)); 43969962b56SMatthew Ahrens 44069962b56SMatthew Ahrens if (delta < 0) 44169962b56SMatthew Ahrens ASSERT3U(-delta, <=, dp->dp_dirty_total); 44269962b56SMatthew Ahrens 44369962b56SMatthew Ahrens dp->dp_dirty_total += delta; 44469962b56SMatthew Ahrens 44569962b56SMatthew Ahrens /* 44669962b56SMatthew Ahrens * Note: we signal even when increasing dp_dirty_total. 44769962b56SMatthew Ahrens * This ensures forward progress -- each thread wakes the next waiter. 44869962b56SMatthew Ahrens */ 44969962b56SMatthew Ahrens if (dp->dp_dirty_total <= zfs_dirty_data_max) 45069962b56SMatthew Ahrens cv_signal(&dp->dp_spaceavail_cv); 45169962b56SMatthew Ahrens } 45269962b56SMatthew Ahrens 453fa9e4066Sahrens void 454fa9e4066Sahrens dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) 455fa9e4066Sahrens { 456c717a561Smaybee zio_t *zio; 457fa9e4066Sahrens dmu_tx_t *tx; 458c717a561Smaybee dsl_dir_t *dd; 459c717a561Smaybee dsl_dataset_t *ds; 460503ad85cSMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 461ce636f8bSMatthew Ahrens list_t synced_datasets; 462ce636f8bSMatthew Ahrens 463ce636f8bSMatthew Ahrens list_create(&synced_datasets, sizeof (dsl_dataset_t), 464ce636f8bSMatthew Ahrens offsetof(dsl_dataset_t, ds_synced_link)); 465fa9e4066Sahrens 466fa9e4066Sahrens tx = dmu_tx_create_assigned(dp, txg); 467fa9e4066Sahrens 46869962b56SMatthew Ahrens /* 46969962b56SMatthew Ahrens * Write out all dirty blocks of dirty datasets. 47069962b56SMatthew Ahrens */ 471c717a561Smaybee zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 47269962b56SMatthew Ahrens while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 47314843421SMatthew Ahrens /* 47414843421SMatthew Ahrens * We must not sync any non-MOS datasets twice, because 47514843421SMatthew Ahrens * we may have taken a snapshot of them. However, we 47614843421SMatthew Ahrens * may sync newly-created datasets on pass 2. 47714843421SMatthew Ahrens */ 47814843421SMatthew Ahrens ASSERT(!list_link_active(&ds->ds_synced_link)); 479ce636f8bSMatthew Ahrens list_insert_tail(&synced_datasets, ds); 480c717a561Smaybee dsl_dataset_sync(ds, zio, tx); 481c717a561Smaybee } 48269962b56SMatthew Ahrens VERIFY0(zio_wait(zio)); 48314843421SMatthew Ahrens 48469962b56SMatthew Ahrens /* 48569962b56SMatthew Ahrens * We have written all of the accounted dirty data, so our 48669962b56SMatthew Ahrens * dp_space_towrite should now be zero. However, some seldom-used 48769962b56SMatthew Ahrens * code paths do not adhere to this (e.g. dbuf_undirty(), also 48869962b56SMatthew Ahrens * rounding error in dbuf_write_physdone). 48969962b56SMatthew Ahrens * Shore up the accounting of any dirtied space now. 49069962b56SMatthew Ahrens */ 49169962b56SMatthew Ahrens dsl_pool_undirty_space(dp, dp->dp_dirty_pertxg[txg & TXG_MASK], txg); 492c717a561Smaybee 493ce636f8bSMatthew Ahrens /* 494ce636f8bSMatthew Ahrens * After the data blocks have been written (ensured by the zio_wait() 495ce636f8bSMatthew Ahrens * above), update the user/group space accounting. 496ce636f8bSMatthew Ahrens */ 49769962b56SMatthew Ahrens for (ds = list_head(&synced_datasets); ds != NULL; 49869962b56SMatthew Ahrens ds = list_next(&synced_datasets, ds)) { 4990a586ceaSMark Shellenbaum dmu_objset_do_userquota_updates(ds->ds_objset, tx); 50069962b56SMatthew Ahrens } 50114843421SMatthew Ahrens 50214843421SMatthew Ahrens /* 50314843421SMatthew Ahrens * Sync the datasets again to push out the changes due to 5043f9d6ad7SLin Ling * userspace updates. This must be done before we process the 505ce636f8bSMatthew Ahrens * sync tasks, so that any snapshots will have the correct 506ce636f8bSMatthew Ahrens * user accounting information (and we won't get confused 507ce636f8bSMatthew Ahrens * about which blocks are part of the snapshot). 