spa_impl.h revision 307122
1231200Smm/* 2231200Smm * CDDL HEADER START 3231200Smm * 4231200Smm * The contents of this file are subject to the terms of the 5231200Smm * Common Development and Distribution License (the "License"). 6231200Smm * You may not use this file except in compliance with the License. 7231200Smm * 8231200Smm * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9231200Smm * or http://www.opensolaris.org/os/licensing. 10231200Smm * See the License for the specific language governing permissions 11231200Smm * and limitations under the License. 12231200Smm * 13231200Smm * When distributing Covered Code, include this CDDL HEADER in each 14231200Smm * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15231200Smm * If applicable, add the following below this CDDL HEADER, with the 16231200Smm * fields enclosed by brackets "[]" replaced with your own identifying 17231200Smm * information: Portions Copyright [yyyy] [name of copyright owner] 18231200Smm * 19231200Smm * CDDL HEADER END 20231200Smm */ 21231200Smm/* 22231200Smm * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23231200Smm * Copyright (c) 2011, 2015 by Delphix. All rights reserved. 24231200Smm * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 25231200Smm * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 26231200Smm * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 27231200Smm * Copyright 2013 Saso Kiselkov. All rights reserved. 28231200Smm */ 29231200Smm 30231200Smm#ifndef _SYS_SPA_IMPL_H 31231200Smm#define _SYS_SPA_IMPL_H 32231200Smm 33231200Smm#include <sys/spa.h> 34231200Smm#include <sys/vdev.h> 35231200Smm#include <sys/metaslab.h> 36231200Smm#include <sys/dmu.h> 37231200Smm#include <sys/dsl_pool.h> 38231200Smm#include <sys/uberblock_impl.h> 39231200Smm#include <sys/zfs_context.h> 40231200Smm#include <sys/avl.h> 41231200Smm#include <sys/refcount.h> 42231200Smm#include <sys/bplist.h> 43231200Smm#include <sys/bpobj.h> 44231200Smm#include <sys/zfeature.h> 45231200Smm#include <zfeature_common.h> 46231200Smm 47231200Smm#ifdef __cplusplus 48231200Smmextern "C" { 49231200Smm#endif 50231200Smm 51231200Smmtypedef struct spa_error_entry { 52231200Smm zbookmark_phys_t se_bookmark; 53231200Smm char *se_name; 54231200Smm avl_node_t se_avl; 55231200Smm} spa_error_entry_t; 56231200Smm 57231200Smmtypedef struct spa_history_phys { 58231200Smm uint64_t sh_pool_create_len; /* ending offset of zpool create */ 59231200Smm uint64_t sh_phys_max_off; /* physical EOF */ 60231200Smm uint64_t sh_bof; /* logical BOF */ 61231200Smm uint64_t sh_eof; /* logical EOF */ 62231200Smm uint64_t sh_records_lost; /* num of records overwritten */ 63231200Smm} spa_history_phys_t; 64231200Smm 65231200Smmstruct spa_aux_vdev { 66231200Smm uint64_t sav_object; /* MOS object for device list */ 67231200Smm nvlist_t *sav_config; /* cached device config */ 68231200Smm vdev_t **sav_vdevs; /* devices */ 69231200Smm int sav_count; /* number devices */ 70231200Smm boolean_t sav_sync; /* sync the device list */ 71231200Smm nvlist_t **sav_pending; /* pending device additions */ 72231200Smm uint_t sav_npending; /* # pending devices */ 73231200Smm}; 74231200Smm 75231200Smmtypedef struct spa_config_lock { 76231200Smm kmutex_t scl_lock; 77231200Smm kthread_t *scl_writer; 78231200Smm int scl_write_wanted; 79231200Smm kcondvar_t scl_cv; 80231200Smm refcount_t scl_count; 81231200Smm} spa_config_lock_t; 82231200Smm 83231200Smmtypedef struct spa_config_dirent { 84231200Smm list_node_t scd_link; 85231200Smm char *scd_path; 86231200Smm} spa_config_dirent_t; 87231200Smm 88231200Smmtypedef enum zio_taskq_type { 89231200Smm ZIO_TASKQ_ISSUE = 0, 90231200Smm ZIO_TASKQ_ISSUE_HIGH, 91231200Smm ZIO_TASKQ_INTERRUPT, 92231200Smm ZIO_TASKQ_INTERRUPT_HIGH, 93231200Smm ZIO_TASKQ_TYPES 94231200Smm} zio_taskq_type_t; 95231200Smm 