1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2011, 2019 by Delphix. All rights reserved. 24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 26 * Copyright 2013 Saso Kiselkov. All rights reserved. 27 * Copyright (c) 2017 Datto Inc. 28 * Copyright (c) 2017, Intel Corporation. 29 */ 30 31 #ifndef _SYS_SPA_IMPL_H 32 #define _SYS_SPA_IMPL_H 33 34 #include <sys/spa.h> 35 #include <sys/spa_checkpoint.h> 36 #include <sys/spa_log_spacemap.h> 37 #include <sys/vdev.h> 38 #include <sys/vdev_removal.h> 39 #include <sys/metaslab.h> 40 #include <sys/dmu.h> 41 #include <sys/dsl_pool.h> 42 #include <sys/uberblock_impl.h> 43 #include <sys/zfs_context.h> 44 #include <sys/avl.h> 45 #include <sys/refcount.h> 46 #include <sys/bplist.h> 47 #include <sys/bpobj.h> 48 #include <sys/dsl_crypt.h> 49 #include <sys/zfeature.h> 50 #include <sys/zthr.h> 51 #include <zfeature_common.h> 52 53 #ifdef __cplusplus 54 extern "C" { 55 #endif 56 57 typedef struct spa_error_entry { 58 zbookmark_phys_t se_bookmark; 59 char *se_name; 60 avl_node_t se_avl; 61 } spa_error_entry_t; 62 63 typedef struct spa_history_phys { 64 uint64_t sh_pool_create_len; /* ending offset of zpool create */ 65 uint64_t sh_phys_max_off; /* physical EOF */ 66 uint64_t sh_bof; /* logical BOF */ 67 uint64_t sh_eof; /* logical EOF */ 68 uint64_t sh_records_lost; /* num of records overwritten */ 69 } spa_history_phys_t; 70 71 /* 72 * All members must be uint64_t, for byteswap purposes. 73 */ 74 typedef struct spa_removing_phys { 75 uint64_t sr_state; /* dsl_scan_state_t */ 76 77 /* 78 * The vdev ID that we most recently attempted to remove, 79 * or -1 if no removal has been attempted. 80 */ 81 uint64_t sr_removing_vdev; 82 83 /* 84 * The vdev ID that we most recently successfully removed, 85 * or -1 if no devices have been removed. 86 */ 87 uint64_t sr_prev_indirect_vdev; 88 89 uint64_t sr_start_time; 90 uint64_t sr_end_time; 91 92 /* 93 * Note that we can not use the space map's or indirect mapping's 94 * accounting as a substitute for these values, because we need to 95 * count frees of not-yet-copied data as though it did the copy. 96 * Otherwise, we could get into a situation where copied > to_copy, 97 * or we complete before copied == to_copy. 98 */ 99 uint64_t sr_to_copy; /* bytes that need to be copied */ 100 uint64_t sr_copied; /* bytes that have been copied or freed */ 101 } spa_removing_phys_t; 102 103 /* 104 * This struct is stored as an entry in the DMU_POOL_DIRECTORY_OBJECT 105 * (with key DMU_POOL_CONDENSING_INDIRECT). It is present if a condense 106 * of an indirect vdev's mapping object is in progress. 107 */ 108 typedef struct spa_condensing_indirect_phys { 109 /* 110 * The vdev ID of the indirect vdev whose indirect mapping is 111 * being condensed. 112 */ 113 uint64_t scip_vdev; 114 115 /* 116 * The vdev's old obsolete spacemap. This spacemap's contents are 117 * being integrated into the new mapping. 118 */ 119 uint64_t scip_prev_obsolete_sm_object; 120 121 /* 122 * The new mapping object that is being created. 