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. 23*2a104a52SAlex Reece * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 24a7a845e4SSteven Hartland * Copyright (c) 2013 Steven Hartland. All rights reserved. 25fa9e4066Sahrens */ 26fa9e4066Sahrens 27fa9e4066Sahrens #include <sys/dsl_pool.h> 28fa9e4066Sahrens #include <sys/dsl_dataset.h> 293f9d6ad7SLin Ling #include <sys/dsl_prop.h> 30fa9e4066Sahrens #include <sys/dsl_dir.h> 311d452cf5Sahrens #include <sys/dsl_synctask.h> 323f9d6ad7SLin Ling #include <sys/dsl_scan.h> 333f9d6ad7SLin Ling #include <sys/dnode.h> 34fa9e4066Sahrens #include <sys/dmu_tx.h> 35fa9e4066Sahrens #include <sys/dmu_objset.h> 36fa9e4066Sahrens #include <sys/arc.h> 37fa9e4066Sahrens #include <sys/zap.h> 38c717a561Smaybee #include <sys/zio.h> 39fa9e4066Sahrens #include <sys/zfs_context.h> 40fa9e4066Sahrens #include <sys/fs/zfs.h> 41088f3894Sahrens #include <sys/zfs_znode.h> 42088f3894Sahrens #include <sys/spa_impl.h> 43cde58dbcSMatthew Ahrens #include <sys/dsl_deadlist.h> 44ad135b5dSChristopher Siden #include <sys/bptree.h> 45ad135b5dSChristopher Siden #include <sys/zfeature.h> 46ce636f8bSMatthew Ahrens #include <sys/zil_impl.h> 473b2aab18SMatthew Ahrens #include <sys/dsl_userhold.h> 48fa9e4066Sahrens 4969962b56SMatthew Ahrens /* 5069962b56SMatthew Ahrens * ZFS Write Throttle 5169962b56SMatthew Ahrens * ------------------ 5269962b56SMatthew Ahrens * 5369962b56SMatthew Ahrens * ZFS must limit the rate of incoming writes to the rate at which it is able 5469962b56SMatthew Ahrens * to sync data modifications to the backend storage. Throttling by too much 5569962b56SMatthew Ahrens * creates an artificial limit; throttling by too little can only be sustained 5669962b56SMatthew Ahrens * for short periods and would lead to highly lumpy performance. On a per-pool 5769962b56SMatthew Ahrens * basis, ZFS tracks the amount of modified (dirty) data. As operations change 5869962b56SMatthew Ahrens * data, the amount of dirty data increases; as ZFS syncs out data, the amount 5969962b56SMatthew Ahrens * of dirty data decreases. When the amount of dirty data exceeds a 6069962b56SMatthew Ahrens * predetermined threshold further modifications are blocked until the amount 6169962b56SMatthew Ahrens * of dirty data decreases (as data is synced out). 6269962b56SMatthew Ahrens * 6369962b56SMatthew Ahrens * The limit on dirty data is tunable, and should be adjusted according to 6469962b56SMatthew Ahrens * both the IO capacity and available memory of the system. The larger the 6569962b56SMatthew Ahrens * window, the more ZFS is able to aggregate and amortize metadata (and data) 6669962b56SMatthew Ahrens * changes. However, memory is a limited resource, and allowing for more dirty 6769962b56SMatthew Ahrens * data comes at the cost of keeping other useful data in memory (for example 6869962b56SMatthew Ahrens * ZFS data cached by the ARC). 6969962b56SMatthew Ahrens * 7069962b56SMatthew Ahrens * Implementation 7169962b56SMatthew Ahrens * 7269962b56SMatthew Ahrens * As buffers are modified dsl_pool_willuse_space() increments both the per- 7369962b56SMatthew Ahrens * txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of 7469962b56SMatthew Ahrens * dirty space used; dsl_pool_dirty_space() decrements those values as data 7569962b56SMatthew Ahrens * is synced out from dsl_pool_sync(). While only the poolwide value is 7669962b56SMatthew Ahrens * relevant, the per-txg value is useful for debugging. The tunable 7769962b56SMatthew Ahrens * zfs_dirty_data_max determines the dirty space limit. Once that value is 7869962b56SMatthew Ahrens * exceeded, new writes are halted until space frees up. 7969962b56SMatthew Ahrens * 8069962b56SMatthew Ahrens * The zfs_dirty_data_sync tunable dictates the threshold at which we 8169962b56SMatthew Ahrens * ensure that there is a txg syncing (see the comment in txg.c for a full 8269962b56SMatthew Ahrens * description of transaction group stages). 8369962b56SMatthew Ahrens * 8469962b56SMatthew Ahrens * The IO scheduler uses both the dirty space limit and current amount of 8569962b56SMatthew Ahrens * dirty data as inputs. Those values affect the number of concurrent IOs ZFS 8669962b56SMatthew Ahrens * issues. See the comment in vdev_queue.c for details of the IO scheduler. 8769962b56SMatthew Ahrens * 8869962b56SMatthew Ahrens * The delay is also calculated based on the amount of dirty data. See the 8969962b56SMatthew Ahrens * comment above dmu_tx_delay() for details. 9069962b56SMatthew Ahrens */ 9169962b56SMatthew Ahrens 9269962b56SMatthew Ahrens /* 9369962b56SMatthew Ahrens * zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory, 9469962b56SMatthew Ahrens * capped at zfs_dirty_data_max_max. It can also be overridden in /etc/system. 9569962b56SMatthew Ahrens */ 9669962b56SMatthew Ahrens uint64_t zfs_dirty_data_max; 9769962b56SMatthew Ahrens uint64_t zfs_dirty_data_max_max = 4ULL * 1024 * 1024 * 1024; 9869962b56SMatthew Ahrens int zfs_dirty_data_max_percent = 10; 9969962b56SMatthew Ahrens 10069962b56SMatthew Ahrens /* 10169962b56SMatthew Ahrens * If there is at least this much dirty data, push out a txg. 10269962b56SMatthew Ahrens */ 10369962b56SMatthew Ahrens uint64_t zfs_dirty_data_sync = 64 * 1024 * 1024; 10469962b56SMatthew Ahrens 10569962b56SMatthew Ahrens /* 10669962b56SMatthew Ahrens * Once there is this amount of dirty data, the dmu_tx_delay() will kick in 10769962b56SMatthew Ahrens * and delay each transaction. 10869962b56SMatthew Ahrens * This value should be >= zfs_vdev_async_write_active_max_dirty_percent. 10969962b56SMatthew Ahrens */ 11069962b56SMatthew Ahrens int zfs_delay_min_dirty_percent = 60; 11169962b56SMatthew Ahrens 11269962b56SMatthew Ahrens /* 11369962b56SMatthew Ahrens * This controls how quickly the delay approaches infinity. 11469962b56SMatthew Ahrens * Larger values cause it to delay less for a given amount of dirty data. 11569962b56SMatthew Ahrens * Therefore larger values will cause there to be more dirty data for a 11669962b56SMatthew Ahrens * given throughput. 11769962b56SMatthew Ahrens * 11869962b56SMatthew Ahrens * For the smoothest delay, this value should be about 1 billion divided 11969962b56SMatthew Ahrens * by the maximum number of operations per second. This will smoothly 12069962b56SMatthew Ahrens * handle between 10x and 1/10th this number. 12169962b56SMatthew Ahrens * 12269962b56SMatthew Ahrens * Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the 12369962b56SMatthew Ahrens * multiply in dmu_tx_delay(). 