]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
e10b0808 | 23 | * Copyright (c) 2011, 2015 by Delphix. All rights reserved. |
3541dc6d | 24 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. |
e10b0808 | 25 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. |
cae5b340 | 26 | * Copyright 2013 Saso Kiselkov. All rights reserved. |
4e820b5a | 27 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. |
cae5b340 | 28 | * Copyright (c) 2017 Datto Inc. |
34dc7c2f BB |
29 | */ |
30 | ||
31 | #ifndef _SYS_SPA_IMPL_H | |
32 | #define _SYS_SPA_IMPL_H | |
33 | ||
34dc7c2f BB |
34 | #include <sys/spa.h> |
35 | #include <sys/vdev.h> | |
36 | #include <sys/metaslab.h> | |
37 | #include <sys/dmu.h> | |
38 | #include <sys/dsl_pool.h> | |
39 | #include <sys/uberblock_impl.h> | |
40 | #include <sys/zfs_context.h> | |
41 | #include <sys/avl.h> | |
42 | #include <sys/refcount.h> | |
43 | #include <sys/bplist.h> | |
428870ff | 44 | #include <sys/bpobj.h> |
ea04106b AX |
45 | #include <sys/zfeature.h> |
46 | #include <zfeature_common.h> | |
34dc7c2f BB |
47 | |
48 | #ifdef __cplusplus | |
49 | extern "C" { | |
50 | #endif | |
51 | ||
52 | typedef struct spa_error_entry { | |
ea04106b AX |
53 | zbookmark_phys_t se_bookmark; |
54 | char *se_name; | |
55 | avl_node_t se_avl; | |
34dc7c2f BB |
56 | } spa_error_entry_t; |
57 | ||
58 | typedef struct spa_history_phys { | |
59 | uint64_t sh_pool_create_len; /* ending offset of zpool create */ | |
60 | uint64_t sh_phys_max_off; /* physical EOF */ | |
61 | uint64_t sh_bof; /* logical BOF */ | |
62 | uint64_t sh_eof; /* logical EOF */ | |
63 | uint64_t sh_records_lost; /* num of records overwritten */ | |
64 | } spa_history_phys_t; | |
65 | ||
66 | struct spa_aux_vdev { | |
67 | uint64_t sav_object; /* MOS object for device list */ | |
68 | nvlist_t *sav_config; /* cached device config */ | |
69 | vdev_t **sav_vdevs; /* devices */ | |
70 | int sav_count; /* number devices */ | |
71 | boolean_t sav_sync; /* sync the device list */ | |
72 | nvlist_t **sav_pending; /* pending device additions */ | |
73 | uint_t sav_npending; /* # pending devices */ | |
74 | }; | |
75 | ||
76 | typedef struct spa_config_lock { | |
77 | kmutex_t scl_lock; | |
78 | kthread_t *scl_writer; | |
b128c09f | 79 | int scl_write_wanted; |
34dc7c2f BB |
80 | kcondvar_t scl_cv; |
81 | refcount_t scl_count; | |
82 | } spa_config_lock_t; | |
83 | ||
b128c09f BB |
84 | typedef struct spa_config_dirent { |
85 | list_node_t scd_link; | |
86 | char *scd_path; | |
87 | } spa_config_dirent_t; | |
88 | ||
c06d4368 | 89 | typedef enum zio_taskq_type { |
b128c09f | 90 | ZIO_TASKQ_ISSUE = 0, |
428870ff | 91 | ZIO_TASKQ_ISSUE_HIGH, |
b128c09f | 92 | ZIO_TASKQ_INTERRUPT, |
428870ff | 93 | ZIO_TASKQ_INTERRUPT_HIGH, |
b128c09f | 94 | ZIO_TASKQ_TYPES |
c06d4368 | 95 | } zio_taskq_type_t; |
b128c09f | 96 | |
428870ff | 97 | /* |
c06d4368 | 98 | * State machine for the zpool-poolname process. The states transitions |
428870ff BB |
99 | * are done as follows: |
100 | * | |
101 | * From To Routine | |
102 | * PROC_NONE -> PROC_CREATED spa_activate() | |
103 | * PROC_CREATED -> PROC_ACTIVE spa_thread() | |