96231200Smm/* 97231200Smm * State machine for the zpool-poolname process. The states transitions 98231200Smm * are done as follows: 99231200Smm * 100231200Smm * From To Routine 101231200Smm * PROC_NONE -> PROC_CREATED spa_activate() 102231200Smm * PROC_CREATED -> PROC_ACTIVE spa_thread() 103231200Smm * PROC_ACTIVE -> PROC_DEACTIVATE spa_deactivate() 104231200Smm * PROC_DEACTIVATE -> PROC_GONE spa_thread() 105231200Smm * PROC_GONE -> PROC_NONE spa_deactivate() 106231200Smm */ 107231200Smmtypedef enum spa_proc_state { 108231200Smm SPA_PROC_NONE, /* spa_proc = &p0, no process created */ 109231200Smm SPA_PROC_CREATED, /* spa_activate() has proc, is waiting */ 110231200Smm SPA_PROC_ACTIVE, /* taskqs created, spa_proc set */ 111231200Smm SPA_PROC_DEACTIVATE, /* spa_deactivate() requests process exit */ 112231200Smm SPA_PROC_GONE /* spa_thread() is exiting, spa_proc = &p0 */ 113231200Smm} spa_proc_state_t; 114231200Smm 115231200Smmtypedef struct spa_taskqs { 116231200Smm uint_t stqs_count; 117231200Smm taskq_t **stqs_taskq; 118231200Smm} spa_taskqs_t; 119231200Smm 120231200Smmstruct spa { 121231200Smm /* 122231200Smm * Fields protected by spa_namespace_lock. 123231200Smm */ 124231200Smm char spa_name[ZFS_MAX_DATASET_NAME_LEN]; /* pool name */ 125231200Smm char *spa_comment; /* comment */ 126231200Smm avl_node_t spa_avl; /* node in spa_namespace_avl */ 127231200Smm nvlist_t *spa_config; /* last synced config */ 128231200Smm nvlist_t *spa_config_syncing; /* currently syncing config */ 129231200Smm nvlist_t *spa_config_splitting; /* config for splitting */ 130231200Smm nvlist_t *spa_load_info; /* info and errors from load */ 131231200Smm uint64_t spa_config_txg; /* txg of last config change */ 132231200Smm int spa_sync_pass; /* iterate-to-convergence */ 133231200Smm pool_state_t spa_state; /* pool state */ 134231200Smm int spa_inject_ref; /* injection references */ 135231200Smm uint8_t spa_sync_on; /* sync threads are running */ 136231200Smm spa_load_state_t spa_load_state; /* current load operation */ 137231200Smm uint64_t spa_import_flags; /* import specific flags */ 138231200Smm spa_taskqs_t spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES]; 139231200Smm dsl_pool_t *spa_dsl_pool; 140231200Smm boolean_t spa_is_initializing; /* true while opening pool */ 141231200Smm metaslab_class_t *spa_normal_class; /* normal data class */ 142231200Smm metaslab_class_t *spa_log_class; /* intent log data class */ 143231200Smm uint64_t spa_first_txg; /* first txg after spa_open() */ 144231200Smm uint64_t spa_final_txg; /* txg of export/destroy */ 145231200Smm uint64_t spa_freeze_txg; /* freeze pool at this txg */ 146231200Smm uint64_t spa_load_max_txg; /* best initial ub_txg */ 147231200Smm uint64_t spa_claim_max_txg; /* highest claimed birth txg */ 148231200Smm timespec_t spa_loaded_ts; /* 1st successful open time */ 149231200Smm objset_t *spa_meta_objset; /* copy of dp->dp_meta_objset */ 150231200Smm kmutex_t spa_evicting_os_lock; /* Evicting objset list lock */ 151231200Smm list_t spa_evicting_os_list; /* Objsets being evicted. */ 152231200Smm kcondvar_t spa_evicting_os_cv; /* Objset Eviction Completion */ 153231200Smm txg_list_t spa_vdev_txg_list; /* per-txg dirty vdev list */ 154231200Smm vdev_t *spa_root_vdev; /* top-level vdev container */ 155231200Smm int spa_min_ashift; /* of vdevs in normal class */ 156231200Smm int spa_max_ashift; /* of vdevs in normal class */ 157231200Smm uint64_t spa_config_guid; /* config pool guid */ 158231200Smm uint64_t spa_load_guid; /* spa_load initialized guid */ 159231200Smm uint64_t spa_last_synced_guid; /* last synced guid */ 160231200Smm list_t spa_config_dirty_list; /* vdevs with dirty config */ 161231200Smm list_t spa_state_dirty_list; /* vdevs with dirty state */ 162231200Smm spa_aux_vdev_t spa_spares; /* hot