12469962b56SMatthew Ahrens */ 12569962b56SMatthew Ahrens uint64_t zfs_delay_scale = 1000 * 1000 * 1000 / 2000; 12605715f94SMark Maybee 1271ab7f2deSmaybee 12869962b56SMatthew Ahrens /* 12969962b56SMatthew Ahrens * XXX someday maybe turn these into #defines, and you have to tune it on a 13069962b56SMatthew Ahrens * per-pool basis using zfs.conf. 13169962b56SMatthew Ahrens */ 13205715f94SMark Maybee 133088f3894Sahrens 1340689f76cSAdam Leventhal hrtime_t zfs_throttle_delay = MSEC2NSEC(10); 1350689f76cSAdam Leventhal hrtime_t zfs_throttle_resolution = MSEC2NSEC(10); 1360689f76cSAdam Leventhal 1373f9d6ad7SLin Ling int 138088f3894Sahrens dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) 139fa9e4066Sahrens { 140fa9e4066Sahrens uint64_t obj; 141fa9e4066Sahrens int err; 142fa9e4066Sahrens 143fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, 144fa9e4066Sahrens dp->dp_root_dir->dd_phys->dd_child_dir_zapobj, 145088f3894Sahrens name, sizeof (obj), 1, &obj); 146ea8dc4b6Seschrock if (err) 147ea8dc4b6Seschrock return (err); 148fa9e4066Sahrens 1493b2aab18SMatthew Ahrens return (dsl_dir_hold_obj(dp, obj, name, dp, ddp)); 150fa9e4066Sahrens } 151fa9e4066Sahrens 152fa9e4066Sahrens static dsl_pool_t * 153fa9e4066Sahrens dsl_pool_open_impl(spa_t *spa, uint64_t txg) 154fa9e4066Sahrens { 155fa9e4066Sahrens dsl_pool_t *dp; 156fa9e4066Sahrens blkptr_t *bp = spa_get_rootblkptr(spa); 157fa9e4066Sahrens 158fa9e4066Sahrens dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); 159fa9e4066Sahrens dp->dp_spa = spa; 160fa9e4066Sahrens dp->dp_meta_rootbp = *bp; 1613b2aab18SMatthew Ahrens rrw_init(&dp->dp_config_rwlock, B_TRUE); 162fa9e4066Sahrens txg_init(dp, txg); 163fa9e4066Sahrens 164fa9e4066Sahrens txg_list_create(&dp->dp_dirty_datasets, 165fa9e4066Sahrens offsetof(dsl_dataset_t, ds_dirty_link)); 166ce636f8bSMatthew Ahrens txg_list_create(&dp->dp_dirty_zilogs, 167ce636f8bSMatthew Ahrens offsetof(zilog_t, zl_dirty_link)); 168fa9e4066Sahrens txg_list_create(&dp->dp_dirty_dirs, 169fa9e4066Sahrens offsetof(dsl_dir_t, dd_dirty_link)); 1701d452cf5Sahrens txg_list_create(&dp->dp_sync_tasks, 1713b2aab18SMatthew Ahrens offsetof(dsl_sync_task_t, dst_node)); 172fa9e4066Sahrens 1731ab7f2deSmaybee mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); 17469962b56SMatthew Ahrens cv_init(&dp->dp_spaceavail_cv, NULL, CV_DEFAULT, NULL); 1751ab7f2deSmaybee 1769d3574bfSNeil Perrin dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri, 1779d3574bfSNeil Perrin 1, 4, 0); 1789d3574bfSNeil Perrin 179fa9e4066Sahrens return (dp); 180fa9e4066Sahrens } 181fa9e4066Sahrens 182ea8dc4b6Seschrock int 183ad135b5dSChristopher Siden dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) 184fa9e4066Sahrens { 185fa9e4066Sahrens int err; 186fa9e4066Sahrens dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 187ad135b5dSChristopher Siden 188ad135b5dSChristopher Siden err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, 189ad135b5dSChristopher Siden &dp->dp_meta_objset); 190ad135b5dSChristopher Siden if (err != 0) 191ad135b5dSChristopher Siden dsl_pool_close(dp); 192ad135b5dSChristopher Siden else 193ad135b5dSChristopher Siden *dpp = dp; 194ad135b5dSChristopher Siden 195ad135b5dSChristopher Siden return (err); 196ad135b5dSChristopher Siden } 197ad135b5dSChristopher Siden 198ad135b5dSChristopher Siden int 199ad135b5dSChristopher Siden dsl_pool_open(dsl_pool_t *dp) 200ad135b5dSChristopher Siden { 201ad135b5dSChristopher Siden int err; 202088f3894Sahrens dsl_dir_t *dd; 203088f3894Sahrens dsl_dataset_t *ds; 204cde58dbcSMatthew Ahrens uint64_t obj; 205fa9e4066Sahrens 2063b2aab18SMatthew Ahrens rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 207fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 208fa9e4066Sahrens DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, 209fa9e4066Sahrens &dp->dp_root_dir_obj); 210ea8dc4b6Seschrock if (err) 211ea8dc4b6Seschrock goto out; 212ea8dc4b6Seschrock 2133b2aab18SMatthew Ahrens err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, 214ea8dc4b6Seschrock NULL, dp, &dp->dp_root_dir); 215ea8dc4b6Seschrock if (err) 216ea8dc4b6Seschrock goto out; 217fa9e4066Sahrens 218088f3894Sahrens err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); 219ea8dc4b6Seschrock if (err) 220ea8dc4b6Seschrock goto out; 221ea8dc4b6Seschrock 222ad135b5dSChristopher Siden if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { 223088f3894Sahrens err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); 224088f3894Sahrens if (err) 225088f3894Sahrens goto out; 226088f3894Sahrens err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj, 227088f3894Sahrens FTAG, &ds); 2288f63aa46SLin Ling if (err == 0) { 2298f63aa46SLin Ling err = dsl_dataset_hold_obj(dp, 2308f63aa46SLin Ling ds->ds_phys->ds_prev_snap_obj, dp, 2318f63aa46SLin Ling &dp->dp_origin_snap); 2328f63aa46SLin Ling dsl_dataset_rele(ds, FTAG); 2338f63aa46SLin Ling } 2343b2aab18SMatthew Ahrens dsl_dir_rele(dd, dp); 235088f3894Sahrens if (err) 236088f3894Sahrens goto out; 237088f3894Sahrens } 238088f3894Sahrens 239ad135b5dSChristopher Siden if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { 240cde58dbcSMatthew Ahrens err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, 241cde58dbcSMatthew Ahrens &dp->dp_free_dir); 242cde58dbcSMatthew Ahrens if (err) 243cde58dbcSMatthew Ahrens goto out; 244cde58dbcSMatthew Ahrens 245cde58dbcSMatthew Ahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 246cde58dbcSMatthew Ahrens DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); 247cde58dbcSMatthew Ahrens if (err) 248cde58dbcSMatthew Ahrens goto out; 2493b2aab18SMatthew Ahrens VERIFY0(bpobj_open(&dp->dp_free_bpobj, 250cde58dbcSMatthew Ahrens dp->dp_meta_objset, obj)); 251cde58dbcSMatthew Ahrens } 252cde58dbcSMatthew Ahrens 2537fd05ac4SMatthew Ahrens /* 2547fd05ac4SMatthew Ahrens * Note: errors ignored, because the leak dir will not exist if we 2557fd05ac4SMatthew Ahrens * have not encountered a leak yet. 2567fd05ac4SMatthew Ahrens */ 2577fd05ac4SMatthew Ahrens (void) dsl_pool_open_special_dir(dp, LEAK_DIR_NAME, 2587fd05ac4SMatthew Ahrens &dp->dp_leak_dir); 2597fd05ac4SMatthew Ahrens 2602acef22dSMatthew Ahrens if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) { 261ad135b5dSChristopher Siden err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 262ad135b5dSChristopher Siden DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, 263ad135b5dSChristopher Siden &dp->dp_bptree_obj); 264ad135b5dSChristopher Siden if (err != 0) 265ad135b5dSChristopher Siden goto out; 266ad135b5dSChristopher Siden } 267ad135b5dSChristopher Siden 2682acef22dSMatthew Ahrens if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMPTY_BPOBJ)) { 269f1745736SMatthew Ahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 270f1745736SMatthew Ahrens DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, 271f1745736SMatthew Ahrens &dp->dp_empty_bpobj); 272f1745736SMatthew Ahrens if (err != 0) 273f1745736SMatthew Ahrens goto out; 274f1745736SMatthew Ahrens } 275f1745736SMatthew Ahrens 276ca45db41SChris Kirby err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 277ca45db41SChris Kirby DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, 278ca45db41SChris Kirby &dp->dp_tmp_userrefs_obj); 279ca45db41SChris Kirby if (err == ENOENT) 280ca45db41SChris Kirby err = 0; 281ca45db41SChris Kirby if (err) 282ca45db41SChris Kirby goto out; 283ca45db41SChris Kirby 284ad135b5dSChristopher Siden err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); 285088f3894Sahrens 286ea8dc4b6Seschrock out: 2873b2aab18SMatthew Ahrens rrw_exit(&dp->dp_config_rwlock, FTAG); 288ea8dc4b6Seschrock return (err); 289fa9e4066Sahrens } 290fa9e4066Sahrens 291fa9e4066Sahrens void 292fa9e4066Sahrens dsl_pool_close(dsl_pool_t *dp) 293fa9e4066Sahrens { 294088f3894Sahrens /* 29569962b56SMatthew Ahrens * Drop our references from dsl_pool_open(). 29669962b56SMatthew Ahrens * 297088f3894Sahrens * Since we held the origin_snap from "syncing" context (which 298088f3894Sahrens * includes pool-opening context), it actually only got a "ref" 299088f3894Sahrens * and not a hold, so just drop that here. 300088f3894Sahrens */ 301088f3894Sahrens if (dp->dp_origin_snap) 3023b2aab18SMatthew Ahrens dsl_dataset_rele(dp->dp_origin_snap, dp); 303ea8dc4b6Seschrock if (dp->dp_mos_dir) 3043b2aab18SMatthew Ahrens dsl_dir_rele(dp->dp_mos_dir, dp); 305cde58dbcSMatthew Ahrens if (dp->dp_free_dir) 3063b2aab18SMatthew Ahrens dsl_dir_rele(dp->dp_free_dir, dp); 3077fd05ac4SMatthew Ahrens if (dp->dp_leak_dir) 3087fd05ac4SMatthew Ahrens dsl_dir_rele(dp->dp_leak_dir, dp); 309ea8dc4b6Seschrock if (dp->dp_root_dir) 3103b2aab18SMatthew Ahrens dsl_dir_rele(dp->dp_root_dir, dp); 311fa9e4066Sahrens 312cde58dbcSMatthew Ahrens bpobj_close(&dp->dp_free_bpobj); 313cde58dbcSMatthew Ahrens 314fa9e4066Sahrens /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ 315ea8dc4b6Seschrock if (dp->dp_meta_objset) 316503ad85cSMatthew Ahrens dmu_objset_evict(dp->dp_meta_objset); 317fa9e4066Sahrens 318fa9e4066Sahrens txg_list_destroy(&dp->dp_dirty_datasets); 319ce636f8bSMatthew Ahrens txg_list_destroy(&dp->dp_dirty_zilogs); 32054a91118SChris Kirby txg_list_destroy(&dp->dp_sync_tasks); 321fa9e4066Sahrens txg_list_destroy(&dp->dp_dirty_dirs); 322fa9e4066Sahrens 323874395d5Smaybee arc_flush(dp->dp_spa); 324fa9e4066Sahrens txg_fini(dp); 3253f9d6ad7SLin Ling dsl_scan_fini(dp); 3263b2aab18SMatthew Ahrens rrw_destroy(&dp->dp_config_rwlock); 3271ab7f2deSmaybee mutex_destroy(&dp->dp_lock); 3289d3574bfSNeil Perrin taskq_destroy(dp->dp_vnrele_taskq); 32988b7b0f2SMatthew Ahrens if (dp->dp_blkstats) 33088b7b0f2SMatthew Ahrens kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); 331fa9e4066Sahrens kmem_free(dp, sizeof (dsl_pool_t)); 332fa9e4066Sahrens } 333fa9e4066Sahrens 334fa9e4066Sahrens dsl_pool_t * 3350a48a24eStimh dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) 336fa9e4066Sahrens { 337fa9e4066Sahrens int err; 338fa9e4066Sahrens dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 339fa9e4066Sahrens dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); 340503ad85cSMatthew Ahrens objset_t *os; 341088f3894Sahrens dsl_dataset_t *ds; 342cde58dbcSMatthew Ahrens uint64_t obj; 343088f3894Sahrens 3443b2aab18SMatthew Ahrens rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 3453b2aab18SMatthew Ahrens 346088f3894Sahrens /* create and open the MOS (meta-objset) */ 347503ad85cSMatthew Ahrens dp->dp_meta_objset = dmu_objset_create_impl(spa, 348503ad85cSMatthew Ahrens NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); 349fa9e4066Sahrens 350fa9e4066Sahrens /* create the pool directory */ 351fa9e4066Sahrens err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 352fa9e4066Sahrens DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); 353fb09f5aaSMadhav Suresh ASSERT0(err); 354fa9e4066Sahrens 3553f9d6ad7SLin Ling /* Initialize scan structures */ 3563b2aab18SMatthew Ahrens VERIFY0(dsl_scan_init(dp, txg)); 3573f9d6ad7SLin Ling 358fa9e4066Sahrens /* create and open the root dir */ 359088f3894Sahrens dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); 3603b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, 361ea8dc4b6Seschrock NULL, dp, &dp->dp_root_dir)); 362fa9e4066Sahrens 363fa9e4066Sahrens /* create and open the meta-objset dir */ 364088f3894Sahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); 3653b2aab18SMatthew Ahrens VERIFY0(dsl_pool_open_special_dir(dp, 366088f3894Sahrens MOS_DIR_NAME, &dp->dp_mos_dir)); 367088f3894Sahrens 368cde58dbcSMatthew Ahrens if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { 369cde58dbcSMatthew Ahrens /* create and open the free dir */ 370cde58dbcSMatthew Ahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, 371cde58dbcSMatthew Ahrens FREE_DIR_NAME, tx); 3723b2aab18SMatthew Ahrens VERIFY0(dsl_pool_open_special_dir(dp, 373cde58dbcSMatthew Ahrens FREE_DIR_NAME, &dp->dp_free_dir)); 374cde58dbcSMatthew Ahrens 375cde58dbcSMatthew Ahrens /* create and open the free_bplist */ 376cde58dbcSMatthew Ahrens obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx); 377cde58dbcSMatthew Ahrens VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 378cde58dbcSMatthew Ahrens DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); 3793b2aab18SMatthew Ahrens VERIFY0(bpobj_open(&dp->dp_free_bpobj, 380cde58dbcSMatthew Ahrens dp->dp_meta_objset, obj)); 381cde58dbcSMatthew Ahrens } 382cde58dbcSMatthew Ahrens 383088f3894Sahrens if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) 384088f3894Sahrens dsl_pool_create_origin(dp, tx); 385088f3894Sahrens 386088f3894Sahrens /* create the root dataset */ 387cde58dbcSMatthew Ahrens obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); 388088f3894Sahrens 389088f3894Sahrens /* create the root objset */ 3903b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); 391503ad85cSMatthew Ahrens os = dmu_objset_create_impl(dp->dp_spa, ds, 392088f3894Sahrens dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx); 393088f3894Sahrens #ifdef _KERNEL 394503ad85cSMatthew Ahrens zfs_create_fs(os, kcred, zplprops, tx); 395088f3894Sahrens #endif 396088f3894Sahrens dsl_dataset_rele(ds, FTAG); 397fa9e4066Sahrens 398fa9e4066Sahrens dmu_tx_commit(tx); 399fa9e4066Sahrens 4003b2aab18SMatthew Ahrens rrw_exit(&dp->dp_config_rwlock, FTAG); 4013b2aab18SMatthew Ahrens 402fa9e4066Sahrens return (dp); 403fa9e4066Sahrens } 404fa9e4066Sahrens 405ce636f8bSMatthew Ahrens /* 406ce636f8bSMatthew