104 | * PROC_ACTIVE -> PROC_DEACTIVATE spa_deactivate() | |
105 | * PROC_DEACTIVATE -> PROC_GONE spa_thread() | |
106 | * PROC_GONE -> PROC_NONE spa_deactivate() | |
107 | */ | |
108 | typedef enum spa_proc_state { | |
109 | SPA_PROC_NONE, /* spa_proc = &p0, no process created */ | |
110 | SPA_PROC_CREATED, /* spa_activate() has proc, is waiting */ | |
111 | SPA_PROC_ACTIVE, /* taskqs created, spa_proc set */ | |
112 | SPA_PROC_DEACTIVATE, /* spa_deactivate() requests process exit */ | |
113 | SPA_PROC_GONE /* spa_thread() is exiting, spa_proc = &p0 */ | |
114 | } spa_proc_state_t; | |
115 | ||
c06d4368 AX |
116 | typedef struct spa_taskqs { |
117 | uint_t stqs_count; | |
118 | taskq_t **stqs_taskq; | |
119 | } spa_taskqs_t; | |
120 | ||
cae5b340 AX |
121 | typedef enum spa_all_vdev_zap_action { |
122 | AVZ_ACTION_NONE = 0, | |
123 | AVZ_ACTION_DESTROY, /* Destroy all per-vdev ZAPs and the AVZ. */ | |
124 | AVZ_ACTION_REBUILD, /* Populate the new AVZ, see spa_avz_rebuild */ | |
125 | AVZ_ACTION_INITIALIZE | |
126 | } spa_avz_action_t; | |
127 | ||
34dc7c2f BB |
128 | struct spa { |
129 | /* | |
130 | * Fields protected by spa_namespace_lock. | |
131 | */ | |
cae5b340 | 132 | char spa_name[ZFS_MAX_DATASET_NAME_LEN]; /* pool name */ |
d96eb2b1 | 133 | char *spa_comment; /* comment */ |
34dc7c2f BB |
134 | avl_node_t spa_avl; /* node in spa_namespace_avl */ |
135 | nvlist_t *spa_config; /* last synced config */ | |
136 | nvlist_t *spa_config_syncing; /* currently syncing config */ | |
428870ff | 137 | nvlist_t *spa_config_splitting; /* config for splitting */ |
572e2857 | 138 | nvlist_t *spa_load_info; /* info and errors from load */ |
34dc7c2f | 139 | uint64_t spa_config_txg; /* txg of last config change */ |
34dc7c2f | 140 | int spa_sync_pass; /* iterate-to-convergence */ |
b128c09f | 141 | pool_state_t spa_state; /* pool state */ |
34dc7c2f | 142 | int spa_inject_ref; /* injection references */ |
34dc7c2f BB |
143 | uint8_t spa_sync_on; /* sync threads are running */ |
144 | spa_load_state_t spa_load_state; /* current load operation */ | |
572e2857 | 145 | uint64_t spa_import_flags; /* import specific flags */ |
c06d4368 | 146 | spa_taskqs_t spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES]; |
34dc7c2f | 147 | dsl_pool_t *spa_dsl_pool; |
9ae529ec | 148 | boolean_t spa_is_initializing; /* true while opening pool */ |
34dc7c2f BB |
149 | metaslab_class_t *spa_normal_class; /* normal data class */ |
150 | metaslab_class_t *spa_log_class; /* intent log data class */ | |
151 | uint64_t spa_first_txg; /* first txg after spa_open() */ | |
152 | uint64_t spa_final_txg; /* txg of export/destroy */ | |
153 | uint64_t spa_freeze_txg; /* freeze pool at this txg */ | |
428870ff BB |
154 | uint64_t spa_load_max_txg; /* best initial ub_txg */ |
155 | uint64_t spa_claim_max_txg; /* highest claimed birth txg */ | |
a07c8b41 | 156 | inode_timespec_t spa_loaded_ts; /* 1st successful open time */ |
34dc7c2f | 157 | objset_t *spa_meta_objset; /* copy of dp->dp_meta_objset */ |
e10b0808 AX |
158 | kmutex_t spa_evicting_os_lock; /* Evicting objset list lock */ |