50814843421SMatthew Ahrens */ 50914843421SMatthew Ahrens zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 51069962b56SMatthew Ahrens while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 51114843421SMatthew Ahrens ASSERT(list_link_active(&ds->ds_synced_link)); 51214843421SMatthew Ahrens dmu_buf_rele(ds->ds_dbuf, ds); 51314843421SMatthew Ahrens dsl_dataset_sync(ds, zio, tx); 51414843421SMatthew Ahrens } 51569962b56SMatthew Ahrens VERIFY0(zio_wait(zio)); 51614843421SMatthew Ahrens 517b24ab676SJeff Bonwick /* 518ce636f8bSMatthew Ahrens * Now that the datasets have been completely synced, we can 519ce636f8bSMatthew Ahrens * clean up our in-memory structures accumulated while syncing: 520ce636f8bSMatthew Ahrens * 521ce636f8bSMatthew Ahrens * - move dead blocks from the pending deadlist to the on-disk deadlist 522ce636f8bSMatthew Ahrens * - release hold from dsl_dataset_dirty() 523b24ab676SJeff Bonwick */ 52469962b56SMatthew Ahrens while ((ds = list_remove_head(&synced_datasets)) != NULL) { 525ce636f8bSMatthew Ahrens objset_t *os = ds->ds_objset; 526cde58dbcSMatthew Ahrens bplist_iterate(&ds->ds_pending_deadlist, 527cde58dbcSMatthew Ahrens deadlist_enqueue_cb, &ds->ds_deadlist, tx); 528ce636f8bSMatthew Ahrens ASSERT(!dmu_objset_is_dirty(os, txg)); 529ce636f8bSMatthew Ahrens dmu_buf_rele(ds->ds_dbuf, ds); 530cde58dbcSMatthew Ahrens } 53169962b56SMatthew Ahrens while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) != NULL) { 532c717a561Smaybee dsl_dir_sync(dd, tx); 53369962b56SMatthew Ahrens } 534fa9e4066Sahrens 535ce636f8bSMatthew Ahrens /* 536ce636f8bSMatthew Ahrens * The MOS's space is accounted for in the pool/$MOS 537ce636f8bSMatthew Ahrens * (dp_mos_dir). We can't modify the mos while we're syncing 538ce636f8bSMatthew Ahrens * it, so we remember the deltas and apply them here. 539ce636f8bSMatthew Ahrens */ 540ce636f8bSMatthew Ahrens if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || 541ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta != 0) { 542ce636f8bSMatthew Ahrens dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, 543ce636f8bSMatthew Ahrens dp->dp_mos_used_delta, 544ce636f8bSMatthew Ahrens dp->dp_mos_compressed_delta, 545ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta, tx); 546ce636f8bSMatthew Ahrens dp->dp_mos_used_delta = 0; 547ce636f8bSMatthew Ahrens dp->dp_mos_compressed_delta = 0; 548ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta = 0; 549ce636f8bSMatthew Ahrens } 550ce636f8bSMatthew Ahrens 551503ad85cSMatthew Ahrens if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || 552503ad85cSMatthew Ahrens list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { 55369962b56SMatthew Ahrens dsl_pool_sync_mos(dp, tx); 554fa9e4066Sahrens } 555fa9e4066Sahrens 556ce636f8bSMatthew Ahrens /* 557ce636f8bSMatthew Ahrens * If we modify a dataset in the same txg that we want to destroy it, 558ce636f8bSMatthew Ahrens * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. 559ce636f8bSMatthew Ahrens * dsl_dir_destroy_check() will fail if there are unexpected holds. 560ce636f8bSMatthew Ahrens * Therefore, we want to sync the MOS (thus syncing the dd_dbuf 561ce636f8bSMatthew Ahrens * and clearing the hold on it) before we process the sync_tasks. 562ce636f8bSMatthew Ahrens * The MOS data dirtied by the sync_tasks will be synced on the next 563ce636f8bSMatthew Ahrens * pass. 564ce636f8bSMatthew Ahrens */ 565ce636f8bSMatthew Ahrens if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { 5663b2aab18SMatthew Ahrens dsl_sync_task_t *dst; 567ce636f8bSMatthew Ahrens /* 568ce636f8bSMatthew Ahrens * No more sync tasks should have been added while we 569ce636f8bSMatthew Ahrens * were syncing. 