Ahrens * Account for the meta-objset space in its placeholder dsl_dir. 407ce636f8bSMatthew Ahrens */ 408ce636f8bSMatthew Ahrens void 409ce636f8bSMatthew Ahrens dsl_pool_mos_diduse_space(dsl_pool_t *dp, 410ce636f8bSMatthew Ahrens int64_t used, int64_t comp, int64_t uncomp) 411ce636f8bSMatthew Ahrens { 412ce636f8bSMatthew Ahrens ASSERT3U(comp, ==, uncomp); /* it's all metadata */ 413ce636f8bSMatthew Ahrens mutex_enter(&dp->dp_lock); 414ce636f8bSMatthew Ahrens dp->dp_mos_used_delta += used; 415ce636f8bSMatthew Ahrens dp->dp_mos_compressed_delta += comp; 416ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta += uncomp; 417ce636f8bSMatthew Ahrens mutex_exit(&dp->dp_lock); 418ce636f8bSMatthew Ahrens } 419ce636f8bSMatthew Ahrens 420cde58dbcSMatthew Ahrens static int 421cde58dbcSMatthew Ahrens deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 422cde58dbcSMatthew Ahrens { 423cde58dbcSMatthew Ahrens dsl_deadlist_t *dl = arg; 424cde58dbcSMatthew Ahrens dsl_deadlist_insert(dl, bp, tx); 425cde58dbcSMatthew Ahrens return (0); 426cde58dbcSMatthew Ahrens } 427cde58dbcSMatthew Ahrens 42869962b56SMatthew Ahrens static void 42969962b56SMatthew Ahrens dsl_pool_sync_mos(dsl_pool_t *dp, dmu_tx_t *tx) 43069962b56SMatthew Ahrens { 43169962b56SMatthew Ahrens zio_t *zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 43269962b56SMatthew Ahrens dmu_objset_sync(dp->dp_meta_objset, zio, tx); 43369962b56SMatthew Ahrens VERIFY0(zio_wait(zio)); 43469962b56SMatthew Ahrens dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); 43569962b56SMatthew Ahrens spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); 43669962b56SMatthew Ahrens } 43769962b56SMatthew Ahrens 43869962b56SMatthew Ahrens static void 43969962b56SMatthew Ahrens dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta) 44069962b56SMatthew Ahrens { 44169962b56SMatthew Ahrens ASSERT(MUTEX_HELD(&dp->dp_lock)); 44269962b56SMatthew Ahrens 44369962b56SMatthew Ahrens if (delta < 0) 44469962b56SMatthew Ahrens ASSERT3U(-delta, <=, dp->dp_dirty_total); 44569962b56SMatthew Ahrens 44669962b56SMatthew Ahrens dp->dp_dirty_total += delta; 44769962b56SMatthew Ahrens 44869962b56SMatthew Ahrens /* 44969962b56SMatthew Ahrens * Note: we signal even when increasing dp_dirty_total. 45069962b56SMatthew Ahrens * This ensures forward progress -- each thread wakes the next waiter. 45169962b56SMatthew Ahrens */ 45269962b56SMatthew Ahrens if (dp->dp_dirty_total <= zfs_dirty_data_max) 45369962b56SMatthew Ahrens cv_signal(&dp->dp_spaceavail_cv); 45469962b56SMatthew Ahrens } 45569962b56SMatthew Ahrens 456fa9e4066Sahrens void 457fa9e4066Sahrens dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) 458fa9e4066Sahrens { 459c717a561Smaybee zio_t *zio; 460fa9e4066Sahrens dmu_tx_t *tx; 461c717a561Smaybee dsl_dir_t *dd; 462c717a561Smaybee dsl_dataset_t *ds; 463503ad85cSMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 464ce636f8bSMatthew Ahrens list_t synced_datasets; 465ce636f8bSMatthew Ahrens 466ce636f8bSMatthew Ahrens list_create(&synced_datasets, sizeof (dsl_dataset_t), 467ce636f8bSMatthew Ahrens offsetof(dsl_dataset_t, ds_synced_link)); 468fa9e4066Sahrens 469fa9e4066Sahrens tx = dmu_tx_create_assigned(dp, txg); 470fa9e4066Sahrens 47169962b56SMatthew Ahrens /* 47269962b56SMatthew Ahrens * Write out all dirty blocks of dirty datasets. 47369962b56SMatthew Ahrens */ 474c717a561Smaybee zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 47569962b56SMatthew Ahrens while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 47614843421SMatthew Ahrens /* 47714843421SMatthew Ahrens * We must not sync any non-MOS datasets twice, because 47814843421SMatthew Ahrens * we may have taken a snapshot of them. However, we 47914843421SMatthew Ahrens * may sync newly-created datasets on pass 2. 48014843421SMatthew Ahrens */ 48114843421SMatthew Ahrens ASSERT(!list_link_active(&ds->ds_synced_link)); 482ce636f8bSMatthew Ahrens list_insert_tail(&synced_datasets, ds); 483c717a561Smaybee dsl_dataset_sync(ds, zio, tx); 484c717a561Smaybee } 48569962b56SMatthew Ahrens VERIFY0(zio_wait(zio)); 48614843421SMatthew Ahrens 48769962b56SMatthew Ahrens /* 48869962b56SMatthew Ahrens * We have written all of the accounted dirty data, so our 48969962b56SMatthew Ahrens * dp_space_towrite should now be zero. However, some seldom-used 49069962b56SMatthew Ahrens * code paths do not adhere to this (e.g. dbuf_undirty(), also 49169962b56SMatthew Ahrens * rounding error in dbuf_write_physdone). 49269962b56SMatthew Ahrens * Shore up the accounting of any dirtied space now. 49369962b56SMatthew Ahrens */ 49469962b56SMatthew Ahrens dsl_pool_undirty_space(dp, dp->dp_dirty_pertxg[txg & TXG_MASK], txg); 495c717a561Smaybee 496ce636f8bSMatthew Ahrens /* 497ce636f8bSMatthew Ahrens * After the data blocks have been written (ensured by the zio_wait() 498ce636f8bSMatthew Ahrens * above), update the user/group space accounting. 499ce636f8bSMatthew Ahrens */ 50069962b56SMatthew Ahrens for (ds = list_head(&synced_datasets); ds != NULL; 50169962b56SMatthew Ahrens ds = list_next(&synced_datasets, ds)) { 5020a586ceaSMark Shellenbaum dmu_objset_do_userquota_updates(ds->ds_objset, tx); 50369962b56SMatthew Ahrens } 50414843421SMatthew Ahrens 50514843421SMatthew Ahrens /* 50614843421SMatthew Ahrens * Sync the datasets again to push out the changes due to 5073f9d6ad7SLin Ling * userspace updates. This must be done before we process the 508ce636f8bSMatthew Ahrens * sync tasks, so that any snapshots will have the correct 509ce636f8bSMatthew Ahrens * user accounting information (and we won't get confused 510ce636f8bSMatthew Ahrens * about which blocks are part of the snapshot). 51114843421SMatthew Ahrens */ 51214843421SMatthew Ahrens zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 51369962b56SMatthew Ahrens while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 51414843421SMatthew Ahrens ASSERT(list_link_active(&ds->ds_synced_link)); 51514843421SMatthew Ahrens dmu_buf_rele(ds->ds_dbuf, ds); 51614843421SMatthew Ahrens dsl_dataset_sync(ds, zio, tx); 51714843421SMatthew Ahrens } 51869962b56SMatthew Ahrens VERIFY0(zio_wait(zio)); 51914843421SMatthew Ahrens 520b24ab676SJeff Bonwick /* 521ce636f8bSMatthew Ahrens * Now that the datasets have been completely synced, we can 522ce636f8bSMatthew Ahrens * clean up our in-memory structures accumulated while syncing: 523ce636f8bSMatthew Ahrens * 524ce636f8bSMatthew Ahrens * - move dead blocks from the pending deadlist to the on-disk deadlist 525ce636f8bSMatthew Ahrens * - release hold from dsl_dataset_dirty() 526b24ab676SJeff Bonwick */ 52769962b56SMatthew Ahrens while ((ds = list_remove_head(&synced_datasets)) != NULL) { 