159 | list_t spa_evicting_os_list; /* Objsets being evicted. */ | |
160 | kcondvar_t spa_evicting_os_cv; /* Objset Eviction Completion */ | |
34dc7c2f BB |
161 | txg_list_t spa_vdev_txg_list; /* per-txg dirty vdev list */ |
162 | vdev_t *spa_root_vdev; /* top-level vdev container */ | |
e10b0808 AX |
163 | int spa_min_ashift; /* of vdevs in normal class */ |
164 | int spa_max_ashift; /* of vdevs in normal class */ | |
3541dc6d GA |
165 | uint64_t spa_config_guid; /* config pool guid */ |
166 | uint64_t spa_load_guid; /* spa_load initialized guid */ | |
3bc7e0fb | 167 | uint64_t spa_last_synced_guid; /* last synced guid */ |
b128c09f BB |
168 | list_t spa_config_dirty_list; /* vdevs with dirty config */ |
169 | list_t spa_state_dirty_list; /* vdevs with dirty state */ | |
cae5b340 AX |
170 | kmutex_t spa_alloc_lock; |
171 | avl_tree_t spa_alloc_tree; | |
34dc7c2f BB |
172 | spa_aux_vdev_t spa_spares; /* hot spares */ |
173 | spa_aux_vdev_t spa_l2cache; /* L2ARC cache devices */ | |
9ae529ec | 174 | nvlist_t *spa_label_features; /* Features for reading MOS */ |
34dc7c2f | 175 | uint64_t spa_config_object; /* MOS object for pool config */ |
428870ff | 176 | uint64_t spa_config_generation; /* config generation number */ |
34dc7c2f | 177 | uint64_t spa_syncing_txg; /* txg currently syncing */ |
428870ff BB |
178 | bpobj_t spa_deferred_bpobj; /* deferred-free bplist */ |
179 | bplist_t spa_free_bplist[TXG_SIZE]; /* bplist of stuff to free */ | |
cae5b340 AX |
180 | zio_cksum_salt_t spa_cksum_salt; /* secret salt for cksum */ |
181 | /* checksum context templates */ | |
182 | kmutex_t spa_cksum_tmpls_lock; | |
183 | void *spa_cksum_tmpls[ZIO_CHECKSUM_FUNCTIONS]; | |
34dc7c2f BB |
184 | uberblock_t spa_ubsync; /* last synced uberblock */ |
185 | uberblock_t spa_uberblock; /* current uberblock */ | |
428870ff | 186 | boolean_t spa_extreme_rewind; /* rewind past deferred frees */ |
572e2857 | 187 | uint64_t spa_last_io; /* lbolt of last non-scan I/O */ |
34dc7c2f | 188 | kmutex_t spa_scrub_lock; /* resilver/scrub lock */ |
34dc7c2f | 189 | uint64_t spa_scrub_inflight; /* in-flight scrub I/Os */ |
34dc7c2f | 190 | kcondvar_t spa_scrub_io_cv; /* scrub I/O completion */ |
34dc7c2f BB |
191 | uint8_t spa_scrub_active; /* active or suspended? */ |
192 | uint8_t spa_scrub_type; /* type of scrub we're doing */ | |
193 | uint8_t spa_scrub_finished; /* indicator to rotate logs */ | |
b128c09f BB |
194 | uint8_t spa_scrub_started; /* started since last boot */ |
195 | uint8_t spa_scrub_reopen; /* scrub doing vdev_reopen */ | |
428870ff | 196 | uint64_t spa_scan_pass_start; /* start time per pass/reboot */ |
cae5b340 AX |
197 | uint64_t spa_scan_pass_scrub_pause; /* scrub pause time */ |
198 | uint64_t spa_scan_pass_scrub_spent_paused; /* total paused */ | |
428870ff | 199 | uint64_t spa_scan_pass_exam; /* examined bytes per pass */ |
34dc7c2f BB |
200 | kmutex_t spa_async_lock; /* protect async state */ |
201 | kthread_t *spa_async_thread; /* thread doing async task */ | |
202 | int spa_async_suspended; /* async tasks suspended */ | |
203 | kcondvar_t spa_async_cv; /* wait for thread_exit() */ | |
