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) 2016 Actifio, Inc. All rights reserved. 28 * Copyright (c) 2017 Datto Inc. 29 * Copyright (c) 2017, Intel Corporation. 30 */ 31 32 #ifndef _SYS_SPA_IMPL_H 33 #define _SYS_SPA_IMPL_H 34 35 #include <sys/spa.h> 36 #include <sys/spa_checkpoint.h> 37 #include <sys/spa_log_spacemap.h> 38 #include <sys/vdev.h> 39 #include <sys/vdev_rebuild.h> 40 #include <sys/vdev_removal.h> 41 #include <sys/metaslab.h> 42 #include <sys/dmu.h> 43 #include <sys/dsl_pool.h> 44 #include <sys/uberblock_impl.h> 45 #include <sys/zfs_context.h> 46 #include <sys/avl.h> 47 #include <sys/zfs_refcount.h> 48 #include <sys/bplist.h> 49 #include <sys/bpobj.h> 50 #include <sys/dsl_crypt.h> 51 #include <sys/zfeature.h> 52 #include <sys/zthr.h> 53 #include <sys/dsl_deadlist.h> 54 #include <zfeature_common.h> 55 56 #ifdef __cplusplus 57 extern "C" { 58 #endif 59 60 typedef struct spa_alloc { 61 kmutex_t spaa_lock; 62 avl_tree_t spaa_tree; 63 } ____cacheline_aligned spa_alloc_t; 64 65 typedef struct spa_error_entry { 66 zbookmark_phys_t se_bookmark; 67 char *se_name; 68 avl_node_t se_avl; 69 } spa_error_entry_t; 70 71 typedef struct spa_history_phys { 72 uint64_t sh_pool_create_len; /* ending offset of zpool create */ 73 uint64_t sh_phys_max_off; /* physical EOF */ 74 uint64_t sh_bof; /* logical BOF */ 75 uint64_t sh_eof; /* logical EOF */ 76 uint64_t sh_records_lost; /* num of records overwritten */ 77 } spa_history_phys_t; 78 79 /* 80 * All members must be uint64_t, for byteswap purposes. 81 */ 82 typedef struct spa_removing_phys { 83 uint64_t sr_state; /* dsl_scan_state_t */ 84 85 /* 86 * The vdev ID that we most recently attempted to remove, 87 * or -1 if no removal has been attempted. 88 */ 89 uint64_t sr_removing_vdev; 90 91 /* 92 * The vdev ID that we most recently successfully removed, 93 * or -1 if no devices have been removed. 94 */ 95 uint64_t sr_prev_indirect_vdev; 96 97 uint64_t sr_start_time; 98 uint64_t sr_end_time; 99 100 /* 101 * Note that we can not use the space map's or indirect mapping's 102 * accounting as a substitute for these values, because we need to 103 * count frees of not-yet-copied data as though it did the copy. 104 * Otherwise, we could get into a situation where copied > to_copy, 105 * or we complete before copied == to_copy. 106 */ 107 uint64_t sr_to_copy; /* bytes that need to be copied */ 108 uint64_t sr_copied; /* bytes that have been copied or freed */ 109 } spa_removing_phys_t; 110 111 /* 112 * This struct is stored as an entry in the DMU_POOL_DIRECTORY_OBJECT 113 * (with key DMU_POOL_CONDENSING_INDIRECT). It is present if a condense 114 * of an indirect vdev's mapping object is in progress. 115 */ 116 typedef struct spa_condensing_indirect_phys { 117 /* 118 * The vdev ID of the indirect vdev whose indirect mapping is 119 * being condensed. 120 */ 121 uint64_t scip_vdev; 122 123 /* 124 * The vdev's old obsolete spacemap. This spacemap's contents are 125 * being integrated into the new mapping. 126 */ 127 uint64_t scip_prev_obsolete_sm_object; 128 129 /* 130 * The new mapping object that is being created. 