570ce636f8bSMatthew Ahrens */ 57169962b56SMatthew Ahrens ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1); 57269962b56SMatthew Ahrens while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)) != NULL) 5733b2aab18SMatthew Ahrens dsl_sync_task_sync(dst, tx); 574ce636f8bSMatthew Ahrens } 575ce636f8bSMatthew Ahrens 576fa9e4066Sahrens dmu_tx_commit(tx); 57705715f94SMark Maybee 57869962b56SMatthew Ahrens DTRACE_PROBE2(dsl_pool_sync__done, dsl_pool_t *dp, dp, uint64_t, txg); 579fa9e4066Sahrens } 580fa9e4066Sahrens 581fa9e4066Sahrens void 582b24ab676SJeff Bonwick dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) 583fa9e4066Sahrens { 584ce636f8bSMatthew Ahrens zilog_t *zilog; 585fa9e4066Sahrens 586ce636f8bSMatthew Ahrens while (zilog = txg_list_remove(&dp->dp_dirty_zilogs, txg)) { 58769962b56SMatthew Ahrens dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); 588ce636f8bSMatthew Ahrens zil_clean(zilog, txg); 589ce636f8bSMatthew Ahrens ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); 590ce636f8bSMatthew Ahrens dmu_buf_rele(ds->ds_dbuf, zilog); 591fa9e4066Sahrens } 592b24ab676SJeff Bonwick ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); 593fa9e4066Sahrens } 594fa9e4066Sahrens 595c717a561Smaybee /* 596c717a561Smaybee * TRUE if the current thread is the tx_sync_thread or if we 597c717a561Smaybee * are being called from SPA context during pool initialization. 598c717a561Smaybee */ 599fa9e4066Sahrens int 600fa9e4066Sahrens dsl_pool_sync_context(dsl_pool_t *dp) 601fa9e4066Sahrens { 602fa9e4066Sahrens return (curthread == dp->dp_tx.tx_sync_thread || 603ad135b5dSChristopher Siden spa_is_initializing(dp->dp_spa)); 604fa9e4066Sahrens } 605fa9e4066Sahrens 606fa9e4066Sahrens uint64_t 607fa9e4066Sahrens dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) 608fa9e4066Sahrens { 609fa9e4066Sahrens uint64_t space, resv; 610fa9e4066Sahrens 611fa9e4066Sahrens /* 612fa9e4066Sahrens * If we're trying to assess whether it's OK to do a free, 613fa9e4066Sahrens * cut the reservation in half to allow forward progress 614fa9e4066Sahrens * (e.g. make it possible to rm(1) files from a full pool). 615fa9e4066Sahrens */ 616485bbbf5SGeorge Wilson space = spa_get_dspace(dp->dp_spa); 617c39f2c8cSChristopher Siden resv = spa_get_slop_space(dp->dp_spa); 618fa9e4066Sahrens if (netfree) 619fa9e4066Sahrens resv >>= 1; 620fa9e4066Sahrens 621fa9e4066Sahrens return (space - resv); 622fa9e4066Sahrens } 6231ab7f2deSmaybee 62469962b56SMatthew Ahrens boolean_t 62569962b56SMatthew Ahrens dsl_pool_need_dirty_delay(dsl_pool_t *dp) 6261ab7f2deSmaybee { 62769962b56SMatthew Ahrens uint64_t delay_min_bytes = 62869962b56SMatthew Ahrens zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100; 62969962b56SMatthew Ahrens boolean_t rv; 6301ab7f2deSmaybee 63169962b56SMatthew Ahrens mutex_enter(&dp->dp_lock); 63269962b56SMatthew Ahrens if (dp->dp_dirty_total > zfs_dirty_data_sync) 63369962b56SMatthew Ahrens txg_kick(dp); 63469962b56SMatthew Ahrens rv = (dp->dp_dirty_total > delay_min_bytes); 63569962b56SMatthew Ahrens mutex_exit(&dp->dp_lock); 63669962b56SMatthew Ahrens return (rv); 6371ab7f2deSmaybee } 6381ab7f2deSmaybee 6391ab7f2deSmaybee void 64069962b56SMatthew Ahrens dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 6411ab7f2deSmaybee { 64269962b56SMatthew Ahrens if (space > 0) { 64369962b56SMatthew Ahrens mutex_enter(&dp->dp_lock); 64469962b56SMatthew Ahrens dp->dp_dirty_pertxg[tx->tx_txg & TXG_MASK] += space; 64569962b56SMatthew Ahrens dsl_pool_dirty_delta(dp, space); 64669962b56SMatthew Ahrens mutex_exit(&dp->dp_lock); 6471ab7f2deSmaybee } 6481ab7f2deSmaybee } 6491ab7f2deSmaybee 6501ab7f2deSmaybee void 65169962b56SMatthew Ahrens dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg) 6521ab7f2deSmaybee { 65369962b56SMatthew Ahrens ASSERT3S(space, >=, 0); 65469962b56SMatthew Ahrens if (space == 0) 65569962b56SMatthew Ahrens return; 65669962b56SMatthew Ahrens mutex_enter(&dp->dp_lock); 65769962b56SMatthew Ahrens if (dp->dp_dirty_pertxg[txg & TXG_MASK] < space) { 65869962b56SMatthew Ahrens /* XXX writing something we didn't dirty? */ 65969962b56SMatthew Ahrens space = dp->dp_dirty_pertxg[txg & TXG_MASK]; 6601ab7f2deSmaybee } 66169962b56SMatthew Ahrens ASSERT3U(dp->dp_dirty_pertxg[txg & TXG_MASK], >=, space); 66269962b56SMatthew Ahrens dp->dp_dirty_pertxg[txg & TXG_MASK] -= space; 66369962b56SMatthew Ahrens ASSERT3U(dp->dp_dirty_total, >=, space); 66469962b56SMatthew Ahrens