123 */ 124 uint64_t scip_next_mapping_object; 125 } spa_condensing_indirect_phys_t; 126 127 struct spa_aux_vdev { 128 uint64_t sav_object; /* MOS object for device list */ 129 nvlist_t *sav_config; /* cached device config */ 130 vdev_t **sav_vdevs; /* devices */ 131 int sav_count; /* number devices */ 132 boolean_t sav_sync; /* sync the device list */ 133 nvlist_t **sav_pending; /* pending device additions */ 134 uint_t sav_npending; /* # pending devices */ 135 }; 136 137 typedef struct spa_config_lock { 138 kmutex_t scl_lock; 139 kthread_t *scl_writer; 140 int scl_write_wanted; 141 kcondvar_t scl_cv; 142 zfs_refcount_t scl_count; 143 } spa_config_lock_t; 144 145 typedef struct spa_config_dirent { 146 list_node_t scd_link; 147 char *scd_path; 148 } spa_config_dirent_t; 149 150 typedef enum zio_taskq_type { 151 ZIO_TASKQ_ISSUE = 0, 152 ZIO_TASKQ_ISSUE_HIGH, 153 ZIO_TASKQ_INTERRUPT, 154 ZIO_TASKQ_INTERRUPT_HIGH, 155 ZIO_TASKQ_TYPES 156 } zio_taskq_type_t; 157 158 /* 159 * State machine for the zpool-poolname process. The states transitions 160 * are done as follows: 161 * 162 * From To Routine 163 * PROC_NONE -> PROC_CREATED spa_activate() 164 * PROC_CREATED -> PROC_ACTIVE spa_thread() 165 * PROC_ACTIVE -> PROC_DEACTIVATE spa_deactivate() 166 * PROC_DEACTIVATE -> PROC_GONE spa_thread() 167 * PROC_GONE -> PROC_NONE spa_deactivate() 168 */ 169 typedef enum spa_proc_state { 170 SPA_PROC_NONE, /* spa_proc = &p0, no process created */ 171 SPA_PROC_CREATED, /* spa_activate() has proc, is waiting */ 172 SPA_PROC_ACTIVE, /* taskqs created, spa_proc set */ 173 SPA_PROC_DEACTIVATE, /* spa_deactivate() requests process exit */ 174 SPA_PROC_GONE /* spa_thread() is exiting, spa_proc = &p0 */ 175 } spa_proc_state_t; 176 177 typedef struct spa_taskqs { 178 uint_t stqs_count; 179 taskq_t **stqs_taskq; 180 } spa_taskqs_t; 181 182 typedef enum spa_all_vdev_zap_action { 183 AVZ_ACTION_NONE = 0, 184 AVZ_ACTION_DESTROY, /* Destroy all per-vdev ZAPs and the AVZ. */ 185 AVZ_ACTION_REBUILD, /* Populate the new AVZ, see spa_avz_rebuild */ 186 AVZ_ACTION_INITIALIZE 187 } spa_avz_action_t; 188 189 typedef enum spa_config_source { 190 SPA_CONFIG_SRC_NONE = 0, 191 SPA_CONFIG_SRC_SCAN, /* scan of path (default: /dev/dsk) */ 192 SPA_CONFIG_SRC_CACHEFILE, /* any cachefile */ 193 SPA_CONFIG_SRC_TRYIMPORT, /* returned from call to tryimport */ 194 SPA_CONFIG_SRC_SPLIT, /* new pool in a pool split */ 195 SPA_CONFIG_SRC_MOS /* MOS, but not always from right txg */ 196 } spa_config_source_t; 197 198 struct spa { 199 /* 200 * Fields protected by spa_namespace_lock. 201 */ 202 char spa_name[ZFS_MAX_DATASET_NAME_LEN]; /* pool name */ 203 char *spa_comment; /* comment */ 204 avl_node_t spa_avl; /* node in spa_namespace_avl */ 205 nvlist_t *spa_config; /* last synced config */ 206 nvlist_t *spa_config_syncing; /* currently syncing config */ 207 nvlist_t *spa_config_splitting; /* config for splitting */ 208 nvlist_t *spa_load_info; /* info and errors from load */ 209 uint64_t spa_config_txg; /* txg of last config change */ 210 int spa_sync_pass; /* iterate-to-convergence */ 211 pool_state_t spa_state; /* pool state */ 212 int spa_inject_ref; /* injection references */ 213 uint8_t spa_sync_on; /* sync threads are running */ 214 spa_load_state_t spa_load_state; /* current load operation */ 215 boolean_t spa_indirect_vdevs_loaded; /* mappings loaded? */ 216 boolean_t spa_trust_config; /* do we trust vdev tree? */ 217 spa_config_source_t spa_config_source; /* where config comes from? */ 218 uint64_t spa_import_flags; /* import specific flags */ 219 spa_taskqs_t spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES]; 220 dsl_pool_t *spa_dsl_pool; 221 boolean_t spa_is_initializing; /* true while opening pool */ 222 metaslab_class_t *spa_normal_class; /* normal data class */ 223 metaslab_class_t *spa_log_class; /* intent log data class */ 224 metaslab_class_t *spa_special_class; /* special