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. 2394c2d0ebSMatthew Ahrens * Copyright (c) 2011, 2017 by Delphix. All rights reserved. 24a7a845e4SSteven Hartland * Copyright (c) 2013 Steven Hartland. All rights reserved. 25bc9014e6SJustin Gibbs * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 26c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 27ff5177eeSAlek Pinchuk * Copyright 2016 Nexenta Systems, Inc. All rights reserved. 28fa9e4066Sahrens */ 29fa9e4066Sahrens 30fa9e4066Sahrens #include <sys/dsl_pool.h> 31fa9e4066Sahrens #include <sys/dsl_dataset.h> 323f9d6ad7SLin Ling #include <sys/dsl_prop.h> 33fa9e4066Sahrens #include <sys/dsl_dir.h> 341d452cf5Sahrens #include <sys/dsl_synctask.h> 353f9d6ad7SLin Ling #include <sys/dsl_scan.h> 363f9d6ad7SLin Ling #include <sys/dnode.h> 37fa9e4066Sahrens #include <sys/dmu_tx.h> 38fa9e4066Sahrens #include <sys/dmu_objset.h> 39fa9e4066Sahrens #include <sys/arc.h> 40fa9e4066Sahrens #include <sys/zap.h> 41c717a561Smaybee #include <sys/zio.h> 42fa9e4066Sahrens #include <sys/zfs_context.h> 43fa9e4066Sahrens #include <sys/fs/zfs.h> 44088f3894Sahrens #include <sys/zfs_znode.h> 45088f3894Sahrens #include <sys/spa_impl.h> 46cde58dbcSMatthew Ahrens #include <sys/dsl_deadlist.h> 47ad135b5dSChristopher Siden #include <sys/bptree.h> 48ad135b5dSChristopher Siden #include <sys/zfeature.h> 49ce636f8bSMatthew Ahrens #include <sys/zil_impl.h> 503b2aab18SMatthew Ahrens #include <sys/dsl_userhold.h> 51fa9e4066Sahrens 5269962b56SMatthew Ahrens /* 5369962b56SMatthew Ahrens * ZFS Write Throttle 5469962b56SMatthew Ahrens * ------------------ 5569962b56SMatthew Ahrens * 5669962b56SMatthew Ahrens * ZFS must limit the rate of incoming writes to the rate at which it is able 5769962b56SMatthew Ahrens * to sync data modifications to the backend storage. Throttling by too much 5869962b56SMatthew Ahrens * creates an artificial limit; throttling by too little can only be sustained 5969962b56SMatthew Ahrens * for short periods and would lead to highly lumpy performance. On a per-pool 6069962b56SMatthew Ahrens * basis, ZFS tracks the amount of modified (dirty) data. As operations change 6169962b56SMatthew Ahrens * data, the amount of dirty data increases; as ZFS syncs out data, the amount 6269962b56SMatthew Ahrens * of dirty data decreases. When the amount of dirty data exceeds a 6369962b56SMatthew Ahrens * predetermined threshold further modifications are blocked until the amount 6469962b56SMatthew Ahrens * of dirty data decreases (as data is synced out). 6569962b56SMatthew Ahrens * 6669962b56SMatthew Ahrens * The limit on dirty data is tunable, and should be adjusted according to 6769962b56SMatthew Ahrens * both the IO capacity and available memory of the system. The larger the 6869962b56SMatthew Ahrens * window, the more ZFS is able to aggregate and amortize metadata (and data) 6969962b56SMatthew Ahrens * changes. However, memory is a limited resource, and allowing for more dirty 7069962b56SMatthew Ahrens * data comes at the cost of keeping other useful data in memory (for example 7169962b56SMatthew Ahrens * ZFS data cached by the ARC). 7269962b56SMatthew Ahrens * 7369962b56SMatthew Ahrens * Implementation 7469962b56SMatthew Ahrens * 7569962b56SMatthew Ahrens * As buffers are modified dsl_pool_willuse_space() increments both the per- 7669962b56SMatthew Ahrens * txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of 7769962b56SMatthew Ahrens * dirty space used; dsl_pool_dirty_space() decrements those values as data 7869962b56SMatthew Ahrens * is synced out from dsl_pool_sync(). While only the poolwide value is 7969962b56SMatthew Ahrens * relevant, the per-txg value is useful for debugging. The tunable 8069962b56SMatthew Ahrens * zfs_dirty_data_max determines the dirty space limit. Once that value is 8169962b56SMatthew Ahrens * exceeded, new writes are halted until space frees up. 8269962b56SMatthew Ahrens * 8369962b56SMatthew Ahrens * The zfs_dirty_data_sync tunable dictates the threshold at which we 8469962b56SMatthew Ahrens * ensure that there is a txg syncing (see the comment in txg.c for a full 8569962b56SMatthew Ahrens * description of transaction group stages). 8669962b56SMatthew Ahrens * 8769962b56SMatthew Ahrens * The IO scheduler uses both the dirty space limit and current amount of 8869962b56SMatthew Ahrens * dirty data as inputs. Those values affect the number of concurrent IOs ZFS 8969962b56SMatthew Ahrens * issues. See the comment in vdev_queue.c for details of the IO scheduler. 9069962b56SMatthew Ahrens * 9169962b56SMatthew Ahrens * The delay is also calculated based on the amount of dirty data. See the 9269962b56SMatthew Ahrens * comment above dmu_tx_delay() for details. 9369962b56SMatthew Ahrens */ 9469962b56SMatthew Ahrens 9569962b56SMatthew Ahrens /* 9669962b56SMatthew Ahrens * zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory, 9769962b56SMatthew Ahrens * capped at zfs_dirty_data_max_max. It can also be overridden in /etc/system. 9869962b56SMatthew Ahrens */ 9969962b56SMatthew Ahrens uint64_t zfs_dirty_data_max; 10069962b56SMatthew Ahrens uint64_t zfs_dirty_data_max_max = 4ULL * 1024 * 1024 * 1024; 10169962b56SMatthew Ahrens int zfs_dirty_data_max_percent = 10; 10269962b56SMatthew Ahrens 10369962b56SMatthew Ahrens /* 10469962b56SMatthew Ahrens * If there is at least this much dirty data, push out a txg. 10569962b56SMatthew Ahrens */ 10669962b56SMatthew Ahrens uint64_t zfs_dirty_data_sync = 64 * 1024 * 1024; 10769962b56SMatthew Ahrens 10869962b56SMatthew Ahrens /* 10969962b56SMatthew Ahrens * Once there is this amount of dirty data, the dmu_tx_delay() will kick in 11069962b56SMatthew Ahrens * and delay each transaction. 11169962b56SMatthew Ahrens * This value should be >= zfs_vdev_async_write_active_max_dirty_percent. 11269962b56SMatthew Ahrens */ 11369962b56SMatthew Ahrens int zfs_delay_min_dirty_percent = 60; 11469962b56SMatthew Ahrens 11569962b56SMatthew Ahrens /* 11669962b56SMatthew Ahrens * This controls how quickly the delay approaches infinity. 117d85a1e96SMatthew Ahrens * Larger values cause it to delay more for a given amount of dirty data. 118d85a1e96SMatthew Ahrens * Therefore larger values will cause there to be less dirty data for a 11969962b56SMatthew Ahrens * given throughput. 12069962b56SMatthew Ahrens * 12169962b56SMatthew Ahrens * For the smoothest delay, this value should be about 1 billion divided 12269962b56SMatthew Ahrens * by the maximum number of operations per second. This will smoothly 12369962b56SMatthew Ahrens * handle between 10x and 1/10th this number. 12469962b56SMatthew Ahrens * 12569962b56SMatthew Ahrens * Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the 12669962b56SMatthew Ahrens * multiply in dmu_tx_delay(). 12769962b56SMatthew Ahrens */ 12869962b56SMatthew Ahrens uint64_t zfs_delay_scale = 1000 * 1000 * 1000 / 2000; 12905715f94SMark Maybee 13094c2d0ebSMatthew Ahrens /* 13194c2d0ebSMatthew Ahrens * This determines the number of threads used by the dp_sync_taskq. 