spares */ 163231200Smm spa_aux_vdev_t spa_l2cache; /* L2ARC cache devices */ 164231200Smm nvlist_t *spa_label_features; /* Features for reading MOS */ 165231200Smm uint64_t spa_config_object; /* MOS object for pool config */ 166231200Smm uint64_t spa_config_generation; /* config generation number */ 167231200Smm uint64_t spa_syncing_txg; /* txg currently syncing */ 168231200Smm bpobj_t spa_deferred_bpobj; /* deferred-free bplist */ 169231200Smm bplist_t spa_free_bplist[TXG_SIZE]; /* bplist of stuff to free */ 170231200Smm zio_cksum_salt_t spa_cksum_salt; /* secret salt for cksum */ 171231200Smm /* checksum context templates */ 172231200Smm kmutex_t spa_cksum_tmpls_lock; 173231200Smm void *spa_cksum_tmpls[ZIO_CHECKSUM_FUNCTIONS]; 174231200Smm uberblock_t spa_ubsync; /* last synced uberblock */ 175231200Smm uberblock_t spa_uberblock; /* current uberblock */ 176231200Smm boolean_t spa_extreme_rewind; /* rewind past deferred frees */ 177231200Smm uint64_t spa_last_io; /* lbolt of last non-scan I/O */ 178231200Smm kmutex_t spa_scrub_lock; /* resilver/scrub lock */ 179231200Smm uint64_t spa_scrub_inflight; /* in-flight scrub I/Os */ 180231200Smm kcondvar_t spa_scrub_io_cv; /* scrub I/O completion */ 181231200Smm uint8_t spa_scrub_active; /* active or suspended? */ 182231200Smm uint8_t spa_scrub_type; /* type of scrub we're doing */ 183231200Smm uint8_t spa_scrub_finished; /* indicator to rotate logs */ 184231200Smm uint8_t spa_scrub_started; /* started since last boot */ 185231200Smm uint8_t spa_scrub_reopen; /* scrub doing vdev_reopen */ 186231200Smm uint64_t spa_scan_pass_start; /* start time per pass/reboot */ 187231200Smm uint64_t spa_scan_pass_exam; /* examined bytes per pass */ 188231200Smm kmutex_t spa_async_lock; /* protect async state */ 189231200Smm kthread_t *spa_async_thread; /* thread doing async task */ 190231200Smm kthread_t *spa_async_thread_vd; /* thread doing vd async task */ 191231200Smm int spa_async_suspended; /* async tasks suspended */ 192231200Smm kcondvar_t spa_async_cv; /* wait for thread_exit() */ 193231200Smm uint16_t spa_async_tasks; /* async task mask */ 194231200Smm char *spa_root; /* alternate root directory */ 195231200Smm uint64_t spa_ena; /* spa-wide ereport ENA */ 196231200Smm int spa_last_open_failed; /* error if last open failed */ 197231200Smm uint64_t spa_last_ubsync_txg; /* "best" uberblock txg */ 198231200Smm uint64_t spa_last_ubsync_txg_ts; /* timestamp from that ub */ 199231200Smm uint64_t spa_load_txg; /* ub txg that loaded */ 200231200Smm uint64_t spa_load_txg_ts; /* timestamp from that ub */ 201231200Smm uint64_t spa_load_meta_errors; /* verify metadata err count */ 202231200Smm uint64_t spa_load_data_errors; /* verify data err count */ 203231200Smm uint64_t spa_verify_min_txg; /* start txg of verify scrub */ 204231200Smm kmutex_t spa_errlog_lock; /* error log lock */ 205231200Smm uint64_t spa_errlog_last; /* last error log object */ 206231200Smm uint64_t spa_errlog_scrub; /* scrub error log object */ 207231200Smm kmutex_t spa_errlist_lock; /* error list/ereport lock */ 208231200Smm avl_tree_t spa_errlist_last; /* last error list */ 209231200Smm avl_tree_t spa_errlist_scrub; /* scrub error list */ 210231200Smm uint64_t spa_deflate; /* should we deflate? */ 211231200Smm uint64_t spa_history; /* history object */ 212231200Smm kmutex_t spa_history_lock; /* history lock */ 213231200Smm vdev_t *spa_pending_vdev; /* pending vdev additions */ 214231200Smm kmutex_t spa_props_lock; /* property lock */ 215231200Smm uint64_t spa_pool_props_object; /* object for properties */ 216231200Smm uint64_t spa_bootfs; /* default boot filesystem */ 217231200Smm uint64_t spa_failmode; /* failure mode for the pool */ 218231200Smm uint64_t spa_delegation; /* delegation on/off */ 219231200Smm list_t spa_config_list; /* previous