204 | uint16_t spa_async_tasks; /* async task mask */ | |
205 | char *spa_root; /* alternate root directory */ | |
34dc7c2f | 206 | uint64_t spa_ena; /* spa-wide ereport ENA */ |
428870ff BB |
207 | int spa_last_open_failed; /* error if last open failed */ |
208 | uint64_t spa_last_ubsync_txg; /* "best" uberblock txg */ | |
209 | uint64_t spa_last_ubsync_txg_ts; /* timestamp from that ub */ | |
210 | uint64_t spa_load_txg; /* ub txg that loaded */ | |
211 | uint64_t spa_load_txg_ts; /* timestamp from that ub */ | |
212 | uint64_t spa_load_meta_errors; /* verify metadata err count */ | |
213 | uint64_t spa_load_data_errors; /* verify data err count */ | |
214 | uint64_t spa_verify_min_txg; /* start txg of verify scrub */ | |
34dc7c2f BB |
215 | kmutex_t spa_errlog_lock; /* error log lock */ |
216 | uint64_t spa_errlog_last; /* last error log object */ | |
217 | uint64_t spa_errlog_scrub; /* scrub error log object */ | |
218 | kmutex_t spa_errlist_lock; /* error list/ereport lock */ | |
219 | avl_tree_t spa_errlist_last; /* last error list */ | |
220 | avl_tree_t spa_errlist_scrub; /* scrub error list */ | |
221 | uint64_t spa_deflate; /* should we deflate? */ | |
222 | uint64_t spa_history; /* history object */ | |
223 | kmutex_t spa_history_lock; /* history lock */ | |
224 | vdev_t *spa_pending_vdev; /* pending vdev additions */ | |
225 | kmutex_t spa_props_lock; /* property lock */ | |
226 | uint64_t spa_pool_props_object; /* object for properties */ | |
227 | uint64_t spa_bootfs; /* default boot filesystem */ | |
b128c09f BB |
228 | uint64_t spa_failmode; /* failure mode for the pool */ |
229 | uint64_t spa_delegation; /* delegation on/off */ | |
230 | list_t spa_config_list; /* previous cache file(s) */ | |
ea04106b AX |
231 | /* per-CPU array of root of async I/O: */ |
232 | zio_t **spa_async_zio_root; | |
b128c09f BB |
233 | zio_t *spa_suspend_zio_root; /* root of all suspended I/O */ |
234 | kmutex_t spa_suspend_lock; /* protects suspend_zio_root */ | |
235 | kcondvar_t spa_suspend_cv; /* notification of resume */ | |
42f7b73b | 236 | zio_suspend_reason_t spa_suspended; /* pool is suspended */ |
428870ff | 237 | uint8_t spa_claiming; /* pool is doing zil_claim() */ |
6d974228 | 238 | boolean_t spa_debug; /* debug enabled? */ |
b128c09f BB |
239 | boolean_t spa_is_root; /* pool is root */ |
240 | int spa_minref; /* num refs when first opened */ | |
fb5f0bc8 | 241 | int spa_mode; /* FREAD | FWRITE */ |
b128c09f | 242 | spa_log_state_t spa_log_state; /* log state */ |
9babb374 | 243 | uint64_t spa_autoexpand; /* lun expansion on/off */ |
428870ff BB |
244 | ddt_t *spa_ddt[ZIO_CHECKSUM_FUNCTIONS]; /* in-core DDTs */ |
245 | uint64_t spa_ddt_stat_object; /* DDT statistics */ | |
cae5b340 | 246 | uint64_t spa_dedup_dspace; /* Cache get_dedup_dspace() */ |
428870ff BB |
247 | uint64_t spa_dedup_ditto; /* dedup ditto threshold */ |
248 | uint64_t spa_dedup_checksum; /* default dedup checksum */ | |
249 | uint64_t spa_dspace; /* dspace in normal class */ | |
250 | kmutex_t spa_vdev_top_lock; /* dueling offline/remove */ | |