13294c2d0ebSMatthew Ahrens */ 13394c2d0ebSMatthew Ahrens int zfs_sync_taskq_batch_pct = 75; 1341ab7f2deSmaybee 1353f9d6ad7SLin Ling int 136088f3894Sahrens dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) 137fa9e4066Sahrens { 138fa9e4066Sahrens uint64_t obj; 139fa9e4066Sahrens int err; 140fa9e4066Sahrens 141fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, 142c1379625SJustin T. Gibbs dsl_dir_phys(dp->dp_root_dir)->dd_child_dir_zapobj, 143088f3894Sahrens name, sizeof (obj), 1, &obj); 144ea8dc4b6Seschrock if (err) 145ea8dc4b6Seschrock return (err); 146fa9e4066Sahrens 1473b2aab18SMatthew Ahrens return (dsl_dir_hold_obj(dp, obj, name, dp, ddp)); 148fa9e4066Sahrens } 149fa9e4066Sahrens 150fa9e4066Sahrens static dsl_pool_t * 151fa9e4066Sahrens dsl_pool_open_impl(spa_t *spa, uint64_t txg) 152fa9e4066Sahrens { 153fa9e4066Sahrens dsl_pool_t *dp; 154fa9e4066Sahrens blkptr_t *bp = spa_get_rootblkptr(spa); 155fa9e4066Sahrens 156fa9e4066Sahrens dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); 157fa9e4066Sahrens dp->dp_spa = spa; 158fa9e4066Sahrens dp->dp_meta_rootbp = *bp; 1593b2aab18SMatthew Ahrens rrw_init(&dp->dp_config_rwlock, B_TRUE); 160fa9e4066Sahrens txg_init(dp, txg); 161fa9e4066Sahrens 162*b7b2590dSMatthew Ahrens txg_list_create(&dp->dp_dirty_datasets, spa, 163fa9e4066Sahrens offsetof(dsl_dataset_t, ds_dirty_link)); 164*b7b2590dSMatthew Ahrens txg_list_create(&dp->dp_dirty_zilogs, spa, 165ce636f8bSMatthew Ahrens offsetof(zilog_t, zl_dirty_link)); 166*b7b2590dSMatthew Ahrens txg_list_create(&dp->dp_dirty_dirs, spa, 167fa9e4066Sahrens offsetof(dsl_dir_t, dd_dirty_link)); 168*b7b2590dSMatthew Ahrens txg_list_create(&dp->dp_sync_tasks, spa, 1693b2aab18SMatthew Ahrens offsetof(dsl_sync_task_t, dst_node)); 170fa9e4066Sahrens 17194c2d0ebSMatthew Ahrens dp->dp_sync_taskq = taskq_create("dp_sync_taskq", 17294c2d0ebSMatthew Ahrens zfs_sync_taskq_batch_pct, minclsyspri, 1, INT_MAX, 17394c2d0ebSMatthew Ahrens TASKQ_THREADS_CPU_PCT); 17494c2d0ebSMatthew Ahrens 1751ab7f2deSmaybee mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); 17669962b56SMatthew Ahrens cv_init(&dp->dp_spaceavail_cv, NULL, CV_DEFAULT, NULL); 1771ab7f2deSmaybee 1789d3574bfSNeil Perrin dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri, 1799d3574bfSNeil Perrin 1, 4, 0); 1809d3574bfSNeil Perrin 181fa9e4066Sahrens return (dp); 182fa9e4066Sahrens } 183fa9e4066Sahrens 184ea8dc4b6Seschrock int 185ad135b5dSChristopher Siden dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) 186fa9e4066Sahrens { 187fa9e4066Sahrens int err; 188fa9e4066Sahrens dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 189ad135b5dSChristopher Siden 190ad135b5dSChristopher Siden err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, 191ad135b5dSChristopher Siden &dp->dp_meta_objset); 192ad135b5dSChristopher Siden if (err != 0) 193ad135b5dSChristopher Siden dsl_pool_close(dp); 194ad135b5dSChristopher Siden else 195ad135b5dSChristopher Siden *dpp = dp; 196ad135b5dSChristopher Siden 197ad135b5dSChristopher Siden return (err); 198ad135b5dSChristopher Siden } 199ad135b5dSChristopher Siden 200ad135b5dSChristopher Siden int 201ad135b5dSChristopher Siden dsl_pool_open(dsl_pool_t *dp) 202ad135b5dSChristopher Siden { 203ad135b5dSChristopher Siden int err; 204088f3894Sahrens dsl_dir_t *dd; 205088f3894Sahrens dsl_dataset_t *ds; 206cde58dbcSMatthew Ahrens uint64_t obj; 207fa9e4066Sahrens 2083b2aab18SMatthew Ahrens rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 209fa9e4066Sahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 210fa9e4066Sahrens DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, 211fa9e4066Sahrens &dp->dp_root_dir_obj); 212ea8dc4b6Seschrock if (err) 213ea8dc4b6Seschrock goto out; 214ea8dc4b6Seschrock 2153b2aab18SMatthew Ahrens err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, 216ea8dc4b6Seschrock NULL, dp, &dp->dp_root_dir); 217ea8dc4b6Seschrock if (err) 218ea8dc4b6Seschrock goto out; 219fa9e4066Sahrens 220088f3894Sahrens err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); 221ea8dc4b6Seschrock if (err) 222ea8dc4b6Seschrock goto out; 223ea8dc4b6Seschrock 224ad135b5dSChristopher Siden if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { 225088f3894Sahrens err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); 226088f3894Sahrens if (err) 227088f3894Sahrens goto out; 228c1379625SJustin T. Gibbs err = dsl_dataset_hold_obj(dp, 229c1379625SJustin T. Gibbs dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds); 2308f63aa46SLin Ling if (err == 0) { 2318f63aa46SLin Ling err = dsl_dataset_hold_obj(dp, 232c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_obj, dp, 2338f63aa46SLin Ling &dp->dp_origin_snap); 2348f63aa46SLin Ling dsl_dataset_rele(ds, FTAG); 2358f63aa46SLin Ling } 2363b2aab18SMatthew Ahrens dsl_dir_rele(dd, dp); 237088f3894Sahrens if (err) 238088f3894Sahrens goto out; 239088f3894Sahrens } 240088f3894Sahrens 241ad135b5dSChristopher Siden if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { 242cde58dbcSMatthew Ahrens err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, 243cde58dbcSMatthew Ahrens &dp->dp_free_dir); 244cde58dbcSMatthew Ahrens if (err) 245cde58dbcSMatthew Ahrens goto out; 246cde58dbcSMatthew Ahrens 247cde58dbcSMatthew Ahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 248cde58dbcSMatthew Ahrens DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); 249cde58dbcSMatthew Ahrens if (err) 250cde58dbcSMatthew Ahrens goto out; 2513b2aab18SMatthew Ahrens VERIFY0(bpobj_open(&dp->dp_free_bpobj, 252cde58dbcSMatthew Ahrens dp->dp_meta_objset, obj)); 253cde58dbcSMatthew Ahrens } 254cde58dbcSMatthew Ahrens 2557fd05ac4SMatthew Ahrens /* 2567fd05ac4SMatthew Ahrens * Note: errors ignored, because the leak dir will not exist if we 2577fd05ac4SMatthew Ahrens * have not encountered a leak yet. 2587fd05ac4SMatthew Ahrens */ 2597fd05ac4SMatthew Ahrens (void) dsl_pool_open_special_dir(dp, LEAK_DIR_NAME, 2607fd05ac4SMatthew Ahrens &dp->dp_leak_dir); 2617fd05ac4SMatthew Ahrens 2622acef22dSMatthew Ahrens if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) { 263ad135b5dSChristopher Siden err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 264ad135b5dSChristopher Siden DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, 265ad135b5dSChristopher Siden &dp->dp_bptree_obj); 266ad135b5dSChristopher Siden if (err != 0) 267ad135b5dSChristopher Siden goto out; 268ad135b5dSChristopher Siden } 269ad135b5dSChristopher Siden 2702acef22dSMatthew Ahrens if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMPTY_BPOBJ)) { 271f1745736SMatthew Ahrens err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 272f1745736SMatthew Ahrens DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, 273f1745736SMatthew Ahrens &dp->dp_empty_bpobj); 274f1745736SMatthew Ahrens if (err != 0) 275f1745736SMatthew Ahrens goto out; 276f1745736SMatthew Ahrens } 277f1745736SMatthew Ahrens 278ca45db41SChris Kirby err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 279ca45db41SChris Kirby DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, 280ca45db41SChris Kirby &dp->dp_tmp_userrefs_obj); 281ca45db41SChris Kirby if (err == ENOENT) 282ca45db41SChris Kirby err = 0; 283ca45db41SChris Kirby if (err) 284ca45db41SChris Kirby goto out; 285ca45db41SChris Kirby 286ad135b5dSChristopher Siden err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); 287088f3894Sahrens 288ea8dc4b6Seschrock out: 2893b2aab18SMatthew Ahrens rrw_exit(&dp->dp_config_rwlock, FTAG); 290ea8dc4b6Seschrock return (err); 291fa9e4066Sahrens } 292fa9e4066Sahrens 293fa9e4066Sahrens void 294fa9e4066Sahrens dsl_pool_close(dsl_pool_t *dp) 295fa9e4066Sahrens { 296088f3894Sahrens /* 29769962b56SMatthew Ahrens * Drop our references from dsl_pool_open(). 