528ce636f8bSMatthew Ahrens objset_t *os = ds->ds_objset; 529cde58dbcSMatthew Ahrens bplist_iterate(&ds->ds_pending_deadlist, 530cde58dbcSMatthew Ahrens deadlist_enqueue_cb, &ds->ds_deadlist, tx); 531ce636f8bSMatthew Ahrens ASSERT(!dmu_objset_is_dirty(os, txg)); 532ce636f8bSMatthew Ahrens dmu_buf_rele(ds->ds_dbuf, ds); 533cde58dbcSMatthew Ahrens } 53469962b56SMatthew Ahrens while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) != NULL) { 535c717a561Smaybee dsl_dir_sync(dd, tx); 53669962b56SMatthew Ahrens } 537fa9e4066Sahrens 538ce636f8bSMatthew Ahrens /* 539ce636f8bSMatthew Ahrens * The MOS's space is accounted for in the pool/$MOS 540ce636f8bSMatthew Ahrens * (dp_mos_dir). We can't modify the mos while we're syncing 541ce636f8bSMatthew Ahrens * it, so we remember the deltas and apply them here. 542ce636f8bSMatthew Ahrens */ 543ce636f8bSMatthew Ahrens if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || 544ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta != 0) { 545ce636f8bSMatthew Ahrens dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, 546ce636f8bSMatthew Ahrens dp->dp_mos_used_delta, 547ce636f8bSMatthew Ahrens dp->dp_mos_compressed_delta, 548ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta, tx); 549ce636f8bSMatthew Ahrens dp->dp_mos_used_delta = 0; 550ce636f8bSMatthew Ahrens dp->dp_mos_compressed_delta = 0; 551ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta = 0; 552ce636f8bSMatthew Ahrens } 553ce636f8bSMatthew Ahrens 554503ad85cSMatthew Ahrens if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || 555503ad85cSMatthew Ahrens list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { 55669962b56SMatthew Ahrens dsl_pool_sync_mos(dp, tx); 557fa9e4066Sahrens } 558fa9e4066Sahrens 559ce636f8bSMatthew Ahrens /* 560ce636f8bSMatthew Ahrens * If we modify a dataset in the same txg that we want to destroy it, 561ce636f8bSMatthew Ahrens * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. 562ce636f8bSMatthew Ahrens * dsl_dir_destroy_check() will fail if there are unexpected holds. 563ce636f8bSMatthew Ahrens * Therefore, we want to sync the MOS (thus syncing the dd_dbuf 564ce636f8bSMatthew Ahrens * and clearing the hold on it) before we process the sync_tasks. 565ce636f8bSMatthew Ahrens * The MOS data dirtied by the sync_tasks will be synced on the next 566ce636f8bSMatthew Ahrens * pass. 567ce636f8bSMatthew Ahrens */ 568ce636f8bSMatthew Ahrens if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { 5693b2aab18SMatthew Ahrens dsl_sync_task_t *dst; 570ce636f8bSMatthew Ahrens /* 571ce636f8bSMatthew Ahrens * No more sync tasks should have been added while we 572ce636f8bSMatthew Ahrens * were syncing. 573ce636f8bSMatthew Ahrens */ 57469962b56SMatthew Ahrens ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1); 57569962b56SMatthew Ahrens while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)) != NULL) 5763b2aab18SMatthew Ahrens dsl_sync_task_sync(dst, tx); 577ce636f8bSMatthew Ahrens } 578ce636f8bSMatthew Ahrens 579fa9e4066Sahrens dmu_tx_commit(tx); 58005715f94SMark Maybee 58169962b56SMatthew Ahrens DTRACE_PROBE2(dsl_pool_sync__done, dsl_pool_t *dp, dp, uint64_t, txg); 582fa9e4066Sahrens } 583fa9e4066Sahrens 584fa9e4066Sahrens void 585b24ab676SJeff Bonwick dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) 586fa9e4066Sahrens { 587ce636f8bSMatthew Ahrens zilog_t *zilog; 588fa9e4066Sahrens 589ce636f8bSMatthew Ahrens while (zilog = txg_list_remove(&dp->dp_dirty_zilogs, txg)) { 59069962b56SMatthew Ahrens dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); 591ce636f8bSMatthew Ahrens zil_clean(zilog, txg); 592ce636f8bSMatthew Ahrens ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); 593ce636f8bSMatthew Ahrens dmu_buf_rele(ds->ds_dbuf, zilog); 594fa9e4066Sahrens } 595b24ab676SJeff Bonwick ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); 596fa9e4066Sahrens } 597fa9e4066Sahrens 598c717a561Smaybee /* 599c717a561Smaybee * TRUE if the current thread is the tx_sync_thread or if we 600c717a561Smaybee * are being called from SPA context during pool initialization. 601c717a561Smaybee */ 602fa9e4066Sahrens int 603fa9e4066Sahrens dsl_pool_sync_context(dsl_pool_t *dp) 604fa9e4066Sahrens { 605fa9e4066Sahrens return (curthread == dp->dp_tx.tx_sync_thread || 606ad135b5dSChristopher Siden spa_is_initializing(dp->dp_spa)); 607fa9e4066Sahrens } 608fa9e4066Sahrens 609fa9e4066Sahrens uint64_t 610fa9e4066Sahrens dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) 611fa9e4066Sahrens { 612fa9e4066Sahrens uint64_t space, resv; 613fa9e4066Sahrens 614fa9e4066Sahrens /* 615fa9e4066Sahrens * If we're trying to assess whether it's OK to do a free, 616fa9e4066Sahrens * cut the reservation in half to allow forward progress 617fa9e4066Sahrens * (e.g. make it possible to rm(1) files from a full pool). 618fa9e4066Sahrens */ 619485bbbf5SGeorge Wilson space = spa_get_dspace(dp->dp_spa); 620c39f2c8cSChristopher Siden resv = spa_get_slop_space(dp->dp_spa); 621fa9e4066Sahrens if (netfree) 622fa9e4066Sahrens resv >>= 1; 623fa9e4066Sahrens 624fa9e4066Sahrens return (space - resv); 625fa9e4066Sahrens } 6261ab7f2deSmaybee 62769962b56SMatthew Ahrens boolean_t 62869962b56SMatthew Ahrens dsl_pool_need_dirty_delay(dsl_pool_t *dp) 6291ab7f2deSmaybee { 63069962b56SMatthew Ahrens uint64_t delay_min_bytes = 63169962b56SMatthew Ahrens zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100; 63269962b56SMatthew Ahrens boolean_t rv; 6331ab7f2deSmaybee 63469962b56SMatthew Ahrens mutex_enter(&dp->dp_lock); 63569962b56SMatthew Ahrens if (dp->dp_dirty_total > zfs_dirty_data_sync) 63669962b56SMatthew Ahrens txg_kick(dp); 63769962b56SMatthew Ahrens rv = (dp->dp_dirty_total > delay_min_bytes); 63869962b56SMatthew Ahrens mutex_exit(&dp->dp_lock); 63969962b56SMatthew Ahrens return (rv); 6401ab7f2deSmaybee } 6411ab7f2deSmaybee 6421ab7f2deSmaybee void 64369962b56SMatthew Ahrens dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 6441ab7f2deSmaybee { 64569962b56SMatthew Ahrens if (space > 0) { 64669962b56SMatthew Ahrens mutex_enter(&dp->dp_lock); 64769962b56SMatthew Ahrens dp->dp_dirty_pertxg[tx->tx_txg & TXG_MASK] += space; 64869962b56SMatthew Ahrens dsl_pool_dirty_delta(dp, space); 64969962b56SMatthew Ahrens mutex_exit(&dp->dp_lock); 6501ab7f2deSmaybee } 6511ab7f2deSmaybee } 6521ab7f2deSmaybee 6531ab7f2deSmaybee void 65469962b56SMatthew Ahrens dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg) 6551ab7f2deSmaybee { 65669962b56SMatthew Ahrens ASSERT3S(space, >=, 0); 65769962b56SMatthew Ahrens if (space == 0) 65869962b56SMatthew Ahrens return; 65969962b56SMatthew Ahrens mutex_enter(&dp->dp_lock); 66069962b56SMatthew