allocation class */ 225 metaslab_class_t *spa_dedup_class; /* dedup allocation class */ 226 uint64_t spa_first_txg; /* first txg after spa_open() */ 227 uint64_t spa_final_txg; /* txg of export/destroy */ 228 uint64_t spa_freeze_txg; /* freeze pool at this txg */ 229 uint64_t spa_load_max_txg; /* best initial ub_txg */ 230 uint64_t spa_claim_max_txg; /* highest claimed birth txg */ 231 timespec_t spa_loaded_ts; /* 1st successful open time */ 232 objset_t *spa_meta_objset; /* copy of dp->dp_meta_objset */ 233 kmutex_t spa_evicting_os_lock; /* Evicting objset list lock */ 234 list_t spa_evicting_os_list; /* Objsets being evicted. */ 235 kcondvar_t spa_evicting_os_cv; /* Objset Eviction Completion */ 236 txg_list_t spa_vdev_txg_list; /* per-txg dirty vdev list */ 237 vdev_t *spa_root_vdev; /* top-level vdev container */ 238 int spa_min_ashift; /* of vdevs in normal class */ 239 int spa_max_ashift; /* of vdevs in normal class */ 240 uint64_t spa_config_guid; /* config pool guid */ 241 uint64_t spa_load_guid; /* spa_load initialized guid */ 242 uint64_t spa_last_synced_guid; /* last synced guid */ 243 list_t spa_config_dirty_list; /* vdevs with dirty config */ 244 list_t spa_state_dirty_list; /* vdevs with dirty state */ 245 /* 246 * spa_alloc_locks and spa_alloc_trees are arrays, whose lengths are 247 * stored in spa_alloc_count. There is one tree and one lock for each 248 * allocator, to help improve allocation performance in write-heavy 249 * workloads. 250 */ 251 kmutex_t *spa_alloc_locks; 252 avl_tree_t *spa_alloc_trees; 253 int spa_alloc_count; 254 255 spa_aux_vdev_t spa_spares; /* hot spares */ 256 spa_aux_vdev_t spa_l2cache; /* L2ARC cache devices */ 257 nvlist_t *spa_label_features; /* Features for reading MOS */ 258 uint64_t spa_config_object; /* MOS object for pool config */ 259 uint64_t spa_config_generation; /* config generation number */ 260 uint64_t spa_syncing_txg; /* txg currently syncing */ 261 bpobj_t spa_deferred_bpobj; /* deferred-free bplist */ 262 bplist_t spa_free_bplist[TXG_SIZE]; /* bplist of stuff to free */ 263 zio_cksum_salt_t spa_cksum_salt; /* secret salt for cksum */ 264 /* checksum context templates */ 265 kmutex_t spa_cksum_tmpls_lock; 266 void *spa_cksum_tmpls[ZIO_CHECKSUM_FUNCTIONS]; 267 uberblock_t spa_ubsync; /* last synced uberblock */ 268 uberblock_t spa_uberblock; /* current uberblock */ 269 boolean_t spa_extreme_rewind; /* rewind past deferred frees */ 270 kmutex_t spa_scrub_lock; /* resilver/scrub lock */ 271 uint64_t spa_scrub_inflight; /* in-flight scrub bytes */ 272 uint64_t spa_load_verify_ios; /* in-flight verification IOs */ 273 kcondvar_t spa_scrub_io_cv; /* scrub I/O completion */ 274 uint8_t spa_scrub_active; /* active or suspended? */ 275 uint8_t spa_scrub_type; /* type of scrub we're doing */ 276 uint8_t spa_scrub_finished; /* indicator to rotate logs */ 277 uint8_t spa_scrub_started; /* started since last boot */ 278 uint8_t spa_scrub_reopen; /* scrub doing vdev_reopen */ 279 uint64_t spa_scan_pass_start; /* start time per pass/reboot */ 280 uint64_t spa_scan_pass_scrub_pause; /* scrub pause time */ 281 uint64_t spa_scan_pass_scrub_spent_paused; /* total paused */ 282 uint64_t spa_scan_pass_exam; /* examined bytes per pass */ 283 uint64_t spa_scan_pass_issued; /* issued bytes per pass */ 284 285 /* 286 * We are in the middle of a resilver, and another resilver 287 * is needed once this one completes. This is set iff any 288 * vdev_resilver_deferred