29869962b56SMatthew Ahrens * 299088f3894Sahrens * Since we held the origin_snap from "syncing" context (which 300088f3894Sahrens * includes pool-opening context), it actually only got a "ref" 301088f3894Sahrens * and not a hold, so just drop that here. 302088f3894Sahrens */ 303088f3894Sahrens if (dp->dp_origin_snap) 3043b2aab18SMatthew Ahrens dsl_dataset_rele(dp->dp_origin_snap, dp); 305ea8dc4b6Seschrock if (dp->dp_mos_dir) 3063b2aab18SMatthew Ahrens dsl_dir_rele(dp->dp_mos_dir, dp); 307cde58dbcSMatthew Ahrens if (dp->dp_free_dir) 3083b2aab18SMatthew Ahrens dsl_dir_rele(dp->dp_free_dir, dp); 3097fd05ac4SMatthew Ahrens if (dp->dp_leak_dir) 3107fd05ac4SMatthew Ahrens dsl_dir_rele(dp->dp_leak_dir, dp); 311ea8dc4b6Seschrock if (dp->dp_root_dir) 3123b2aab18SMatthew Ahrens dsl_dir_rele(dp->dp_root_dir, dp); 313fa9e4066Sahrens 314cde58dbcSMatthew Ahrens bpobj_close(&dp->dp_free_bpobj); 315cde58dbcSMatthew Ahrens 316fa9e4066Sahrens /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ 317ea8dc4b6Seschrock if (dp->dp_meta_objset) 318503ad85cSMatthew Ahrens dmu_objset_evict(dp->dp_meta_objset); 319fa9e4066Sahrens 320fa9e4066Sahrens txg_list_destroy(&dp->dp_dirty_datasets); 321ce636f8bSMatthew Ahrens txg_list_destroy(&dp->dp_dirty_zilogs); 32254a91118SChris Kirby txg_list_destroy(&dp->dp_sync_tasks); 323fa9e4066Sahrens txg_list_destroy(&dp->dp_dirty_dirs); 324fa9e4066Sahrens 32594c2d0ebSMatthew Ahrens taskq_destroy(dp->dp_sync_taskq); 32694c2d0ebSMatthew Ahrens 327244781f1SPrakash Surya /* 328244781f1SPrakash Surya * We can't set retry to TRUE since we're explicitly specifying 329244781f1SPrakash Surya * a spa to flush. This is good enough; any missed buffers for 330244781f1SPrakash Surya * this spa won't cause trouble, and they'll eventually fall 331244781f1SPrakash Surya * out of the ARC just like any other unused buffer. 332244781f1SPrakash Surya */ 333244781f1SPrakash Surya arc_flush(dp->dp_spa, FALSE); 334244781f1SPrakash Surya 335fa9e4066Sahrens txg_fini(dp); 3363f9d6ad7SLin Ling dsl_scan_fini(dp); 337bc9014e6SJustin Gibbs dmu_buf_user_evict_wait(); 338bc9014e6SJustin Gibbs 3393b2aab18SMatthew Ahrens rrw_destroy(&dp->dp_config_rwlock); 3401ab7f2deSmaybee mutex_destroy(&dp->dp_lock); 3419d3574bfSNeil Perrin taskq_destroy(dp->dp_vnrele_taskq); 34288b7b0f2SMatthew Ahrens if (dp->dp_blkstats) 34388b7b0f2SMatthew Ahrens kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); 344fa9e4066Sahrens kmem_free(dp, sizeof (dsl_pool_t)); 345fa9e4066Sahrens } 346fa9e4066Sahrens 347fa9e4066Sahrens dsl_pool_t * 3480a48a24eStimh dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) 349fa9e4066Sahrens { 350fa9e4066Sahrens int err; 351fa9e4066Sahrens dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); 352fa9e4066Sahrens dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); 353503ad85cSMatthew Ahrens objset_t *os; 354088f3894Sahrens dsl_dataset_t *ds; 355cde58dbcSMatthew Ahrens uint64_t obj; 356088f3894Sahrens 3573b2aab18SMatthew Ahrens rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 3583b2aab18SMatthew Ahrens 359088f3894Sahrens /* create and open the MOS (meta-objset) */ 360503ad85cSMatthew Ahrens dp->dp_meta_objset = dmu_objset_create_impl(spa, 361503ad85cSMatthew Ahrens NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); 362fa9e4066Sahrens 363fa9e4066Sahrens /* create the pool directory */ 364fa9e4066Sahrens err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 365fa9e4066Sahrens DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); 366fb09f5aaSMadhav Suresh ASSERT0(err); 367fa9e4066Sahrens 3683f9d6ad7SLin Ling /* Initialize scan structures */ 3693b2aab18SMatthew Ahrens VERIFY0(dsl_scan_init(dp, txg)); 3703f9d6ad7SLin Ling 371fa9e4066Sahrens /* create and open the root dir */ 372088f3894Sahrens dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); 3733b2aab18SMatthew Ahrens VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, 374ea8dc4b6Seschrock NULL, dp, &dp->dp_root_dir)); 375fa9e4066Sahrens 376fa9e4066Sahrens /* create and open the meta-objset dir */ 377088f3894Sahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); 3783b2aab18SMatthew Ahrens VERIFY0(dsl_pool_open_special_dir(dp, 379088f3894Sahrens MOS_DIR_NAME, &dp->dp_mos_dir)); 380088f3894Sahrens 381cde58dbcSMatthew Ahrens if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { 382cde58dbcSMatthew Ahrens /* create and open the free dir */ 383cde58dbcSMatthew Ahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, 384cde58dbcSMatthew Ahrens FREE_DIR_NAME, tx); 3853b2aab18SMatthew Ahrens VERIFY0(dsl_pool_open_special_dir(dp, 386cde58dbcSMatthew Ahrens FREE_DIR_NAME, &dp->dp_free_dir)); 387cde58dbcSMatthew Ahrens 388cde58dbcSMatthew Ahrens /* create and open the free_bplist */ 389b5152584SMatthew Ahrens obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx); 390cde58dbcSMatthew Ahrens VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 391cde58dbcSMatthew Ahrens DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); 3923b2aab18SMatthew Ahrens VERIFY0(bpobj_open(&dp->dp_free_bpobj, 393cde58dbcSMatthew Ahrens dp->dp_meta_objset, obj)); 394cde58dbcSMatthew Ahrens } 395cde58dbcSMatthew Ahrens 396088f3894Sahrens if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) 397088f3894Sahrens dsl_pool_create_origin(dp, tx); 398088f3894Sahrens 399088f3894Sahrens /* create the root dataset */ 400cde58dbcSMatthew Ahrens obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); 401088f3894Sahrens 402088f3894Sahrens /* create the root objset */ 4033b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); 404c166b69dSPaul Dagnelie rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 405503ad85cSMatthew Ahrens os = dmu_objset_create_impl(dp->dp_spa, ds, 406088f3894Sahrens dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx); 407c166b69dSPaul Dagnelie rrw_exit(&ds->ds_bp_rwlock, FTAG); 408088f3894Sahrens #ifdef _KERNEL 409503ad85cSMatthew Ahrens zfs_create_fs(os, kcred, zplprops, tx); 410088f3894Sahrens #endif 411088f3894Sahrens dsl_dataset_rele(ds, FTAG); 412fa9e4066Sahrens 413fa9e4066Sahrens dmu_tx_commit(tx); 414fa9e4066Sahrens 4153b2aab18SMatthew Ahrens rrw_exit(&dp->dp_config_rwlock, FTAG); 4163b2aab18SMatthew Ahrens 417fa9e4066Sahrens return (dp); 418fa9e4066Sahrens } 419fa9e4066Sahrens 420ce636f8bSMatthew