dsl_pool_dirty_delta(dp, -space); 66569962b56SMatthew Ahrens mutex_exit(&dp->dp_lock); 6661ab7f2deSmaybee } 667088f3894Sahrens 668088f3894Sahrens /* ARGSUSED */ 669088f3894Sahrens static int 6703b2aab18SMatthew Ahrens upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) 671088f3894Sahrens { 672088f3894Sahrens dmu_tx_t *tx = arg; 673088f3894Sahrens dsl_dataset_t *ds, *prev = NULL; 674088f3894Sahrens int err; 675088f3894Sahrens 6763b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); 677088f3894Sahrens if (err) 678088f3894Sahrens return (err); 679088f3894Sahrens 680c1379625SJustin T. Gibbs while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 681c1379625SJustin T. Gibbs err = dsl_dataset_hold_obj(dp, 682c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 683088f3894Sahrens if (err) { 684088f3894Sahrens dsl_dataset_rele(ds, FTAG); 685088f3894Sahrens return (err); 686088f3894Sahrens } 687088f3894Sahrens 688c1379625SJustin T. Gibbs if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) 689088f3894Sahrens break; 690088f3894Sahrens dsl_dataset_rele(ds, FTAG); 691088f3894Sahrens ds = prev; 692088f3894Sahrens prev = NULL; 693088f3894Sahrens } 694088f3894Sahrens 695088f3894Sahrens if (prev == NULL) { 696088f3894Sahrens prev = dp->dp_origin_snap; 697088f3894Sahrens 698088f3894Sahrens /* 699088f3894Sahrens * The $ORIGIN can't have any data, or the accounting 700088f3894Sahrens * will be wrong. 701088f3894Sahrens */ 702c1379625SJustin T. Gibbs ASSERT0(dsl_dataset_phys(prev)->ds_bp.blk_birth); 703088f3894Sahrens 704088f3894Sahrens /* The origin doesn't get attached to itself */ 705088f3894Sahrens if (ds->ds_object == prev->ds_object) { 706088f3894Sahrens dsl_dataset_rele(ds, FTAG); 707088f3894Sahrens return (0); 708088f3894Sahrens } 709088f3894Sahrens 710088f3894Sahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 711c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_obj = prev->ds_object; 712c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_txg = 713c1379625SJustin T. Gibbs dsl_dataset_phys(prev)->ds_creation_txg; 714088f3894Sahrens 715088f3894Sahrens dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 716c1379625SJustin T. Gibbs dsl_dir_phys(ds->ds_dir)->dd_origin_obj = prev->ds_object; 717088f3894Sahrens 718088f3894Sahrens dmu_buf_will_dirty(prev->ds_dbuf, tx); 719c1379625SJustin T. Gibbs dsl_dataset_phys(prev)->ds_num_children++; 720088f3894Sahrens 721c1379625SJustin T. Gibbs if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) { 722088f3894Sahrens ASSERT(ds->ds_prev == NULL); 7233b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, 724c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_obj, 725c1379625SJustin T. Gibbs ds, &ds->ds_prev)); 726088f3894Sahrens } 727088f3894Sahrens } 728088f3894Sahrens 729c1379625SJustin T. Gibbs ASSERT3U(dsl_dir_phys(ds->ds_dir)->dd_origin_obj, ==, prev->ds_object); 730c1379625SJustin T. Gibbs ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_obj, ==, prev->ds_object); 731088f3894Sahrens 732c1379625SJustin T. Gibbs if (dsl_dataset_phys(prev)->ds_next_clones_obj == 0) { 733c33e334fSMatthew Ahrens dmu_buf_will_dirty(prev->ds_dbuf, tx); 734c1379625SJustin T. Gibbs dsl_dataset_phys(prev)->ds_next_clones_obj = 735088f3894Sahrens zap_create(dp->dp_meta_objset, 736088f3894Sahrens DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 737088f3894Sahrens } 7383b2aab18SMatthew Ahrens VERIFY0(zap_add_int(dp->dp_meta_objset, 739c1379625SJustin T. Gibbs dsl_dataset_phys(prev)->ds_next_clones_obj, ds->ds_object, tx)); 740088f3894Sahrens 741088f3894Sahrens dsl_dataset_rele(ds, FTAG); 742088f3894Sahrens if (prev != dp->dp_origin_snap) 743088f3894Sahrens dsl_dataset_rele(prev, FTAG); 744088f3894Sahrens return (0); 745088f3894Sahrens } 746088f3894Sahrens 747088f3894Sahrens void 748088f3894Sahrens dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) 749088f3894Sahrens { 750088f3894Sahrens ASSERT(dmu_tx_is_syncing(tx)); 751088f3894Sahrens