is set. 289 */ 290 boolean_t spa_resilver_deferred; 291 kmutex_t spa_async_lock; /* protect async state */ 292 kthread_t *spa_async_thread; /* thread doing async task */ 293 int spa_async_suspended; /* async tasks suspended */ 294 kcondvar_t spa_async_cv; /* wait for thread_exit() */ 295 uint16_t spa_async_tasks; /* async task mask */ 296 uint64_t spa_missing_tvds; /* unopenable tvds on load */ 297 uint64_t spa_missing_tvds_allowed; /* allow loading spa? */ 298 299 spa_removing_phys_t spa_removing_phys; 300 spa_vdev_removal_t *spa_vdev_removal; 301 302 spa_condensing_indirect_phys_t spa_condensing_indirect_phys; 303 spa_condensing_indirect_t *spa_condensing_indirect; 304 zthr_t *spa_condense_zthr; /* zthr doing condense. */ 305 306 uint64_t spa_checkpoint_txg; /* the txg of the checkpoint */ 307 spa_checkpoint_info_t spa_checkpoint_info; /* checkpoint accounting */ 308 zthr_t *spa_checkpoint_discard_zthr; 309 310 space_map_t *spa_syncing_log_sm; /* current log space map */ 311 avl_tree_t spa_sm_logs_by_txg; 312 kmutex_t spa_flushed_ms_lock; /* for metaslabs_by_flushed */ 313 avl_tree_t spa_metaslabs_by_flushed; 314 spa_unflushed_stats_t spa_unflushed_stats; 315 list_t spa_log_summary; 316 uint64_t spa_log_flushall_txg; 317 318 char *spa_root; /* alternate root directory */ 319 uint64_t spa_ena; /* spa-wide ereport ENA */ 320 int spa_last_open_failed; /* error if last open failed */ 321 uint64_t spa_last_ubsync_txg; /* "best" uberblock txg */ 322 uint64_t spa_last_ubsync_txg_ts; /* timestamp from that ub */ 323 uint64_t spa_load_txg; /* ub txg that loaded */ 324 uint64_t spa_load_txg_ts; /* timestamp from that ub */ 325 uint64_t spa_load_meta_errors; /* verify metadata err count */ 326 uint64_t spa_load_data_errors; /* verify data err count */ 327 uint64_t spa_verify_min_txg; /* start txg of verify scrub */ 328 kmutex_t spa_errlog_lock; /* error log lock */ 329 uint64_t spa_errlog_last; /* last error log object */ 330 uint64_t spa_errlog_scrub; /* scrub error log object */ 331 kmutex_t spa_errlist_lock; /* error list/ereport lock */ 332 avl_tree_t spa_errlist_last; /* last error list */ 333 avl_tree_t spa_errlist_scrub; /* scrub error list */ 334 uint64_t spa_deflate; /* should we deflate? */ 335 uint64_t spa_history; /* history object */ 336 kmutex_t spa_history_lock; /* history lock */ 337 vdev_t *spa_pending_vdev; /* pending vdev additions */ 338 kmutex_t spa_props_lock; /* property lock */ 339 uint64_t spa_pool_props_object; /* object for properties */ 340 uint64_t spa_bootfs; /* default boot filesystem */ 341 uint64_t spa_failmode; /* failure mode for the pool */ 342 uint64_t spa_delegation; /* delegation on/off */ 343 list_t spa_config_list; /* previous cache file(s) */ 344 /* per-CPU array of root of async I/O: */ 345 zio_t **spa_async_zio_root; 346 zio_t *spa_suspend_zio_root; /* root of all suspended I/O */ 347 zio_t *spa_txg_zio[TXG_SIZE]; /* spa_sync() waits for this */ 348 kmutex_t spa_suspend_lock; /* protects suspend_zio_root */ 349 kcondvar_t spa_suspend_cv; /* notification of resume */ 350 zio_suspend_reason_t spa_suspended; /* pool is suspended */ 351 uint8_t spa_claiming; /* pool is doing zil_claim() */ 352 boolean_t spa_is_root; /* pool is root */ 353 int spa_minref; /* num refs when first opened */ 354 int spa_mode; /* FREAD | FWRITE */ 355 spa_log_state_t spa_log_state; /* log state */ 356 uint64_t spa_autoexpand; /* lun expansion on/off */ 357 uint64_t spa_bootsize; /* efi