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, 2018 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 */ 29 30 #ifndef _SYS_SPA_IMPL_H 31 #define _SYS_SPA_IMPL_H 32 33 #include <sys/spa.h> 34 #include <sys/spa_checkpoint.h> 35 #include <sys/vdev.h> 36 #include <sys/vdev_removal.h> 37 #include <sys/metaslab.h> 38 #include <sys/dmu.h> 39 #include <sys/dsl_pool.h> 40 #include <sys/uberblock_impl.h> 41 #include <sys/zfs_context.h> 42 #include <sys/avl.h> 43 #include <sys/refcount.h> 44 #include <sys/bplist.h> 45 #include <sys/bpobj.h> 46 #include <sys/zfeature.h> 47 #include <sys/zthr.h> 48 #include <zfeature_common.h> 49 50 #ifdef __cplusplus 51 extern "C" { 52 #endif 53 54 typedef struct spa_error_entry { 55 zbookmark_phys_t se_bookmark; 56 char *se_name; 57 avl_node_t se_avl; 58 } spa_error_entry_t; 59 60 typedef struct spa_history_phys { 61 uint64_t sh_pool_create_len; /* ending offset of zpool create */ 62 uint64_t sh_phys_max_off; /* physical EOF */ 63 uint64_t sh_bof; /* logical BOF */ 64 uint64_t sh_eof; /* logical EOF */ 65 uint64_t sh_records_lost; /* num of records overwritten */ 66 } spa_history_phys_t; 67 68 /* 69 * All members must be uint64_t, for byteswap purposes. 70 */ 71 typedef struct spa_removing_phys { 72 uint64_t sr_state; /* dsl_scan_state_t */ 73 74 /* 75 * The vdev ID that we most recently attempted to remove, 76 * or -1 if no removal has been attempted. 77 */ 78 uint64_t sr_removing_vdev; 79 80 /* 81 * The vdev ID that we most recently successfully removed, 82 * or -1 if no devices have been removed. 83 */ 84 uint64_t sr_prev_indirect_vdev; 85 86 uint64_t sr_start_time; 87 uint64_t sr_end_time; 88 89 /* 90 * Note that we can not use the space map's or indirect mapping's 91 * accounting as a substitute for these values, because we need to 92 * count frees of not-yet-copied data as though it did the copy. 93 * Otherwise, we could get into a situation where copied > to_copy, 94 * or we complete before copied == to_copy. 95 */ 96 uint64_t sr_to_copy; /* bytes that need to be copied */ 97 uint64_t sr_copied; /* bytes that have been copied or freed */ 98 } spa_removing_phys_t; 99 100 /* 101 * This struct is stored as an entry in the DMU_POOL_DIRECTORY_OBJECT 102 * (with key DMU_POOL_CONDENSING_INDIRECT). It is present if a condense 103 * of an indirect vdev's mapping object is in progress. 104 */ 105 typedef struct spa_condensing_indirect_phys { 106 /* 107 * The vdev ID of the indirect vdev whose indirect mapping is 108 * being condensed. 109 */ 110 uint64_t scip_vdev; 111 112 /* 113 * The vdev's old obsolete spacemap. This spacemap's contents are 114 * being integrated into the new mapping. 115 */ 116 uint64_t scip_prev_obsolete_sm_object; 117 118 /* 119 * The new mapping object that is being created. 