Ahrens /* 421ce636f8bSMatthew Ahrens * Account for the meta-objset space in its placeholder dsl_dir. 422ce636f8bSMatthew Ahrens */ 423ce636f8bSMatthew Ahrens void 424ce636f8bSMatthew Ahrens dsl_pool_mos_diduse_space(dsl_pool_t *dp, 425ce636f8bSMatthew Ahrens int64_t used, int64_t comp, int64_t uncomp) 426ce636f8bSMatthew Ahrens { 427ce636f8bSMatthew Ahrens ASSERT3U(comp, ==, uncomp); /* it's all metadata */ 428ce636f8bSMatthew Ahrens mutex_enter(&dp->dp_lock); 429ce636f8bSMatthew Ahrens dp->dp_mos_used_delta += used; 430ce636f8bSMatthew Ahrens dp->dp_mos_compressed_delta += comp; 431ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta += uncomp; 432ce636f8bSMatthew Ahrens mutex_exit(&dp->dp_lock); 433ce636f8bSMatthew Ahrens } 434ce636f8bSMatthew Ahrens 43569962b56SMatthew Ahrens static void 43669962b56SMatthew Ahrens dsl_pool_sync_mos(dsl_pool_t *dp, dmu_tx_t *tx) 43769962b56SMatthew Ahrens { 43869962b56SMatthew Ahrens zio_t *zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 43969962b56SMatthew Ahrens dmu_objset_sync(dp->dp_meta_objset, zio, tx); 44069962b56SMatthew Ahrens VERIFY0(zio_wait(zio)); 44169962b56SMatthew Ahrens dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); 44269962b56SMatthew Ahrens spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); 44369962b56SMatthew Ahrens } 44469962b56SMatthew Ahrens 44569962b56SMatthew Ahrens static void 44669962b56SMatthew Ahrens dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta) 44769962b56SMatthew Ahrens { 44869962b56SMatthew Ahrens ASSERT(MUTEX_HELD(&dp->dp_lock)); 44969962b56SMatthew Ahrens 45069962b56SMatthew Ahrens if (delta < 0) 45169962b56SMatthew Ahrens ASSERT3U(-delta, <=, dp->dp_dirty_total); 45269962b56SMatthew Ahrens 45369962b56SMatthew Ahrens dp->dp_dirty_total += delta; 45469962b56SMatthew Ahrens 45569962b56SMatthew Ahrens /* 45669962b56SMatthew Ahrens * Note: we signal even when increasing dp_dirty_total. 45769962b56SMatthew Ahrens * This ensures forward progress -- each thread wakes the next waiter. 45869962b56SMatthew Ahrens */ 459313ae1e1SAndriy Gapon if (dp->dp_dirty_total < zfs_dirty_data_max) 46069962b56SMatthew Ahrens cv_signal(&dp->dp_spaceavail_cv); 46169962b56SMatthew Ahrens } 46269962b56SMatthew Ahrens 463fa9e4066Sahrens void 464fa9e4066Sahrens dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) 465fa9e4066Sahrens { 466c717a561Smaybee zio_t *zio; 467fa9e4066Sahrens dmu_tx_t *tx; 468c717a561Smaybee dsl_dir_t *dd; 469c717a561Smaybee dsl_dataset_t *ds; 470503ad85cSMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 471ce636f8bSMatthew Ahrens list_t synced_datasets; 472ce636f8bSMatthew Ahrens 473ce636f8bSMatthew Ahrens list_create(&synced_datasets, sizeof (dsl_dataset_t), 474ce636f8bSMatthew Ahrens offsetof(dsl_dataset_t, ds_synced_link)); 475fa9e4066Sahrens 476fa9e4066Sahrens tx = dmu_tx_create_assigned(dp, txg); 477fa9e4066Sahrens 47869962b56SMatthew Ahrens /* 47969962b56SMatthew Ahrens * Write out all dirty blocks of dirty datasets. 48069962b56SMatthew Ahrens */ 481c717a561Smaybee zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 48269962b56SMatthew Ahrens while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 48314843421SMatthew Ahrens /* 48414843421SMatthew Ahrens * We must not sync any non-MOS datasets twice, because 48514843421SMatthew Ahrens * we may have taken a snapshot of them. However, we 48614843421SMatthew Ahrens * may sync newly-created datasets on pass 2. 48714843421SMatthew Ahrens */ 48814843421SMatthew Ahrens ASSERT(!list_link_active(&ds->ds_synced_link)); 489ce636f8bSMatthew Ahrens list_insert_tail(&synced_datasets, ds); 490c717a561Smaybee dsl_dataset_sync(ds, zio, tx); 491c717a561Smaybee } 49269962b56SMatthew Ahrens VERIFY0(zio_wait(zio)); 49314843421SMatthew Ahrens 49469962b56SMatthew Ahrens /* 49569962b56SMatthew Ahrens * We have written all of the accounted dirty data, so our 49669962b56SMatthew Ahrens * dp_space_towrite should now be zero. However, some seldom-used 49769962b56SMatthew Ahrens * code paths do not adhere to this (e.g. dbuf_undirty(), also 49869962b56SMatthew Ahrens * rounding error in dbuf_write_physdone). 49969962b56SMatthew Ahrens * Shore up the accounting of any dirtied space now. 50069962b56SMatthew Ahrens */ 50169962b56SMatthew Ahrens dsl_pool_undirty_space(dp, dp->dp_dirty_pertxg[txg & TXG_MASK], txg); 502c717a561Smaybee 503ff5177eeSAlek Pinchuk /* 504ff5177eeSAlek Pinchuk * Update the long range free counter after 505ff5177eeSAlek Pinchuk * we're done syncing user data 506ff5177eeSAlek Pinchuk */ 507ff5177eeSAlek Pinchuk mutex_enter(&dp->dp_lock); 508ff5177eeSAlek Pinchuk ASSERT(spa_sync_pass(dp->dp_spa) == 1 || 509ff5177eeSAlek Pinchuk dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] == 0); 510ff5177eeSAlek Pinchuk dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] = 0; 511ff5177eeSAlek Pinchuk mutex_exit(&dp->dp_lock); 512ff5177eeSAlek Pinchuk 513ce636f8bSMatthew Ahrens /* 514ce636f8bSMatthew Ahrens * After the data blocks have been written (ensured by the zio_wait() 51594c2d0ebSMatthew Ahrens * above), update the user/group space accounting. This happens 51694c2d0ebSMatthew Ahrens * in tasks dispatched to dp_sync_taskq, so wait for them before 51794c2d0ebSMatthew Ahrens * continuing. 518ce636f8bSMatthew Ahrens */ 51969962b56SMatthew Ahrens for (ds = list_head(&synced_datasets); ds != NULL; 52069962b56SMatthew Ahrens ds = list_next(&synced_datasets, ds)) { 5210a586ceaSMark Shellenbaum dmu_objset_do_userquota_updates(ds->ds_objset, tx); 52269962b56SMatthew Ahrens } 52394c2d0ebSMatthew Ahrens taskq_wait(dp->dp_sync_taskq); 52414843421SMatthew Ahrens 52514843421SMatthew Ahrens /* 52614843421SMatthew Ahrens * Sync the datasets again to push out the changes due to 5273f9d6ad7SLin Ling * userspace updates. This must be done before we process the 528ce636f8bSMatthew Ahrens * sync tasks, so that any snapshots will have the correct 529ce636f8bSMatthew Ahrens * user accounting information (and we won't get confused 530ce636f8bSMatthew Ahrens * about which blocks are part of the snapshot). 