ASSERT(dp->dp_origin_snap != NULL); 752088f3894Sahrens 7533b2aab18SMatthew Ahrens VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb, 754c33e334fSMatthew Ahrens tx, DS_FIND_CHILDREN)); 755088f3894Sahrens } 756088f3894Sahrens 757cde58dbcSMatthew Ahrens /* ARGSUSED */ 758cde58dbcSMatthew Ahrens static int 7593b2aab18SMatthew Ahrens upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 760cde58dbcSMatthew Ahrens { 761cde58dbcSMatthew Ahrens dmu_tx_t *tx = arg; 762cde58dbcSMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 763cde58dbcSMatthew Ahrens 764c1379625SJustin T. Gibbs if (dsl_dir_phys(ds->ds_dir)->dd_origin_obj != 0) { 765cde58dbcSMatthew Ahrens dsl_dataset_t *origin; 766cde58dbcSMatthew Ahrens 7673b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, 768c1379625SJustin T. Gibbs dsl_dir_phys(ds->ds_dir)->dd_origin_obj, FTAG, &origin)); 769cde58dbcSMatthew Ahrens 770c1379625SJustin T. Gibbs if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 771cde58dbcSMatthew Ahrens dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 772c1379625SJustin T. Gibbs dsl_dir_phys(origin->ds_dir)->dd_clones = 773c1379625SJustin T. Gibbs zap_create(mos, DMU_OT_DSL_CLONES, DMU_OT_NONE, 774c1379625SJustin T. Gibbs 0, tx); 775cde58dbcSMatthew Ahrens } 776cde58dbcSMatthew Ahrens 7773b2aab18SMatthew Ahrens VERIFY0(zap_add_int(dp->dp_meta_objset, 778c1379625SJustin T. Gibbs dsl_dir_phys(origin->ds_dir)->dd_clones, 779c1379625SJustin T. Gibbs ds->ds_object, tx)); 780cde58dbcSMatthew Ahrens 781cde58dbcSMatthew Ahrens dsl_dataset_rele(origin, FTAG); 782cde58dbcSMatthew Ahrens } 783cde58dbcSMatthew Ahrens return (0); 784cde58dbcSMatthew Ahrens } 785cde58dbcSMatthew Ahrens 786cde58dbcSMatthew Ahrens void 787cde58dbcSMatthew Ahrens dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) 788cde58dbcSMatthew Ahrens { 789cde58dbcSMatthew Ahrens ASSERT(dmu_tx_is_syncing(tx)); 790cde58dbcSMatthew Ahrens uint64_t obj; 791cde58dbcSMatthew Ahrens 792cde58dbcSMatthew Ahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); 7933b2aab18SMatthew Ahrens VERIFY0(dsl_pool_open_special_dir(dp, 794cde58dbcSMatthew Ahrens FREE_DIR_NAME, &dp->dp_free_dir)); 795cde58dbcSMatthew Ahrens 796cde58dbcSMatthew Ahrens /* 797cde58dbcSMatthew Ahrens * We can't use bpobj_alloc(), because spa_version() still 798cde58dbcSMatthew Ahrens * returns the old version, and we need a new-version bpobj with 799cde58dbcSMatthew Ahrens * subobj support. So call dmu_object_alloc() directly. 800cde58dbcSMatthew Ahrens */ 801cde58dbcSMatthew Ahrens obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, 802b5152584SMatthew Ahrens SPA_OLD_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); 8033b2aab18SMatthew Ahrens VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 804cde58dbcSMatthew Ahrens DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); 8053b2aab18SMatthew Ahrens VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); 806cde58dbcSMatthew Ahrens 8073b2aab18SMatthew Ahrens VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 808cde58dbcSMatthew Ahrens upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN)); 809cde58dbcSMatthew Ahrens } 810cde58dbcSMatthew Ahrens 811088f3894Sahrens void 812088f3894Sahrens dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) 813088f3894Sahrens { 814088f3894Sahrens uint64_t dsobj; 815088f3894Sahrens dsl_dataset_t *ds; 816088f3894Sahrens 817088f3894Sahrens ASSERT(dmu_tx_is_syncing(tx)); 818088f3894Sahrens ASSERT(dp->dp_origin_snap == NULL); 8193b2aab18SMatthew Ahrens ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER)); 820088f3894Sahrens 821088f3894Sahrens /* create the origin dir, ds, & snap-ds */ 822088f3894Sahrens dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, 823088f3894Sahrens NULL, 0, kcred, tx); 8243b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 8253b2aab18SMatthew Ahrens dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx); 826c1379625SJustin T. Gibbs VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, 827088f3894Sahrens dp, &dp->dp_origin_snap)); 828088f3894Sahrens dsl_dataset_rele(ds, FTAG); 829088f3894Sahrens } 8309d3574bfSNeil Perrin 8319d3574bfSNeil Perrin taskq_t * 8329d3574bfSNeil Perrin dsl_pool_vnrele_taskq(dsl_pool_t *dp) 8339d3574bfSNeil Perrin { 8349d3574bfSNeil Perrin return (dp->dp_vnrele_taskq); 8359d3574bfSNeil Perrin } 836ca45db41SChris Kirby 837ca45db41SChris Kirby /* 838ca45db41SChris Kirby * Walk through the pool-wide zap object of temporary snapshot user holds 839ca45db41SChris Kirby * and release them. 840ca45db41SChris Kirby */ 841ca45db41SChris Kirby void 842ca45db41SChris Kirby dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) 843ca45db41SChris Kirby { 844ca45db41SChris Kirby zap_attribute_t za; 845ca45db41SChris Kirby zap_cursor_t zc; 846ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 847ca45db41SChris Kirby uint64_t zapobj = dp->dp_tmp_userrefs_obj; 848a7a845e4SSteven Hartland nvlist_t *holds; 849ca45db41SChris Kirby 850ca45db41SChris Kirby if (zapobj == 0) 851ca45db41SChris Kirby return; 852ca45db41SChris Kirby ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 853ca45db41SChris Kirby 854a7a845e4SSteven Hartland holds = fnvlist_alloc(); 855a7a845e4SSteven Hartland 856ca45db41SChris Kirby for (zap_cursor_init(&zc, mos, zapobj); 857ca45db41SChris Kirby zap_cursor_retrieve(&zc, &za) == 0; 858ca45db41SChris Kirby zap_cursor_advance(&zc)) { 859ca45db41SChris Kirby char *htag; 860a7a845e4SSteven Hartland nvlist_t *tags; 861ca45db41SChris Kirby 862ca45db41SChris Kirby htag = strchr(za.za_name, '-'); 863ca45db41SChris Kirby *htag = '\0'; 864ca45db41SChris Kirby ++htag; 865a7a845e4SSteven Hartland if (nvlist_lookup_nvlist(holds, za.za_name, &tags) != 0) { 866a7a845e4SSteven Hartland tags = fnvlist_alloc(); 867a7a845e4SSteven Hartland fnvlist_add_boolean(tags, htag); 868a7a845e4SSteven Hartland fnvlist_add_nvlist(holds, za.za_name, tags); 869a7a845e4SSteven Hartland fnvlist_free(tags); 870a7a845e4SSteven Hartland } else { 871a7a845e4SSteven Hartland fnvlist_add_boolean(tags, htag); 872a7a845e4SSteven Hartland } 873ca45db41SChris Kirby } 874a7a845e4SSteven Hartland dsl_dataset_user_release_tmp(dp, holds); 875a7a845e4SSteven Hartland fnvlist_free(holds); 876ca45db41SChris Kirby zap_cursor_fini(&zc); 877ca45db41SChris Kirby } 878ca45db41SChris Kirby 879ca45db41SChris Kirby /* 880ca45db41SChris Kirby * Create the pool-wide zap object for storing temporary snapshot holds. 881ca45db41SChris Kirby */ 882ca45db41SChris Kirby void 883ca45db41SChris Kirby dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) 884ca45db41SChris Kirby { 885ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 886ca45db41SChris Kirby 887ca45db41SChris Kirby ASSERT(dp->dp_tmp_userrefs_obj == 0); 888ca45db41SChris Kirby ASSERT(dmu_tx_is_syncing(tx)); 889ca45db41SChris Kirby 890ad135b5dSChristopher Siden dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, 891ad135b5dSChristopher Siden DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); 892ca45db41SChris Kirby } 893ca45db41SChris Kirby 894ca45db41SChris Kirby static int 895ca45db41SChris Kirby dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, 8963b2aab18SMatthew Ahrens const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding) 897ca45db41SChris Kirby { 898ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 899ca45db41SChris Kirby uint64_t zapobj = dp->dp_tmp_userrefs_obj; 900ca45db41SChris Kirby char *name; 901ca45db41SChris Kirby int error; 902ca45db41SChris Kirby 903ca45db41SChris Kirby ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 904ca45db41SChris Kirby ASSERT(dmu_tx_is_syncing(tx)); 905ca45db41SChris Kirby 906ca45db41SChris Kirby /* 907ca45db41SChris Kirby * If the pool was created prior to SPA_VERSION_USERREFS, the 908ca45db41SChris Kirby * zap object for temporary holds might not exist yet. 909ca45db41SChris Kirby */ 910ca45db41SChris Kirby if (zapobj == 0) { 