120 */ 121 uint64_t scip_next_mapping_object; 122 } spa_condensing_indirect_phys_t; 123 124 struct spa_aux_vdev { 125 uint64_t sav_object; /* MOS object for device list */ 126 nvlist_t *sav_config; /* cached device config */ 127 vdev_t **sav_vdevs; /* devices */ 128 int sav_count; /* number devices */ 129 boolean_t sav_sync; /* sync the device list */ 130 nvlist_t **sav_pending; /* pending device additions */ 131 uint_t sav_npending; /* # pending devices */ 132 }; 133 134 typedef struct spa_config_lock { 135 kmutex_t scl_lock; 136 kthread_t *scl_writer; 137 int scl_write_wanted; 138 kcondvar_t scl_cv; 139 zfs_refcount_t scl_count; 140 } spa_config_lock_t; 141 142 typedef struct spa_config_dirent { 143 list_node_t scd_link; 144 char *scd_path; 145 } spa_config_dirent_t; 146 147 typedef enum zio_taskq_type { 148 ZIO_TASKQ_ISSUE = 0, 149 ZIO_TASKQ_ISSUE_HIGH, 150 ZIO_TASKQ_INTERRUPT, 151 ZIO_TASKQ_INTERRUPT_HIGH, 152 ZIO_TASKQ_TYPES 153 } zio_taskq_type_t; 154 155 /* 156 * State machine for the zpool-poolname process. The states transitions 157 * are done as follows: 158 * 159 * From To Routine 160 * PROC_NONE -> PROC_CREATED spa_activate() 161 * PROC_CREATED -> PROC_ACTIVE spa_thread() 162 * PROC_ACTIVE -> PROC_DEACTIVATE spa_deactivate() 163 * PROC_DEACTIVATE -> PROC_GONE spa_thread() 164 * PROC_GONE -> PROC_NONE spa_deactivate() 165 */ 166 typedef enum spa_proc_state { 167 SPA_PROC_NONE, /* spa_proc = &p0, no process created */ 168 SPA_PROC_CREATED, /* spa_activate() has proc, is waiting */ 169 SPA_PROC_ACTIVE, /* taskqs created, spa_proc set */ 170 SPA_PROC_DEACTIVATE, /* spa_deactivate() requests process exit */ 171 SPA_PROC_GONE /* spa_thread() is exiting, spa_proc = &p0 */ 172 } spa_proc_state_t; 173 174 typedef struct spa_taskqs { 175 uint_t stqs_count; 176 taskq_t **stqs_taskq; 177 } spa_taskqs_t; 178 179 typedef enum spa_all_vdev_zap_action { 180 AVZ_ACTION_NONE = 0, 181 AVZ_ACTION_DESTROY, /* Destroy all per-vdev ZAPs and the AVZ. */ 182 AVZ_ACTION_REBUILD, /* Populate the new AVZ, see spa_avz_rebuild */ 183 AVZ_ACTION_INITIALIZE 184 } spa_avz_action_t; 185 186 typedef enum spa_config_source { 187 SPA_CONFIG_SRC_NONE = 0, 188 SPA_CONFIG_SRC_SCAN, /* scan of path (default: /dev/dsk) */ 189 SPA_CONFIG_SRC_CACHEFILE, /* any cachefile */ 190 SPA_CONFIG_SRC_TRYIMPORT, /* returned from call to tryimport */ 191 SPA_CONFIG_SRC_SPLIT, /* new pool in a pool split */ 192 SPA_CONFIG_SRC_MOS /* MOS, but not always from right txg */ 193 } spa_config_source_t; 194 195 struct spa { 196 /* 197 * Fields protected by spa_namespace_lock. 198 */ 199 char spa_name[ZFS_MAX_DATASET_NAME_LEN]; /* pool name */ 200 char *spa_comment; /* comment */ 201 avl_node_t spa_avl; /* node in spa_namespace_avl */ 202 nvlist_t *spa_config; /* last synced config */ 203 nvlist_t *spa_config_syncing; /* currently syncing config */ 204 nvlist_t *spa_config_splitting; /* config for splitting */ 205 nvlist_t *spa_load_info; /* info and errors from load */ 206 uint64_t spa_config_txg; /* txg of last config change */ 207 int spa_sync_pass; /* iterate-to-convergence */ 208 pool_state_t spa_state; /* pool state */ 209 int spa_inject_ref; /* injection references */ 210 uint8_t spa_sync_on; /* sync threads are running */ 211 spa_load_state_t spa_load_state; /* current load operation */ 212 boolean_t spa_indirect_vdevs_loaded; /* mappings loaded? */ 213 boolean_t spa_trust_config; /* do we trust vdev tree? */ 214 spa_config_source_t spa_config_source; /* where config comes from? */ 215 uint64_t spa_import_flags; /* import specific flags */ 216 spa_taskqs_t spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES]; 217 dsl_pool_t *spa_dsl_pool; 218 boolean_t spa_is_initializing; /* true while opening pool */ 219 metaslab_class_t *spa_normal_class; /* normal data class */ 220 metaslab_class_t *spa_log_class; /* intent log data class */ 221 uint64_t spa_first_txg; /* first txg after spa_open() */ 222 uint64_t spa_final_txg; /* txg of export/destroy */ 223 uint64_t spa_freeze_txg; /* freeze pool at this txg */ 224 uint64_t spa_load_max_txg; /* best initial ub_txg */ 225 uint64_t spa_claim_max_txg; /* highest claimed birth txg */ 226 timespec_t spa_loaded_ts; /* 1st successful open time */ 227 objset_t *spa_meta_objset; /* copy of dp->dp_meta_objset */ 228 kmutex_t spa_evicting_os_lock; /* Evicting objset list lock */ 229 list_t spa_evicting_os_list; /* Objsets being evicted. */ 230 kcondvar_t spa_evicting_os_cv; /* Objset Eviction Completion */ 231 txg_list_t spa_vdev_txg_list; /* per-txg dirty vdev list */ 232 vdev_t *spa_root_vdev; /* top-level vdev container */ 233 int spa_min_ashift; /* of vdevs in normal class */ 234 int spa_max_ashift; /* of vdevs in normal class */ 235 uint64_t spa_config_guid; /* config pool guid */ 236 uint64_t spa_load_guid; /* spa_load initialized guid */ 237 uint64_t spa_last_synced_guid; /* last synced guid */ 238 list_t spa_config_dirty_list; /* vdevs with dirty config */ 239 list_t spa_state_dirty_list; /* vdevs with dirty state */ 240 /* 241 * spa_alloc_locks and spa_alloc_trees are arrays, whose lengths are 242 * stored in spa_alloc_count. There is one tree and one lock for each 243 * allocator, to help improve allocation performance in write-heavy 244 * workloads. 245 */ 246 kmutex_t *spa_alloc_locks; 247 avl_tree_t *spa_alloc_trees; 248 int spa_alloc_count; 249 250 spa_aux_vdev_t spa_spares; /* hot spares */ 251 spa_aux_vdev_t spa_l2cache; /* L2ARC cache devices */ 252 nvlist_t *spa_label_features; /* Features for reading MOS */ 253 uint64_t spa_config_object; /* MOS object for pool config */ 254 uint64_t spa_config_generation; /* config generation number */ 255 uint64_t spa_syncing_txg; /* txg currently syncing */ 256 bpobj_t spa_deferred_bpobj; /* deferred-free bplist */ 257 bplist_t spa_free_bplist[TXG_SIZE]; /* bplist of stuff to free */ 258 zio_cksum_salt_t spa_cksum_salt; /* secret salt for cksum */ 259 /* checksum context templates */ 260 kmutex_t spa_cksum_tmpls_lock; 261 void *spa_cksum_tmpls[ZIO_CHECKSUM_FUNCTIONS]; 262 uberblock_t spa_ubsync; /* last synced uberblock */ 263 uberblock_t spa_uberblock; /* current uberblock */ 264 boolean_t spa_extreme_rewind; /* rewind past deferred frees */ 265 uint64_t spa_last_io; /* lbolt of last non-scan I/O */ 266 kmutex_t spa_scrub_lock; /* resilver/scrub lock */ 267 uint64_t spa_scrub_inflight; /* in-flight scrub I/Os */ 268 kcondvar_t spa_scrub_io_cv; /* scrub I/O completion */ 269 uint8_t spa_scrub_active; /* active or suspended? */ 270 uint8_t spa_scrub_type; /* type of scrub we're doing */ 271 uint8_t spa_scrub_finished; /* indicator to rotate logs */ 272 uint8_t spa_scrub_started; /* started since last boot */ 273 uint8_t spa_scrub_reopen; /* scrub doing vdev_reopen */ 274 uint64_t spa_scan_pass_start; /* start time per pass/reboot */ 275 uint64_t spa_scan_pass_scrub_pause; /* scrub pause time */ 276 