53114843421SMatthew Ahrens */ 53214843421SMatthew Ahrens zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); 53369962b56SMatthew Ahrens while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { 53414843421SMatthew Ahrens ASSERT(list_link_active(&ds->ds_synced_link)); 53514843421SMatthew Ahrens dmu_buf_rele(ds->ds_dbuf, ds); 53614843421SMatthew Ahrens dsl_dataset_sync(ds, zio, tx); 53714843421SMatthew Ahrens } 53869962b56SMatthew Ahrens VERIFY0(zio_wait(zio)); 53914843421SMatthew Ahrens 540b24ab676SJeff Bonwick /* 541ce636f8bSMatthew Ahrens * Now that the datasets have been completely synced, we can 542ce636f8bSMatthew Ahrens * clean up our in-memory structures accumulated while syncing: 543ce636f8bSMatthew Ahrens * 544ce636f8bSMatthew Ahrens * - move dead blocks from the pending deadlist to the on-disk deadlist 545ce636f8bSMatthew Ahrens * - release hold from dsl_dataset_dirty() 546b24ab676SJeff Bonwick */ 54769962b56SMatthew Ahrens while ((ds = list_remove_head(&synced_datasets)) != NULL) { 548bfaed0b9SAndriy Gapon dsl_dataset_sync_done(ds, tx); 549cde58dbcSMatthew Ahrens } 55069962b56SMatthew Ahrens while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) != NULL) { 551c717a561Smaybee dsl_dir_sync(dd, tx); 55269962b56SMatthew Ahrens } 553fa9e4066Sahrens 554ce636f8bSMatthew Ahrens /* 555ce636f8bSMatthew Ahrens * The MOS's space is accounted for in the pool/$MOS 556ce636f8bSMatthew Ahrens * (dp_mos_dir). We can't modify the mos while we're syncing 557ce636f8bSMatthew Ahrens * it, so we remember the deltas and apply them here. 558ce636f8bSMatthew Ahrens */ 559ce636f8bSMatthew Ahrens if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || 560ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta != 0) { 561ce636f8bSMatthew Ahrens dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, 562ce636f8bSMatthew Ahrens dp->dp_mos_used_delta, 563ce636f8bSMatthew Ahrens dp->dp_mos_compressed_delta, 564ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta, tx); 565ce636f8bSMatthew Ahrens dp->dp_mos_used_delta = 0; 566ce636f8bSMatthew Ahrens dp->dp_mos_compressed_delta = 0; 567ce636f8bSMatthew Ahrens dp->dp_mos_uncompressed_delta = 0; 568ce636f8bSMatthew Ahrens } 569ce636f8bSMatthew Ahrens 57094c2d0ebSMatthew Ahrens if (!multilist_is_empty(mos->os_dirty_dnodes[txg & TXG_MASK])) { 57169962b56SMatthew Ahrens dsl_pool_sync_mos(dp, tx); 572fa9e4066Sahrens } 573fa9e4066Sahrens 574ce636f8bSMatthew Ahrens /* 575ce636f8bSMatthew Ahrens * If we modify a dataset in the same txg that we want to destroy it, 576ce636f8bSMatthew Ahrens * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. 577ce636f8bSMatthew Ahrens * dsl_dir_destroy_check() will fail if there are unexpected holds. 578ce636f8bSMatthew Ahrens * Therefore, we want to sync the MOS (thus syncing the dd_dbuf 579ce636f8bSMatthew Ahrens * and clearing the hold on it) before we process the sync_tasks. 580ce636f8bSMatthew Ahrens * The MOS data dirtied by the sync_tasks will be synced on the next 581ce636f8bSMatthew Ahrens * pass. 582ce636f8bSMatthew Ahrens */ 583ce636f8bSMatthew Ahrens if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { 5843b2aab18SMatthew Ahrens dsl_sync_task_t *dst; 585ce636f8bSMatthew Ahrens /* 586ce636f8bSMatthew Ahrens * No more sync tasks should have been added while we 587ce636f8bSMatthew Ahrens * were syncing. 588ce636f8bSMatthew Ahrens */ 58969962b56SMatthew Ahrens ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1); 59069962b56SMatthew Ahrens while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)) != NULL) 5913b2aab18SMatthew Ahrens dsl_sync_task_sync(dst, tx); 592ce636f8bSMatthew Ahrens } 593ce636f8bSMatthew Ahrens 594fa9e4066Sahrens dmu_tx_commit(tx); 59505715f94SMark Maybee 59669962b56SMatthew Ahrens DTRACE_PROBE2(dsl_pool_sync__done, dsl_pool_t *dp, dp, uint64_t, txg); 597fa9e4066Sahrens } 598fa9e4066Sahrens 599fa9e4066Sahrens void 600b24ab676SJeff Bonwick dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) 601fa9e4066Sahrens { 602ce636f8bSMatthew Ahrens zilog_t *zilog; 603fa9e4066Sahrens 60443297f97SGeorge Wilson while (zilog = txg_list_head(&dp->dp_dirty_zilogs, txg)) { 60569962b56SMatthew Ahrens dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); 60643297f97SGeorge Wilson /* 60743297f97SGeorge Wilson * We don't remove the zilog from the dp_dirty_zilogs 60843297f97SGeorge Wilson * list until after we've cleaned it. This ensures that 60943297f97SGeorge Wilson * callers of zilog_is_dirty() receive an accurate 61043297f97SGeorge Wilson * answer when they are racing with the spa sync thread. 61143297f97SGeorge Wilson */ 612ce636f8bSMatthew Ahrens zil_clean(zilog, txg); 61343297f97SGeorge Wilson (void) txg_list_remove_this(&dp->dp_dirty_zilogs, zilog, txg); 614ce636f8bSMatthew Ahrens ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); 615ce636f8bSMatthew Ahrens dmu_buf_rele(ds->ds_dbuf, zilog); 616fa9e4066Sahrens } 617b24ab676SJeff Bonwick ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); 618fa9e4066Sahrens } 619fa9e4066Sahrens 620c717a561Smaybee /* 621c717a561Smaybee * TRUE if the current thread is the tx_sync_thread or if we 622c717a561Smaybee * are being called from SPA context during pool initialization. 623c717a561Smaybee */ 624fa9e4066Sahrens int 625fa9e4066Sahrens dsl_pool_sync_context(dsl_pool_t *dp) 626fa9e4066Sahrens { 627fa9e4066Sahrens return (curthread == dp->dp_tx.tx_sync_thread || 62894c2d0ebSMatthew Ahrens spa_is_initializing(dp->dp_spa) || 62994c2d0ebSMatthew Ahrens taskq_member(dp->dp_sync_taskq, curthread)); 630fa9e4066Sahrens } 631fa9e4066Sahrens 632fa9e4066Sahrens uint64_t 633fa9e4066Sahrens dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) 634fa9e4066Sahrens { 635fa9e4066Sahrens uint64_t space, resv; 636fa9e4066Sahrens 637fa9e4066Sahrens /* 638fa9e4066Sahrens * If we're trying to assess whether it's OK to do a free, 639fa9e4066Sahrens * cut the reservation in half to allow forward progress 640fa9e4066Sahrens * (e.g. make it possible to rm(1) files from a full pool). 641fa9e4066Sahrens */ 642485bbbf5SGeorge Wilson space = spa_get_dspace(dp->dp_spa); 643c39f2c8cSChristopher Siden resv = spa_get_slop_space(dp->dp_spa); 644fa9e4066Sahrens if (netfree) 645fa9e4066Sahrens resv >>= 1; 646fa9e4066Sahrens 647fa9e4066Sahrens return (space - resv); 648fa9e4066Sahrens } 6491ab7f2deSmaybee 65069962b56SMatthew Ahrens boolean_t 65169962b56SMatthew Ahrens dsl_pool_need_dirty_delay(dsl_pool_t *dp) 6521ab7f2deSmaybee { 65369962b56SMatthew Ahrens uint64_t delay_min_bytes = 65469962b56SMatthew Ahrens zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100; 65569962b56SMatthew Ahrens boolean_t rv; 6561ab7f2deSmaybee 65769962b56SMatthew Ahrens mutex_enter(&dp->dp_lock); 65869962b56SMatthew Ahrens if (dp->dp_dirty_total > zfs_dirty_data_sync) 65969962b56SMatthew Ahrens txg_kick(dp); 66069962b56SMatthew Ahrens rv = (dp->dp_dirty_total > delay_min_bytes); 66169962b56SMatthew Ahrens mutex_exit(&dp->dp_lock); 66269962b56SMatthew Ahrens return (rv); 6631ab7f2deSmaybee } 6641ab7f2deSmaybee 6651ab7f2deSmaybee void 66669962b56SMatthew Ahrens dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) 6671ab7f2deSmaybee { 66869962b56SMatthew Ahrens if (space > 0) { 66969962b56SMatthew Ahrens mutex_enter(&dp->dp_lock); 67069962b56SMatthew Ahrens dp->dp_dirty_pertxg[tx->tx_txg & TXG_MASK] += space; 67169962b56SMatthew Ahrens dsl_pool_dirty_delta(dp, space); 67269962b56SMatthew Ahrens mutex_exit(&dp->dp_lock); 6731ab7f2deSmaybee } 6741ab7f2deSmaybee } 6751ab7f2deSmaybee 6761ab7f2deSmaybee void 67769962b56SMatthew Ahrens dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg) 6781ab7f2deSmaybee { 67969962b56SMatthew Ahrens ASSERT3S(space, >=, 0); 68069962b56SMatthew Ahrens if (space == 0) 68169962b56SMatthew Ahrens return; 68269962b56SMatthew Ahrens mutex_enter(&dp->dp_lock); 68369962b56SMatthew Ahrens if (dp->dp_dirty_pertxg[txg & TXG_MASK] < space) { 68469962b56SMatthew Ahrens /* XXX writing something we didn't