Ahrens if (dp->dp_dirty_pertxg[txg & TXG_MASK] < space) { 66169962b56SMatthew Ahrens /* XXX writing something we didn't dirty? */ 66269962b56SMatthew Ahrens space = dp->dp_dirty_pertxg[txg & TXG_MASK]; 6631ab7f2deSmaybee } 66469962b56SMatthew Ahrens ASSERT3U(dp->dp_dirty_pertxg[txg & TXG_MASK], >=, space); 66569962b56SMatthew Ahrens dp->dp_dirty_pertxg[txg & TXG_MASK] -= space; 66669962b56SMatthew Ahrens ASSERT3U(dp->dp_dirty_total, >=, space); 66769962b56SMatthew Ahrens dsl_pool_dirty_delta(dp, -space); 66869962b56SMatthew Ahrens mutex_exit(&dp->dp_lock); 6691ab7f2deSmaybee } 670088f3894Sahrens 671088f3894Sahrens /* ARGSUSED */ 672088f3894Sahrens static int 6733b2aab18SMatthew Ahrens upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) 674088f3894Sahrens { 675088f3894Sahrens dmu_tx_t *tx = arg; 676088f3894Sahrens dsl_dataset_t *ds, *prev = NULL; 677088f3894Sahrens int err; 678088f3894Sahrens 6793b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); 680088f3894Sahrens if (err) 681088f3894Sahrens return (err); 682088f3894Sahrens 683088f3894Sahrens while (ds->ds_phys->ds_prev_snap_obj != 0) { 684088f3894Sahrens err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, 685088f3894Sahrens FTAG, &prev); 686088f3894Sahrens if (err) { 687088f3894Sahrens dsl_dataset_rele(ds, FTAG); 688088f3894Sahrens return (err); 689088f3894Sahrens } 690088f3894Sahrens 691088f3894Sahrens if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) 692088f3894Sahrens break; 693088f3894Sahrens dsl_dataset_rele(ds, FTAG); 694088f3894Sahrens ds = prev; 695088f3894Sahrens prev = NULL; 696088f3894Sahrens } 697088f3894Sahrens 698088f3894Sahrens if (prev == NULL) { 699088f3894Sahrens prev = dp->dp_origin_snap; 700088f3894Sahrens 701088f3894Sahrens /* 702088f3894Sahrens * The $ORIGIN can't have any data, or the accounting 703088f3894Sahrens * will be wrong. 704088f3894Sahrens */ 7053b2aab18SMatthew Ahrens ASSERT0(prev->ds_phys->ds_bp.blk_birth); 706088f3894Sahrens 707088f3894Sahrens /* The origin doesn't get attached to itself */ 708088f3894Sahrens if (ds->ds_object == prev->ds_object) { 709088f3894Sahrens dsl_dataset_rele(ds, FTAG); 710088f3894Sahrens return (0); 711088f3894Sahrens } 712088f3894Sahrens 713088f3894Sahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 714088f3894Sahrens ds->ds_phys->ds_prev_snap_obj = prev->ds_object; 715088f3894Sahrens ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg; 716088f3894Sahrens 717088f3894Sahrens dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 718088f3894Sahrens ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object; 719088f3894Sahrens 720088f3894Sahrens dmu_buf_will_dirty(prev->ds_dbuf, tx); 721088f3894Sahrens prev->ds_phys->ds_num_children++; 722088f3894Sahrens 723088f3894Sahrens if (ds->ds_phys->ds_next_snap_obj == 0) { 724088f3894Sahrens ASSERT(ds->ds_prev == NULL); 7253b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, 726088f3894Sahrens ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); 727088f3894Sahrens } 728088f3894Sahrens } 729088f3894Sahrens 7303b2aab18SMatthew Ahrens ASSERT3U(ds->ds_dir->dd_phys->dd_origin_obj, ==, prev->ds_object); 7313b2aab18SMatthew Ahrens ASSERT3U(ds->ds_phys->ds_prev_snap_obj, ==, prev->ds_object); 732088f3894Sahrens 733088f3894Sahrens if (prev->ds_phys->ds_next_clones_obj == 0) { 734c33e334fSMatthew Ahrens dmu_buf_will_dirty(prev->ds_dbuf, tx); 735088f3894Sahrens prev->ds_phys->ds_next_clones_obj = 736088f3894Sahrens zap_create(dp->dp_meta_objset, 737088f3894Sahrens DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 738088f3894Sahrens } 7393b2aab18SMatthew Ahrens VERIFY0(zap_add_int(dp->dp_meta_objset, 740088f3894Sahrens prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx)); 741088f3894Sahrens 742088f3894Sahrens dsl_dataset_rele(ds, FTAG); 743088f3894Sahrens if (prev != dp->dp_origin_snap) 744088f3894Sahrens dsl_dataset_rele(prev, FTAG); 745088f3894Sahrens return (0); 746088f3894Sahrens } 747088f3894Sahrens 748088f3894Sahrens void 749088f3894Sahrens dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) 750088f3894Sahrens { 751088f3894Sahrens ASSERT(dmu_tx_is_syncing(tx)); 752088f3894Sahrens ASSERT(dp->dp_origin_snap != NULL); 753088f3894Sahrens 7543b2aab18SMatthew Ahrens VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb, 755c33e334fSMatthew Ahrens tx, DS_FIND_CHILDREN)); 756088f3894Sahrens } 757088f3894Sahrens 758cde58dbcSMatthew Ahrens /* ARGSUSED */ 759cde58dbcSMatthew Ahrens static int 7603b2aab18SMatthew Ahrens upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 761cde58dbcSMatthew Ahrens { 762cde58dbcSMatthew Ahrens dmu_tx_t *tx = arg; 763cde58dbcSMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 764cde58dbcSMatthew Ahrens 7653b2aab18SMatthew Ahrens if (ds->ds_dir->dd_phys->dd_origin_obj != 0) { 766cde58dbcSMatthew Ahrens dsl_dataset_t *origin; 767cde58dbcSMatthew Ahrens 7683b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, 769cde58dbcSMatthew Ahrens ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin)); 770cde58dbcSMatthew Ahrens 771cde58dbcSMatthew Ahrens if (origin->ds_dir->dd_phys->dd_clones == 0) { 772cde58dbcSMatthew Ahrens dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 773cde58dbcSMatthew Ahrens origin->ds_dir->dd_phys->dd_clones = zap_create(mos, 774cde58dbcSMatthew Ahrens DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); 775cde58dbcSMatthew Ahrens } 776cde58dbcSMatthew Ahrens 7773b2aab18SMatthew Ahrens VERIFY0(zap_add_int(dp->dp_meta_objset, 7783b2aab18SMatthew Ahrens origin->ds_dir->dd_phys->dd_clones, ds->ds_object, tx)); 779cde58dbcSMatthew Ahrens 780cde58dbcSMatthew Ahrens dsl_dataset_rele(origin, FTAG); 781cde58dbcSMatthew Ahrens } 782cde58dbcSMatthew Ahrens return (0); 783cde58dbcSMatthew Ahrens } 784cde58dbcSMatthew Ahrens 785cde58dbcSMatthew Ahrens void 786cde58dbcSMatthew Ahrens dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) 787cde58dbcSMatthew Ahrens { 788cde58dbcSMatthew Ahrens ASSERT(dmu_tx_is_syncing(tx)); 789cde58dbcSMatthew Ahrens uint64_t obj; 790cde58dbcSMatthew Ahrens 791cde58dbcSMatthew Ahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); 7923b2aab18SMatthew Ahrens VERIFY0(dsl_pool_open_special_dir(dp, 793cde58dbcSMatthew Ahrens FREE_DIR_NAME, &dp->dp_free_dir)); 794cde58dbcSMatthew Ahrens 795cde58dbcSMatthew Ahrens /* 796cde58dbcSMatthew Ahrens * We can't use bpobj_alloc(), because spa_version() still 797cde58dbcSMatthew Ahrens * returns the old version, and we need a new-version bpobj with 798cde58dbcSMatthew Ahrens * subobj support. So call dmu_object_alloc() directly. 