uint64_t spa_scan_pass_scrub_spent_paused; /* total paused */ 277 uint64_t spa_scan_pass_exam; /* examined bytes per pass */ 278 kmutex_t spa_async_lock; /* protect async state */ 279 kthread_t *spa_async_thread; /* thread doing async task */ 280 int spa_async_suspended; /* async tasks suspended */ 281 kcondvar_t spa_async_cv; /* wait for thread_exit() */ 282 uint16_t spa_async_tasks; /* async task mask */ 283 uint64_t spa_missing_tvds; /* unopenable tvds on load */ 284 uint64_t spa_missing_tvds_allowed; /* allow loading spa? */ 285 286 spa_removing_phys_t spa_removing_phys; 287 spa_vdev_removal_t *spa_vdev_removal; 288 289 spa_condensing_indirect_phys_t spa_condensing_indirect_phys; 290 spa_condensing_indirect_t *spa_condensing_indirect; 291 zthr_t *spa_condense_zthr; /* zthr doing condense. */ 292 293 uint64_t spa_checkpoint_txg; /* the txg of the checkpoint */ 294 spa_checkpoint_info_t spa_checkpoint_info; /* checkpoint accounting */ 295 zthr_t *spa_checkpoint_discard_zthr; 296 297 char *spa_root; /* alternate root directory */ 298 uint64_t spa_ena; /* spa-wide ereport ENA */ 299 int spa_last_open_failed; /* error if last open failed */ 300 uint64_t spa_last_ubsync_txg; /* "best" uberblock txg */ 301 uint64_t spa_last_ubsync_txg_ts; /* timestamp from that ub */ 302 uint64_t spa_load_txg; /* ub txg that loaded */ 303 uint64_t spa_load_txg_ts; /* timestamp from that ub */ 304 uint64_t spa_load_meta_errors; /* verify metadata err count */ 305 uint64_t spa_load_data_errors; /* verify data err count */ 306 uint64_t spa_verify_min_txg; /* start txg of verify scrub */ 307 kmutex_t spa_errlog_lock; /* error log lock */ 308 uint64_t spa_errlog_last; /* last error log object */ 309 uint64_t spa_errlog_scrub; /* scrub error log object */ 310 kmutex_t spa_errlist_lock; /* error list/ereport lock */ 311 avl_tree_t spa_errlist_last; /* last error list */ 312 avl_tree_t spa_errlist_scrub; /* scrub error list */ 313 uint64_t spa_deflate; /* should we deflate? */ 314 uint64_t spa_history; /* history object */ 315 kmutex_t spa_history_lock; /* history lock */ 316 vdev_t *spa_pending_vdev; /* pending vdev additions */ 317 kmutex_t spa_props_lock; /* property lock */ 318 uint64_t spa_pool_props_object; /* object for properties */ 319 uint64_t spa_bootfs; /* default boot filesystem */ 320 uint64_t spa_failmode; /* failure mode for the pool */ 321 uint64_t spa_delegation; /* delegation on/off */ 322 list_t spa_config_list; /* previous cache file(s) */ 323 /* per-CPU array of root of async I/O: */ 324 zio_t **spa_async_zio_root; 325 zio_t *spa_suspend_zio_root; /* root of all suspended I/O */ 326 zio_t *spa_txg_zio[TXG_SIZE]; /* spa_sync() waits for this */ 327 kmutex_t spa_suspend_lock; /* protects suspend_zio_root */ 328 kcondvar_t spa_suspend_cv; /* notification of resume */ 329 zio_suspend_reason_t spa_suspended; /* pool is suspended */ 330 uint8_t spa_claiming; /* pool is doing zil_claim() */ 331 boolean_t spa_is_root; /* pool is root */ 332 int spa_minref; /* num refs when first opened */ 333 int spa_mode; /* FREAD | FWRITE */ 334 spa_log_state_t spa_log_state; /* log state */ 335 uint64_t spa_autoexpand; /* lun expansion on/off */ 336 uint64_t spa_bootsize; /* efi system