dirty? */ 68569962b56SMatthew Ahrens space = dp->dp_dirty_pertxg[txg & TXG_MASK]; 6861ab7f2deSmaybee } 68769962b56SMatthew Ahrens ASSERT3U(dp->dp_dirty_pertxg[txg & TXG_MASK], >=, space); 68869962b56SMatthew Ahrens dp->dp_dirty_pertxg[txg & TXG_MASK] -= space; 68969962b56SMatthew Ahrens ASSERT3U(dp->dp_dirty_total, >=, space); 69069962b56SMatthew Ahrens dsl_pool_dirty_delta(dp, -space); 69169962b56SMatthew Ahrens mutex_exit(&dp->dp_lock); 6921ab7f2deSmaybee } 693088f3894Sahrens 694088f3894Sahrens /* ARGSUSED */ 695088f3894Sahrens static int 6963b2aab18SMatthew Ahrens upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) 697088f3894Sahrens { 698088f3894Sahrens dmu_tx_t *tx = arg; 699088f3894Sahrens dsl_dataset_t *ds, *prev = NULL; 700088f3894Sahrens int err; 701088f3894Sahrens 7023b2aab18SMatthew Ahrens err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); 703088f3894Sahrens if (err) 704088f3894Sahrens return (err); 705088f3894Sahrens 706c1379625SJustin T. Gibbs while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { 707c1379625SJustin T. Gibbs err = dsl_dataset_hold_obj(dp, 708c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); 709088f3894Sahrens if (err) { 710088f3894Sahrens dsl_dataset_rele(ds, FTAG); 711088f3894Sahrens return (err); 712088f3894Sahrens } 713088f3894Sahrens 714c1379625SJustin T. Gibbs if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) 715088f3894Sahrens break; 716088f3894Sahrens dsl_dataset_rele(ds, FTAG); 717088f3894Sahrens ds = prev; 718088f3894Sahrens prev = NULL; 719088f3894Sahrens } 720088f3894Sahrens 721088f3894Sahrens if (prev == NULL) { 722088f3894Sahrens prev = dp->dp_origin_snap; 723088f3894Sahrens 724088f3894Sahrens /* 725088f3894Sahrens * The $ORIGIN can't have any data, or the accounting 726088f3894Sahrens * will be wrong. 727088f3894Sahrens */ 728c166b69dSPaul Dagnelie rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); 729c1379625SJustin T. Gibbs ASSERT0(dsl_dataset_phys(prev)->ds_bp.blk_birth); 730c166b69dSPaul Dagnelie rrw_exit(&ds->ds_bp_rwlock, FTAG); 731088f3894Sahrens 732088f3894Sahrens /* The origin doesn't get attached to itself */ 733088f3894Sahrens if (ds->ds_object == prev->ds_object) { 734088f3894Sahrens dsl_dataset_rele(ds, FTAG); 735088f3894Sahrens return (0); 736088f3894Sahrens } 737088f3894Sahrens 738088f3894Sahrens dmu_buf_will_dirty(ds->ds_dbuf, tx); 739c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_obj = prev->ds_object; 740c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_txg = 741c1379625SJustin T. Gibbs dsl_dataset_phys(prev)->ds_creation_txg; 742088f3894Sahrens 743088f3894Sahrens dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); 744c1379625SJustin T. Gibbs dsl_dir_phys(ds->ds_dir)->dd_origin_obj = prev->ds_object; 745088f3894Sahrens 746088f3894Sahrens dmu_buf_will_dirty(prev->ds_dbuf, tx); 747c1379625SJustin T. Gibbs dsl_dataset_phys(prev)->ds_num_children++; 748088f3894Sahrens 749c1379625SJustin T. Gibbs if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) { 750088f3894Sahrens ASSERT(ds->ds_prev == NULL); 7513b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, 752c1379625SJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_obj, 753c1379625SJustin T. Gibbs ds, &ds->ds_prev)); 754088f3894Sahrens } 755088f3894Sahrens } 756088f3894Sahrens 757c1379625SJustin T. Gibbs ASSERT3U(dsl_dir_phys(ds->ds_dir)->dd_origin_obj, ==, prev->ds_object); 758c1379625SJustin T. Gibbs ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_obj, ==, prev->ds_object); 759088f3894Sahrens 760c1379625SJustin T. Gibbs if (dsl_dataset_phys(prev)->ds_next_clones_obj == 0) { 761c33e334fSMatthew Ahrens dmu_buf_will_dirty(prev->ds_dbuf, tx); 762c1379625SJustin T. Gibbs dsl_dataset_phys(prev)->ds_next_clones_obj = 763088f3894Sahrens zap_create(dp->dp_meta_objset, 764088f3894Sahrens DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); 765088f3894Sahrens } 7663b2aab18SMatthew Ahrens VERIFY0(zap_add_int(dp->dp_meta_objset, 767c1379625SJustin T. Gibbs dsl_dataset_phys(prev)->ds_next_clones_obj, ds->ds_object, tx)); 768088f3894Sahrens 769088f3894Sahrens dsl_dataset_rele(ds, FTAG); 770088f3894Sahrens if (prev != dp->dp_origin_snap) 771088f3894Sahrens dsl_dataset_rele(prev, FTAG); 772088f3894Sahrens return (0); 773088f3894Sahrens } 774088f3894Sahrens 775088f3894Sahrens void 776088f3894Sahrens dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) 777088f3894Sahrens { 778088f3894Sahrens ASSERT(dmu_tx_is_syncing(tx)); 779088f3894Sahrens ASSERT(dp->dp_origin_snap != NULL); 780088f3894Sahrens 7813b2aab18SMatthew Ahrens VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb, 78212380e1eSArne Jansen tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); 783088f3894Sahrens } 784088f3894Sahrens 785cde58dbcSMatthew Ahrens /* ARGSUSED */ 786cde58dbcSMatthew Ahrens static int 7873b2aab18SMatthew Ahrens upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 788cde58dbcSMatthew Ahrens { 789cde58dbcSMatthew Ahrens dmu_tx_t *tx = arg; 790cde58dbcSMatthew Ahrens objset_t *mos = dp->dp_meta_objset; 791cde58dbcSMatthew Ahrens 792c1379625SJustin T. Gibbs if (dsl_dir_phys(ds->ds_dir)->dd_origin_obj != 0) { 793cde58dbcSMatthew Ahrens dsl_dataset_t *origin; 794cde58dbcSMatthew Ahrens 7953b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, 796c1379625SJustin T. Gibbs dsl_dir_phys(ds->ds_dir)->dd_origin_obj, FTAG, &origin)); 797cde58dbcSMatthew Ahrens 798c1379625SJustin T. Gibbs if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { 799cde58dbcSMatthew Ahrens dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); 800c1379625SJustin T. Gibbs dsl_dir_phys(origin->ds_dir)->dd_clones = 801c1379625SJustin T. Gibbs zap_create(mos, DMU_OT_DSL_CLONES, DMU_OT_NONE, 802c1379625SJustin T. Gibbs 0, tx); 803cde58dbcSMatthew Ahrens } 804cde58dbcSMatthew Ahrens 8053b2aab18SMatthew Ahrens VERIFY0(zap_add_int(dp->dp_meta_objset, 806c1379625SJustin T. Gibbs dsl_dir_phys(origin->ds_dir)->dd_clones, 807c1379625SJustin T. Gibbs ds->ds_object, tx)); 808cde58dbcSMatthew Ahrens 809cde58dbcSMatthew Ahrens dsl_dataset_rele(origin, FTAG); 810cde58dbcSMatthew Ahrens } 811cde58dbcSMatthew Ahrens return (0); 812cde58dbcSMatthew Ahrens } 813cde58dbcSMatthew Ahrens 814cde58dbcSMatthew Ahrens void 815cde58dbcSMatthew Ahrens dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) 816cde58dbcSMatthew Ahrens { 817cde58dbcSMatthew Ahrens ASSERT(dmu_tx_is_syncing(tx)); 818cde58dbcSMatthew Ahrens uint64_t obj; 819cde58dbcSMatthew Ahrens 820cde58dbcSMatthew Ahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); 8213b2aab18SMatthew Ahrens VERIFY0(dsl_pool_open_special_dir(dp, 822cde58dbcSMatthew Ahrens FREE_DIR_NAME, &dp->dp_free_dir)); 823cde58dbcSMatthew Ahrens 824cde58dbcSMatthew Ahrens /* 825cde58dbcSMatthew Ahrens * We can't use bpobj_alloc(), because spa_version() still 826cde58dbcSMatthew Ahrens * returns the old version, and we need a new-version bpobj with 827cde58dbcSMatthew Ahrens * subobj support. So call dmu_object_alloc() directly. 