131 */ 132 uint64_t scip_next_mapping_object; 133 } spa_condensing_indirect_phys_t; 134 135 struct spa_aux_vdev { 136 uint64_t sav_object; /* MOS object for device list */ 137 nvlist_t *sav_config; /* cached device config */ 138 vdev_t **sav_vdevs; /* devices */ 139 int sav_count; /* number devices */ 140 boolean_t sav_sync; /* sync the device list */ 141 nvlist_t **sav_pending; /* pending device additions */ 142 uint_t sav_npending; /* # pending devices */ 143 }; 144 145 typedef struct spa_config_lock { 146 kmutex_t scl_lock; 147 kthread_t *scl_writer; 148 int scl_write_wanted; 149 int scl_count; 150 kcondvar_t scl_cv; 151 } ____cacheline_aligned spa_config_lock_t; 152 153 typedef struct spa_config_dirent { 154 list_node_t scd_link; 155 char *scd_path; 156 } spa_config_dirent_t; 157 158 typedef enum zio_taskq_type { 159 ZIO_TASKQ_ISSUE = 0, 160 ZIO_TASKQ_ISSUE_HIGH, 161 ZIO_TASKQ_INTERRUPT, 162 ZIO_TASKQ_INTERRUPT_HIGH, 163 ZIO_TASKQ_TYPES 164 } zio_taskq_type_t; 165 166 /* 167 * State machine for the zpool-poolname process. The states transitions 168 * are done as follows: 169 * 170 * From To Routine 171 * PROC_NONE -> PROC_CREATED spa_activate() 172 * PROC_CREATED -> PROC_ACTIVE spa_thread() 173 * PROC_ACTIVE -> PROC_DEACTIVATE spa_deactivate() 174 * PROC_DEACTIVATE -> PROC_GONE spa_thread() 175 * PROC_GONE -> PROC_NONE spa_deactivate() 176 */ 177 typedef enum spa_proc_state { 178 SPA_PROC_NONE, /* spa_proc = &p0, no process created */ 179 SPA_PROC_CREATED, /* spa_activate() has proc, is waiting */ 180 SPA_PROC_ACTIVE, /* taskqs created, spa_proc set */ 181 SPA_PROC_DEACTIVATE, /* spa_deactivate() requests process exit */ 182 SPA_PROC_GONE /* spa_thread() is exiting, spa_proc = &p0 */ 183 } spa_proc_state_t; 184 185 typedef struct spa_taskqs { 186 uint_t stqs_count; 187 taskq_t **stqs_taskq; 188 } spa_taskqs_t; 189 190 typedef enum spa_all_vdev_zap_action { 191 AVZ_ACTION_NONE = 0, 192 AVZ_ACTION_DESTROY, /* Destroy all per-vdev ZAPs and the AVZ. */ 193 AVZ_ACTION_REBUILD, /* Populate the new AVZ, see spa_avz_rebuild */ 194 AVZ_ACTION_INITIALIZE 195 } spa_avz_action_t; 196 197 typedef enum spa_config_source { 198 SPA_CONFIG_SRC_NONE = 0, 199 SPA_CONFIG_SRC_SCAN, /* scan of path (default: /dev/dsk) */ 200 SPA_CONFIG_SRC_CACHEFILE, /* any cachefile */ 201 SPA_CONFIG_SRC_TRYIMPORT, /* returned from call to tryimport */ 202 SPA_CONFIG_SRC_SPLIT, /* new pool in a pool split */ 203 SPA_CONFIG_SRC_MOS /* MOS, but not always from right txg */ 204 } spa_config_source_t; 205 206 struct spa { 207 /* 208 * Fields protected by spa_namespace_lock. 209 */ 210 char spa_name[ZFS_MAX_DATASET_NAME_LEN]; /* pool name */ 211 char *spa_comment; /* comment */ 212 avl_node_t spa_avl; /* node in spa_namespace_avl */ 213 nvlist_t *spa_config; /* last synced config */ 214 nvlist_t *spa_config_syncing; /* currently syncing config */ 215 nvlist_t *spa_config_splitting; /* config for splitting */ 216 nvlist_t *spa_load_info; /* info and errors from load */ 217 uint64_t spa_config_txg; /* txg of last config change */ 218 int spa_sync_pass; /* iterate-to-convergence */ 219 pool_state_t spa_state; /* pool state */ 220 int spa_inject_ref; /* injection references */ 221 uint8_t