911ca45db41SChris Kirby if (holding) { 912ca45db41SChris Kirby dsl_pool_user_hold_create_obj(dp, tx); 913ca45db41SChris Kirby zapobj = dp->dp_tmp_userrefs_obj; 914ca45db41SChris Kirby } else { 915be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 916ca45db41SChris Kirby } 917ca45db41SChris Kirby } 918ca45db41SChris Kirby 919ca45db41SChris Kirby name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); 920ca45db41SChris Kirby if (holding) 9213b2aab18SMatthew Ahrens error = zap_add(mos, zapobj, name, 8, 1, &now, tx); 922ca45db41SChris Kirby else 923ca45db41SChris Kirby error = zap_remove(mos, zapobj, name, tx); 924ca45db41SChris Kirby strfree(name); 925ca45db41SChris Kirby 926ca45db41SChris Kirby return (error); 927ca45db41SChris Kirby } 928ca45db41SChris Kirby 929ca45db41SChris Kirby /* 930ca45db41SChris Kirby * Add a temporary hold for the given dataset object and tag. 931ca45db41SChris Kirby */ 932ca45db41SChris Kirby int 933ca45db41SChris Kirby dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 9343b2aab18SMatthew Ahrens uint64_t now, dmu_tx_t *tx) 935ca45db41SChris Kirby { 93615508ac0SChris Kirby return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); 937ca45db41SChris Kirby } 938ca45db41SChris Kirby 939ca45db41SChris Kirby /* 940ca45db41SChris Kirby * Release a temporary hold for the given dataset object and tag. 941ca45db41SChris Kirby */ 942ca45db41SChris Kirby int 943ca45db41SChris Kirby dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 944ca45db41SChris Kirby dmu_tx_t *tx) 945ca45db41SChris Kirby { 946ca45db41SChris Kirby return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL, 947ca45db41SChris Kirby tx, B_FALSE)); 948ca45db41SChris Kirby } 9493b2aab18SMatthew Ahrens 9503b2aab18SMatthew Ahrens /* 9513b2aab18SMatthew Ahrens * DSL Pool Configuration Lock 9523b2aab18SMatthew Ahrens * 9533b2aab18SMatthew Ahrens * The dp_config_rwlock protects against changes to DSL state (e.g. dataset 9543b2aab18SMatthew Ahrens * creation / destruction / rename / property setting). It must be held for 9553b2aab18SMatthew Ahrens * read to hold a dataset or dsl_dir. I.e. you must call 9563b2aab18SMatthew Ahrens * dsl_pool_config_enter() or dsl_pool_hold() before calling 9573b2aab18SMatthew Ahrens * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock 9583b2aab18SMatthew Ahrens * must be held continuously until all datasets and dsl_dirs are released. 9593b2aab18SMatthew Ahrens * 9603b2aab18SMatthew Ahrens * The only exception to this rule is that if a "long hold" is placed on 9613b2aab18SMatthew Ahrens * a dataset, then the dp_config_rwlock may be dropped while the dataset 9623b2aab18SMatthew Ahrens * is still held. The long hold will prevent the dataset from being 9633b2aab18SMatthew Ahrens * destroyed -- the destroy will fail with EBUSY. A long hold can be 9643b2aab18SMatthew Ahrens * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset 9653b2aab18SMatthew Ahrens * (by calling dsl_{dataset,objset}_{try}own{_obj}). 9663b2aab18SMatthew Ahrens * 9673b2aab18SMatthew Ahrens * Legitimate long-holders (including owners) should be long-running, cancelable 9683b2aab18SMatthew Ahrens * tasks that should cause "zfs destroy" to fail. This includes DMU 9693b2aab18SMatthew Ahrens * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open), 9703b2aab18SMatthew Ahrens * "zfs send", and "zfs diff". There are several other long-holders whose 9713b2aab18SMatthew Ahrens * uses are suboptimal (e.g. "zfs promote", and zil_suspend()). 9723b2aab18SMatthew Ahrens * 9733b2aab18SMatthew Ahrens * The usual formula for long-holding would be: 9743b2aab18SMatthew Ahrens * dsl_pool_hold() 9753b2aab18SMatthew Ahrens * dsl_dataset_hold() 9763b2aab18SMatthew Ahrens * ... perform checks ... 9773b2aab18SMatthew Ahrens * dsl_dataset_long_hold() 9783b2aab18SMatthew Ahrens * dsl_pool_rele() 9793b2aab18SMatthew Ahrens * ... perform long-running task ... 