828cde58dbcSMatthew Ahrens */ 829cde58dbcSMatthew Ahrens obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, 830b5152584SMatthew Ahrens SPA_OLD_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); 8313b2aab18SMatthew Ahrens VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 832cde58dbcSMatthew Ahrens DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); 8333b2aab18SMatthew Ahrens VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); 834cde58dbcSMatthew Ahrens 8353b2aab18SMatthew Ahrens VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 83612380e1eSArne Jansen upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); 837cde58dbcSMatthew Ahrens } 838cde58dbcSMatthew Ahrens 839088f3894Sahrens void 840088f3894Sahrens dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) 841088f3894Sahrens { 842088f3894Sahrens uint64_t dsobj; 843088f3894Sahrens dsl_dataset_t *ds; 844088f3894Sahrens 845088f3894Sahrens ASSERT(dmu_tx_is_syncing(tx)); 846088f3894Sahrens ASSERT(dp->dp_origin_snap == NULL); 8473b2aab18SMatthew Ahrens ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER)); 848088f3894Sahrens 849088f3894Sahrens /* create the origin dir, ds, & snap-ds */ 850088f3894Sahrens dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, 851088f3894Sahrens NULL, 0, kcred, tx); 8523b2aab18SMatthew Ahrens VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); 8533b2aab18SMatthew Ahrens dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx); 854c1379625SJustin T. Gibbs VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, 855088f3894Sahrens dp, &dp->dp_origin_snap)); 856088f3894Sahrens dsl_dataset_rele(ds, FTAG); 857088f3894Sahrens } 8589d3574bfSNeil Perrin 8599d3574bfSNeil Perrin taskq_t * 8609d3574bfSNeil Perrin dsl_pool_vnrele_taskq(dsl_pool_t *dp) 8619d3574bfSNeil Perrin { 8629d3574bfSNeil Perrin return (dp->dp_vnrele_taskq); 8639d3574bfSNeil Perrin } 864ca45db41SChris Kirby 865ca45db41SChris Kirby /* 866ca45db41SChris Kirby * Walk through the pool-wide zap object of temporary snapshot user holds 867ca45db41SChris Kirby * and release them. 868ca45db41SChris Kirby */ 869ca45db41SChris Kirby void 870ca45db41SChris Kirby dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) 871ca45db41SChris Kirby { 872ca45db41SChris Kirby zap_attribute_t za; 873ca45db41SChris Kirby zap_cursor_t zc; 874ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 875ca45db41SChris Kirby uint64_t zapobj = dp->dp_tmp_userrefs_obj; 876a7a845e4SSteven Hartland nvlist_t *holds; 877ca45db41SChris Kirby 878ca45db41SChris Kirby if (zapobj == 0) 879ca45db41SChris Kirby return; 880ca45db41SChris Kirby ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 881ca45db41SChris Kirby 882a7a845e4SSteven Hartland holds = fnvlist_alloc(); 883a7a845e4SSteven Hartland 884ca45db41SChris Kirby for (zap_cursor_init(&zc, mos, zapobj); 885ca45db41SChris Kirby zap_cursor_retrieve(&zc, &za) == 0; 886ca45db41SChris Kirby zap_cursor_advance(&zc)) { 887ca45db41SChris Kirby char *htag; 888a7a845e4SSteven Hartland nvlist_t *tags; 889ca45db41SChris Kirby 890ca45db41SChris Kirby htag = strchr(za.za_name, '-'); 891ca45db41SChris Kirby *htag = '\0'; 892ca45db41SChris Kirby ++htag; 893a7a845e4SSteven Hartland if (nvlist_lookup_nvlist(holds, za.za_name, &tags) != 0) { 894a7a845e4SSteven Hartland tags = fnvlist_alloc(); 895a7a845e4SSteven Hartland fnvlist_add_boolean(tags, htag); 896a7a845e4SSteven Hartland fnvlist_add_nvlist(holds, za.za_name, tags); 897a7a845e4SSteven Hartland fnvlist_free(tags); 898a7a845e4SSteven Hartland } else { 899a7a845e4SSteven Hartland fnvlist_add_boolean(tags, htag); 900a7a845e4SSteven Hartland } 901ca45db41SChris Kirby } 902a7a845e4SSteven Hartland dsl_dataset_user_release_tmp(dp, holds); 903a7a845e4SSteven Hartland fnvlist_free(holds); 904ca45db41SChris Kirby zap_cursor_fini(&zc); 905ca45db41SChris Kirby } 906ca45db41SChris Kirby 907ca45db41SChris Kirby /* 908ca45db41SChris Kirby * Create the pool-wide zap object for storing temporary snapshot holds. 909ca45db41SChris Kirby */ 910ca45db41SChris Kirby void 911ca45db41SChris Kirby dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) 912ca45db41SChris Kirby { 913ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 914ca45db41SChris Kirby 915ca45db41SChris Kirby ASSERT(dp->dp_tmp_userrefs_obj == 0); 916ca45db41SChris Kirby ASSERT(dmu_tx_is_syncing(tx)); 917ca45db41SChris Kirby 918ad135b5dSChristopher Siden dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, 919ad135b5dSChristopher Siden DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); 920ca45db41SChris Kirby } 921ca45db41SChris Kirby 922ca45db41SChris Kirby static int 923ca45db41SChris Kirby dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, 9243b2aab18SMatthew Ahrens const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding) 925ca45db41SChris Kirby { 926ca45db41SChris Kirby objset_t *mos = dp->dp_meta_objset; 927ca45db41SChris Kirby uint64_t zapobj = dp->dp_tmp_userrefs_obj; 928ca45db41SChris Kirby char *name; 929ca45db41SChris Kirby int error; 930ca45db41SChris Kirby 931ca45db41SChris Kirby ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); 932ca45db41SChris Kirby ASSERT(dmu_tx_is_syncing(tx)); 933ca45db41SChris Kirby 934ca45db41SChris Kirby /* 935ca45db41SChris Kirby * If the pool was created prior to SPA_VERSION_USERREFS, the 936ca45db41SChris Kirby * zap object for temporary holds might not exist yet. 937ca45db41SChris Kirby */ 938ca45db41SChris Kirby if (zapobj == 0) { 939ca45db41SChris Kirby if (holding) { 940ca45db41SChris Kirby dsl_pool_user_hold_create_obj(dp, tx); 941ca45db41SChris Kirby zapobj = dp->dp_tmp_userrefs_obj; 942ca45db41SChris Kirby } else { 943be6fd75aSMatthew Ahrens return (SET_ERROR(ENOENT)); 944ca45db41SChris Kirby } 945ca45db41SChris Kirby } 946ca45db41SChris Kirby 947ca45db41SChris Kirby name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); 948ca45db41SChris Kirby if (holding) 9493b2aab18SMatthew Ahrens error = zap_add(mos, zapobj, name, 8, 1, &now, tx); 950ca45db41SChris Kirby else 951ca45db41SChris Kirby error = zap_remove(mos, zapobj, name, tx); 952ca45db41SChris Kirby strfree(name); 953ca45db41SChris Kirby 954ca45db41SChris Kirby return (error); 955ca45db41SChris Kirby } 956ca45db41SChris Kirby 957ca45db41SChris Kirby /* 958ca45db41SChris Kirby * Add a temporary hold for the given dataset object and tag. 959ca45db41SChris Kirby */ 960ca45db41SChris Kirby int 961ca45db41SChris Kirby dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 9623b2aab18SMatthew Ahrens uint64_t now, dmu_tx_t *tx) 963ca45db41SChris Kirby { 96415508ac0SChris Kirby return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); 965ca45db41SChris Kirby } 966ca45db41SChris Kirby 967ca45db41SChris Kirby /* 968ca45db41SChris Kirby * Release a temporary hold for the given dataset object and tag. 969ca45db41SChris Kirby */ 970ca45db41SChris Kirby int 971ca45db41SChris Kirby dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, 972ca45db41SChris Kirby dmu_tx_t *tx) 973ca45db41SChris Kirby { 974ca45db41SChris Kirby return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL, 975ca45db41SChris Kirby tx, B_FALSE)); 976ca45db41SChris Kirby } 9773b2aab18SMatthew Ahrens 9783b2aab18SMatthew Ahrens /* 9793b2aab18SMatthew Ahrens * DSL Pool Configuration Lock 9803b2aab18SMatthew Ahrens * 9813b2aab18SMatthew Ahrens * The dp_config_rwlock protects against changes to DSL state (e.g. dataset 9823b2aab18SMatthew Ahrens * creation / destruction / rename / property setting). It must be held for 9833b2aab18SMatthew Ahrens * read to hold a dataset or dsl_dir. I.e. you must call 9843b2aab18SMatthew Ahrens * dsl_pool_config_enter() or dsl_pool_hold() before calling 9853b2aab18SMatthew Ahrens * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock 9863b2aab18SMatthew Ahrens * must be held continuously until all datasets and dsl_dirs are released. 