spa_sync_on; /* sync threads are running */ 222 spa_load_state_t spa_load_state; /* current load operation */ 223 boolean_t spa_indirect_vdevs_loaded; /* mappings loaded? */ 224 boolean_t spa_trust_config; /* do we trust vdev tree? */ 225 boolean_t spa_is_splitting; /* in the middle of a split? */ 226 spa_config_source_t spa_config_source; /* where config comes from? */ 227 uint64_t spa_import_flags; /* import specific flags */ 228 spa_taskqs_t spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES]; 229 dsl_pool_t *spa_dsl_pool; 230 boolean_t spa_is_initializing; /* true while opening pool */ 231 boolean_t spa_is_exporting; /* true while exporting pool */ 232 metaslab_class_t *spa_normal_class; /* normal data class */ 233 metaslab_class_t *spa_log_class; /* intent log data class */ 234 metaslab_class_t *spa_embedded_log_class; /* log on normal vdevs */ 235 metaslab_class_t *spa_special_class; /* special allocation class */ 236 metaslab_class_t *spa_dedup_class; /* dedup allocation class */ 237 uint64_t spa_first_txg; /* first txg after spa_open() */ 238 uint64_t spa_final_txg; /* txg of export/destroy */ 239 uint64_t spa_freeze_txg; /* freeze pool at this txg */ 240 uint64_t spa_load_max_txg; /* best initial ub_txg */ 241 uint64_t spa_claim_max_txg; /* highest claimed birth txg */ 242 inode_timespec_t spa_loaded_ts; /* 1st successful open time */ 243 objset_t *spa_meta_objset; /* copy of dp->dp_meta_objset */ 244 kmutex_t spa_evicting_os_lock; /* Evicting objset list lock */ 245 list_t spa_evicting_os_list; /* Objsets being evicted. */ 246 kcondvar_t spa_evicting_os_cv; /* Objset Eviction Completion */ 247 txg_list_t spa_vdev_txg_list; /* per-txg dirty vdev list */ 248 vdev_t *spa_root_vdev; /* top-level vdev container */ 249 uint64_t spa_min_ashift; /* of vdevs in normal class */ 250 uint64_t spa_max_ashift; /* of vdevs in normal class */ 251 uint64_t spa_min_alloc; /* of vdevs in normal class */ 252 uint64_t spa_config_guid; /* config pool guid */ 253 uint64_t spa_load_guid; /* spa_load initialized guid */ 254 uint64_t spa_last_synced_guid; /* last synced guid */ 255 list_t spa_config_dirty_list; /* vdevs with dirty config */ 256 list_t spa_state_dirty_list; /* vdevs with dirty state */ 257 /* 258 * spa_allocs is an array, whose lengths is stored in spa_alloc_count. 259 * There is one tree and one lock for each allocator, to help improve 260 * allocation performance in write-heavy workloads. 261 */ 262 spa_alloc_t *spa_allocs; 263 int spa_alloc_count; 264 265 spa_aux_vdev_t spa_spares; /* hot spares */ 266 spa_aux_vdev_t spa_l2cache; /* L2ARC cache devices */ 267 nvlist_t *spa_label_features; /* Features for reading MOS */ 268 uint64_t spa_config_object; /* MOS object for pool config */ 269 uint64_t spa_config_generation; /* config generation number */ 270 uint64_t spa_syncing_txg; /* txg currently syncing */ 271 bpobj_t spa_deferred_bpobj; /* deferred-free bplist */ 272 bplist_t spa_free_bplist[TXG_SIZE]; /* bplist of stuff to free */ 273 zio_cksum_salt_t spa_cksum_salt; /* secret salt for cksum */ 274 /* checksum context templates */ 275 kmutex_t spa_cksum_tmpls_lock; 276 void *spa_cksum_tmpls[ZIO_CHECKSUM_FUNCTIONS]; 277 uberblock_t spa_ubsync; /* last synced uberblock */ 278 uberblock_t spa_uberblock; /* current uberblock */ 279 boolean_t spa_extreme_rewind; /* rewind past deferred frees */ 280 kmutex_t spa_scrub_lock; /* resilver/scrub lock */ 281 uint64_t spa_scrub_inflight; /* in-flight scrub bytes */ 282 283 /* in-flight verification bytes */ 284 uint64_t spa_load_verify_bytes; 285 kcondvar_t spa_scrub_io_cv; /* scrub I/O completion */ 286 uint8_t spa_scrub_active; /* active or suspended? */ 287 uint8_t spa_scrub_type; /* type of scrub we're doing */ 288 uint8_t spa_scrub_finished; /* indicator to rotate logs */ 289 uint8_t spa_scrub_started; /* started since last boot */ 290 uint8_t spa_scrub_reopen; /* scrub doing vdev_reopen */ 291 uint64_t spa_scan_pass_start; /* start time per pass/reboot */ 292 uint64_t spa_scan_pass_scrub_pause; /* scrub pause time */ 293 uint64_t spa_scan_pass_scrub_spent_paused; /* total paused */ 294 uint64_t spa_scan_pass_exam; /* examined bytes per pass */ 295 uint64_t spa_scan_pass_issued; /* issued bytes per pass */ 296 297 /* 298 * We are in the middle of a resilver, and another resilver 299 * is needed once this one completes. This is set iff any 300 * vdev_resilver_deferred is set. 301 */ 302 boolean_t spa_resilver_deferred; 303 kmutex_t spa_async_lock; /* protect async state */ 304 kthread_t *spa_async_thread; /* thread doing async task */ 305 int spa_async_suspended; /* async tasks suspended */ 306 kcondvar_t spa_async_cv; /* wait for thread_exit() */ 307 uint16_t spa_async_tasks; /* async task mask */ 308 uint64_t spa_missing_tvds; /* unopenable tvds on load */ 309 uint64_t spa_missing_tvds_allowed; /* allow loading spa? */ 310 311 spa_removing_phys_t spa_removing_phys; 312 spa_vdev_removal_t *spa_vdev_removal; 313 314 spa_condensing_indirect_phys_t spa_condensing_indirect_phys; 315 spa_condensing_indirect_t *spa_condensing_indirect; 316 zthr_t *spa_condense_zthr; /* zthr doing condense. */ 317 318 uint64_t spa_checkpoint_txg; /* the txg of the checkpoint */ 319 spa_checkpoint_info_t spa_checkpoint_info; /* checkpoint accounting */ 320 zthr_t *spa_checkpoint_discard_zthr; 321 322 space_map_t *spa_syncing_log_sm; /* current log space map */ 323 avl_tree_t spa_sm_logs_by_txg; 324 kmutex_t spa_flushed_ms_lock; /* for metaslabs_by_flushed */ 325 avl_tree_t spa_metaslabs_by_flushed; 326 spa_unflushed_stats_t spa_unflushed_stats; 327 list_t spa_log_summary; 328 uint64_t spa_log_flushall_txg; 329 330 zthr_t *spa_livelist_delete_zthr; /* deleting livelists */ 331 zthr_t *spa_livelist_condense_zthr; /* condensing livelists */ 332 uint64_t spa_livelists_to_delete; /* set of livelists to free */ 333 livelist_condense_entry_t spa_to_condense; /* next to condense */ 334 335 char *spa_root; /* alternate root directory */ 336 uint64_t spa_ena; /* spa-wide ereport ENA */ 337 int spa_last_open_failed; /* error if last open failed */ 338 uint64_t spa_last_ubsync_txg; /* "best" uberblock txg */ 339 uint64_t spa_last_ubsync_txg_ts; /* timestamp from that ub */ 340 uint64_t spa_load_txg; /* ub txg that loaded */ 341 uint64_t spa_load_txg_ts; /* timestamp from that