9803b2aab18SMatthew Ahrens * dsl_dataset_long_rele() 9813b2aab18SMatthew Ahrens * dsl_dataset_rele() 9823b2aab18SMatthew Ahrens * 9833b2aab18SMatthew Ahrens * Note that when the long hold is released, the dataset is still held but 9843b2aab18SMatthew Ahrens * the pool is not held. The dataset may change arbitrarily during this time 9853b2aab18SMatthew Ahrens * (e.g. it could be destroyed). Therefore you shouldn't do anything to the 9863b2aab18SMatthew Ahrens * dataset except release it. 9873b2aab18SMatthew Ahrens * 9883b2aab18SMatthew Ahrens * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only 9893b2aab18SMatthew Ahrens * or modifying operations. 9903b2aab18SMatthew Ahrens * 9913b2aab18SMatthew Ahrens * Modifying operations should generally use dsl_sync_task(). The synctask 9923b2aab18SMatthew Ahrens * infrastructure enforces proper locking strategy with respect to the 9933b2aab18SMatthew Ahrens * dp_config_rwlock. See the comment above dsl_sync_task() for details. 9943b2aab18SMatthew Ahrens * 9953b2aab18SMatthew Ahrens * Read-only operations will manually hold the pool, then the dataset, obtain 9963b2aab18SMatthew Ahrens * information from the dataset, then release the pool and dataset. 9973b2aab18SMatthew Ahrens * dmu_objset_{hold,rele}() are convenience routines that also do the pool 9983b2aab18SMatthew Ahrens * hold/rele. 9993b2aab18SMatthew Ahrens */ 10003b2aab18SMatthew Ahrens 10013b2aab18SMatthew Ahrens int 10023b2aab18SMatthew Ahrens dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp) 10033b2aab18SMatthew Ahrens { 10043b2aab18SMatthew Ahrens spa_t *spa; 10053b2aab18SMatthew Ahrens int error; 10063b2aab18SMatthew Ahrens 10073b2aab18SMatthew Ahrens error = spa_open(name, &spa, tag); 10083b2aab18SMatthew Ahrens if (error == 0) { 10093b2aab18SMatthew Ahrens *dp = spa_get_dsl(spa); 10103b2aab18SMatthew Ahrens dsl_pool_config_enter(*dp, tag); 10113b2aab18SMatthew Ahrens } 10123b2aab18SMatthew Ahrens return (error); 10133b2aab18SMatthew Ahrens } 10143b2aab18SMatthew Ahrens 10153b2aab18SMatthew Ahrens void 10163b2aab18SMatthew Ahrens dsl_pool_rele(dsl_pool_t *dp, void *tag) 10173b2aab18SMatthew Ahrens { 10183b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, tag); 10193b2aab18SMatthew Ahrens spa_close(dp->dp_spa, tag); 10203b2aab18SMatthew Ahrens } 10213b2aab18SMatthew Ahrens 10223b2aab18SMatthew Ahrens void 10233b2aab18SMatthew Ahrens dsl_pool_config_enter(dsl_pool_t *dp, void *tag) 10243b2aab18SMatthew Ahrens { 10253b2aab18SMatthew Ahrens /* 10263b2aab18SMatthew Ahrens * We use a "reentrant" reader-writer lock, but not reentrantly. 10273b2aab18SMatthew Ahrens * 10283b2aab18SMatthew Ahrens * The rrwlock can (with the track_all flag) track all reading threads, 10293b2aab18SMatthew Ahrens * which is very useful for debugging which code path failed to release 10303b2aab18SMatthew Ahrens * the lock, and for verifying that the *current* thread does hold 10313b2aab18SMatthew Ahrens * the lock. 10323b2aab18SMatthew Ahrens * 10333b2aab18SMatthew Ahrens * (Unlike a rwlock, which knows that N threads hold it for 10343b2aab18SMatthew Ahrens * read, but not *which* threads, so rw_held(RW_READER) returns TRUE 10353b2aab18SMatthew Ahrens * if any thread holds it for read, even if this thread doesn't). 10363b2aab18SMatthew Ahrens */ 10373b2aab18SMatthew Ahrens ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); 10383b2aab18SMatthew Ahrens rrw_enter(&dp->dp_config_rwlock, RW_READER, tag); 10393b2aab18SMatthew Ahrens } 10403b2aab18SMatthew Ahrens 10413b2aab18SMatthew Ahrens void 10423b2aab18SMatthew Ahrens dsl_pool_config_exit(dsl_pool_t *dp, void *tag) 10433b2aab18SMatthew Ahrens { 10443b2aab18SMatthew Ahrens rrw_exit(&dp->dp_config_rwlock, tag); 10453b2aab18SMatthew Ahrens } 10463b2aab18SMatthew Ahrens 10473b2aab18SMatthew Ahrens boolean_t 10483b2aab18SMatthew Ahrens dsl_pool_config_held(dsl_pool_t *dp) 10493b2aab18SMatthew Ahrens { 10503b2aab18SMatthew Ahrens return (RRW_LOCK_HELD(&dp->dp_config_rwlock)); 10513b2aab18SMatthew Ahrens } 1052