9873b2aab18SMatthew Ahrens * 9883b2aab18SMatthew Ahrens * The only exception to this rule is that if a "long hold" is placed on 9893b2aab18SMatthew Ahrens * a dataset, then the dp_config_rwlock may be dropped while the dataset 9903b2aab18SMatthew Ahrens * is still held. The long hold will prevent the dataset from being 9913b2aab18SMatthew Ahrens * destroyed -- the destroy will fail with EBUSY. A long hold can be 9923b2aab18SMatthew Ahrens * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset 9933b2aab18SMatthew Ahrens * (by calling dsl_{dataset,objset}_{try}own{_obj}). 9943b2aab18SMatthew Ahrens * 9953b2aab18SMatthew Ahrens * Legitimate long-holders (including owners) should be long-running, cancelable 9963b2aab18SMatthew Ahrens * tasks that should cause "zfs destroy" to fail. This includes DMU 9973b2aab18SMatthew Ahrens * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open), 9983b2aab18SMatthew Ahrens * "zfs send", and "zfs diff". There are several other long-holders whose 9993b2aab18SMatthew Ahrens * uses are suboptimal (e.g. "zfs promote", and zil_suspend()). 10003b2aab18SMatthew Ahrens * 10013b2aab18SMatthew Ahrens * The usual formula for long-holding would be: 10023b2aab18SMatthew Ahrens * dsl_pool_hold() 10033b2aab18SMatthew Ahrens * dsl_dataset_hold() 10043b2aab18SMatthew Ahrens * ... perform checks ... 10053b2aab18SMatthew Ahrens * dsl_dataset_long_hold() 10063b2aab18SMatthew Ahrens * dsl_pool_rele() 10073b2aab18SMatthew Ahrens * ... perform long-running task ... 10083b2aab18SMatthew Ahrens * dsl_dataset_long_rele() 10093b2aab18SMatthew Ahrens * dsl_dataset_rele() 10103b2aab18SMatthew Ahrens * 10113b2aab18SMatthew Ahrens * Note that when the long hold is released, the dataset is still held but 10123b2aab18SMatthew Ahrens * the pool is not held. The dataset may change arbitrarily during this time 10133b2aab18SMatthew Ahrens * (e.g. it could be destroyed). Therefore you shouldn't do anything to the 10143b2aab18SMatthew Ahrens * dataset except release it. 10153b2aab18SMatthew Ahrens * 10163b2aab18SMatthew Ahrens * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only 10173b2aab18SMatthew Ahrens * or modifying operations. 10183b2aab18SMatthew Ahrens * 10193b2aab18SMatthew Ahrens * Modifying operations should generally use dsl_sync_task(). The synctask 10203b2aab18SMatthew Ahrens * infrastructure enforces proper locking strategy with respect to the 10213b2aab18SMatthew Ahrens * dp_config_rwlock. See the comment above dsl_sync_task() for details. 10223b2aab18SMatthew Ahrens * 10233b2aab18SMatthew Ahrens * Read-only operations will manually hold the pool, then the dataset, obtain 10243b2aab18SMatthew Ahrens * information from the dataset, then release the pool and dataset. 10253b2aab18SMatthew Ahrens * dmu_objset_{hold,rele}() are convenience routines that also do the pool 10263b2aab18SMatthew Ahrens * hold/rele. 10273b2aab18SMatthew Ahrens */ 10283b2aab18SMatthew Ahrens 10293b2aab18SMatthew Ahrens int 10303b2aab18SMatthew Ahrens dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp) 10313b2aab18SMatthew Ahrens { 10323b2aab18SMatthew Ahrens spa_t *spa; 10333b2aab18SMatthew Ahrens int error; 10343b2aab18SMatthew Ahrens 10353b2aab18SMatthew Ahrens error = spa_open(name, &spa, tag); 10363b2aab18SMatthew Ahrens if (error == 0) { 10373b2aab18SMatthew Ahrens *dp = spa_get_dsl(spa); 10383b2aab18SMatthew Ahrens dsl_pool_config_enter(*dp, tag); 10393b2aab18SMatthew Ahrens } 10403b2aab18SMatthew Ahrens return (error); 10413b2aab18SMatthew Ahrens } 10423b2aab18SMatthew Ahrens 10433b2aab18SMatthew Ahrens void 10443b2aab18SMatthew Ahrens dsl_pool_rele(dsl_pool_t *dp, void *tag) 10453b2aab18SMatthew Ahrens { 10463b2aab18SMatthew Ahrens dsl_pool_config_exit(dp, tag); 10473b2aab18SMatthew Ahrens spa_close(dp->dp_spa, tag); 10483b2aab18SMatthew Ahrens } 10493b2aab18SMatthew Ahrens 10503b2aab18SMatthew Ahrens void 10513b2aab18SMatthew Ahrens dsl_pool_config_enter(dsl_pool_t *dp, void *tag) 10523b2aab18SMatthew Ahrens { 10533b2aab18SMatthew Ahrens /* 10543b2aab18SMatthew Ahrens * We use a "reentrant" reader-writer lock, but not reentrantly. 10553b2aab18SMatthew Ahrens * 10563b2aab18SMatthew Ahrens * The rrwlock can (with the track_all flag) track all reading threads, 10573b2aab18SMatthew Ahrens * which is very useful for debugging which code path failed to release 10583b2aab18SMatthew Ahrens * the lock, and for verifying that the *current* thread does hold 10593b2aab18SMatthew Ahrens * the lock. 10603b2aab18SMatthew Ahrens * 10613b2aab18SMatthew Ahrens * (Unlike a rwlock, which knows that N threads hold it for 10623b2aab18SMatthew Ahrens * read, but not *which* threads, so rw_held(RW_READER) returns TRUE 10633b2aab18SMatthew Ahrens * if any thread holds it for read, even if this thread doesn't). 10643b2aab18SMatthew Ahrens */ 10653b2aab18SMatthew Ahrens ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); 10663b2aab18SMatthew Ahrens rrw_enter(&dp->dp_config_rwlock, RW_READER, tag); 10673b2aab18SMatthew Ahrens } 10683b2aab18SMatthew Ahrens 10691d3f896fSArne Jansen void 10701d3f896fSArne Jansen dsl_pool_config_enter_prio(dsl_pool_t *dp, void *tag) 10711d3f896fSArne Jansen { 10721d3f896fSArne Jansen ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); 10731d3f896fSArne Jansen rrw_enter_read_prio(&dp->dp_config_rwlock, tag); 10741d3f896fSArne Jansen } 10751d3f896fSArne Jansen 10763b2aab18SMatthew Ahrens void 10773b2aab18SMatthew Ahrens dsl_pool_config_exit(dsl_pool_t *dp, void *tag) 10783b2aab18SMatthew Ahrens { 10793b2aab18SMatthew Ahrens rrw_exit(&dp->dp_config_rwlock, tag); 10803b2aab18SMatthew Ahrens } 10813b2aab18SMatthew Ahrens 10823b2aab18SMatthew Ahrens boolean_t 10833b2aab18SMatthew Ahrens dsl_pool_config_held(dsl_pool_t *dp) 10843b2aab18SMatthew Ahrens { 10853b2aab18SMatthew Ahrens return (RRW_LOCK_HELD(&dp->dp_config_rwlock)); 10863b2aab18SMatthew Ahrens } 108712380e1eSArne Jansen 108812380e1eSArne Jansen boolean_t 108912380e1eSArne Jansen dsl_pool_config_held_writer(dsl_pool_t *dp) 109012380e1eSArne Jansen { 109112380e1eSArne Jansen return (RRW_WRITE_HELD(&dp->dp_config_rwlock)); 109212380e1eSArne Jansen } 1093