251 | kmutex_t spa_proc_lock; /* protects spa_proc* */ | |
252 | kcondvar_t spa_proc_cv; /* spa_proc_state transitions */ | |
253 | spa_proc_state_t spa_proc_state; /* see definition */ | |
60101509 | 254 | proc_t *spa_proc; /* "zpool-poolname" process */ |
428870ff BB |
255 | uint64_t spa_did; /* if procp != p0, did of t1 */ |
256 | boolean_t spa_autoreplace; /* autoreplace set in open */ | |
257 | int spa_vdev_locks; /* locks grabbed */ | |
258 | uint64_t spa_creation_version; /* version at pool creation */ | |
9ae529ec CS |
259 | uint64_t spa_prev_software_version; /* See ub_software_version */ |
260 | uint64_t spa_feat_for_write_obj; /* required to write to pool */ | |
261 | uint64_t spa_feat_for_read_obj; /* required to read from pool */ | |
262 | uint64_t spa_feat_desc_obj; /* Feature descriptions */ | |
ea04106b AX |
263 | uint64_t spa_feat_enabled_txg_obj; /* Feature enabled txg */ |
264 | kmutex_t spa_feat_stats_lock; /* protects spa_feat_stats */ | |
265 | nvlist_t *spa_feat_stats; /* Cache of enabled features */ | |
266 | /* cache feature refcounts */ | |
267 | uint64_t spa_feat_refcount_cache[SPA_FEATURES]; | |
c06d4368 AX |
268 | taskqid_t spa_deadman_tqid; /* Task id */ |
269 | uint64_t spa_deadman_calls; /* number of deadman calls */ | |
a08ee875 | 270 | hrtime_t spa_sync_starttime; /* starting time of spa_sync */ |
c06d4368 | 271 | uint64_t spa_deadman_synctime; /* deadman expiration timer */ |
cae5b340 AX |
272 | uint64_t spa_all_vdev_zaps; /* ZAP of per-vd ZAP obj #s */ |
273 | spa_avz_action_t spa_avz_action; /* destroy/rebuild AVZ? */ | |
ea04106b | 274 | uint64_t spa_errata; /* errata issues detected */ |
a08ee875 | 275 | spa_stats_t spa_stats; /* assorted spa statistics */ |
cae5b340 AX |
276 | hrtime_t spa_ccw_fail_time; /* Conf cache write fail time */ |
277 | taskq_t *spa_zvol_taskq; /* Taskq for minor management */ | |
42f7b73b | 278 | taskq_t *spa_prefetch_taskq; /* Taskq for prefetch threads */ |
cae5b340 AX |
279 | uint64_t spa_multihost; /* multihost aware (mmp) */ |
280 | mmp_thread_t spa_mmp; /* multihost mmp thread */ | |
a08ee875 | 281 | |
34dc7c2f | 282 | /* |
a08ee875 | 283 | * spa_refcount & spa_config_lock must be the last elements |
34dc7c2f BB |
284 | * because refcount_t changes size based on compilation options. |
285 | * In order for the MDB module to function correctly, the other | |
286 | * fields must remain in the same location. | |
287 | */ | |
b128c09f | 288 | spa_config_lock_t spa_config_lock[SCL_LOCKS]; /* config changes */ |
34dc7c2f | 289 | refcount_t spa_refcount; /* number of opens */ |
cae5b340 AX |
290 | |
291 | taskq_t *spa_upgrade_taskq; /* taskq for upgrade jobs */ | |
34dc7c2f BB |
292 | }; |
293 | ||
c28b2279 | 294 | extern char *spa_config_path; |
b128c09f | 295 | |
c06d4368 AX |
296 | extern 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 | extern void spa_taskq_dispatch_sync(spa_t *, zio_type_t t, zio_taskq_type_t q, | |
299 | task_func_t *func, void *arg, uint_t flags); | |
300 | ||
301 | ||
34dc7c2f BB |
302 | #ifdef __cplusplus |
303 | } | |
304 | #endif | |
305 | ||
306 | #endif /* _SYS_SPA_IMPL_H */ |