799cde58dbcSMatthew Ahrens */ 800cde58dbcSMatthew Ahrens obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, 801cde58dbcSMatthew Ahrens SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); 8023b2aab18SMatthew Ahrens VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 803cde58dbcSMatthew Ahrens DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); 8043b2aab18SMatthew Ahrens VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); 805cde58dbcSMatthew Ahrens 8063b2aab18SMatthew Ahrens VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 807cde58dbcSMatthew Ahrens upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN)); 808cde58dbcSMatthew Ahrens } 809cde58dbcSMatthew Ahrens 810088f3894Sahrens void 811088f3894Sahrens dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) 812088f3894Sahrens { 813088f3894Sahrens uint64_t dsobj; 814088f3894Sahrens dsl_dataset_t *ds; 815088f3894Sahrens 816088f3894Sahrens ASSERT(dmu_tx_is_syncing(tx)); 817088f3894Sahrens ASSERT(dp->dp_origin_snap == NULL); 8183b2aab18SMatthew Ahrens ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER)); 819088f3894Sahrens 820088f3894Sahrens /* create the origin dir, ds, & snap-ds */ 821088f3894Sahrens dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, 822088f3894Sahrens NULL, 0, kcred, tx); 8233b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 8243b2aab18SMatthew Ahrens dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx); 8253b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, 826088f3894Sahrens dp, &dp->dp_origin_snap)); 827088f3894Sahrens dsl_dataset_rele(ds, FTAG); 828088f3894Sahrens } 8299d3574bfSNeil Perrin 8309d3574bfSNeil Perrin taskq_t * 8319d3574bfSNeil Perrin dsl_pool_vnrele_taskq(dsl_pool_t *dp) 8329d3574bfSNeil Perrin { 8339d3574bfSNeil Perrin return (dp->dp_vnrele_taskq); 8349d3574bfSNeil Perrin } 835ca45db41SChris Kirby 836ca45db41SChris Kirby /* 837ca45db41SChris Kirby * Walk through the pool-wide zap object of temporary snapshot user holds 838ca45db41SChris Kirby * and release them. 839ca45db41SChris Kirby */ 840ca45db41SChris Kirby void 841ca45db41SChris Kirby dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) 842ca45db41SChris Kirby { 843ca45db41SChris Kirby zap_attribute_t za; 844ca45db41SChris Kirby zap_cursor_t zc; 845ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 846ca45db41SChris Kirby uint64_t zapobj = dp->dp_tmp_userrefs_obj; 847a7a845e4SSteven Hartland nvlist_t *holds; 848ca45db41SChris Kirby 849ca45db41SChris Kirby if (zapobj == 0) 850ca45db41SChris Kirby return; 851ca45db41SChris Kirby ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 852ca45db41SChris Kirby 853a7a845e4SSteven Hartland holds = fnvlist_alloc(); 854a7a845e4SSteven Hartland 855ca45db41SChris Kirby for (zap_cursor_init(&zc, mos, zapobj); 856ca45db41SChris Kirby zap_cursor_retrieve(&zc, &za) == 0; 857ca45db41SChris Kirby zap_cursor_advance(&zc)) { 858ca45db41SChris Kirby char *htag; 859a7a845e4SSteven Hartland nvlist_t *tags; 860ca45db41SChris Kirby 861ca45db41SChris Kirby htag = strchr(za.za_name, '-'); 862ca45db41SChris Kirby *htag = '\0'; 863ca45db41SChris Kirby ++htag; 864a7a845e4SSteven Hartland if (nvlist_lookup_nvlist(holds, za.za_name, &tags) != 0) { 865a7a845e4SSteven Hartland tags = fnvlist_alloc(); 866a7a845e4SSteven Hartland fnvlist_add_boolean(tags, htag); 867a7a845e4SSteven Hartland fnvlist_add_nvlist(holds, za.za_name, tags); 868a7a845e4SSteven Hartland fnvlist_free(tags); 869a7a845e4SSteven Hartland } else { 870a7a845e4SSteven Hartland fnvlist_add_boolean(tags, htag); 871a7a845e4SSteven Hartland } 872ca45db41SChris Kirby } 873a7a845e4SSteven Hartland dsl_dataset_user_release_tmp(dp, holds); 874a7a845e4SSteven Hartland fnvlist_free(holds); 875ca45db41SChris Kirby zap_cursor_fini(&zc); 876ca45db41SChris Kirby } 877ca45db41SChris Kirby 878ca45db41SChris Kirby /* 879ca45db41SChris Kirby * Create the pool-wide zap object for storing temporary snapshot holds. 880ca45db41SChris Kirby */ 881ca45db41SChris Kirby void 882ca45db41SChris Kirby dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) 883ca45db41SChris Kirby { 884ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 885ca45db41SChris Kirby 886ca45db41SChris Kirby ASSERT(dp->dp_tmp_userrefs_obj == 0); 887ca45db41SChris Kirby ASSERT(dmu_tx_is_syncing(tx)); 888ca45db41SChris Kirby 889ad135b5dSChristopher Siden dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, 890ad135b5dSChristopher Siden DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); 891ca45db41SChris Kirby } 892ca45db41SChris Kirby 893ca45db41SChris Kirby static int 894ca45db41SChris Kirby dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, 8953b2aab18SMatthew Ahrens const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding) 896ca45db41SChris Kirby { 897ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 898ca45db41SChris Kirby uint64_t zapobj = dp->dp_tmp_userrefs_obj; 899ca45db41SChris Kirby char *name; 900ca45db41SChris Kirby int error; 901ca45db41SChris Kirby 902ca45db41SChris Kirby ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 903ca45db41SChris Kirby ASSERT(dmu_tx_is_syncing(tx)); 904ca45db41SChris Kirby 905ca45db41SChris Kirby /* 906ca45db41SChris Kirby * If the pool was created prior to SPA_VERSION_USERREFS, the 907ca45db41SChris Kirby * zap object for temporary holds might not exist yet. 908ca45db41SChris Kirby */ 909ca45db41SChris Kirby if (zapobj == 0) { 910ca45db41SChris Kirby if (holding) { 911ca45db41SChris Kirby dsl_pool_user_hold_create_obj(dp, tx); 912ca45db41SChris Kirby zapobj = dp->dp_tmp_userrefs_obj; 913ca45db41SChris Kirby } else { 914be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 915ca45db41SChris Kirby } 916ca45db41SChris Kirby } 917ca45db41SChris Kirby 918ca45db41SChris Kirby name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); 919ca45db41SChris Kirby if (holding) 9203b2aab18SMatthew Ahrens error = zap_add(mos, zapobj, name, 8, 1, &now, tx); 921ca45db41SChris Kirby else 922ca45db41SChris Kirby error = zap_remove(mos, zapobj, name, tx); 923ca45db41SChris Kirby strfree(name); 924ca45db41SChris Kirby 925ca45db41SChris Kirby return (error); 926ca45db41SChris Kirby } 927ca45db41SChris Kirby 928ca45db41SChris Kirby /* 929ca45db41SChris Kirby * Add a temporary hold for the given dataset object and tag. 930ca45db41SChris Kirby */ 931ca45db41SChris Kirby int 932ca45db41SChris Kirby dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 9333b2aab18SMatthew Ahrens uint64_t now, dmu_tx_t *tx) 934ca45db41SChris Kirby { 93515508ac0SChris Kirby return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); 936ca45db41SChris Kirby } 937ca45db41SChris Kirby 938ca45db41SChris Kirby /* 939ca45db41SChris