system partition size */ 358 ddt_t *spa_ddt[ZIO_CHECKSUM_FUNCTIONS]; /* in-core DDTs */ 359 uint64_t spa_ddt_stat_object; /* DDT statistics */ 360 uint64_t spa_dedup_ditto; /* dedup ditto threshold */ 361 uint64_t spa_dedup_checksum; /* default dedup checksum */ 362 uint64_t spa_dspace; /* dspace in normal class */ 363 kmutex_t spa_vdev_top_lock; /* dueling offline/remove */ 364 kmutex_t spa_proc_lock; /* protects spa_proc* */ 365 kcondvar_t spa_proc_cv; /* spa_proc_state transitions */ 366 spa_proc_state_t spa_proc_state; /* see definition */ 367 struct proc *spa_proc; /* "zpool-poolname" process */ 368 uint64_t spa_did; /* if procp != p0, did of t1 */ 369 boolean_t spa_autoreplace; /* autoreplace set in open */ 370 int spa_vdev_locks; /* locks grabbed */ 371 uint64_t spa_creation_version; /* version at pool creation */ 372 uint64_t spa_prev_software_version; /* See ub_software_version */ 373 uint64_t spa_feat_for_write_obj; /* required to write to pool */ 374 uint64_t spa_feat_for_read_obj; /* required to read from pool */ 375 uint64_t spa_feat_desc_obj; /* Feature descriptions */ 376 uint64_t spa_feat_enabled_txg_obj; /* Feature enabled txg */ 377 /* cache feature refcounts */ 378 uint64_t spa_feat_refcount_cache[SPA_FEATURES]; 379 cyclic_id_t spa_deadman_cycid; /* cyclic id */ 380 uint64_t spa_deadman_calls; /* number of deadman calls */ 381 hrtime_t spa_sync_starttime; /* starting time fo spa_sync */ 382 uint64_t spa_deadman_synctime; /* deadman expiration timer */ 383 uint64_t spa_all_vdev_zaps; /* ZAP of per-vd ZAP obj #s */ 384 spa_avz_action_t spa_avz_action; /* destroy/rebuild AVZ? */ 385 uint64_t spa_autotrim; /* automatic background trim? */ 386 spa_keystore_t spa_keystore; /* loaded crypto keys */ 387 uint64_t spa_errata; /* errata issues detected */ 388 389 /* 390 * spa_iokstat_lock protects spa_iokstat and 391 * spa_queue_stats[]. 392 */ 393 kmutex_t spa_iokstat_lock; 394 struct kstat *spa_iokstat; /* kstat of io to this pool */ 395 struct { 396 int spa_active; 397 int spa_queued; 398 } spa_queue_stats[ZIO_PRIORITY_NUM_QUEUEABLE]; 399 400 /* arc_memory_throttle() parameters during low memory condition */ 401 uint64_t spa_lowmem_page_load; /* memory load during txg */ 402 uint64_t spa_lowmem_last_txg; /* txg window start */ 403 404 hrtime_t spa_ccw_fail_time; /* Conf cache write fail time */ 405 406 uint64_t spa_multihost; /* multihost aware (mmp) */ 407 mmp_thread_t spa_mmp; /* multihost mmp thread */ 408 list_t spa_leaf_list; /* list of leaf vdevs */ 409 uint64_t spa_leaf_list_gen; /* track leaf_list changes */ 410 411 /* 412 * spa_refcount & spa_config_lock must be the last elements 413 * because refcount_t changes size based on compilation options. 414 * because zfs_refcount_t changes size based on compilation options. 415 * In order for the MDB module to function correctly, the other 416 * fields must remain in the same location. 417 */ 418 spa_config_lock_t spa_config_lock[SCL_LOCKS]; /* config changes */ 419 zfs_refcount_t spa_refcount; /* number of opens */ 420 421 taskq_t *spa_upgrade_taskq; /* taskq for upgrade jobs */ 422 }; 423 424 extern const char *spa_config_path; 425 426 extern void spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 427 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent); 428 extern void spa_load_spares(spa_t *spa); 429 extern void spa_load_l2cache(spa_t *spa); 430 431 #ifdef __cplusplus 432 } 433 #endif 434 435 #endif /* _SYS_SPA_IMPL_H */ 436