ub */ 342 uint64_t spa_load_meta_errors; /* verify metadata err count */ 343 uint64_t spa_load_data_errors; /* verify data err count */ 344 uint64_t spa_verify_min_txg; /* start txg of verify scrub */ 345 kmutex_t spa_errlog_lock; /* error log lock */ 346 uint64_t spa_errlog_last; /* last error log object */ 347 uint64_t spa_errlog_scrub; /* scrub error log object */ 348 kmutex_t spa_errlist_lock; /* error list/ereport lock */ 349 avl_tree_t spa_errlist_last; /* last error list */ 350 avl_tree_t spa_errlist_scrub; /* scrub error list */ 351 uint64_t spa_deflate; /* should we deflate? */ 352 uint64_t spa_history; /* history object */ 353 kmutex_t spa_history_lock; /* history lock */ 354 vdev_t *spa_pending_vdev; /* pending vdev additions */ 355 kmutex_t spa_props_lock; /* property lock */ 356 uint64_t spa_pool_props_object; /* object for properties */ 357 uint64_t spa_bootfs; /* default boot filesystem */ 358 uint64_t spa_failmode; /* failure mode for the pool */ 359 uint64_t spa_deadman_failmode; /* failure mode for deadman */ 360 uint64_t spa_delegation; /* delegation on/off */ 361 list_t spa_config_list; /* previous cache file(s) */ 362 /* per-CPU array of root of async I/O: */ 363 zio_t **spa_async_zio_root; 364 zio_t *spa_suspend_zio_root; /* root of all suspended I/O */ 365 zio_t *spa_txg_zio[TXG_SIZE]; /* spa_sync() waits for this */ 366 kmutex_t spa_suspend_lock; /* protects suspend_zio_root */ 367 kcondvar_t spa_suspend_cv; /* notification of resume */ 368 zio_suspend_reason_t spa_suspended; /* pool is suspended */ 369 uint8_t spa_claiming; /* pool is doing zil_claim() */ 370 boolean_t spa_is_root; /* pool is root */ 371 int spa_minref; /* num refs when first opened */ 372 spa_mode_t spa_mode; /* SPA_MODE_{READ|WRITE} */ 373 boolean_t spa_read_spacemaps; /* spacemaps available if ro */ 374 spa_log_state_t spa_log_state; /* log state */ 375 uint64_t spa_autoexpand; /* lun expansion on/off */ 376 ddt_t *spa_ddt[ZIO_CHECKSUM_FUNCTIONS]; /* in-core DDTs */ 377 uint64_t spa_ddt_stat_object; /* DDT statistics */ 378 uint64_t spa_dedup_dspace; /* Cache get_dedup_dspace() */ 379 uint64_t spa_dedup_checksum; /* default dedup checksum */ 380 uint64_t spa_dspace; /* dspace in normal class */ 381 kmutex_t spa_vdev_top_lock; /* dueling offline/remove */ 382 kmutex_t spa_proc_lock; /* protects spa_proc* */ 383 kcondvar_t spa_proc_cv; /* spa_proc_state transitions */ 384 spa_proc_state_t spa_proc_state; /* see definition */ 385 proc_t *spa_proc; /* "zpool-poolname" process */ 386 uintptr_t spa_did; /* if procp != p0, did of t1 */ 387 boolean_t spa_autoreplace; /* autoreplace set in open */ 388 int spa_vdev_locks; /* locks grabbed */ 389 uint64_t spa_creation_version; /* version at pool creation */ 390 uint64_t spa_prev_software_version; /* See ub_software_version */ 391 uint64_t spa_feat_for_write_obj; /* required to write to pool */ 392 uint64_t spa_feat_for_read_obj; /* required to read from pool */ 393 uint64_t spa_feat_desc_obj; /* Feature descriptions */ 394 uint64_t spa_feat_enabled_txg_obj; /* Feature enabled txg */ 395 kmutex_t spa_feat_stats_lock; /* protects spa_feat_stats */ 396 nvlist_t *spa_feat_stats; /* Cache of enabled features */ 397 /* cache feature refcounts */ 398 uint64_t spa_feat_refcount_cache[SPA_FEATURES]; 399 taskqid_t spa_deadman_tqid; /* Task id */ 400 uint64_t spa_deadman_calls; /* number of deadman calls */ 401 hrtime_t spa_sync_starttime; /* starting time of spa_sync */ 402 uint64_t spa_deadman_synctime; /* deadman sync expiration */ 403 uint64_t spa_deadman_ziotime; /* deadman zio expiration */ 404 uint64_t spa_all_vdev_zaps; /* ZAP of per-vd ZAP obj #s */ 405 spa_avz_action_t spa_avz_action; /* destroy/rebuild AVZ? */ 406 uint64_t spa_autotrim; /* automatic background trim? */ 407 uint64_t spa_errata; /* errata issues detected */ 408 spa_stats_t spa_stats; /* assorted spa statistics */ 409 spa_keystore_t spa_keystore; /* loaded crypto keys */ 410 411 /* arc_memory_throttle() parameters during low memory condition */ 412 uint64_t spa_lowmem_page_load; /* memory load during txg */ 413 uint64_t spa_lowmem_last_txg; /* txg window start */ 414 415 hrtime_t spa_ccw_fail_time; /* Conf cache write fail time */ 416 taskq_t *spa_zvol_taskq; /* Taskq for minor management */ 417 taskq_t *spa_prefetch_taskq; /* Taskq for prefetch threads */ 418 uint64_t spa_multihost; /* multihost aware (mmp) */ 419 mmp_thread_t spa_mmp; /* multihost mmp thread */ 420 list_t spa_leaf_list; /* list of leaf vdevs */ 421 uint64_t spa_leaf_list_gen; /* track leaf_list changes */ 422 uint32_t spa_hostid; /* cached system hostid */ 423 424 /* synchronization for threads in spa_wait */ 425 kmutex_t spa_activities_lock; 426 kcondvar_t spa_activities_cv; 427 kcondvar_t spa_waiters_cv; 428 int spa_waiters; /* number of waiting threads */ 429 boolean_t spa_waiters_cancel; /* waiters should return */ 430 431 char *spa_compatibility; /* compatibility file(s) */ 432 433 /* 434 * spa_refcount & spa_config_lock must be the last elements 435 * because zfs_refcount_t changes size based on compilation options. 436 * In order for the MDB module to function correctly, the other 437 * fields must remain in the same location. 438 */ 439 spa_config_lock_t spa_config_lock[SCL_LOCKS]; /* config changes */ 440 zfs_refcount_t spa_refcount; /* number of opens */ 441 442 taskq_t *spa_upgrade_taskq; /* taskq for upgrade jobs */ 443 }; 444 445 extern char *spa_config_path; 446 extern char *zfs_deadman_failmode; 447 extern int spa_slop_shift; 448 extern void spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 449 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent); 450 extern void spa_taskq_dispatch_sync(spa_t *, zio_type_t t, zio_taskq_type_t q, 451 task_func_t *func, void *arg, uint_t flags); 452 extern void spa_load_spares(spa_t *spa); 453 extern void spa_load_l2cache(spa_t *spa); 454 extern sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, 455 const char *name); 456 extern void spa_event_post(sysevent_t *ev); 457 extern int param_set_deadman_failmode_common(const char *val); 458 extern void spa_set_deadman_synctime(hrtime_t ns); 459 extern void spa_set_deadman_ziotime(hrtime_t ns); 460 extern const char *spa_history_zone(void); 461 462 #ifdef __cplusplus 463 } 464 #endif 465 466 #endif /* _SYS_SPA_IMPL_H */ 467