partition size */ 337 ddt_t *spa_ddt[ZIO_CHECKSUM_FUNCTIONS]; /* in-core DDTs */ 338 uint64_t spa_ddt_stat_object; /* DDT statistics */ 339 uint64_t spa_dedup_ditto; /* dedup ditto threshold */ 340 uint64_t spa_dedup_checksum; /* default dedup checksum */ 341 uint64_t spa_dspace; /* dspace in normal class */ 342 kmutex_t spa_vdev_top_lock; /* dueling offline/remove */ 343 kmutex_t spa_proc_lock; /* protects spa_proc* */ 344 kcondvar_t spa_proc_cv; /* spa_proc_state transitions */ 345 spa_proc_state_t spa_proc_state; /* see definition */ 346 struct proc *spa_proc; /* "zpool-poolname" process */ 347 uint64_t spa_did; /* if procp != p0, did of t1 */ 348 boolean_t spa_autoreplace; /* autoreplace set in open */ 349 int spa_vdev_locks; /* locks grabbed */ 350 uint64_t spa_creation_version; /* version at pool creation */ 351 uint64_t spa_prev_software_version; /* See ub_software_version */ 352 uint64_t spa_feat_for_write_obj; /* required to write to pool */ 353 uint64_t spa_feat_for_read_obj; /* required to read from pool */ 354 uint64_t spa_feat_desc_obj; /* Feature descriptions */ 355 uint64_t spa_feat_enabled_txg_obj; /* Feature enabled txg */ 356 /* cache feature refcounts */ 357 uint64_t spa_feat_refcount_cache[SPA_FEATURES]; 358 cyclic_id_t spa_deadman_cycid; /* cyclic id */ 359 uint64_t spa_deadman_calls; /* number of deadman calls */ 360 hrtime_t spa_sync_starttime; /* starting time fo spa_sync */ 361 uint64_t spa_deadman_synctime; /* deadman expiration timer */ 362 uint64_t spa_all_vdev_zaps; /* ZAP of per-vd ZAP obj #s */ 363 spa_avz_action_t spa_avz_action; /* destroy/rebuild AVZ? */ 364 365 /* 366 * spa_iokstat_lock protects spa_iokstat and 367 * spa_queue_stats[]. 368 */ 369 kmutex_t spa_iokstat_lock; 370 struct kstat *spa_iokstat; /* kstat of io to this pool */ 371 struct { 372 int spa_active; 373 int spa_queued; 374 } spa_queue_stats[ZIO_PRIORITY_NUM_QUEUEABLE]; 375 376 /* arc_memory_throttle() parameters during low memory condition */ 377 uint64_t spa_lowmem_page_load; /* memory load during txg */ 378 uint64_t spa_lowmem_last_txg; /* txg window start */ 379 380 hrtime_t spa_ccw_fail_time; /* Conf cache write fail time */ 381 382 uint64_t spa_multihost; /* multihost aware (mmp) */ 383 mmp_thread_t spa_mmp; /* multihost mmp thread */ 384 list_t spa_leaf_list; /* list of leaf vdevs */ 385 uint64_t spa_leaf_list_gen; /* track leaf_list changes */ 386 387 /* 388 * spa_refcount & spa_config_lock must be the last elements 389 * because refcount_t changes size based on compilation options. 390 * because zfs_refcount_t changes size based on compilation options. 391 * In order for the MDB module to function correctly, the other 392 * fields must remain in the same location. 393 */ 394 spa_config_lock_t spa_config_lock[SCL_LOCKS]; /* config changes */ 395 zfs_refcount_t spa_refcount; /* number of opens */ 396 }; 397 398 extern const char *spa_config_path; 399 400 extern void spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 401 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent); 402 extern void spa_load_spares(spa_t *spa); 403 extern void spa_load_l2cache(spa_t *spa); 404 405 #ifdef __cplusplus 406 } 407 #endif 408 409 #endif /* _SYS_SPA_IMPL_H */ 410