cache file(s) */ 220231200Smm /* per-CPU array of root of async I/O: */ 221231200Smm zio_t **spa_async_zio_root; 222231200Smm zio_t *spa_suspend_zio_root; /* root of all suspended I/O */ 223231200Smm kmutex_t spa_suspend_lock; /* protects suspend_zio_root */ 224231200Smm kcondvar_t spa_suspend_cv; /* notification of resume */ 225231200Smm uint8_t spa_suspended; /* pool is suspended */ 226231200Smm uint8_t spa_claiming; /* pool is doing zil_claim() */ 227231200Smm boolean_t spa_debug; /* debug enabled? */ 228231200Smm boolean_t spa_is_root; /* pool is root */ 229231200Smm int spa_minref; /* num refs when first opened */ 230231200Smm int spa_mode; /* FREAD | FWRITE */ 231231200Smm spa_log_state_t spa_log_state; /* log state */ 232231200Smm uint64_t spa_autoexpand; /* lun expansion on/off */ 233231200Smm ddt_t *spa_ddt[ZIO_CHECKSUM_FUNCTIONS]; /* in-core DDTs */ 234231200Smm uint64_t spa_ddt_stat_object; /* DDT statistics */ 235231200Smm uint64_t spa_dedup_ditto; /* dedup ditto threshold */ 236231200Smm uint64_t spa_dedup_checksum; /* default dedup checksum */ 237231200Smm uint64_t spa_dspace; /* dspace in normal class */ 238231200Smm kmutex_t spa_vdev_top_lock; /* dueling offline/remove */ 239231200Smm kmutex_t spa_proc_lock; /* protects spa_proc* */ 240231200Smm kcondvar_t spa_proc_cv; /* spa_proc_state transitions */ 241231200Smm spa_proc_state_t spa_proc_state; /* see definition */ 242231200Smm struct proc *spa_proc; /* "zpool-poolname" process */ 243 uint64_t spa_did; /* if procp != p0, did of t1 */ 244 kthread_t *spa_trim_thread; /* thread sending TRIM I/Os */ 245 kmutex_t spa_trim_lock; /* protects spa_trim_cv */ 246 kcondvar_t spa_trim_cv; /* used to notify TRIM thread */ 247 boolean_t spa_autoreplace; /* autoreplace set in open */ 248 int spa_vdev_locks; /* locks grabbed */ 249 uint64_t spa_creation_version; /* version at pool creation */ 250 uint64_t spa_prev_software_version; /* See ub_software_version */ 251 uint64_t spa_feat_for_write_obj; /* required to write to pool */ 252 uint64_t spa_feat_for_read_obj; /* required to read from pool */ 253 uint64_t spa_feat_desc_obj; /* Feature descriptions */ 254 uint64_t spa_feat_enabled_txg_obj; /* Feature enabled txg */ 255 /* cache feature refcounts */ 256 uint64_t spa_feat_refcount_cache[SPA_FEATURES]; 257#ifdef illumos 258 cyclic_id_t spa_deadman_cycid; /* cyclic id */ 259#else /* !illumos */ 260#ifdef _KERNEL 261 struct callout spa_deadman_cycid; /* callout id */ 262#endif 263#endif /* illumos */ 264 uint64_t spa_deadman_calls; /* number of deadman calls */ 265 hrtime_t spa_sync_starttime; /* starting time fo spa_sync */ 266 uint64_t spa_deadman_synctime; /* deadman expiration timer */ 267#ifdef illumos 268 /* 269 * spa_iokstat_lock protects spa_iokstat and 270 * spa_queue_stats[]. 271 */ 272 kmutex_t spa_iokstat_lock; 273 struct kstat *spa_iokstat; /* kstat of io to this pool */ 274 struct { 275 int spa_active; 276 int spa_queued; 277 } spa_queue_stats[ZIO_PRIORITY_NUM_QUEUEABLE]; 278#endif 279 hrtime_t spa_ccw_fail_time; /* Conf cache write fail time */ 280 281 /* 282 * spa_refcount & spa_config_lock must be the last elements 283 * because refcount_t changes size based on compilation options. 284 * In order for the MDB module to function correctly, the other 285 * fields must remain in the same location. 286 */ 287 spa_config_lock_t spa_config_lock[SCL_LOCKS]; /* config changes */ 288 refcount_t spa_refcount; /* number of opens */ 289#ifndef illumos 290 boolean_t spa_splitting_newspa; /* creating new spa in split */ 291#endif 292}; 293 294extern const char *spa_config_path; 295 296extern void spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 297 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent); 298 299#ifdef __cplusplus 300} 301#endif 302 303#endif /* _SYS_SPA_IMPL_H */ 304