Kirby * Release a temporary hold for the given dataset object and tag. 940ca45db41SChris Kirby */ 941ca45db41SChris Kirby int 942ca45db41SChris Kirby dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 943ca45db41SChris Kirby dmu_tx_t *tx) 944ca45db41SChris Kirby { 945ca45db41SChris Kirby return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL, 946ca45db41SChris Kirby tx, B_FALSE)); 947ca45db41SChris Kirby } 9483b2aab18SMatthew Ahrens 9493b2aab18SMatthew Ahrens /* 9503b2aab18SMatthew Ahrens * DSL Pool Configuration Lock 9513b2aab18SMatthew Ahrens * 9523b2aab18SMatthew Ahrens * The dp_config_rwlock protects against changes to DSL state (e.g. dataset 9533b2aab18SMatthew Ahrens * creation / destruction / rename / property setting). It must be held for 9543b2aab18SMatthew Ahrens * read to hold a dataset or dsl_dir. I.e. you must call 9553b2aab18SMatthew Ahrens * dsl_pool_config_enter() or dsl_pool_hold() before calling 9563b2aab18SMatthew Ahrens * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock 9573b2aab18SMatthew Ahrens * must be held continuously until all datasets and dsl_dirs are released. 9583b2aab18SMatthew Ahrens * 9593b2aab18SMatthew Ahrens * The only exception to this rule is that if a "long hold" is placed on 9603b2aab18SMatthew Ahrens * a dataset, then the dp_config_rwlock may be dropped while the dataset 9613b2aab18SMatthew Ahrens * is still held. The long hold will prevent the dataset from being 9623b2aab18SMatthew Ahrens * destroyed -- the destroy will fail with EBUSY. A long hold can be 9633b2aab18SMatthew Ahrens * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset 9643b2aab18SMatthew Ahrens * (by calling dsl_{dataset,objset}_{try}own{_obj}). 9653b2aab18SMatthew Ahrens * 9663b2aab18SMatthew Ahrens * Legitimate long-holders (including owners) should be long-running, cancelable 9673b2aab18SMatthew Ahrens * tasks that should cause "zfs destroy" to fail. This includes DMU 9683b2aab18SMatthew Ahrens * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open), 9693b2aab18SMatthew Ahrens * "zfs send", and "zfs diff". There are several other long-holders whose 9703b2aab18SMatthew Ahrens * uses are suboptimal (e.g. "zfs promote", and zil_suspend()). 9713b2aab18SMatthew Ahrens * 9723b2aab18SMatthew Ahrens * The usual formula for long-holding would be: 9733b2aab18SMatthew Ahrens * dsl_pool_hold() 9743b2aab18SMatthew Ahrens * dsl_dataset_hold() 9753b2aab18SMatthew Ahrens * ... perform checks ... 9763b2aab18SMatthew Ahrens * dsl_dataset_long_hold() 9773b2aab18SMatthew Ahrens * dsl_pool_rele() 9783b2aab18SMatthew Ahrens * ... perform long-running task ... 9793b2aab18SMatthew Ahrens * dsl_dataset_long_rele() 9803b2aab18SMatthew Ahrens * dsl_dataset_rele() 9813b2aab18SMatthew Ahrens * 9823b2aab18SMatthew Ahrens * Note that when the long hold is released, the dataset is still held but 9833b2aab18SMatthew Ahrens * the pool is not held. The dataset may change arbitrarily during this time 9843b2aab18SMatthew Ahrens * (e.g. it could be destroyed). Therefore you shouldn't do anything to the 9853b2aab18SMatthew Ahrens * dataset except release it. 9863b2aab18SMatthew Ahrens * 9873b2aab18SMatthew Ahrens * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only 9883b2aab18SMatthew Ahrens * or modifying operations. 9893b2aab18SMatthew Ahrens * 9903b2aab18SMatthew Ahrens * Modifying operations should generally use dsl_sync_task(). The synctask 9913b2aab18SMatthew Ahrens * infrastructure enforces proper locking strategy with respect to the 9923b2aab18SMatthew Ahrens * dp_config_rwlock. See the comment above dsl_sync_task() for details. 9933b2aab18SMatthew Ahrens * 9943b2aab18SMatthew Ahrens * Read-only operations will manually hold the pool, then the dataset, obtain 9953b2aab18SMatthew Ahrens * information from the dataset, then release the pool and dataset. 9963b2aab18SMatthew Ahrens * dmu_objset_{hold,rele}() are convenience routines that also do the pool 9973b2aab18SMatthew Ahrens * hold/rele. 9983b2aab18SMatthew Ahrens */ 9993b2aab18SMatthew Ahrens 10003b2aab18SMatthew Ahrens int 10013b2aab18SMatthew Ahrens dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp) 10023b2aab18SMatthew Ahrens { 10033b2aab18SMatthew Ahrens spa_t *spa; 10043b2aab18SMatthew Ahrens int error; 10053b2aab18SMatthew Ahrens 10063b2aab18SMatthew Ahrens error = spa_open(name, &spa, tag); 10073b2aab18SMatthew Ahrens if (error == 0) { 10083b2aab18SMatthew Ahrens *dp = spa_get_dsl(spa); 10093b2aab18SMatthew Ahrens dsl_pool_config_enter(*dp, tag); 10103b2aab18SMatthew Ahrens } 10113b2aab18SMatthew Ahrens return (error); 10123b2aab18SMatthew Ahrens } 10133b2aab18SMatthew Ahrens 10143b2aab18SMatthew Ahrens void 10153b2aab18SMatthew Ahrens dsl_pool_rele(dsl_pool_t *dp, void *tag) 10163b2aab18SMatthew Ahrens { 10173b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, tag); 10183b2aab18SMatthew Ahrens spa_close(dp->dp_spa, tag); 10193b2aab18SMatthew Ahrens } 10203b2aab18SMatthew Ahrens 10213b2aab18SMatthew Ahrens void 10223b2aab18SMatthew Ahrens dsl_pool_config_enter(dsl_pool_t *dp, void *tag) 10233b2aab18SMatthew Ahrens { 10243b2aab18SMatthew Ahrens /* 10253b2aab18SMatthew Ahrens * We use a "reentrant" reader-writer lock, but not reentrantly. 10263b2aab18SMatthew Ahrens * 10273b2aab18SMatthew Ahrens * The rrwlock can (with the track_all flag) track all reading threads, 10283b2aab18SMatthew Ahrens * which is very useful for debugging which code path failed to release 10293b2aab18SMatthew Ahrens * the lock, and for verifying that the *current* thread does hold 10303b2aab18SMatthew Ahrens * the lock. 10313b2aab18SMatthew Ahrens * 10323b2aab18SMatthew Ahrens * (Unlike a rwlock, which knows that N threads hold it for 10333b2aab18SMatthew Ahrens * read, but not *which* threads, so rw_held(RW_READER) returns TRUE 10343b2aab18SMatthew Ahrens * if any thread holds it for read, even if this thread doesn't). 10353b2aab18SMatthew Ahrens */ 10363b2aab18SMatthew Ahrens ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); 10373b2aab18SMatthew Ahrens rrw_enter(&dp->dp_config_rwlock, RW_READER, tag); 10383b2aab18SMatthew Ahrens } 10393b2aab18SMatthew Ahrens 10403b2aab18SMatthew Ahrens void 10413b2aab18SMatthew Ahrens dsl_pool_config_exit(dsl_pool_t *dp, void *tag) 10423b2aab18SMatthew Ahrens { 10433b2aab18SMatthew Ahrens rrw_exit(&dp->dp_config_rwlock, tag); 10443b2aab18SMatthew Ahrens } 10453b2aab18SMatthew Ahrens 10463b2aab18SMatthew Ahrens boolean_t 10473b2aab18SMatthew Ahrens dsl_pool_config_held(dsl_pool_t *dp) 10483b2aab18SMatthew Ahrens { 10493b2aab18SMatthew Ahrens return (RRW_LOCK_HELD(&dp->dp_config_rwlock)); 10503b2aab18SMatthew Ahrens } 1051